Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Consulting Editor
THOMAS A. LIPO
FRED C. SCHWEPPE
MICHAEL C. CARAMANIS
Boston University
RICHARD D. TABORS
ROGER E. BOHN
e-ISBN-13: 978-1-4613-1683-1
DOl: 10.1007/978-1-4613-1683-1
I. Electric utilities-Rates. 2. Commodity exchanges.
I. Schweppe, Fred c., 1933II. Series.
HG6047.E43S66 1987
338.4'336362 - dcl9
87-36643
CIP
DEDICATION
To Mary Alice Sanderson, long live the sponge;
and to our children, Carl, Christina, Constantine, Daniel, David, Edmund, Fritz (Charles),
Kristen, and Malaika.
Shortly before completion of this book Fred C.
Schweppe, our friend, colleague and senior
author died suddenly. Fred created spot pricing
and proved, again, that "The forecast is always
wrong!"
T ABLE OF CONTENTS
Preface
,,-,
J.I
1.2
1.3
1.4
1.5
1.6
XVII
Overview
Goal of Book
Three Steps to an Energy Marketplace
How Will Customers Respond?
Encl"li,+, Marketplace Operation: A Developed Country
Energy Marketplace Operation: A Developing Country
Discussion of Chapter I
Supplemcnt to Chapter I: Summary of Issues
Historical Notes and References: Chapter I
Notes
3
9
II
15
19
19
20
26
27
29
29
31
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
32
vii
34
35
38
41
42
44
45
47
viii Contents
49
51
51
53
55
56
57
58
63
66
68
Implementation
81
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
82
87
95
97
102
106
107
108
III
5.1
5.2
5.3
5.4
5.5
112
Generation Only
Generation Fuel and Variable Maintenace: ).(t)
Generation Quality of Supply, YQs(t): Cost Function Approach
Generation Quality of Supply, YQs(t): Market Clearing Approach
Generation Self-Dispatch
Multiple Time Periods
Discussion of Chapter 6
Historical Notes and References: Chapter 6
Notes
69
71
75
76
76
77
77
80
90
109
109
115
117
123
125
126
127
129
131
132
137
143
146
147
148
149
149
ix
151
152
154
159
160
161
162
168
170
172
175
175
Revenue Reconciliation
177
8.1
8.2
8.3
8.4
8.5
8.6
8.7
8.8
8.9
178
184
188
189
191
195
195
199
200
201
202
205
9.'1
9.2
9.3
9.4
9.5
9.6
206
213
222
227
231
235
236
236
]0
237
10.1
10.2
10.3
10.4
10.5
10.6
10.7
238
240
245
247
248
248
252
253
254
"
REFERENCES
175
176
255
x Contents
261
APPENDICES
267
269
A.I
A.2
A.3
AA
Network Flows
Local Controllers
Mathematical Models for System Dynamics
Power System Dynamics
Further Reading
Notes
269
275
277
277
279
279
281
281
283
288
291
292
294
294
295
CI
C2
C3
CA
C5
C.6
C.7
295
299
301
305
306
308
309
310
311
DC Load "low
313
B
B.I
B.2
B.3
313
317
320
325
327
327
331
333
FAPER
335
F.I
F.2
F.3
FA
F.5
F.6
Basic Concepts
FAPER Designs
Variations on the Basic Concept
Analysis of Multiple FAPER Response
Incentives to Install FAPERS
Discussion
335
336
338
339
339
339
341
xi
G.l
G.2
G.3
G.4
Introduction
24-Hour Trajectories
Price Duration Curves
Impact of Customer Response on Variable Energy Costs
341
342
345
345
347
INDEX
349
ACKNOWLEDGMENTS
Xhe .title page of this book lists four names, but as the bibliography shows,
spOt pricing has had many more authors. The folIowing is an attempt to
acl<nowledge partially the contributions of others, in roughly chronological
order. N~tu\l"alIy, none of those mentioned here is responsible for what we have
done with their ideas.
The impodance of changing the coupling that exists between a utility and its
industrial customers emerged during a research project (1977-78) done for New
England Electric System in close cooperation with some of its industrial
customers. Ed Gulachenski of New England Electric was the prime mover
behind the research effort.
Many of the initial formulations and papers associated with spot pricing
arose during a series of informal "Homeostatic Control" workshops conducted
at MIT during 1977-78. James Kirtley of MIT's Electric Power Systems Engineering Laboratory, EPSEL (now part of the Laboratory Electromagnetic and
Electronic Systems, LEES), was one of the key early contributors with his, Tom
Sterling's and Ron William's local communication network methodology and
microshedding concepts (which are called price-quantity transactions in this
book). Hugh Outhred of the University of New South Wales was another key
initial idea generator, and he has continued to evolve the ideas in a faraway land
south of the equator. Fred Pickel and Alan Cox helped co-author the initial
Homeostatic Control paper. Other members of the original Homeostatic
Control working group included Lenny Gould and Steve Umans.
xiii
xiv Acknowledgments
The overall effort was given a key boost when Henry Jacoby and Loren Cox,
Directors of MIT's Center for Energy Policy Research, organized the Boxborough Conference on Homeostatic Control in 1979. If we had listened more
carefully to Loren's advice over the years, spot pricing might have progressed
much faster than it has. David White, Director of MIT's Energy Laboratory,
was an early and continuously sustaining supporter of the ideas.
Early research efforts were supported under the sponsorship of Charlie Smith
and Ray Dunlop at DOE, which provided a quick, if short, start. The PSC of
Wisconsin offered the next installment. Rod Stevenson at the University of
Wisconsin helped formulate and complete this early study on the impact of spot
pricing for industrial applications.
In California, the moral support of Rusty Schweickart, then Chairman of the
California Energy Commission, and John Bryson, then Chairman of the California Department of Public Utilities, led to two key projects, one funded jointly
by Pacific Gas and Electric and Southern California Edison with AI Garcia and
Jack Runnels as two extremely helpful project monitors. A foll\ow-up, was
funded by the California Energy Commission guided and assisted by John
Wilson.
The California studies started a continuing and extremely productive collaboration with John Flory of Utility Customer Interfaces. The world needs more
individuals of his caliber.
At one point in the spot pricing evolution (as developed for regulated utilities), we were pulled into the great debate on deregulation. This led to interesting
and stimulating discussions with a variety of people such as Leonard Hyman of
Merrill Lynch Pierce Fenner and Smith, Joe Pace ofNERA, and William Berry,
president of Virginia Electric. Interactions with Richard Schmalensee and Paul
Joskow of MIT helped formulate our approach to deregulation.
When we first met Robert Peddie, then Chairman of the Southeastern Distribution Board in England, he was already well along with his exciting, visionary
approach to residential customer-utility interactions, called the CALMU
system. Subsequent discussions and interactions with him and his CALMU
system had a significant impact on our own efforts.
,
Bernie Hastings of Detroit Edison did not sit down to write theoretical spot
pricing papers or to do simulation studies. Instead he simply went ahead and
saw to the actual implementation of spot prices for some of Detroit Edison's
industrial customers. This was the first implementation of the basic ideas of spot
pricing.
John Phillips, Chairman of the California Energy Coalition, is a very special
individual whose support, ideas, and enthusiasm, combined with his strong
advocacy of utility-customer partnering, aided spot pricing in ways that can
never be quantified.
The Integrated Communications Systems Inc. (ICS) demonstration of their
Transtext system for Georgia Power residential customers provided us with an
opportunity to evaluate the variable energy costs (i.e. spot prices) for many of
xv
xviii
Preface
written to serve as a single, integrated sourcebook on the theory and implementation of a spot price based energy marketplace.
The book is written for electric power engineers interested in operation,
planning, and load management; for economists interested in electric power
regulation and pricing; and for utility regulators and overall policy makers. It
is a "how to" book, written so it can be used by those who are mainly interested
in the application of the concepts and techniques. Detailed mathematical derivations are also provided for those who are interested.
A spot price based energy marketplace has many benefits for both the electric
utility and its customers. These benefits include improvements in operating
efficiency, reductions in needed capital investments, and customer options on
the type (reliability) of electricity to be bought. A spot priced based energy
marketplace is a win-win situation for both the regulated utility and its
customers. The customer's lifestyles improve because the customers are receiving more service from the use of electric energy per dollar spent. The utility has
a more controllable, less uncertain world in which to operate.
A spot price based energy marketplace can be implemented using today's
proven technologies. However, its existence stimulates the development of new
microelectronic technologies and hence enables further exploitation of the
microelectronic revolution in communication and computation.
The spot price based energy marketplace concepts were originally developed
to meet the present and future needs of the complex, interconnected, sophisticated power systems of developed countries. However, the basic ideas are also
applicable to the smaller, rapidly growing, less sophisticated power systems
often found in other parts of the world.
The spot price based energy marketplace was developed to be applicable to
present-day structures wherein a privately owned utility is regulated by some
government agency, or the utility is government owned and operated. However,
the energy marketplace introduces the possibility of various degrees of deregulation wherein some generation is provided by privately owned, less regulated
companies.
Spot pricing is the natural evolution of existing techniques for power system
operation, planning, load management and the economic theory of marginal
cost pricing. This book relates spot pricing to existing rate structures, direct load
control techniques, interruptible contracts, interutility sales, and power system
operation.
Organization of Book
Chapter I provides an overview of spot pricing and the issues to be covered. The
rest of the main text is subdivided into two parts.
Part I (Chapters 2 through 5) consists of three main chapters that show how
hourly spot prices behave, discuss various types of energy marketplace transactions, and address implementation of a spot price based energy marketplace
from both the utility and customer points of view. The final chapter of Part I
xix
briefly explores the possible evolution of the energy marketplace (as developed
for a regulated utility) into a system of deregulated generation.
Part II is a sequence of chapters (6 through 10) that develop the theory of spot
prices starting with simple cases and progressing toward more complex and
realistic situations.
Appendices provide background material and supplemental detail.
The book is written so that it can be read in different sequences. All readers
should begin with Chapter I. The text is organized to discuss behavior and
implementation (Part I) before going through the technical theory (Part II). This
is the appropriate sequence for readers interested primarily in broad applicability and implementation issues, since they do not have to wade through a lot of
equations. However, some readers may find it preferable to read Part n before
Part I to get the detailed theory first.
Readers with a limited background in power systems operation and planning
should begin by glancing at Appendices A, B, and C. These appendices provide
a brief overview of the main power system concepts affecting an energy marketplace.
J. OVERVIEW
This chapter provides an overview of the book and the spot price based energy
marketplace.
Section 1.1 discuspes the goals of the book. The three steps to an energy
marketplace are introduced in Section 1.2. (These three steps are expanded in
Chapters 2, 3 and 4 respectively.) Section 1.3 addresses customer response to an
energy marijetplace. Sections 1.4 and 1.5 discuss energy marketplace operation
for developed and developing countries. A supplement to Chapter 1 provides a
summary of issues relevant to a spot price based energy marketplace.
SECTION 1.1. GOAL OF BOOK
The electric utility industry today is undergoing rapid and irreversible changes.
Volatile fuel costs, less predictable load growth, a more complex regulatory
environment and a deceleration in conventional technical progress are important examples of these changes. Yet the need for growth in productivity and
efficiency, and for increased flexibility to handle future uncertainties, is stronger
and more challenging than ever. The utility industry, which has matured into a
toO-billion-dollar industry in the United States and constitutes a substantial
economic sector in all industrialized countries, must evolve to meet this challenge.
New directions for the utility industry are being sought by many interested
parties in the government, the private sector, and the universities. One such
direction has been widespread interest in utility-customer cooperation through
Freedom of Choice: Provide customers with options on the cost and reliability of supply and how they choose to use electric energy.
Economic Efficiency: Motivate customers to adjust their own electric energy
usage patterns to match utility marginal costs.
Equity:l Reduce customer cross-subsidies-i.e. a customer's charges are based
on the utility's costs to serve that customer.
Utility Control, Operation and Planning: Consider the engineering requirements for controlling, operating and planning an electric power system.
Present-Day Transactions
Most transactions between utilities and their customers today fall far short of
meeting these four criteria. Flat and time-of-use rates are related to actual
marginal costs in only a gross, average sense. The demand charge is an anachronism with at most a tenuous relationship to marginal costs; it usually encourages counterproductive customer behavior. Load management approaches
often involve direct utility control (i.e., the big brother approach) and
sometimes encourage wasteful customer behavior. Cross-subsidization between
customer classes and within a given class is rampant. Utility-customer transactions are typically specified without considering the engineering problems of
controlling, operating, and planning an electri\ power system.
The failure of most present-day utility-customer transactions to meet today's
needs can be traced to their historical foundations. They were established by
individuals unconcerned with power system control and operation, during times
when communication and computation were very expensive, when there was less
incentive to use electricity in an efficient fashion, and when cross-subsidies were
of limited concern to society.
The four basic criteria cannot be achieved by putting "band-aids" on the
present-day approaches. They cannot be achieved by adding modern microelectronics whose functions are based on present-day approaches.
The four basic criteria can be achieved only by returning to the first principles
of economics and engineering and by viewing the utility and its customers as a
single integrated system. The result of this integration is the spot price based
energy marketplace, which is the subject of this book.
Energy Marketplace Transactions
Students taking an introductory course in economics are taught that the best
way to achieve
I. Overview
o
o
o
Economic efficiency
Equity
Customer freedom of choice
is the use of marketplace where market clearing prices are determined by supply
and demand. For example, when a lot of fresh produce (e.g., lettuce) is available,
the prices go down and consumption goes up. During other times of the year,
or after a hard freeze, the supply goes down and prices go up to reduce
consumption.
Five ingredients for a successful marketplace are
I.
2.
3.
4.
5.
A supply side with varying supply costs that increase with demand
A demand side with varying demands which can adapt to price changes
A market mechanism for buying and selling
No monopsonistic behavior on the demand side
No monopolistic behavior on the supply side
Week 1
20
40
eo
eo
100
1~
100
120
l~
1~
Hou~
Week 2
:~
20
40
eo
80
\40
leo
Energy Marketplace The buying and selling of electric energy between independent
customers and a regulated or government owned utility.
The one exception is Chapter 5, where we discuss the possibilities of deregulation of generation.
The energy marketplace is designed explicitly to include engineering issues
associated with power system control and operation. Hence all four of the
original criteria (efficiency, freedom of choice, equity, and control/operation)
are met.
In the energy marketplace, all utility-customer transactions are based in a
self-consistent fashion on a single quantity, the hourly spot price. The hourly
spot price is determined by the demand at that hour and the hourly varying costs
and capabilities of the generation, transmission and distribution systems. The
1. Overview 7
MUs/kWh .-----y-EA-R-'-PE-'K-O-'y-S-S-UM-M-AR-Y----.
500
MIlS/kWh.-----y-E-=AR-'-Ty-,.P-,.'CA-:L-:cOA-y-,S-UM-M-=AR-:y---,
300
'00
200
Summer
300
Wmter
200
100
'00
10
12
16
19
22
24
LY2; \1\:\
'2
,s
22
24
Hours
Hours
Mils/kWh .-----y-E'-R-2P-=-:AK-=D-:Ay....,S-:UM-M-:A=-Ry,-----,
MUs/kWh .-----=yE-:AR:-2-T-=YP-::'CA-L-::"DA-:y-,S-UM-:M-:AR:-Y----,
300
300
200
200
'00
100
'2
Hours
"
2'
HOUTS
The initial reaction of many people to spot pricing is that the major difference
between spot prices and present-day transactions is that spot prices have
complex time variations. Actually, present-day prices also exhibit complex time
variations, so the major difference is in the nature of the price variations, not
their presence. As one example, consider Figure 1.1.3, which plots the historical
variation of prices (jkWh) for one utility. The hourly cost (spot price) variations of Figures 1.1.1 and 1.1.2 simply exhibit finer time detail. As a second
example, consider an industrial customer with a 5/kWh energy charge and a
II
/kwh
"'8.0
r-
....
7.0
I..r-
,""r-
I-
r-
6.0
....
:,""I""
5.0
...
I4.0
3.0
hrrfT
1979
1980
1981
Figure 1.1.3. Example of monthly variations in residential prices, under conventional rates.
5 $/kW demand charge based on the energy used during the customer's peak
hour during the month. The corresponding energy rate paid by the industrial
customer is plotted in Figure 1.1.4. It displays a dramatic time variation which
bears little resemblance to either Figure 1.1.1 or 1.1.2.
A second difference lies in the nature of the, relationship between the utility
and its customers. In a spot price based energy marketplace, the utility and its
customers are partners working together to achieve the maximum benefit from
electric energy usage at minimum cost. The amount of such partnering found in
present-day utility-customer relationships is small at best.
Present-day flat and time-of-use rates are specified by formulas based on
expected costs, rates of return on equity, etc. An hourly spot price is based on
the same principles, but the formula is solved every hour by computers instead
of once a year or once per month (for fuel adjustment clauses). Present-day
direct load control and present-day interruptable contracts are special cases of
energy marketplace transactions. Present-day power system operation often
involves a real-time spot market for purchases and sales between utilities. The
spot price based energy marketplace simply extends this spot market to include
the customers. Thus it can be concluded that
The spot price based energy marketplace is the logical evolution of present-day rates and
load management techniques, married with present-day practices of power system operation, the concept of utility--customer partnering, and the availability of inexpensive
communications and computation equipment.
I. Overview
Rate Paid
by Customer
/kWh
505
5r-----------,~------------------------o
720
Hours
A spot pric~ based energy marketplace that meets the four criteria of Section 1.1
can be achieved in three steps:
Step I: Define hourly spot prices and evaluate their behavior.
Step II: Specify an appropriate set of utility-<:ustomer transactions based on
the hourly spot price and associated transactions costs.
Step III: Implement the energy marketplace considering the needs and capabilities of both the utility and the customers.
Step I: Define Hourly Spot Prices
The fundamental quantity underlying the energy marketplace is the hourly spot
price. 3
The hourly spot price is defined in terms of marginal costs subject to revenue
reconciliation (i.e. recovery of operating and embedded capital costs).
The value of the spot price at any hour depends on the hourly variations of
o
o
o
The hourly spot price is a random process (e.g. see Figures 1.1.1 and 1.1.2). Its
future value cannot be predicted perfectly because of random equipment
outages and demand variations.
An hourly spot price can be determined for a utility which is buying energy
from, as well as seJling energy to its cutomers. The buy-back hourly spot price
can be either greater or less than the selling hourly spot price.
Step II: Specify Utility-Customer Transactions
All utility-<:ustomer transactions are based on the hourly spot price defined in
Step I. Three general types of transactions are
o
o
o
Price-Only
Price-Quantity
Long-Term Contracts
One-Hour Update: Customers see prices ($/kWh) varying each hour, predicted and communicated one hour in advance.
24-Hour Update: Customers see prices ($jkWh) varying every hour, predicted
and communicated 24 hours in advance.
Time-of-Use (TOU) and Flat Rates: 4 Rates ($jkWh) are calculated using
predictions of hourly spot price behavior one billing period in advance.
A need for fast acting, accurate load control (seconds to minutes rather than
hours) to maintain power system security.
A desire to reduce transactions costs below those of rapidly varying price-only
I. Overview
Thus the incremental cost of the customer's usage that hour is 9/kWh. Such
transactions enable a customer to buy an insurance policy by locking in
1000kWh at 1O/kWh, even though the customer still sees the spot price for
actual usage.
A variation on this long-term contract is the option wherein the customer buys
the.'fight to buy up to a fixed amount of energy at a fixed price. The customer
ex~rcises the option ,if the hourly spot price turns out to be greater than the
o(9tiqn price.
'
Step III: fmplement Energy Marketplace
Implementation of the energy marketplace involves the utility and its customers
.
.;
operatmg as' partners.
Utility implementation concerns include real-time calculation/prediction of
hourly spot prices, metering-{;ommunications-billing, and system control
center operation using the new control signal called price. The impact of the
energy marketplace on utility long-term investment decisions is also of concern.
Customers who choose to exploit the energy marketplace potentials must
implement the appropriate response systems, which could range from simple
manual response to sophisticated digital control. The utility can provide the
control mechanism as a service to customers.
SECTION 1.3. HOW WILL CUSTOMERS RESPOND?
A key question for a spot price based energy marketplace is: How will customers
respond (change their usage patterns) to price changes? Two ways to try to
answer this question are
o
o
i2
Unfortunately, not enough direct observations of the right type are available
today. There are a lot of data (and studies) on customer response to time-of-use
rates, which are a special case of a spot price based transaction. However, there
is a large difference between a time-of-use rate's behavior and that of, for
example, Figures 1.1.1 and 1.1.2. This difference combined with the nonlinear
nature of customer response makes an extrapolation of response from time-ofuse results a vague guideline at best. Interruptible rates and certain types of
direct load control are also special cases of spot price based transactions, but
their method of implementation makes it very difficult to draw conclusions for
an energy marketplace type operation. There are various implementations of
one- and 24-hour-type spot prices in operation, but the data coming from them
are still too limited to provide a basis for any general conclusion.
When enough direct observations are not yet available, it is necessary to
resort to the use of models. Models can range in sophistication from statistically
based computer simulations or optimizations to simple mental pictures obtained
by having an industrial plant manager spend a few hours showing one around
the plant and explaining the processes and operating constraints.
We will now provide some discussion on what we know of customer response
in the industrial and residential sectors (discussions on response mechanisms are
given in Section 4.3).
Industrial Response
o
o
The
The
The
The
The
I. Overview
13
one-half-hour update of spot prices complained that during some seasons of the
year the price didn't vary enough to make any money.
When we started looking at industrial response, we naively assumed that
response depended only on the physical character of the plant and the
economics of its operation. However, in the real world other factors also play
a key role. These factors included the nature of union contracts, the management of the company (locally owned or part of a big corporation), and the
character of the work force (social background), the location of the plant (was
it safe to work there at night). We studied two plants with identical SIC codes
located within twenty miles of each other. However, they had completely
different response capabilities.
It became very clear during our interactions that industrial response to a
one-year test or demonstration program would be much less than for a realworld spot price based energy marketplace that won't go away in 12 months.
Many customers could greatly improve their response capabilities by relatively
minor additions of new control, process, etc., equipment. However, such modifications could not be cost justified for a mere demonstration or test.
Industrial response can be greatly increased if the customers know a few
hours to a day in advance when exceptionally high or low prices will or are
expected to occur. Such lead or warning time allows rescheduling of highenergy-consuming processes with much less cost. The use of a 24-hour update
pr:~vides a "hard price" a day in advance. For a one-hour spot price, it is
i~()rtant for the utility to provide a forecast of future prices (the same forecasts
which formed the blise for updating the 24-hour spot price). The importance of
w?!r~ing.or lead time.is another reason why it is difficult to extrapolate from time
of use or interruptible response to energy marketplace response.
Industrial response is usually a very nonlinear function of price. Doubling the
price from $ to 10/kWh may create little response, while doubling it from 10
to 20 /kWh could cause sizable reaction because of the threshold effects. It is
also clear from our studies that price differentials during a day can be equally
if not more important than the absolute value of the price.
Residential Response
14
electronics become less than the benefits (to the customer and/or utility) of the
resulting response. Some might argue that such a time has already arrived.
The importance of warning time and the nonlinearity of response apply to
both residential and industrial customers.
Example of Customer Benefits
Why is it that putting customers onto a spot price makes them better off and
simultaneously makes the utility (or the utility's other customers) better off? Of
course one of the precepts of free market economies is that prices based on a
marketplace send better signals than do fixed price. But that is a theoretical
argument, and while it can be made rigorous (as we do in Chapter 6), it is not
very intuitive initially.
Therefore let's look at a simple example: an industrial customer who has to
pump a lot of water (or other fluids) into storage tanks. For the sake of
familiarity, consider a municipal water department which has to fill its water
tanks daily. Suppose the customer is initially on a two level time-of-day rate with
prices which vary over time but in a predetermined and rigid way. (If the
customer is initially on a flat rate or a rate with a demand charge, then there are
even more benefits from spot pricing in the following example.) Then it will
pump water every night and every weekend, at the same hours every week. If
it cannot fill the tanks completely during the low price period (which in some
time-of-day rates is necessarily less than 12 hours long), it will wait until the
tanks get to the low limit, then resume pumping. This will often mean that it
resumes pumping in the afternoon or early evening, just at the tail end of the
high price period. The customer will keep the same timing of usage every week
that a particular rate is in effect, and every day of the week, without adjusting
for generator outages, changes in electricity generating costs, etc.
Now switch this customer to a 24 hour upaate spot price. At worst, the
customer can continue its old pumping schedule as if it were still on predetermined time-of-day prices. Assuming that the old time-of-day rate and the spot
price based rate were calculated based on comparable formulas, its bill for
electricity would then stay the same and the cost for the utility to serve it would
stay the same. (If the old rate had a hidden cross subsidy, in either direction,
then the bill could change by the amount of that cross-subsidy.)
However, our spot pricing customer will start adjusting its pumping
schedule each day, based on the hourly prices for the following day. It will plan
to pump in all the lowest cost hours, which may not necessarily be consecutive.
Because each day's schedule will be different, it wilI start heavy pumping at a
different time each day. On days when the overall spot price turns out to be low
(often but not always these will be weekends) it may pump almost all day, to fill
up longer term storage areas, and give more flexibility on high priced days. (This
requires some kind of forecasting, but no matter how unsophisticated the
forecasting method, over the course of a year the customer is never worse off
than just ignoring this opportunity to store for more than a day.) Different
I. Overview
15
seasons of the year and different generator outage situations will automatically
lead to appropriate behavior by the customer, without elaborate formal mechanisms. (e.g. look at Figure 1.1.1 and 1.1.2.)
On some days the customer may have unusually high demand for pumping
services (e.g. due to a fire). On these days it will know that it can get energy when
it needs it, albeit perhaps at a high price. To the extent it has some discretion
(e.g. the fire has been put out, but the tanks are still below a safe level) it can
choose to reduce demand during the peak peak priced hours. All such decisions
are entirely up to the customer, not the utility.
Obviously the customer's benefits from spot pricing depend on how much
operating flexibility it has (or can get, by making selective investments). They
also depend on the hour to hour and day to day price variation - the higher the
variation, the more the customer stands to save by shifting its usage.
How about the utility's costs? If the customer does not respond to spot prices,
the utility sees the same demand pattern as before and is no worse off. However
if the customer makes any response at all, the utility is guaranteed to be made
better off because all responses will shift energy use from times of high spot
prices to times of low spot prices.
We could construct similar examples for other situations, but the basic point
is clear. If the customer does not change its behavior, then neither it nor the
utility are worse off. (Except perhaps for shifts in cross subsidies. In principle
. thl:lSpot prices could be adjusted to preserve cross subsidies, and we will mention
thltt at an appropria,te point.) The more the customer changes behavior, the
better off both parties are. And highly variable spot prices make the benefits for
bo~hparties bigger.
Why No Numbers?
We have devbted a lot of time and effort to trying to understand how industrial
and residential customers will respond. We firmly believe large response will
occur. (Otherwise we wouldn't have written this book.) However, readers who
go beyond Chapter I will find that we have chosen not to present explicit
numbers or even case studies on customer response. One reason for this decision
was the desire to keep the size of the book under contro\. However, the main
reason is that our understandings have not reached the point where we can
provide numerics that can be extrapolated to cover a utility service territory. We
are like the classical blind men feeling different parts of an elephant. We have
a lot of information but still are not certain what the overall elephant looks like.
However, we do know that it is big.
SECTION 1.4. ENERGY MARKETPLACE OPERATION: A DEVELOPED COUNTRY
I~-------------------------I
~I____~
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
:L
Price Only
Transactions
Power
System
Operation
Price Quantity
Transactions
Customer's
Scheduling
and
Control
Computer
Short Term
Price
Forecast
~m~
________________________
JI
Long Term Price Forecast
Customer
I. Overview
17
o
o
A large industrial customer who sees one-hour update spot prices, purchases
interruptible energy, and has a long-term contract.
A sophisticated residential customer seeing 24-hour update spot prices.
A simple residential customer seeing 24-hour update spot prices. 6
HI
I. Overview
19
manual control. This customer learns of such times by reading the newspaper
and/or listening to announcements made by the utility over the radio and TV.
SECTION 1.5. ENERGY MARKETPLACE OPERATION: A DEVELOPING COUNTRY
This chapter has provided an overview of a spot pricing system. The authors of
the book strongly believe that a spot price based energy marketplace is the
logical and inevitable successor of the present-day system. The key question is
not if it will happen, but when, how and to what degree.
We would like to make a general comment on the appearance of some of our
results. Certain of our formulas lead to price behavior which is quite different
than conventional utility rates, and sometimes is even counter-intuitive. For
example, line overload conditions on a single transmission line can affect spot
prices throughout a network, raising some prices and lowering others. Another
example is that spot prices for customers and generators are entirely symmetric.
Results such as this reflect the physical and economic reality of electric power
systems. In fact in many cases power system practice has been similar to spot
prices for inter-utility sales, but customers have not been allowed to participate
on the same basis. For example, inter-utility sales of economy energy have long
been based on hourly prices, and have treated buying and selling symmetrically.
In other cases, the anomalies are due to regulatory issues such as revenue
reconciliation. They occur in any pricing system, although spot pricing
sometimes makes them more visible by removing hidden cross-subsidies.
One side effect of the way spot prices reflect physical and economic reality is
that spot pricing can be used to evaluate non-spot rates and transactions. For
example, if spot prices paid to a small power producer would over time give it
higher payments than the present-day rate, then at present the small power
producer is cross-subsidizing the utility's other customers.
SUPPLEMENT TO CHAPTER I: SUMMARY OF ISSUES
1. Overview
21
Transactions Costs
.h
The costs of changing the status quo will be large, but transient in nature. The
transactions costs of the additional metering, communication and computation
needed for an energy marketplace .will continue through time. Thus one basic
principle underlying the specification of energy marketplace transactions is that
the associated costs never exceed the additional resulting benefits, i.e., transac-
22
tional costs determine how customers are matched to types of transactions. The
microelectronic revolution is driving transactional costs down at a rapid rate.
Utility Operating Costs
These costs include fuel, losses and maintenance associated with generation,
transmission and distribution. Ideally, customer-utility transactions result in
customers shifting some usage to times when the operating costs are low. Flat
rates provide no such motivation. TOU rates can result in some desirable
demand shifting, but because of the unpredictability of the actual times of utility
low and high costs a year in advance (see Figures 1.1.1 and 1.1.2), TOU rates
can also cause undesirable load shifts. OLe could have a sizeable impact on
utility operating costs if it can be exercised as a part of the normal system
operation (i.e. if there is no limit on its usage). However, OLe often involves
contracts with customers which limit the number of utility control actions, so
the reduction in annual operating costs is greatly reduced. Demand charges
motivate customers to reduce their own monthly peak in a manner which
usually has little or no effect on the utility's operating costs.
A spot price based energy marketplace can yield major reductions in utility
operating costs because customers see prices directly related to the actual
operating costs occurring at that time.
Utility System Operation
I. Overview 23
I,
y
S
11
11
S
:)
s
h
I.
I,
11
11
11
Uncertainty in long-term load growth, fuel costs and availability, capital costs,
environment'al standards, etc. cause longfalls or shortfalls in utility capital
investments. Spot pricing provides feedback that acts as a buffer against uncertainty. Spot pricing is effective during times of capacity excess as well as shortage
(short- or long-term) because low prices can stimulate use of electric energy.
d
s
e
An electric utility needs to predict future load behavior on time scales ranging
from hours to years.
Predictions of short-term (hours) response to DLC requires a sophisticated
model for each individual applicance type under control (e.g., an air conditionmodel depends on time of day, temperature, and season of year). Developof DLC response models is difficult because DLe tends to operate only
and hence only a limited data base is available. Prediction of short-term
response and of short-term spot pricing response is very similar. Spot price
does require a somewhat more sophisticated model than TOU
but the marketplace transactions provide extensive data on both
:usltonler desires and usage patterns which enables the modeling to be done.
Even more important is the fact that spot prices provide closed loop feedback
control which reduces the impact of uncertainty.1O
The major determinants of long-term (many years) load behavior are exogenous quantities such as the general state of the economy, demographic trends,
and the price of electricity relative to gas and oil. Long-term predictions of OLC
impacts present special problems because many of the OLC techniques are
vulnerable to customer countermeasures, which customers learn over time as the
OLC is exercised more and more. The detailed information on customer usage
patterns and priorities provided by marketplace transactions is a major aid to
the long-term modeling problem.
Utility Revenue Stability
In an ideal world, a regulated utility receives revenues which cover its costs plus
a reasonable rate of return on investment. In the real world, life is not so simple.
TOU rates have caused revenue stability problems because utilities have found
it difficult to specify TOU rates that yield the specified revenue targets. An
energy marketplace introduces similar problems. However, the ability to continuously adjust prices to market conditions makes it possible to greatly reduce
revenue stability problems associated with modeling customer response. These
problems will decline as data becomes available to develop the necessary
models.
Customer Options
Customers have a growing desire to make their own decisions on the type of
electric service they purchase. OLC and interruptible contracts are steps in this
direction. However, an energy marketplace provides customers with a richer
spectrum of options and allows more freedo,m of choice.
Customer Education
Customers almost always say they want to reduce their monthly electricity bills.
Actually they are willing to pay a higher bill if it is associated with tangible
benefits they understand, such as increased comfort or production. Spot pricing
gives customers maximum control over their bills.
The time average of TOU rates is higher than the time average of spot prices.
Thus customers with flat usage see a reduction of their bills in the energy
marketplace.
I. Overview 25
s
d
t1
e
e
y
Customers need to understand what is going on. Customers can readily understand TOU rates for energy, but are often confused by demand charges and
DLC contracts which include limitations and conditions on utility control.
Spot pricing is a concept customers can understand because they are used to
seeing prices which fluctuate in response to supply and demand conditions (e.g.,
the price offresh produce and airline tickets). An energy marketplace can result
in a smaller and more easily understood set of transactions than that presently
used by many utilities. Today, many utility rate books are thick documents that
take training (and a lot of patience) to understand.
<;ustomer Response
er
of
iclis
rs,
lIs.
ble
ng
:es.
rgy
re~ond,., DLC has the advantage that the customer does not have to make or
implement'real-time decisions. However, DLC can have a major negative
impact on ,c:ustomer lifestyle unless override capabilities are provided (e.g.,
residential atr conditioner cycling during a hot day when the customer is having
an important party at home). Modern electronics can make it easy for a
customer to respond to spot price variations. Since the customer is in final
control, major impacts on lifestyle such as might occur due to the loss of air
conditioning at particularly inappropriate times do not happen under spot
pricing. The utility can provide control services for residential and small commercial customers where the customer chooses the control strategy to be
.T/"\II/"\\l"Pfl while the utility provides the hardware and software for its imple-
Investor-owned utilities operate under regulatory commissions. An energy marketplace changes the role of the regulatory commission.
In an energy marketplace, the regulatory commission no longer determines
the actual values of the rates. Instead the commission determines the formulas
which are used to compute the spot prices. This is a logical evolution from
present practice, such as fuel adjustment clauses. It has the advantage that the
commission is oper~ting at a higher level of decision making which is more
appropriate to its basic function.
An energy marketplace gives the commission more options on how to provide
the utility with incentives for efficient operation.
Impact on Microelectronics Industry
The microelectronic revolution has given birth to many new, dynamic companies which are trying to introduce new communication, computation and
control systems into the electricity market. However, many vendors have
become extremely frustrated with the present-day electric utility situation
because it is not clear what sort of product is marketable. The establishment of
an energy marketplace will specify the ground rules for such vendors and
provide them the opportunity to exploit their capabilities and offer innovative
products to the customers. This will result in reduced costs for transactions and
customer response.
Discussion of Supplement
All of the above utility, customer, regulatory and vendor issues have to be
considered when deciding whether to follow the energy marketplace path and,
if so, what directions to take. We believe that careful weighing of the costs and
benefits will show an energy marketplace to be a clear winner in virtually all
situations.
HISTORICAL NOTES AND REFERENCES-CHAPTER I
I. Overview
\n
an
be
ir-
es
as
m
1e
re
Ie
1-
d
'e
n
)f
d
e
d
27
NOTES
d
e
h
:r
:e
o
y
28
8. A variation on this approach is to replace the use of prices to allocate a scarce resource by
allocating each industry energy credits, where a kWh used during critical times requires the use
of more credits. This leads to the possibility of bartering among customers (see Williams, 1984).
9. Such times often do not correspond to times of peak system demand because of maintenance
scheduling and random equipment outages.
10. In theory, feedback can sometimes aggravate the effect of uncertainty and even lead to an
unstable system. However, spot prices are computed in a fashion which provides desirable
feedback.
e
).
:e
n
Ie
29
The hourly spot price for electric energy is the foundation of our approach
an energy marketplace. This chapter provides a review of hourly spot price
ci)ncepts. Mathematical details can be found in Part II. The Annotated Bibliogiaphy'contains an extensive list of references and discusses their contents.
The hourly spot price is determined by the supply/demand conditions that
exist at thai hour. In particular it depends on that hour's:
to
o
o
This chapter provides the reader with an understanding of the composition and
interpretation of hourly spot prices.
Sections 2.1 through 2.4 define the hourly spot price and its components.
Section 2.5 discusses the special case where an aggregated network model is used
instead of explicitly considering individual network lines. Section 2.6 discusses
revenue reconciliation. Section 2.7 discusses buy-back rates. Sections 2.8 and 2.9
discuss two measures of hourly spot price behavior: expected price trajectories
and the price duration curve. The chapter concludes with Section 2.10, which
considers customer response.
31
Define
Pk(t): Hourly spot price for kth customer during hour t ($/kwh)
dk(t): Demand of kth customer during hour t (kwh)
d(t): Total demand of all customers during hour t (kwh)
del) = Lk dk (t)
An hourly spot price can be quantified in various ways. The basic approach
used in this book is
Pk (I): Marginal (or incremental') cost of providing electric energy to cllstomer k during
hour ( taking into consideration both operating and capital costs ($/kwh).
The hourly spot price (without revenue reconciliation) is given by the marginal
cost, i.e.
(Total cost of providing electric energy to all customers
now and through the future]
(2.1. I)
The basic definition of (2.1.1) involves only marginal costs without consideration of revenue reconciliation - i.e. without consideration of embedded capital
costs and rate of return on investment.
A key property of marginal cost based spot prices (equation 2.1.1) is
They tend to recover both operating and capital costs.
Since generation is assumed to be dispatched optimally, marginal costs exceed
average variable operating costs. Thus, charging customers at marginal costs
yields revenues that exceed total variable operating costs; and this difference can
be applied towards the capital costs. An optimum power system's marginal costs
yields revenues which exactly match operating and capital costs. Unfortunately,
in the real world, this difference will usually either over- or underrecover capital
costS.2 Mechanisms for revenue reconciliation are discussed in Section 2.6.
NUMERICAL EXAMPLE. Consider a utility with two generators where
Capacity
IOOOMW
IOOMW
Generator I
Generator 2
Operating Costs
2/kWh
IO/kWh
Assume the generators are optimally dispatched (i.e. Generator I is used until
demand exceeds 1000 MW). Then, ignoring losses and capital costs, it follows
that
If Demand
If Demand
1000MW, then:
1100 MW, then:
Utility
Operating Cost
($(hr)
20,000
30,000
Average Op.
Cost
((kWh)
2
Spot Price
((kWh)
2.72
10
Assume
[)
gefine
short-term social welfare as
.
f
.I
Assume customers behave in their own best interest, i.e. are not willing to pay
1O(kWh fCir a benefit worth only 5 (kWh. Then it follows that
.II
:d
t5
t5
y,
Demand
(MW)
Customer
Benefit
($)
Minus
Costs
($)
Equals
Short-Term
Social Welfare
($)
1000
50,000
20,000
30,000
1100
55,000
30,000
25,000
Hence short-term social welfare is higher if customers see spot prices instead of
average operating costs. Now change the conditions and assume the customer
benefit is 15 (kWh instead of 5 jkWh. Then for both pricing schemes
Demand = 1100 MW
Customer Benefit = $165,000
Utility Revenue
($)
Utility Revenue
Minus Costs ($)
110,000
80,000
30,000
The $80,000 profit made by the utility under spot pricing can be used to pay the
capital costs of the two generators; however, this may either over- or underrecover the actual capital costs of the plants. We will return to this issue in
Section 2.6.
SECTION 2.2. COMPONENTS OF HOURLY SPOT PRICES
The hourly spot price associated with the kth customer during hour t is viewed
as the sum of individual components defined byJ,4
[Generation Marginal Fuel]
Pk(t) = Yr(t)
+
+
+
+
+
+
YM(t)
YQs(t)
YR (t)
~L.k (t)
~QS.k(t)
~R.k(t)
(2.2.1)
A(t)
YF(t)
yet)
A(t)
'lk (I)
I/u(t)
YM(t)
[System Lambda]s
YQs(I)
~QS.k(t)
(2.2.2)
For example, the loss component 'fJL.k(t) depends on system lambda, A(t).
The network components of the hourly spot price depend on the customer
index k because different customers are located at different parts of the network.
This spatial pricing results from the differences in line losses and the fact that
individual lines can become overloaded in one part of the network while over
the remaining lines flows are sustainable. In practice, spot prices will be the same
for most customers in a specific class, e.g., service at 13 kV in the Sanderson
Township. Section 2.5 introduces aggregated network models which eliminate
some of the complications associated with spatial pricing.
SECTION 2.3. OPERATING COST COMPONENTS
The operating cost components of the hourly spot prices are usually the largest.
They are:
Generation Fuel and Operations: A(t)
Network Losses: 1)L,k (I)
The system lambda, A(t), component of the hourly spot price is the derivative
of generation total fuel and maintenance costs with respect to demand this hour.
Ideally it is obtainecLas the output of the unit commitment generation dispatch
logics used in many modern electric power system control centers. System
lam~da (as it is defirted in this book) includes the effect of purchases and sales
th' utili!y may make with external, interconnected utilities.
In general, A(t) tends to increase with increasing total demand d(t). The A vs.
demand curve varies over time since it depends on forced (and scheduled) power
plant outag1s, water availability, purchase-sale opportunities, load-following
costs as affected by ramp and must-run constraints, etc. When these intertempora! constraints are important, A(t) may be heavily influenced by planned
future events. For example, in a system with a lot of hydro storage, the incremental generator is sometimes a hydro unit. The system lambda at those times
is calculated based on the anticipated running cost of a fossil unit which will
need to generate more when the reservoirs are empty.
NUMERICAL EXAMPLE. Consider a 14-generator system (no transmission losses)
with maximum demand of 2000 MW where
Plant
1 Nuclear
I Coal
12 Gas Turbines
Capacity
MW
1000
800
100 each
Operating Cost
/kWh
2
3
10
Figure 2.3. J shows how the availability of the nuclear and coal plants affects the
36
$/kWh
Gas Turbine
10 f-
3
2
Coal
Nuclear
1000
18002000
d
kWh
Gas Turbine
10
Nuclear
2
I
1000
18002000
Gas Turbine
10
Coal
31-------'
2
800 1000
1800 2000
This component arises from the energy losses resulting from transmission and
distribution. It is shown in Chapter 7 that, assuming a quadratic dependence of
losses on line flows, 1]L,k(t) is given by
[A.(t)
oL(t)
}'Qs(t)] fJdk(t)
(2.3.1)
[A(t)
"
az;(t)
I'Qs(t)] ~ 2R;z;(t) adk(t)
L, L,[z,(t)]
losses in line
L;[z,(t)]
R;z7(t)
6
aztO '=
ad(t)
L(t)
d(t)
,
Ri.~t)
.~
[A(t)
I'Qs(t)]2Rd(t)
Assume annual line losses over a year are 5% of annual flows. Thus
Annual Losses = 5% Annual Flows
So
8760
8760
:l
1=1
Ri(t)
0.05
L z(t)
(=1
f
Assume the demand variation over a year is such that
z(t)
5.5(10- 5 )
[A(t)
[A(t)
[A(t)
1500 MWh,
+ 1'Qs(I)]2(5.5)(l0-5)(1500)
+ 1'Qs(I)][O.165]
500 MWh,
1'QS (1))[0.055]
Thus for annual losses of 5%, the marginal network loss component ranges
from 5.5% to 16.5% of A(t) + YQs(t). Note that if line load were constant over
the years, marginal network losses would be twice the average loss, or 10%.
SECTION 2.4. QUALITY OF SUPPLY COMPONENTS
The generation and network quality of supply components, YQs(!) and IJQS.k(t),
can be quantified in different ways. All approaches yield behaviors characterized
by very small or zero levels most of the time with a large, rapid increase when
the generation or network capacity is being ,approached. During such critical
times, these quality of supply components dominate the hourly spot price.
Generation Quality of Supply: l'Qs(t)
Define
get): Total generation during hour t (kWh)
get) = Total Demand + Total Losses
= d(t) + L(t)
gcrit.y(t): A critical generation level based on available generation capacity and
operating reserve requirements
Market Clearing Price: Set I'Qs(t) to be the value that causes customers to
reduce usage until g(t) = gcrit.y(t). This value depends on the amount of load
reduction required.
Value of Unserved Energy: Set I'Qs(t) such that the resulting spot prices equal
the cost to the customer of unserved energy, averaged over customers end
uses.
Allocation of'Peaking Plant Capital: Base YQs(t) on the annualized capital
cost of a new peaking plant.
For the peaking plant cost allocation method, one approach discussed in
Chapter 6 yields
(2.4.1 )
')'os(l)
H760
L (1;(1)
G;,
f~
es
er
t),
Note that multiplication of av(t) by any constant does not change YQs(t)
'.Jhe loss of load probability LOLPy(t) in (2.4.1) is evaluated for hour t at the
be-ginning of hour t. ~f it is assumed that all events can be predicted perfectly one
h(),ur in advance, then
~d
LOLP .. (I) =
al
ld
otherwise
900 MWh
g(t)
1000 MWh
To illustrate the market clearing price approach, assume all customers are seeing
spot prices and that a 10% drop in load requires a 20jkWh increase in price
(i.e., an elasticity of - 0.05). Then YQs (t) = 20 jkWh. To illustrate the value of
40
the unserved energy approach, assume the cost to the customer of unserved
energy is 100 /kWh. Then when generation starts to exceed its critical level,
')IQs (t) goes up to 90 /kWh so thaL1.(t) + ')IQs (I) = 100 /kWh. Finally, consider
an illustration of the peaking plant capital approach. If Annualized Plant
Cost = 100$/kW
and
LOLH y = 200 hours,
then
')IQs(t) = [50/
kWh] x LOLPy(t).
Network Quality of Supply:
'1Q5,/I)
By analogy with 1Qs(t), I]QS.k(t) becomes large in magnitude when the capacity
of the network to transport energy is being approached. Assume one particular
line, say line i, with flow z;(t) is overloading. One structural form for 'lQS,k(t)
derived in Chapter 7 is
110s,k ()
t =
()
05.",,(1
)OZ;(I)
odk (l)
(2.4.2)
where two ways to model the marginal cost 8Qs .ry,,(t) are
o
Market clearing: 8Qs .,I ,; is adjusted until customers and generators respond to
change the usage and generation patterns so that the line overload goes away.
The derivative of a cost function with respect to z;(1) where the cost function
is zero until z;(t) approaches the line overload level.
Spot prices are affected at locations throughout the network even if only one line
(line i) is being overloaded,
NUMERICAL EXAMPLE. Consider two buses with one lossless line connecting
them. Assume Bus 1 has a base-load generator with marginal fuel costs of
5/kWh while Bus 2 has a peaking plant with marginal fuel costs of 1O/kWh.,
Assume all load is at Bus 2 and the load is less than the base load unit's capacity.
Thus if the line can carryall of the demand, the peaking plant generation is zero
and
o
o
A. = 5/kWh, I]QS.k = 0
At both busses p = 5/kWh
If the line capacity is less than the demand and the peaking plant has to be run
to meet the demand, the resulting prices are
o
o
p = 5/kWh
p
10/kWh
'lQS,2
= O.
rved
~vel,
ider
lant
50/
The generation and network quality of supply components can yield very high
spot prices under some conditions. If the energy marketplace contains a mixture
of customers seeing hourly spot prices and customers seeing long-term,
predetermined prices, then the outage costs of the predetermined price
customers set an upper bound on the hourly spot price. Thus
When the hourly spot price reaches their outage costs, the predetermined price customers
are "dropped" rather than raising the hourly spot price any further.
lcity
:ular
;,k (t)
.4.2)
j to
lay.
Ion
The network loss '1L,k (t) of (2.3.1) and network quality of supply '1QS,k (t) of
(2.4,2) are based on an explicit network model which considers individual lines
(and transformers, etc,). In practice it is often desirable to use a more aggregated
network model for the transmission system. Even if all individual transmission
lines are modeled, some aggregation of lower voltage distribution lines will
almost always be desired.
To illustrate the approach, assume all transmission and distribution lines are
t(),.be aggregated together. This can be viewed as a hypothetical two-bus system
'~here all generation is on one bus and all demand d(t) is on the other. Therefore
"ill customers see ~the same spot price; i.e., Pk(t) = pet), '1L.k (I) = '1L (t),
.llqSk(t)
'1Q5(t)
109
of
Vh.,
ity.
ero
where the losses L(t) are the function of total demand d(t), total generation get),
and possibly other quantities such as purchases and sales from other utilities.
The network quality of supply term of (2.4.2) is now based on the use of an
aggregate cost function. Since individual line flows are not modeled, market
clearing or a cost function based on individual lines cannot be used. One
approach discussed in Chapter 7 is closely related in philosophy to the "annualized cost of a peaking plant" method of computing the generation quality of
supply YQs(t). It is
un
11Qs(1) =
(2.5.2)
AQs ,),: Annualized capital cost of network expansion made to maintain system reliab
ility ($jkW)
a,,(1)
42
8760
L a~(t)
ii"
t~1
a~(t)
is
k[d(t) - dcd',~]
otherwise
(2,5,3)
k:
If the demand d(t) is viewed as a random variable with a flat probability density
between d crit." and dma., then it is shown in Section 7.9 that (2.5.2) (2.5.3) become
for del) ~ dcrit.q
(2.5.4)
NUMERICAL EXAMPLE.
Assume
100$/kW
~
=
~ dcrit.~,
(2.5.4) yields
so for del)
~Qs (t)
0.05
[::~ -
0.9] $/kWh
dmax
45 /k Wh
o
o
I)
For an ideal world, the use of revolving funds or certain types of surcharge/
refund is recommended, as they do not change the hourly spot price and hence
attain the maximum possible social benefits. However, such approaches present
many practical implementation problems. Therefore we will discuss here the
approach of modifying the spot price through the use of revenue reconciliation
components.
Define
pdt)
yet)
y(t)
[Generation Components]
[Network Components)
II! (I)
A(t)
(2.6.1 )
YQs(l)
The basic idea is to modify the prices paid by the customers so that the
utility's revenue over some time interval (say one year) covers its operating and
eniJ';,edded capital costs plus a reasonable rate of return on investment. This
gives rise to the gen~ration and network revenue reconciliation components,
y~~O'and IJR.k(t) respectively.
0ne sttucture that~is a special case of the Ramsey pricing or weighted least
squares theory of Chapter 8 is a multiplicative modification, i.e .
YR (I)
1111;(1)
m'lA (I)
'1R,k(l)
J:l.760
(l
m)
I I
I =:: I
fh(l)ddt)
(2.6.3)
I)
where the left-hand side should actually be written in terms of the expected value
of lJdt)dk(t). If demand response to price is considered, dk(t) is a function of Pk(t)
which is a function of m, so an iterative solution of (2.6.3) for m is required.
44
where m and m~ can have different signs. Another decomposes revenue reconciliation "even further down to individual lines or voltage classes of lines. Still
another approach yields revenue reconciliation multipliers that vary with k, e.g.
depends on the kth customer's demand elasticity. Some forms discussed in
Chapter 8 have m's which depend on dk(t), i.e., represent nonlinear pricing
modifications. We emphasize use of constant multipliers in this book because
they simplify the discussions. In practice, a more sophisticated approach can be
chosen for actual implementation.
SECTION 2.7. BUY-BACK RATES
Ifa utility buys back electric energy from its customers, an hourly buy-back spot
price is needed. An important point developed in Part n is
The operating and quality of supply hourly spot price components are independent of
whether the customer is buying from or selling to the utility.
However, revenue reconciliation destroys this symmetry. Revenue reconciliation increases the hourly buy-back spot price when the utility is overrecovering
revenue and vice versa.
In Section 2.6, one hourly spot price with revenue reconciliation is given by
Pk(t) =
(I
m)[(y(t)
IJk(t)]
(2.7.1)
This same basic formula applies to both buying and selling, except the value of
m changes. Define
msell:
mbuy:
For the special case leading to (2.6.2), it is shown in Chapter 8 that after a few
approximations
so that if
Spot price for selling energy to customer
Pbuy.k(t): Spot price for buying energy from customers
Psell.k (t):
then
P"II.,(t) =
(\
m)[y(t)
(1 - m)[y(l)
Pbuy.k(l) =
NUMERICAL EXAMPLE.
P,dl = ~ =
.8
(2.7.2)
til(t)]
If m
11k(t)]
0.2,
1.5
Phu)
An Unfortunate Complication
The use of (2.7.2) can lead to unpleasant complications in some situations. As
an example, consider a customer with demand d(l) and own generation get)
where d(t) > g(t). If the customer has a single meter which measures a net
purchase of d(t) - get) from the utility, the customer's bill is
Bill = (\
m)[y(t)
(2.7.3)
tik(t)][d(t) - g(t)]
If the customer has two meters, one for demand d(t) and one for generation get),
the
customer's net bill is
#:,\:,
0;
Net Bill
= (I
m)(1!(t)
- (I - m)(y(t)
'1k(t)]d(t)
[Usage]
(Generator Buyback]
'1k(t)]g(f)
(2.7.4)
Unfortunately (2.7.3) and (2.7.4) are not equal unless m = 0, i.e., unless revenue
reconciliatiOin components are not needed.
This problem is not basic to spot pricing; it is simply a consequence of using
a simple approach to revenue reconciliation. Other approaches to revenue
reconciliation discussed in Chapter 8 such as the use of revolving funds solve this
problem but introduce other complications. The world would be nicer if revenue
reconciliation could be ignored.
SECTION 2.8. EXPECTED PRICE TRAJECTORIES
The preceding sections discussed the individual components of the hourly spot
price. It is now time to put them together and look at the behavior of the spot
price itself.
The hourly spot price is a random process because demand and outages are
random. Figure 2.8.1 presents expected spot price trajectories for 24 hours
assuming no spatial (k) dependence, and that demand is independent of price.
These trajectories are the conditional expectation of the spot price given a
specific demand level; i.e. the effects of generation outages have been averaged
out.
46
Mils/kWh .-----y-EA-A-'-=-=PECCAK"'"'O-AY-,:S-:S....,UM-:M-:AR::c
y- - - - - ,
500
MIIS/kWh~
300
400
300
W,n'.,
200,
200
;wnme,
'I~L2?(,~~,.
100
12
18
22
24
Hours
MUs/kWh
Mils/kWh
300
300
200
200
100
'00
'2
SUM~lAAY
18
24
Hours
The expected hourly spot prices exhibit very wide variations with time.
The behavior of spot prices may change significantly with changes in installed
capacity.
It is tempting to look at Figure 2.8.1 and also conclude that hourly spot prices
Price
(/kWh)
50
10
2000
Hours
>
te,nQ to be smallest -in the Spring and Fall. However, such a conclusion is not
n~essarily true for1wo reasons:
Maintenance Scheduling Was Ignored: Most maintenance is done in Spring
and Fa\l~Therefore prices during Spring and Fall will usually be higher than
shown here.
" Averages Over Generation Outages Were Used: The actual p(t) can get quite
large in Spring or Fall if major forced outages occur.
o
Because of generation outages, the highest spot prices will usually not occur
at the hour of the highest demand. Furthermore, actual spot price trajectories
will not always be as smooth as those of Figure 2.8.1, since a generator outage
during hour 1 can cause the spot price to "jump" the next hour.
SECTION 2.9. PRICE DURATION CURVES
The 24-hour price trajectories of Section 2.8 give one picture of hourly spot price
behavior. A different picture (taken from a different angle) is given by looking
at a price duration curve which defines the probability distribution of the hourly
prices.
An annual price duration curve can be thought of as summarizing the
information included in 365, 24-hour expected price trajectories; see Figure
2.9.1. It is analogous to the load duration curve. Price duration curves can be
obtained for a year or a subyearly period.
A price duration curve can be calculated using three different methodologies:
o
o
o
The conceptually simplest way to compute an annual price duration curve starts
by computing expected hourly price trajectories for 365 days and then summing
the number of hours in the year that the expected value exceeds some specified
level.
Monte Carlo Simulation
The hourly trajectories of Section 2.8 are computed using the expected value of
the marginal costs where the expectation is over generation outages. Thus a
price duration curve computed directly from such hourly trajectories (as above)
does not display all of the actual characteristics of the hourly spot price variations.
A more accurate procedure is to compute hourly spot price trajectories by
Monte Carlo simulation, wherein the explicit values for the generation outages
for a given day are first determined using random numbers and then the 24-hour
price trajectory for that day is computed. Plotting the resulting 24-hour trajectories would not be very meaningful, but by doing enough random sampling, the
random hourly price trajectories can be averaged into the actual price duration
curve. The Monte Carlo price curve will have a higher peak than the one which
just summarizes 365 expected 24-hour trajectories.
Monte Carlo is the most versatile and powerful approach because detailed
time dynamics (such as customer response to spot prices using thermal storage)
can be incorporated. Unfortunately, the Monte Carlo approach can require a
large amount of computer time.
Use of Probability Convolution
n be
gies:
the price duration curve. The remaining components are obtained by considering their conditional expectation given the level of system lambda (since their
correlation with system lambda is generally high, the error introduced is usually
small).
In the 24-hour price trajectories of Section 2.8, spot prices were based on the
expected value of system lambda conditional upon the load level. Here spot
prices are based on the expected value of load conditional upon system lambda
level.
SECTION 2.10. CUSTOMER RESPONSE
arts
ling
fied
e of
IS a
Ive)
Customers have no desire for electric energy per se (although people can get a
charge out of it). Customers do desire the services provided by electric energy.
Customers respond to different prices in two basic ways:
naby
ges
Dur
jec-
the
IOn
ich
led
ge)
ea
)th
t is
ent
)n)
lat
t is
mg
o+;Modify Usage: ltprice in a given hour is very high, they may reduce usage
.:at that hour because the value of some of the services is less than the price.
. Alternatively, if price is very low, they may increase usage to receive services
~::l:h~y !,ormally wouldn't buy.
Reschedule Usage: If the price is high during some hours of the day and low
during other hours, customers may reschedule usage so their desires are met
but at dift!erent times. Such rescheduling usually imposes some customer costs
(or reduction in benefits received from the service) so the amount of rescheduling depends on the price difference.
Some examples of customer response are
o Space Conditioning: Reset thermostat. Pre-heat or cool at times of low price
to make use of a building's thermal storage capability. Use thermal bricks or
ice to store thermal energy purchased at times of low prices.
o Water Heating: Heat at times of low price.
o Water Pumping: Fill t.anks at times of low price.
o Reschedule Production: Reschedule hours of industrial production for
processes with high energy usage and low labor cost to times of low price
(assuming sufficient product storage is available).
o Fuel Switch: Use electric energy when its price is low and switch to oil or
natural gas when the price of electric energy is high.
o Cooking: Change the menu if the price is high or defer the meal until the price
drops (not necessarily a "cost-effective" response).
50
Price
($/kWh)
Resulting Demand
Demand (kWh)
Supply Curve: Plot of p vs. d showing how supply costs and hence prices
increase with demand.
Demand Curve: Plot of d vs. p showing how demand decreases as price
increases.
If both curves are plotted on the same axis, their intersection yields the resulting
values of both price p and demand d. This is illustrated in Figure 2.10.1. Both
curves change each hour in response to weather, time of day, outages, purchases
from other utilities, etc.
!ar,
md
the
lith
~Ia
aglasmd
ces
ice
ing
)th
ses
This chapter only presents overall ideas. Details and the underlying theory are
given in Chapters 6, 7, and 8 of Part II. For further reading on our approach,
see the notes and references at the end of these chapters and also the Annotated
Bibliography.
Our basic approach defines an hourly spot price which depends on equipment
ot1~ges and load levels at that hour, as well as long-run revenue reconciliation.
T~'se hourly spot pt;ices, which are set after events are revealed, provide the
ba~is.for the actual energy marketplace transactions.
~any.authors have rejected the idea that prices can be set after events are
revealed. Fbr example, Crew and Kleindorfer [1980, p. 55] write:
Where the fin~ sets the price, it may do so either exallte or expost ... For the case of the
regulator setting the price expost, he or she would either have to allow a market-clearing
price or have some deliberate arrangement for setting the price above or below the price.
Were the regulator [to allow) the market clearing price, he would, in effect, be giving up
his right to regulate price.
Turvey and Anderson [1977, p. 298] are even more adamant in their rejection
of spot pricing:
'" for a wide class of random disturbances (but not for all), it is not possible to respond
to the resultant random excess or shortage of capacity by adjusting prices ... Failure of
a generating plant on Thursday cannot be followed by a higher price on Friday, and the
price in January cannot be raised when it becomes apparent that January is colder than
usual. Even though telecontrol makes the necessary metering technically possible, it
would be expensive, and ... there would be difficulties in informing consumers of the new
price. It would also be scarcely possible to estimate its market clearing level. Sudden and
random price fluctuations would in any case impose considerable costs and irritations on
consumers. Hence responsive pricing that always restrains demand to capacity is not
practicable, and some interruptions [are thus desirable].
Their rejection thus appears to be based on the belief that the transactions costs
of spot pricing would outweigh any possible benefits. At least for large
customers, we obviously do not agree. Of course, the technology for decentralized communication and control has improved drastically since these words
were written.
Other authors have accepted to some extent the concept of setting prices after
events have occurred. As discussed at the end of Chapter I, spot pricing of
public utility services was apparently first proposed by Vickrey [1971], under the
name "responsive pricing." His original article presented general discussions
using as examples mainly long-distance telephones and airlines. The,emphasis
was on curtailment premiums, rather than on marginal production cost changes
over time. Later manuscripts on electricity developed the ideas in more detail,
including some discussion of optimal investment criteria [Vickrey, 1979, p. 12],
metering requirements and designs, pricing of reactive energy, and short-run
marginal operating costs (system lambda). He proposed that utilities be free to
set prices however they want over time, subject only to limits on total profits.
In retrospect, many of his ideas were quite far sighted.
Vickrey's essential insight was that prices can be set after some variables are
observed, and optimal prices should reflect this. Since his original article,
different versions of this basic idea have been developed independently and
under different names, with varying levels of rigor. The other approaches
include:
o
o
o
53
:osts
arge
cenords
display system designed for residential use. A new version can accept a spot
price data stream.
Blackmon [1985] evaluates feasibility and potential benefits of establishing a
futures market for electricity related to the ideas of Section 3.5.
tfter
NOTES
g of
. the
ions
lasis
1ges
tail,
12],
,run
e to
fits.
are
de,
and
:hes
mal
oth
ous
ork
:or~nt.
:ific
etic
,ith
for
(er.
din
tem
has
oad
and
\. At this level of the discussion, marginal means the same thing as incremental. In actual imple
mentation there is a difference between the two terms (see Chapter 4, Section 4,5),
2, Considering the massive uncertainty utility planners have to face, it is almost an accident if the
generation mix, etc, happen to be "optimum" at any given time,
3. A notational convention should be clear; y's and IJ'S are used for generation and networkrelated
quantities respectively,
4, The theory of Part II includes another component related to network maintenance, but it is
ignored here because it is difficult to model and usually not very important (see Section 7.4),
5, }.(I) as defined in this book includes the effect of purchases and sales with other utilities, Hence
it is not always equivalent to the "lambda" used in economic dispatch logics (see Appendix B),
6. R, has units or (MWh) - I and not the resistance units of ohms because of the use of "per unit
voltages," See Appendix D.
7. In general, the market clearing price is less than the cost of unserved energy because market
clearing implies that the customers select the least valuable load first. Unserved energy costs
assume nonselective blackout costs. which include very valuable loads,
8. The curves of Figure 2,8, I are the result of actual studies of a particular utility, Other utilities we
have studied have the same amount of variation but of course with different characteristics,
56 . I:Tlleenergimarketplace
energy marketplace. Then Section 3.8 relates spot price based transactions to:
present-day transactions. The chapter concludes with Sections 3.8,3.9 and 3.10,
which discuss customer-owned generation, special rates for special customers,
and wheeling rates.
SECTION 3.1. CRITERIA FOR CHOICE OF TRANSACTIONS
Of course such a criterion cannot be used until the costs of transactions and their
benefits are understood and estimated.
The costs of electricity sales fall into two categories: physical costs, and
transactions costs. Physical costs were dealt with in Chapter 2, and are for
energy itself (per kWh). Transactions costs are the costs of computing and
communicating the information and money that go with the transactions.
Transactions Costs
Imers,
is
their
and
: for
and
oth
ual
ing
ter
rs.
lill
,'s
r's
Role of Forecasts
The last two criteria deal with issues concerned with the operation and control
of the e1ectriQ power system and the customer's own usage devices. An ability
to forecast the future with some degree of accuracy is critical for both functions.
In particular,
o
nm
LIS
~e
n
n
v
57
The electric utility must be able to forecast approximately how the aggregate
customer demand will respond to price.
The customer must have available a forecast of how prices will behave in the
future.
The use of forecasts available does not imply that perfect foresight is required,
i.e. that the forecasts have to be error free. It does imply that information on
expected future behavior is essential for making control, operating and planning
decisions. This is also true today.
SECTION 3.2. CUSTOMER CLASSES
In Chapter 2, a spatially varying (due to the network effects) hourly spot price
was discussed. Thus, in theory, each customer could have a different hourly spot
price. In practice, some large industrial customers may have their own prices,
but most customers will just be divided into classes. Subsequent discussions
often refer to the "price seen by the kth customer." This price will be the same
for all customers in their class.
Three customer classes used in present-day rates are residential, commercial,
and industrial. In an energy marketplace, one logical basis for defining customer
classes is the voltage level at which the customer receives service. Customers who
receive service at 13 kV impose fewer capital costs and cause fewer network
losses than customers fed at a line voltage level such as 440 V. Thus the higher
voltage customers should see lower prices (all other things being equal).
A related basis for specifying customer classes could be geographical location,
if the network capital costs and losses vary widely throughout the utility's
service territory.
As will be seen, a customer's demand characteristics can also form the basis
for the definition of customer classes.
SECTION 3.3. PRICE-ONLY TRANSACTIONS
For a price-only sale of electric energy, the utility quotes, in advance, a fixed
price for energy ($/kWh) where the quote is valid for some specified period of
time. The customer can buy any amount of electric energy at the quoted price.
Most present-day transactions are of this type, and this will continue in virtually
any implementation of spot pricing.
Characteristics of Price-Only
There are many possible price-only transactions types. The four discussed most
often in this book are
o
One-Hour Update: An energy price valid Jor the next hour is quoted at the
beginning of the hour.
24-Hour Update: On the afternoon of the present day (say at 4 PM), the 24
prices are quoted that will hold each hour for a period starting early the next
morning and lasting until the same time the subsequent morning (say from
3AM to 3AM).
Billing Period Update Flat: A flat energy rate (i.e. constant in time) valid for
the subsequent billing period is quoted at the beginning of the billing period,
e.g. each month.
Billing Period Update TOU: A time-of-use (TOU) type energy rate (i.e. varies
at prespecified times of day and days of week) is quoted at the beginning of
the billing period.
Yearly update flat and/or TOU rates could be defined analogously to the above.
The preceding four types illustrate two key characteristics of price-only
transactions:
o
Update Cycle Length: The length of time a quoted price or set of prices is
valid. The interval between new price announcements.
Period Definition: The number of separate prices that are quoted within the
update cycle.
example, a 24-hour update has an update cycle length of 24 hours and
definition of24. A billing period TOU update has a period definition that
on the definition of peak, off-peak and shoulder times, I and an update
of one billing period.
Two other general characteristics of price-only transactions are
:ial,
ner
,ho
:)rk
her
o
on,
:y's
Advance Warning: Length of time before the start of an update cycle that the
prices are quoted.
Number of Price Levels: Prices may be constrained so that they can be set
only at prespecified levels.
lsis
:ed
of
ceo
lIy
T~:'mustrate
)st
he
24
:xt
'm
Pk(tIT)
or
d,
es
of
'e.
Iy
is
(3.3.1 )
information at hour
T.
In less mathematical terminology, E{Pk(t)lr} is the best guess of the value of the
hourly spot price at t that can be made at the earlier time r. This is not
unexpected. Of course with a one hour update, t = rand Pk(t/r) = Pk(t).
The second term of (3.3.1) is called the covariance term. It is much less
intuitive, very messy in appearance, and in general less important. In fact, it can
often be dropped. However, it is part of the general theory so it will be discussed.
Cov (1)
E{df,(t)/r)
E{[Pk(t) -
d: (t)
E{Pk(t)lr}][d;(t) -
(3.3.2)
E{d;(t)lr}]lr}
10/kWh
Define
q: Probability coal plant is available during hour t given all information available at hour
1".
Then
E{p(t)lr}
q5
(I -
q) 10
10 - 5q
(/kWh)
so (3.3.1) yields
.3.2)
= 5.25 jkWh .
Assume the only uncertainty at time r is the outside
temperature that will exist during hour t. Assume that if the temperature is high,
the overall demand will be high and the gas turbine will have to be used. If the
temperature is low, only the coal plant will be needed. Thus
If q
(/kWh)
10 - 5q
p(tlr) =
NUMERICAL EXAMPLE.
pet) =
e of
tive
the
e is
)vasee
Define
q: Probability temperature is low during hour t given all weather information available
at hour r.
10 - 5q
na
ltes
r IS
our
s to
H&i>wever, now assU}J1e the kth customer has an air conditioner which is run
w.!i'cn the temperature is high, so during hour t
). . II
dkrt)
='
IS on
2. kWh'f'
I air cond
Itlonmg
Then
E{dk(t)I,}
2(1 - q) =
2 - q
5q(l - q)
So (3.3.1) becomes
is
= 7.5 +
(1.25)/(1.5)
= 8.33/kWh,
so the covariance
62
term is not negligible. However, if the derivative dk(t) had been assumed to be
a constant, independent of dk(t), then from (3.3.2), Cov (t) = 0 and
PkCt/r) = 7.5/kWh. 2
Specification of Price-Only: Effect of Period Definition
Equation (3.3. J) is fine as long as the price changes each hour, i.e. has a very
detailed period definition. However, billing period flat or TOU prices do not
vary each hour. Thus (3.3.1) has to be modified for many types of period
definitions.
The principles are easiest seen for a flat billing period update. This is not a
recommended type of transaction, since it loses most of the value of time varying
prices. However it is very common today, hence is an example of how to use spot
prices to calculate present day rates. Define
POat:
Constant price specified at beginning of month to hold for all hours of the month.
I
= 720
720
(3.3.3)
1=1
PHat
720
is computed
In general, both transactions costs and the benefits associated with meeting the
other criteria tend to increase as transactions become closer to continuous
hourly spot prices, i.e.:
o
o
o
o be
and
very
not
riod
ot a
ying
,pot
>nth.
the
.3.3)
). It
mth
iled
the
ous
A power system can theoretically be controlled through the use of prices or the
use of quantity control. Theoretically, the desired customer behavior could be
obtained by the use of quantity control wherein the utility directly controls all
electric energy usage based on the hourly spot price and customer-provided
information on how they value the services provided by electric energy.
However, under the conditions of the energy marketplace where there are a huge
number of customers with large diversity in usage patterns and needs, the use
of prices is far superior to a quantity control with respect to both reducing transactions costs and increasing benefits. There is, however, a potential role for
certain price-quantity transactions in the energy marketplace.
Price-quantity transactions can sometimes be used to reduce transactions
costs. For example, a 24-hour update price-only transaction has lower transactions costs than a one-hour update but is vulnerable to plant outages or errors
made in forecasting the weather. (If the customers respond to the wrong price
signal if a plant outage occurs or a major weather forecasting mistake is made
at the time the 24-hour update prices were computed and quoted.) The use of
3:i~~-hour update spot price combined with an interruptable contract (a type of
flfice-quantity transaction) enables corrections to be made for especially bad
weaJher forecasts oj. plant outages, with transactions costs that might be lower
t~n those of a one>-hour update price. Whether the transactions costs would
really be much lower is an empirical question. We are skeptical.
A one-hQur update price can be combined with an interruptible contract to
provide the~utility with a way of modifying demand on time intervals of less than
one hour.' This enables the customers to carry the operating reserve necessary
for power system security control, i.e., to respond satisfactorily to sudden loss
of a major generation plant or tie line support. If the customers are to carry the
operating reserve, the utility has to have fast and predictable customer response.
Characteristics of Price-Quantity
of
off,
, go
md,
Type of Pledge:
Fixed amount of energy reduction over some time interval
- Fixed amount of power level reduction
- All energy above a prespecified level
- Reduction of power to a prespecified level, or
where q is the probability the coal plant will be available as estimated using all
the information available at time t. If q = 0.95 then as earlier (assuming the
covariance term is zero), (3.3.1) yields p(tlt) = 5.25/kWh. However, if the
possible event that the coal plant is forced out is ignored, p(tlt) = 5jkWh.
Hence, if a customer agrees a priori to reduce demand if the coal plant forced
outage event occurs, that customer can be offered the lower rate of 5 jkWh. If
the coal plant turns out to be available at time t, the customer has saved
0.25 /k Wh for all of the energy that was purchased.
NUMERICAL EXAMPLE. Consider a situation where it costs the utility 0.5 jkWh
to maintain the necessary operating reserves. In this case, the utility could offer
a O.5jkWh discount on its regular energy rate to customers who are willing to
help carry the operating reserve by rapidly reducing load when a sudden,
unexpected loss of a major generator occurs.
In the above two examples, the price-quantity transaction's value was specified by the utility's costs. An alternate procedure is for the utility to auction off
price-quantity contracts so that customers themselves determine the prices. For
the operating reserve example, the utility might start out by offering a 0.1 jkWh
discount; then if not enough customers sign up to meet the utility's needs, the
utility could raise the discount to 0.2 jkWh. This process would continue until
, i.e.
sort
either the customers have pledged enough load to meet the utility'S needs or
until the offered discount has reached O.5/kWh, in which case the utility carries
the operating reserve on its own generators.
A Possible Trap
lues
ltrol
mly
ling
)nly
ansthe
1 of
eral
gl!en .....
~
all
the
the
Nh.
ced
1. If
ved
Wh
ffer
~ to
len,
ecioff
For
Wh
the
ntil
66
If Customers Uses
If Q(t) = 9/kWh,
the customer
If Q(t) = II /kWh,
the customer
IMWh
2MWh
OMWh
Pays $100
Pays $100
Pays $10
Pays $100
Pays $100 + 110
Receives $\0
90
when making such tradeoff studies. All three argue for the use of prices rather
than price quantity transactions.
First, although one of the motivations for the use of price-quantity was to
reduce transactions costs relative to price-only, it is possible to get so carried
away in designing fancy price-quantity transactions that the resulting transactions costs exceed those of price-only with a fast update cycle length.
Second, another motivation for the use of price-quantity was a desire to be
able to predict customer response more accurately. However, with certain types
of price-quantity transactions, such predictions become rather difficult. For
example, if the price-quantity transaction gives the utility the right to turn off
a customer's air conditioning system, it is necessary for the utility to develop a
model which gives the probability the air conditioning system is on as a function
of time of day, day of week, season of year, and of course, outside temperature.
In general, as the nature of the price-quantity transaction becomes more
sophisticated, it becomes more difficult to predict customer response.
Third, important criteria involve issues of customer understanding and acceptance. A price-only transaction is inherently easier for customers to understand
than most, if not all, price-quantity transactions. Also, a customer might
initially accept the idea of lower prices under price-quantity but then become
very disenchanted after the utility has exercised its control options a few times.
Price-quantity transactions can have a unique role in security control when
some or all of the operating reserve is carried by the customers.
SECTION 3.5. LONG-TERM CONTRACTS
One of the criteria listed in Section 3.1 for evaluating the properties of energy
marketplace transactions is the extent to which they facilitate customer operating and planning decisions. It might be very important for certain customers to
know what their future energy costs are going to be hours, days, months or years
in advance. This desire could be met by offering transactions with long update
cycle times, but this would result in an efficiency reduction. Fortunately there
is a type of long-term contract which enables customers to buy rights to future
energy at a fixed price (e.g. to buy an insurance policy) while still maintaining
the efficiency of short update cycle lengths ranging from hours to days. It is no
coincidence that the following discussion of long term contracts sounds like the
way agricultural forward contracts work. The underlying issues are very similar.
Wh,
10
ther
s to
ried
lns, be
pes
For
off
pa
ion
Ire.
ore
eptnd
ght
me
les.
len
Whether the risk hedging provided by electrical contracts is worth their transactions costs remains to be seen.
The basic approach is to offer fixed-price, fixed-quantity long-term contracts
for specific future time intervals. Consider a customer who has bought such a
contract. When the future time finally arrives, the customer can definitely have
the agreed amount of energy at the specified price, independent of the actual
spot price at that time. However, if the hourly spot price turns out to be above
the value in the customer's contract, the customer might choose to use less
energy than in the contract and effectively sell back his/her rights to the utility
at a profit. If the hourly spot price turns out to be much lower than that specified
in the contract, the customer has paid a penalty for having replaced uncertainty
with certainty.
NUMERICAL EXAMPLE. Assume that on January I, an industrial customer
purchases 1 mWh of energy to be delivered between 10 and II AM on July I for
10/kWh. Assume that when July 1st finally comes the price p(t) between 10 and
II AM is either 9 /kWh II /kWh. The actual cash flow depends on the
customer's actual usage on July 1st and is given in Table 3.5.1.
Futures Market
rgy
'at, to
ars
ate
ere
LIre
ing
no
the
ar.
68
Many types of transactions are possible in a spot price based energy marketplace. A key question is "Which transaction will a particular customer see?"
Three approaches are
o
If a customer has the option to choose among types of transactions, we recommend the following approach.
Specify a particular default price-only transaction for a particular class of
customers on the basis of trading off the transactions costs versus some measure
of the benefits evaluated for the class as a whole. Place all customers in this class
under the default transaction unless they choose to deal under a more sophisticated rate (such as one with a shorter update cycle time). More sophisticated
rates have higher transactions costs and customers who exercise this option have
to pay the additional costs over and above those of the default rate.
As an example of the above, consider a utility whose cost-benefit analysis
indicates industrial and commercial customers should see one- and 24-hour
update prices respectively while residential customers should, as a class, see
billing period update prices. Furthermore, assume that all residential customers
with air conditioners are put into a special class which pays higher monthly rates
because the covariance term of (3.3.1) is positive. Some of these air conditioner
customers might choose to see 24-hour update prices so they could reschedule
their air conditioning away from times of high prices and hence save money on
their electric bills (while also making things better for the utility). Such
customers would have to pay an additional charge to cover the extra communication and metering costs.
In a similar fashion, price-quantity and futures market transactions could be
made available as options for any customer willing to cover the additional
transactions cost.
Naturally the real world is complex enough that other, less simple, approach-
, the
, the
) one
rket;ee?"
;tion
lions
ored
om,s of
sure
:lass
listiated
lave
lysis
lOur
see
ners
'ates
)ner
dule
von
;uch
omd be
:mal
ach-
In some parts of the world, large industrial customer rates are individually
negotiated between the customer and the utility. Thus each customer sees rates
that are custom tailored to their needs, subject of course to utility cost constraints. Such custom tailoring is not part of most U.s.A. practices, but the
theory of spot pricing makes it a viable alternative with some desirable features.
Under the custom tailoring approach, a large industrial or commercial
customer specifies the type of transaction that best meets its needs. For example,
a customer might choose a 24-hour update with only three pricing periods
corresponding to the shift change times of the customer. As another example,
a customer might choose a price-quantity (interruptable) contract with warning
times, quantities, etc. that match the customer's own mode of operation. One
of the beauties of the overall spot pricing theory is that starting with basic hourly
spC,H prices, the utility can compute the value of a host of different types of
{f\nsactions in a consistent fashion. Thus even if each large industrial-comi:ftercial customer cijooses a different type of transaction, all of the transactions
v.onld be consistent and thus, hopefully, acceptable to the regulatory com~
.
.
rtnsslon.
With custom tailoring, it becomes more practical to include the various
'''covarianc; terms" that are part of the basic theory but that are difficult to
compute for a class of customers.
Custom tailoring implies a lot of overhead administrative expenses. Thus it
is difficult to envision its application to residential or small industrial-<::ommercia} customers.
SECTION 3.7. WHY NO DEMAND CHARGE?
Demand charges are presently an important source of revenue for many utilities.
However, demand charges have essentially no role in the energy marketplace. s
An explicit example of a demand charge is used to elaborate its disadvantages.
Define
rc: Flat rate energy charge ($/kWh)
Demand charge based on one-hour peak demand during billing period
I'd:
($jkW)
70
Billk = re
L dk(t)
I-I
+ rAmax dk(t)]
(3.7.1)
Define
Pequil(t): Time-varying energy prices which are equivalent to (3.7.1) in the
sense that
Billk = '1:.;:1 PCquil(t)dk(t)
Assume re = 5jkWh and rd = 5$jkW. The resulting Pequil(t) was plotted in
Figure 1.1.4 of Chapter I. It bears very little resemblance to say Figure 1.1.1 or
1.1.2. Hence demand charges yield price signals that do not look like the actual
hourly spot price.
Since the price signal sent by a demand charge bears little relation to hourly
spot prices, the criterion of economic efficiency is violated. Customers are not
motivated to adjust their usage patterns to match the utility's capabilities; i.e.,
A demand charge motivates customers to take actions which reduce their bills while
leading to little or no reduction in the utility's costs.
(3.7.1)
:val is
~s are
Demand charges are often used to recover capital costs, while energy charges are
used to recover fuel costs. This approach is not applicable in the energy marketplace, which employs rates based on marginal costs. Rates based on marginal
costs can lead to revenues which overrecover (as well as underrecover) relative
to embedded capital costs, rate of return, etc. Thus association of demand
charges with capital can result in negative demand charges - a concept which is
difficult to accept (at least by the authors of this book).
Attempts to Justify a Demand Charge
n the
~ed in
1.lor
lctual
One could try to relate demand charges to the covariance term of (3.3.1) and
argue that customers with a higher monthly one-hour peak usage also have a
demand behavior that yields a positive covariance term. However, we find little
justification for such an argument.
A demand charge can be justified in an energy marketplace if the size of a
customer's peak usage has an impact on the utility's distribution costs, e.g. if the
utility has to install a special transformer or distribution line. Of course such
distribution costs could also be incorporated into an energy charge via the
revenue reconciliation term, or treated as a fixed charge.
SECTION 3.8. RELATIONSHIP TO PRESENT-DAY TRANSACTIONS
oUrly
e not
;; i.e.,
while
Introl
hour
Such
peak
.mers
omer
Jas &
uring
a two
ityof
uring
year.
liance
. very
Section 3.7, we <:Iiscussed the role (or lack thereof) of demand charges. We
discuss relatiQnships between the energy marketplace and other types of
"tle'Sent-day transa~tions.
Table 3.8.1. Characteristics of Generic Present-Day Transactions
Type of Pre~nt-Day Transactions
Assumed Characteristics'
Energy Usage
. Flat
"TOU
Demand Charge
Block Rate
lnterruptable Contacts
Demand Limiter
" Fixed Limit
,. Variable Limit
c Interlock
Direct Appliance Control
,; Air Conditioners
;. Water Heating
Life-Line Rates
All values except fuel adjustment charges are assumed to be specified one year in advance.
We consider only the generic types of transactions listed in Table 3.8.1, which
gives their assumed names and characteristics. Table 3.8.1 leaves out many
interesting types of present-day transactions, but the scope is sufficient to
provide the reader with a good feel for the relationships between present-day
transactions and those of the energy marketplace. Note that unless the prices are
calculated using the formulas of this book, i.e. are based on hourly spot prices,
the rates are not consistent with our approach, even if the type of transaction
is similar.
Two particular existing innovative rates not included in Table 3.8.1 are rates
depending on the outside temperature or demand charges imposed at the time
of the utility's peak demand rather than the customer's peak. They can be
viewed as a step from the present-day transactions of Table 3.8.1 toward energy
marketplace transactions.
Energy Usage
If a fuel adjustment clause is not used, present-day TOU and flat rates are
price-only transactions with an update cycle interval of one year. A retrospective
(i.e. based on past fuel costs) fuel adjustment clause that is adjusted each billing
period is not the same as a billing period update cycle interval, because a
price-only transaction is based on predictions of future costs.
When a fuel adjustment clause is being used, we believe it is illogical to specify
one part of the rate a year in advance. The customers receive more accurate price
signals if the overall rate is on a billing period update cycle, and the fuel
adjustment clause is dropped as a separate item.
Average Spot Price Versus Flat Rate
Define
rAJ!:
Flat energy rate ($/kWh) that would recover the required revenue.
It follows that if
H760
1=1
then
rAa<
8760
8760
I-I
L d(t)
I~
L p(t)d(t)
vhich
nany
1t to
t-day
~s are
rices,
ction
rates
time
n be
lergy
; are
ctive
lIing
se a
~cify
)rice
fuel
customers with flat usage patterns (such as some industrial customers) will
spot prices since their bills will go down relative to the flat rate (as defined
above).
Block rates are nonlinear pricing schemes wherein the energy price per kWh
on the customer's total kWh usage.
The revenue reconciliation discussions of Section 8.7 show how nonlinear
price structures can arise from the theory of spot pricing, but the motivation for
present-day block rates does not appear to be based on the same logic.
Present-day block rates are sometimes justified by arguing that large
lOT/'fYI.pro should pay lower rates because they are fed at high voltage and
not pay the capital costs and losses of the lower voltage distribution
However, the use of customer classes based on voltage level of service
is a more desirable approach.
The covariance term of (3.3. J) can also be used as a way to try to define a
rate. For example, if the class of customers with high monthly energy
also tend to h~ve a negative covariance term, then a declining block rate
Unfortunately, monthly energy usage by itself is not a very good
of customer hourly usage patterns.
.. """1-'''11'';)
Interruptable Contracts
Most presedt-day interruptable contracts are special cases of an energy marketplace price-quantity transaction or are disguised price-only transactions.
However, most present-day interruptable contracts are prespecified a year in
advance, while energy marketplace transactions allow for much more frequent
updates.
Thinking in terms of an energy marketplace opens up the range of possible
interruptable contracts. With the energy marketplace framework, it becomes
possible to offer customers a flexible set of transactions. Thus customers can
match their needs and ability to respond. This flexibility can be extremely
important for industrial customers.
Demand limiters can have constant settings (e.g. fuses) or the limits can be
imposed under utility control in real time (e.g. Demand Subscription Service).
They are related in concept to demand charges, and many of the critical
discussions given on demand charges also apply here.
Real-time demand limits based on the utility's needs can be effective.
However, like demand charges, they have the disadvantage that as customers
become more sophisticated, they can install control hardware or modify their
behavior to reduce the plan's overall effectiveness.
It is extremely difficult to see how applicance interlocks lead to an appreciable
reduction in the utility's costs. At best, there is an extremely weak link.
The basic felicity underlying many demand-limiting schemes is the belief that
reduction in an individual's peak demand results in a reduction in the peak
demand seen by the utility. Diversity and the law of large numbers destroy this
belief. A general rule, which is true in almost but not all cases, is
o
Demand limiters reduce the utility's demand only if they reduce the energy
used by individual customers over a period of several hours.
A customer selling the utility the right to directly control individual appliances
is usually a special type of energy marketplace price-quantity transaction or a
price-only transaction in disguise. The key question is whether or not it is a
desirable type of transaction. The answer depends on the type of appliance. Air
conditioner cycling and water heater controls are discussed to illustrate the
issues.
Under air conditioner cycling, the utility can turn the air conditioner off if
deemed necessary. This has three major shortcomings. First, the utility is put in
the position of appearing to play big brother. The utility has no way of knowing
how important the air conditioning is to anyone customer at anyone time. Thus
control might be exercised at a particularly bad time (such as when the customer
is holding a very important dinner party); the result is a very unhappy customer.
Second, the utility has to build a model for the probability that the air conditioner is on as a function of time of day, week and season and of outside
temperature and humidity in order to predict what type of response to expect.
Third, most present-day air conditioning cycling contracts have a clause which
limits the number of times the utility can exercise control. This places a very
heavy burden on the utility system operator, who has to gamble on whether or
not to exercise control at any given time. Air conditioner cycling does not have
a role in an energy marketplace.
Direct utility control of water heaters is, however, quite different in character.
The large amount of thermal storage in most hot water tanks enables the design
of systems where the utility can exercise direct control (both on and oft), which
helps the utility reduce costs without causing much or even any customer
inconvenience.
Life-Line Rates
Life-line (or base-line) rates refer to the use of low, non-cost-based prices for
particular classes of low income or otherwise needy customers. The class is
usually distinguished by a low monthly use of electric energy.
ners
:heir
able
that
)eak
this
~rgy
lces
::>r a
is a
Air
the
ff if
tin
ling
hus
mer
ner.
:on;ide
ecL
lich
'ery
r or
ave
ter.
ign
ich
ner
for
; is
The use of such rates is socially motivated by the belief that today, enough
energy to meet certain basic needs should be made available to everyone,
loe'perloent of ability to pay. This could be done by using general tax reven ues.
rates are an alternate procedure wherein needy customers pay artificilow rates and the other customers make up the difference - i.e., a hidden tax.
Life-line rates can definitely be incorporated into an energy marketplace if so
Note that even customers who are on life-line rates should see a billing
update cycle. Nobody should be denied the right to achieve the benefits
matching their usage patterns to actual costs.
'T:'r'T ......,"" 3,9. CUSTOMER-OWNED GENERATION: AVOIDED COSTS
revenue reconciliation is ignored, hourly spot prices for the utility selling to
customer are the same as when the utility is buying electric energy generated
a customer (see Section 2.7). Revenue reconciliation can destroy the
lImmptr\l but does not change the basic price behavior. Thus most of the
discussions of this chapter also apply to the transformation of a basic
hourly spot price into a set of transactions which provide good
benefit tradeoffs relative to the criteria of Section 3.1.
price-only and long-term contract transactions discussions apply almost
for word to the buy-back case. The details of some of the price-quantity
"."'!l\t~""".v .. S may chanJ?;e, but the principles do not. For example, during hours
when a private cogenerator is generating electric energy below
because of a lew internal demand for steam, the cogenerator might sell
}~tility<the extra capacity for use in emergency situations.
fUV-t.J<t ......
Capacity Credits
the United~States, the PURPA legislation stated that buy-back rates should
based on avoided costs without clearly defining what avoided costs are.
spot prices provide such a definition.
In present-day transactions, avoided costs are often subdivided into an
.
energy cost and a capacity credit which is used to reflect impacts on the
ity's capital costs. In the energy marketplace, capacity credits are not used.
pacity credits can be viewed as being incorporated into either the hourly spot
gen~ration quality of supply component, or the revenue reconciliation
01111pOlnem. Use of capacity credits for customer generation is analogous to the
of demand charges for customer usage, and is not appropriate.
These discussions on buy-back rates have implicitly assumed that each incustomer-owned generator is small enough relative to the utility's total
tion that diversity arguments can be used. Thus they can be treated just
individual customer loads where diversity is critical; i.e., no individual load
customer generator) behavior has a noticable effect on the overall spot price.
individual customer-owned generator is too large, the conclusion no longer
(The definition of "too large" is very situation-specific.)
The basis of the argument is that if they enter or do not leave, the capital costs
of the generators are spread over a larger base and hence regular customers also
benefit.
Similar types of arguments can be applied to the case of a utility with
insufficient generating capacity, albeit the sign of rate differential changes.
Similar types of arguments are also applied to customer-owned generation.
Under relatively reasonable assumptions (as discussed in Section 9.4), spot
pricing theory provides a rational basis for establishing such special rates for
special customers. The theory of Section 9.4 combines revenue reconciliation
and long-term customer demand elasticities to show that such special rates are
self-consistent with the spot price based rates being discussed in this chapter.
The concept of charging special customers special rates is controversial, and
leads to lots of complications in actual implementation.
SECTION 3.11. WHEELING RATES
neraspot
'gued
I
ry, or
osing
eave.
costs
; also
with
rlges.
,n.
spot
s for
ltion
s are
)ter.
and
lity's
two
~ctric
s are
. the
f the
to be
tions
4 on
chapter has outlined the very rich menu of transactions that are possible
an hourly spot price based energy marketplace. Different transactions can
because they all have the same common basis, the hourly spot price. The
of which transactions to offer particular classes of customers is based on
between the cost of the transactions and the benefits of having transthat are close to the hourly spot prices.
One key conclusion is that there is symmetry between customer owned loads
customer owned generation. There is no reason to discriminate against one
the other by charging special rates. Also, there are good reasons to let
see hourly spot prices so that they can adjust their generation
to better help the utility.
An obvious question is, what should be the relative roles of price-only,
and fixed-price-fixed-quantity long-term contract transactions?
presently recommend major reliance on price-only transactions with pricetransactions playing only a support role to deal with particular needs
operating reserves. This recommendation applies today particularly to
and large commercial customers, and in the future to much of the
class as well. One reason for this recommendation is the complexity
by price-quantity transactions into the lives of both the customers
utility operators.
;1'-"'iiYP(1-nr,'l'p-fixed-quantity contracts playa different type of role than priceand price-quanttty, if they are applied over a long term such as months or
:cThey provided customers with risk hedging. Second in our opinion most
uantity transactions should be converted to price-only transactions
give customers the ability to change their mind, if they are willing to pay
penalte'. Final comparisons must await more field experience.
- of energy marketplace transactions with present-day transshows both similarities and fundamental differences. The price-only and
energy marketplace transactions can be viewed as the logical
VolilltiCln of present-day time-of-use (TOU) rates and interruptable contracts.
fixed-price-fixed-quantity long-term contracts/futures market seems to
no present-day counterpart. Present-day demand charges, capacity credits,
many demand limiting schemes have essentially no role in an hourly spot
based energy marketplace.
if only the very largest customers are actually on hourly spot prices, spot
principles can be used to set rates for other customers .
IST,ORICA.L NOTES AND REFERENCES - CHAPTER 3
chapter discussed in general terms the mapping of hourly spot prices into
marketplace transactions. Details and theory are presented in Chapter 9
II. Readers who want to pursue our ideas further should start with
9. There appears to be little literature on systematic rules for how often
78
to change prices over time, despite the vast literature on how to set the level of
time varying prices. Bohn [1982, Section 2.5] presents a simple model and
reviews the existing economics literature. A key insight is the difference between
prespecified changes (which can change once per period) and new price calculations, which occur once each update cycle. Other key issues are transactions
costs and the impact of self-selection, which are analyzed for TOU rates by
Acton and Mitchell [1980]. Various papers discussed in the Annotated Bibliography develop the framework used here and in Chapter 9.
A time-of-use (TOU) rate is a special type of price only transaction as
discussed in Section 3.3. Hence, this is as good a place as any to review this TOU
literature. TOU rates can have many periods, but are updated only annually or
irregularly. They are an important intermediate rate between flat prices and spot
prices.
The desirability of TOU rates has been the topic of major research by the
Electric Power Research Institute [1979]. For a good summary of this effort and
discussions of associated problems, see MaIko and Faruqui [1980] and Faruqui
and Maiko [198Ia]. For a good review of U.S. Department of Energy sponsored
residential TOU experiments, see Faruqui and MaIko [I98Jb].
The idea of time-differentiated prices goes back at least to Boiteux [1949] (see
also Vickrey [1955] and Steiner [1957]). Until Brown and Johnson (1969] the
models were purely static and deterministic. During the 1970s various authors
presented prescriptions for TOU pricing in static models with demand uncertainty.
The "standard" TOU pricing models are surveyed in Gellerson and Grosskopf (1980] and Crew and Kleindorfer (1979]. They include Wenders (1976],
Crew and Kleindorfer [1976, 1979, Ch. 4 and 5], Turvey and Anderson [1977,
Ch. 14], and various predecessors. These models include multiple types of
generators and stochastic demand. Some of the limitations of these models are
as follows:
o
vel of
I and
tween
Icula:tions
es by
iblio-
)n
as
TOU
lIyor
I spot
'y the
t and
ruqui
sored
,] (see
I] the
thors
ncer-
ross976],
1977,
~s of
s are
mply
. and
:d. It
: and
hese
I.
By
they
lily a
ck of
;tock
tock,
oited
practical value. In fact long-run marginal costs can only be calculated con:ditional on a particular scenario or probability distribution of demand and
factor prices. This problem is addressed by Ellis [1981].
The models assume that demands and generating costs are independent from
one hour to another. This is very convenient, since it allows the use of a single
load duration curve or price duration curve. Nonetheless, the availability of
storage [Nguyen, 1976] startup/shutdown costs, etc., and demand rescheduling can have a major impact on optimal investment policies.
The models ignore transmission, which is equivalent to assuming an infinitely
strong transmission system. These models give no insight into how to price
over space.
The models do not use the device of state contingent prices. Therefore, the
investment conditions derived in the models are hard to interpret, although
they are correct (given the limiting assumptions above). For example, Crew
and Kleindorfer [1979, p.77] interpret their results only for the case of
interchanging units which are adjacent in the loading order. Littlechild [1972]
showed the way out of this problem, but his point was apparently missed by
subsequent authors.
Several authors present deterministic, explicitly dynamic models which can be
as deterministic versions of price-only transactions combined with
decisions. Crew and Kleindorfer [1979, Ch. 7] give a continuous time
control model with one type of capital. They get the result that
level of capacity, price is to be set to maximize instantaneous [short run)
returns subject to the given capacity restriction [po 113]. That is, price should
SRMC. 9f course, at optimum capital stock is adjusted so as to equate SRMC and
C .... In the event of ... a fall in demand, [optimal] price is less than LRMC, then
ty would be allowed to decline until equality between price and LRMC were
Vh"!:,t"'\I"r'th",
are thinking here on a time scale of years, not hours; they reject continuous
of prices to reflect the actual level of demand. Nonetheless, their
can be interpreted in terms of hourly price adjustments.
Turvey and Anderson [1978, Ch.17] have a discrete time dynamic model
leads to discontinuous prices, as capital investment is made in lumps.
....""'pvpr, they reject this approach: "It is apparent that, for one reason or
such fluctuations are unacceptable." They also acknowledge that
Imles1tIDIem decisions must be made before price decisions, and with more
about future demands, but they do not incorporate this into their
[p.305].
Ellis [1981] explicitly models sequential investment and pricing decisions. He
""U\o,IU'.1"" that" ... welfare optimal pricing rules differ according to whether
prices must be set either before or after investment decisions are made" [p.2].
80
He uses dynamic programming to look at how the character of optimal sequential investment depends on capital stock irreversibility and the sequential revelation of information about future demands.
Price-quantity transactions as discussed in Section 3.4 can be viewed as
procedures which allows customers to choose their own reliability levels.
Marchand [1974] has a model in which customers select and pay for different
reliability. The utility allocates shortages accordingly, when curtailment is
necessary. His approach differs from (and is, except for transactions costs,
inferior to) spot pricing because customers must contract in advance, and
therefore have no real time control over their level of service. Also, customers
not curtailed by the utility have no incentive to adjust demands.
Marchand's proposal has been greatly extended under the name "priority
service". [Oren et ai, 1986; Chao and Wilson, 1987] Although discussed as an
alternative to spot pricing, priority service in fact addresses a very different set
of issues. Customers select priority levels; higher priority customers pay more,
but are cut off later in the event of supply interruptions. Such a scheme is
obviously not designed to deal with cyclic and hourly changes in the system
lambda, but instead to deal with occasional situations where the generation
quality of supply component becomes positive and there is not enough generation capacity available to meet the demand. The existing literature on priority
service also seems oriented toward residential customers. The implicit but
unquantified assumption is that metering and other transactions costs will be
high relative to energy used. We do not attempt a detailed analysis of priority
service in this book. Since it is a form of price-quantity transactions it has the
virtues and defects of that class.
NOTES
I. Many present-day rates have the appearance of being updated yearly but are not because of the
presence of fuel adjustment clauses, which changes upredictably every quarter or month, and in
many cases are retroactive.
2. Unfortunately, a good model for the derivative of demand with respect to price is not available.
This problem will raise its head many times throughout this book. Conventional rates avoid this
problem by ignoring the covariance issue, causing cross-subsidies and inefficiencies.
3. The use of onc hour as the basic time unit in this book is not essential. For example, it is possible
to use a five-minute update price if desired. However the update cycle length cannot be reduced
too much as then the power system dynamics start to become important and more advanced
pricing theory is required. Berger [1983] discusses some of the issues which arise when system
dynamics come into play. A special case arises when customers receive incentives to install devices
which temporarily control certain of their loads as a function of system frequency. This concept,
which makes use of a FAPER (Frequency Adaptive Power Energy Rescheduler), is discussed in
Appendix F.
4. An implicit underlying assumption is that all transactions are based on hourly spot prices; i.e. are
consistent. A choice between an hourly spot price and a flat rate computed on a different basis
should not be allowed.
5. Recent historical research by Hausman and Neufeld [1984J points out that a strong advocacy of
marginal cost based time-dependent electricity rates existed in the United States as early as the
1882-1915 period when a wide-ranging debate over electricity rates took place. They suggest that
the final adoption of "demand charges" instead of time-of-use rates can be found in the
"economic conditions faced by electric utilities in their early days and in the nature of rate
regulation" [Hausman and Neufeld, 1984, p. 125J.
6. Demand charges are often computed on a 15- or 30-minute basis. This increases the problem.
quenevelaed as
evels.
ferent
~nt is
costs,
, and
)mers
iority
as an
n.t set
nore,
ne is
'stem
ation
neraority
t but
ill be
ority
s the
IMPLEMENT AnON
first two steps leading to an energy marketplace are to evaluate the behaY'iRUJ: of the hourly spot prices (as discussed in Chapter 2) and to determine the
ty~es of,transactions'to be offered (as discussed in Chapter 3). The third step is
the actual implementation, which is the subject of the present chapter.
The discussions are divided into four general areas:
.~
of the
md in
ilable.
ld this
o
o
<l
.ssible
duced
anced
ystem
evices
lcept,
sed in
e. are
basis
ICY of
IS the
t that
n the
The exact pattern is as follows. Sections 4.1 and 4.2 discuss operation and
planning for the energy marketplace, while Sections 4.3 and 4.4 discuss operation and planning issues relative to the customer. Section 4.5 deviates from the
pattern and discusses some of the practical issues associated with calculating an
hourly spot price. Sections 4.6 and 4.7 then discuss operation and planning for
r rate
81
82
Number
Periods
Number
Levels
Price-Only
Flat Rate
Peak-Offpeak Rates
Month
Month
Day
Day
Infinite
Day
24
Day
24
Infinite
Hour
Hour
Inllnite
the utility. The chapter concludes with section 4.8, which discusses operation
and planning issues faced by a regulatory commission.
Note that this chapter is, deliberately, quite comprehensive. For example, it
talks about some types of transactions which we personally doubt will be used
by many utilities. In contrast, most utilities will start with a simple, price-only
hourly spot price or a 24 hour update price. This will be tried with a few large
customers who volunteer for the rate. The experience of implementing these
transactions will give indications about which way to go with broader application. Although in principle one can argue for starting with a clean slate,
regulatory tradition and the inherent uncertainties of trying something new
suggest that this is unlikely, at least in the U.S.
SECTION 4.1. ENERGY MARKETPLACE: OPERATION
The operation of an energy marketplace is characterized by the types of transactions offered and by the communication and metering-billing systems used to
implement them.
Types of Transactions
Price-Only: An energy rate that is quoted in advance. Customers may use all
that they desire at that rate.
Price-Quantity: Customers agree to adjust their usage in some pattern on a
signal from the utility.
4. Implementation
)Ius
,e
83
:lg
days
's
day
19 peak
j
s
hour
ration
pIe, it
. used
:-only
large
these
plicaslate,
: new
.nsaced to
ns:
Ise all
on a
duality in Table 4.1.1 does not imply similar customer response, acceptance,
One aspect of price-quantity transactions which is not found in their pricearts is the issue of what happens if the customer does not fulfill the
quantity change when so requested by the utility. Three possible
:,,,%
~,
84
Existing Systems
Mail
o Newspaper
o Commercial Radio/TV Programs
o Telephone Calls
- Customer-instigated
- Utility-instigated
- Digital data, synthesized voice, or person to person
o
Special Systems
u
Table 4.1.2 lists some of the different types of communication media that could
be used. The Existing Systems of Table 4.1.2 require no special hardware while
the Special Systems do. Communication can be either one-way or two-way.
4. Implementation
85
Day
Hour
x
x
'"''~'''~'''
ommu,nic:ati(m to Customers
For price-only transactions, the utility has to monitor the customer's usage and
then compute a monthly bill (assuming the billing period is one month) given
720
Pk(t) dk(t)
k =
1 ... 720
{ = I
could
while
-way.
86
Store the energy usage for each hour of the month (720 storage registers)
which are then retrieved by a meter reader and returned to a central utility
computer to calculate the monthly bill.
a Communicate the hourly price to the meter, which has one storage register
that accumulates the bill (product of price times demand); retrieved by the
meter reader once a month.
There are many possible variations between these extremes. For example,
transactions Types 3, 5 and 7 of Table 4.1.1 can be implemented by a meter
which has two storage registers and simply receives a binary signal by some
special communication system. If a two-way electronic communication system
exists, the need for a meter reader can be eliminated.
o
Discussions thus far have assumed that the energy marketplace communications
and metering-billing functions are to be implemented as a stand-alone system.
However, in practice they could be integrated with other types of functions.
Many utilities are considering distribution automation and control whereby
electronic communication systems are established on the distribution system for
monitoring flows, controlling switches and setting relays, etc. If such a communication system reaches the customer's meter, it could conceivably also be
used for the energy marketplace transactions.
There is a major revolution brewing in establishing electronic communication
between an individual residence and the outside world. The functions being
considered include
o
o
o
a
a
a
Such applications can have a major effect on how energy marketplace transac-
4. Implementation 87
.ters)
tility
are implemented in the residential sector. In many cases, the same com'on medium and terminal equipment can be shared.
'IDlllll\;ct
~ister
{ the
nple,
neter
lome
stem
''''\'"'~t''r ...... contracts place no extra burden on the real-time communicationmetering and billing systems (unless, of course, the length of the contract is only
a few hours or days). As with the bidding for price-quantity contracts, there
many possible ways the market mechanisms can be developed. For example,
I~ ~ 'f6~'''' contracts could be offered that
ated
tible
uces
lOne
lica-
take
ions
tem.
1S.
reby
for
om) be
I
tion
~ing
om-
sac-
88
Demand
(MWH)
r-------------------....,
Peak
Valley
24
Hours
If Utility's Situation Is
A: Generation Capacity Shortage
B: Generation Fuel Shortage
C: Fuel Cost Varies Widely with Demand Level
D: Excess Base Load Generation Capacity
E: Excess All Type Generation Capacity
Then Optimum Load Modification is to
Move Peak
Move Valley
Down
Up
A, B, C
C, D, E
4. Implementation 89
."t,"\mpr"
include
hourly spot price, Pk(t), is the spot price for customer k during hour t. In
!)rac:uce, the k index usually refers to customer classes and rarely to individual
(except possibly for some very large industrials). For example, a given
cover all residential customers in some geographic region. For commerand industrial customers, the index k may refer to voltage level of service,
again with possible geographic dependence .
. definition of these customer classes influences the degree of sophistication
in calculating the hourly spot price (see Section 4.5).
We Get There From Here?
lIar
lual
ons
o
red
ner
ing
der
ted
the
to
90
Control and decision making devices-logics that customers can implement to'
exploit the potential of the energy marketplace are now discussed.
Approaches to Control and Decision Making
Customer response to energy marketplace signals can be divided into two levels:
o
4. Implementation 91
finished or finished p r o d u c t s . ) ;
Working Hours: Shifting work hours or assigning workers to other tasks for'!
a few hours
"
Product storage is a very important type of energy storage which is sometimes]!
overlooked. Shift rescheduling is energy storge that is like product storage, butJ
it is usually much less c o s t - e f f e c t i v e . , ;
o
Ignoring the Variations: Customer finds that the normal variations are insufficient to justify any response.
Treating Them as Prespecified Time-of-Use: Customer responds to normal
variations only in an expected value sense, where expectation is over months
and makes no attempt to respond to the day-to-day variations.
Real-Time Response: Customer responds in real time to both changes in the
forecasts of future spot prices and the}r actual variations.
The customers who respond in real time to normal spot price variations often
view their tactical control as being divided into two parts.
o
4. Implementation 93
lsks
ge,
spot
'oducj
)mers,
jigital
costs
etitive
lmanj such
cause
One basic premise of the spot price based energy marketplace is that the utility
not act like Big Brother and take over customer decision-making prerogaHowever, a utility can provide tactical control hardware and services for
the customers (particularly residential and small commercia\). For example, the
utility could offer to provide digital computers, switches and internal communication systems which will respond to hourly spot prices and other inputs as a
tactical control service. This does not constitute utility Big Brother action
provided the customer specifies the strategic decision-making logics to be used
by the hardware. The utility-provided system simply implements the customer's
desires. The customer would probably want to keep override power realizing
that each override costs money.
A Residential Example
Help Customer Understand: The customer learns how much electrial enel
is used for the various services it provides and develops an understanding
cost per service associated with use of electric energy.
Learn Customer's Desires: The customer states desired service levels (e.,
ideal temperature of house) as well as cost versus service tradeoffs (e.g., he
many dollars savings are required before they will accept deviations from tl
ideal temperature).
Warn Customer: The customer is warned when the spot price is high enoug
to cause a cost per service that exceeds a customer's prespecified critical valul
e.g., when it is going to cost more than 50 to run the dishwasher. If tb
customer desires, use of specific appliances (e.g., dishwasher, washin
machine, dryer) is automatically inhibited when cost per service gets too higl
Provide Routine Control: Space conditioning, thermal storage, water heating
refrigerator-freezer defrosting, swimming pool pumping and heating, etc. arl
routinely optimized by combining forecasted future spot prices and weatheJ
conditions with models for house thermodynamics, hot water usage patterns
etc., using the customer's desires and cost-service tradeoffs.
Provide Special Control: Implement special customer requests with minimum
costs such as have living-dining room at nOF from 7-11 PM for a party;
having swimming pool clean and heated next Saturday afternoon; run dishwasher and washing machine before 6.30 AM tomorrow; etc.
Suggest Heroic Control: Help the customer decide what to do when faced
with very high spot prices due to quality of supply components. Can be done
beforehand with a simulated high-price scenario (weeks to months) or in
response to an actual high price.
Develop Mathematical Models: Use observed behavior to estimate mathematical models for house thermodynamics, hot water usage patterns, etc. for use
in control logics and to help customers understand how they use electric
energy.
Communication with Utility: Inform the utility of the customer's desires,
control logics, etc., so utility can better forecast the impact of spot price
changes on aggregate customer response.
The existence of the energy marketplace can cause the residential customer to
purchase new appliances, etc., that are better able to respond. Examples are
4. Implementation
95
thermal storage and dual firing (electric and natural gas) space conAs time goes by, appliance manufacturers start to produce appliances
to be able to exploit time-varying prices.
Indlustrial Example
"
iesires,
t price
customers were asked what type of electricity rates they would like to see,
answer would usually be that the bill should be low and have no uncertainty,
the rates should be simple. If the desire to minimize uncertainty and
JIJllfJ""''')' were the only criteria, customers would rarely choose to go on a
price-only transaction. However, the potential of a reduction in the bill
increasing the difference between the benefits received and the bill) can
. very strong motivation to choose a more sophisticated transaction.
Relative to the uncertainty issue, many customers are initially appalled when
think about the possibility of having to face very high prices (e.g. 50 cents
one dollar per kilowatt hour) at some times. However, after consideration,
cus;tolnelrs realize that their prime concern with uncertainty is not the hourly
price itself but rather the uncertainty in their monthly and yearly electric
bills.7 A highly fluctuating hourly spot price (due to both normal and
ty of supply components) can result in monthly and yearly energy bills
are quite predictable because of time-averaging effects. In fact, customers
use their capability to respond to reduce the uncertainty in the energy bills.
96
Customers who appreciate the potential of the energy marketplace will VV .....V.,"
wide variations in normal spot price behavior, and the times when the q
of supply components become very large. Variations provide the customers
greatest opportunity to save money and reduce their monthly bills
increase the benefits they receive from electric energy.
A Necessary Condition for Capital Investment
Customers will usually find it much more advantageous to make capital investments to exploit the energy marketplace if they are constructing a new plant or
building (or are already planning a major retrofit). However, the key issues are
essentiaIIy the same whether new construction or retrofitting is considered, and
hence the foIIowing discussions apply to both.
Examples of possible capital investment's are
o
o
o
4. Implementation 97
value of the spot price at time t, given the information available at time r,
)'Qs(t)
t rJu(t)
+ rJQsJ(tlr) + )'R(t) +
rJR.k(t)
(4.5.1)
System lambda has been defined as the partial derivative of generation fuel and
Marginal
Costs
($/kWh)
maintenance with respect to demand at hour t. There are various ways it can be
implemented.
In the best of all possible worlds, lambda would be obtained as one of the
direct outputs of a unit commitment logic which automatically takes care of all
multiple time period dependence associated with unit startup and shutdown
costs, hydro storage and dispatch, purchases and sales with other utilities, etc.
Unfortunately, in many real-world applic~tions, life will not be so easy since the
existing unit commitment logics may not be appropriately defined and in fact
may not even be computerized. Therefore other approaches, which are not as
elegant, will be discussed.
There is an incremental-decremental approach, wherein the demand in the
existing system dispatch logics is varied at a given hour t to see how the total
costs at hour t change. Because of the shapes of the heat-rate curves, unit
shutdown costs, etc., a plot of the change in costs with respect to size of
increment or decrement of demand will often not be a smooth or even monotonic function as illustrated by Figure 4.5.1. A reasonable approach is to use a range
of incremental and decremental costs and do some averaging to obtain the
quantity to be called A(t).
A planning type production cost model (deterministic or probabilistic) can be
used instead of an on-line program in the control center. This production cost
model could be used, for example, once a day to generate a A vs. demand curve
for use during that day's operation. The inputs to this production cost model
are, of course, based on the generators available that day, the demand characteristics predicted for that day, and predicted purchases and sales from other
4. Implementation
99
The separate model approach has many advantages during the initial
phases of an energy marketplace.
Many control centers have a computerized economic dispatch logic which
tes a quantity called system lambda every five to ten minutes. This
does not include multiple-period effects but is still a reasonable quantity
use. However, care is needed since the lambda generated by an existing
program may not include, for example, the cost of gas turbines and/or
effects of purchases and sales. Also such economic dispatch codes are usually
set up to predict lambda 24 hours in advance.
Many control centers engage in purchases and sales each hour. Examples of
sales structures are the Florida and California Power Broker
which develop a frequently updated price quote that drives interutility
exchange. This quote could be used as the needed lambda in some cases.
In all cases, the relationship between the computed lambda and the loss
:onl0(me:nt of the hourly spot price has to be considered. For example, a
)Urcn:ise or sale quote to another utility may include losses and so is not the true
at the swing generator.
The conclusion of the above discussion is that there are various ways to get
the quantity we call lambda. The choice for any particular implementation is
dependent on the explicit capability of the existing central control
Usually an adequate approximation is readily calculable .
,n"~"''''nlr''tion
eutl~'atic)n
oton-
be
cost
:urve
IOdel
arac)ther
10
A second approach is to remember that the use of the LOLP measure is really!
an arbitrary (albeit reasonable) choice. Simply to define the generation qualit~
of supply component to be a function of operating reserve margin, where thi~'
function is zero unless the operating reserve margin is below a certain level after,
which the generation quality of supply component rises smoothly.9 This genera"
tion reserve margin could be defined from outputs from the unit commitment
dispatch logics, or could be determined simply as some deterministic functionl
of demand and available generation.
An example of this second approach is to use (2.4.1), i.e.,
YQs(/)
a;,
(1,,(/)
A Qs .;. -'---
(4.5.2)
G;,
H760
L (1;,(/)
(r
with
k[g(/) -
[gcri,.;.(/) -
6gJJ
g(t)
> gcril.;,(t) - 6g
otherwise
If desired, one could view the ar (t) of (4.5.2) as an approximate LOLP, (/). Note
the relationship between the YQs(t) of (4.5.2) and the aggregate I'/Qs(t) of (2.5.2)
(2.5.3). Initially most utilities will want only a crude model.
Network Losses:
'1L,k (t)
If an on-line load flow is available covering all buses, it can be used to obtain
line flows and their partial derivatives with respect to bus injections, thus
enabling the quantification of the network loss components I'/L,k' However, such
an on-line load flow for all transmission-level buses is not always available and
is rarely if ever available (today) at the distribution network level. Hence
aggregate network models of some type are required. The level of detail needed
depends heavily on how the customer classes (i.e., the k index) have been
defined.
Total line losses L for the DC load flow approximation presented in Appendix
D can be expressed as a quadratic function of bus injections y by
L = Lo + y' By.IO The dimension of the bus injection vector y and the corresponding B-matrix can be reduced by aggregating several or many buses
together, by assuming the demand (generation) at each individual bus is a fixed
percentage of the sum of demands (generation) at all of the buses to be
combined. Present-day economic dispatch is often done by use of a B matrix
with each generation bus incorporated, but all demand aggregated into one bus,
A two-bus equivalent can be obtained by aggregating all generation into one
4. Implementation 101
and all load into a second bus (Le., no customer class distinction) to yield
loss model such as
(4.5.3)
a,d(t)
+ Qsd(t)g(t)
+ a 2 d 2 (t) +
+ Q6 S (t)
a3g(t)
+ Q4i(t)
(4.5.4)
then to estimate values for the ao, ai' ... coefficients by using regression
. on whatever historical data or load flow planning studies are available.
a structural form such as (4.5.4), it is easy to compute its derivative with
to demand" and hence the network loss component of the hourly spot
Quality of Supply: '1QS.k(t)
Thus far the calculation of the components of Pk(t) have been discussed
assuming information at hour t is available. For actual implementation, it is also
necessary to calculate the best guesses of the value at hour t given information
available at hour r, t > r. Thus it is necessary to compute Jc(tlr) = E{A(t)lr},
etc.
A unit commitment study done at time r will yield Jc{tlr) directly. If other
procedures are used, best guesses of generation availability and load forecasts
(see Section 4.6) can be used. Such best guesses can also be used to yield the loss
and quality of supply component guesses.
If generation quality of supply is computed using a loss of load probability
type forecast as in (2.4.1), it is theoretically necessary to incorporate, probabilistically, the effects of possible generation outages between time r and hour t.
However, a rigorous treatment of such effects in probably not justified, especially in initial implementation. A heuristic approach is to use (4.5.2) where ~g
increases as t - r increases, but even such modifications may not be justifiable
in practice.
Covariance Term
4. Implementation
Time
----
Load
Model/Forecast
Economic Security
Functions
Evaluate
Spot Price
Based Transactions
Calculate
Hourly
Spot Prices
103
System
Dispatch
f-
short-term operational needs, the utility needs a load model which can be
to forec~st the hour-by-hour demands for the next day to one week as a
of time of day, expected weather patterns, and expected prices. Presentload models incorporate only weather and time dependence. The modeling
price effects cannot ignore the impact of rescheduling of demand; in other
the demand at hour t depends on both the price at hour t and the prices
demands at other times before and after the hour t.
,', The development of such a price-dependent load model could appear to
a major obstacl~ to the implementation of the energy marketplace,
the necessary data to specify and develop such a model is simply not
ble today. However, this data availability problem will resolve itself
provided the energy marketplace is introduced in a gradual fashion.
particular, for the first few customers seeing spot prices it will not be
to build a price-dependent load model into the utility operations,
I)CI~i:1ll:'C the response of just a few customers will not be sufficient to influence
the spot price significantly. As the degree of penetration of spot price based rates
increases, the need for an approximate price-dependent load model will become
real, but by then sufficient data will have been obtained to develop it. As the
penetration of spot price based transactions increases further, more accurate
load models will be needed but the necessary data to develop them will have
become available. A related phenomenon with similar consequences is that
given customer's response level will increase with time, therefore leaving
time to gather data. Thus the availability of data to develop the needed
model will occur automatically provided energy marketplace transactions
introduced gradually.
Given the availability of data, it is a nontrivial but relatively
exercise to do the statistical manipulation needed to obtain a
load model.
Economic Security Functions
Given the Pk(tlr) (as discussed in Section 4.5), it is necessary to compute the
values of the spot price based transactions.
4. Implementation
105
p(8It)
p(9It)]
Op~rator
power systems are operated by highly trained human beings who use
computers tOl)fovide the needed information in a usable form. The price
of Figure 4.6.1 has two different effects on the system operators:
It makes their lives more difficult by introducing a new set of numbers (e.g.
prices) wlilich they have to handle
It makes their lives easier by giving them new control capabilities which
reduce the impact of uncertainty and reduce (eventually eliminate) the trauma
of having to resort to rotating blackouts or other unpleasant control actions
more
be
ill
sively
n the
e the
The discussions thus far have tacitly assumed the implementation is being done
for a single utility which has its own central dispatch and control system. If the
utility is part of a centrally dispatched power pool involving other utilities,
additional implementation issues arise. The actual problems to be solved depend
on the specific nature of the pool. However, some generic issues can at least be
listed.
The actual operating costs of a given utility within a pool may be determined
by the own-load dispatch logics (see Appendix B), combined with formulas
which distribute savings of pool operations among pool members.
i06
Weather
Load
Model/Forecast
Time
Supply
Models
r'""""'""
Prices
Financial
Model
f-
The pool lambda may require correction for losses to translate it into a given
utility lambda.
Network and generation quality of supply components may be defined differently for different members of the pool.
Revenue reconiliation components will definitely be different for different
members of the pool.
It should be emphasized that a utility's membership in a power pool does not
prevent it from implementing a spot based energy marketplace even if the other
pool members do not.
SECTION 4.7. UTILITY: PLANNING
For planning, the utility needs a load model which can be used to forecast future
4. Implementation
107
Models
Financial Models
Many aspects of present-day financial models apply directly to an energy
marketplace utility. Additions to existing programs will be required to predict
how future spot prices will behave given a particular planning scenario.
SECTION 4.8. REGULATORY COMMISSION: OPERATION AND PLANNING
Jture
108
4. Implementation
109
ases
. The ideas underlying Table 4.1.1 are taken from Flory [1984] which contains a much more
complete development.
If operating reserves are to be carried by the customer's load, much shorter time intervals are
required, but the ideas extend in a fairly straightforward fashion.
Robert Peddie, originator of the CALMU system and former Chairman of the Southeast
Distribution Board in England, taught us the importance of this point of view.
Any industria!' plant or commercial building which already has a PC and modem in its control
room can implement such a communications system with no hardware cost. Such customers
usually already have recording meters to take care of billing communications.
We are not considering the case where the network quality of supply is negative.
Turning ofT the TV does not reduce usage very much but is a way to get the kids away from the
tube.
Fuel adjustment clauses cause today's customers uncertainty in their annual energy bills.
Present-day control centers (see Appendix B) are very sophisticated systems, designed to improve
the economic efficiency of generating electric power (subject to security constraints). However,
their implementation also involves some arbitrary choices and the use of approximation. For
example, the fuel cost of a coal-fired plant that buys from various sources (at different prices)
is not uniquely specified and the mUltiple-period time couplings (e.g., unit commitment and
maintenance scheduling) are rarely if ever handled in the theoretically optimum fashion.
This approach is used in many present-day direct load control and interruptible contract
implementations.
110
10. Lil accounts for losses such as in transformers which occur even if there is no demand.
II. Once a class of customers had been on spot pricing. the utility could examine the data to see
whether the covariance was indeed small.
.
12. Some present-day load models have become very complex because of a requirement to incor-.
porate the effects of particular types of price-quantity transactions.
13. Utility planners can, after all, initially assume little or no response to spot prices. This will give
them a conservative, worst case, plan since any response to spot prices will be a pleasant surprise.
They can also work directly with large customers to see that new investments are planned to
enhance response.
14. It is important to realize that a small number of customers account for the majority of electricity
use. Exact data is hard to get, but Bohn [1982 page 326] estimates for example that in two
utilities 0.1 % of customers account for about 30% of energy used. These customers are logical
candidates for spot pricing since their energy bills are large, they already have good metering.
and are quite sophisticated (i.e. profit maximizing) in their choices about energy use.
112
Regulated
T&D
Company
Market Coordinator
-"
------- r-------
~-~
'"
Information
Consultants
~
Deregulated
Independent
Entities
Generators
f-
<:;:::=::>
Energy Brokers
-'7
f-
Users
Section 5.1 discusses the different elements of Figure 5.1.1. Sections 5.2 and 5.3
discuss how the deregulated marketplace of Figure 5.1.1 fulfills the basic
functions of an electric power system that are summarized in Table 5.1.1 (See
Appendices A, B, and C for background on the functions of Table 5.1.1.).
Section 5.4 concludes the chapter with a scenario that might lead to Figure 5.1.1.
SECTION 5.1. A DEREGULATED ENERGY MARKETPLACE
113
and operating the system and the marketplace. Its specific duties
on and
lace.
o the T
:>f each
~
T and
lnd 5.3
: basic
.1 (See
5.1.1.).
e5.1.1.
,Generation represents the bulk of the assets in most power systems. In the
o
o
Are barred by antitrust laws from explicitly cooperating with other generating
companies in their areas, or owning too many units in one region 2
Are motivated by profit
Are forced by competition to act in socially beneficial ways
From a functional point of view, the electric energy users of Figure 5.1.1 behave
in much the same way, independent of whether the energy marketplace is
regulated or deregulated. Large industrial--<:ommercial and sophisticated residential customer see one-hour and 24-hour update spot prices while small
residential customers see billing period updates. However, the deregulated
energy marketplace employs a significant number of price-quantity transactions
involving short-term (seconds to minutes) transactions required for emergency
state control (e.g. to carry operating reserve on the load). Such short-term
actions are also desirable in a regulated marketplace but are not as essential.
Although going from a regulated to a deregulated energy marketplace has
little functional impact on the users, the hope of those who advocate deregulation is that the prices themselves will be lower (in the long run).
Information Consultants and Energy Brokers
The structure of Figure 5. I. I. combines the bulk transmission and the local
distribution into one company. An alternative structure which has both advantages and disadvantages involves a single bulk transmission company and many
individual local distribution companies, all of whom are regulated. We will only
discuss the combined T and D company since it is somewhat simpler, but almost
all of the ideas apply equally to the decomposed structure.
<:>'<'1''1'1,''''''
U5
the deregulated world of Figure 5. I.l, the generators are not centrally dis. Instead the Market Coordinator of Figure 5.1.1 sends each generator
pot price and each generator self-dispatches itself by generating if the spot
paid for electric energy exceeds the plant's marginal operating costs. A
ve reader might say, "Such generator self-dispatch and central utility
are theoretically equivalent." Such a reader would be right.
The Market Coordinator of Figure 5.1.1 keeps supply and demand in balance
continuously adjusting the spot price. On nights when demand is minimal,
spot price declines until only generating units with low operating costs
ain on-line (e.g., nuclear or wind-powered units and perhaps some coal-fired
. As demand increases during the morning, the spot price may rise to the
where owners of electricity storage units, which had purchased and stored
during the night, begin selling back energy.
On the demand side of the market, users reschedule their electricity-intensive
to times of low spot prices, and generally repond to spot prices to
their net benefits from electric power usage.
Ifil sudden tempOra'ry outage of large generating units occurs, the spot price
.
increases.,Generating units already on-line increase their output to
mum. Users' 'process-control computers automatically delay the on
of air conditioners, furnaces, heaters, pumps and similar equipment.
generators bring their units on-line. Owners of reservoir hydro units open
sluices to,increase production. If the outage is severe, the regulated T and
company exercises some of its operating reserve price-quantity contracts for
first few minutes. These measures gradually reduce the spot price, which
to decline as more units come on-line. If the spot price remains quite
some customers close down portions of their operations to save money.
y, equilibrium is restored at a higher spot price than before the outage.
{~pjlcililcation
of Spot Prices
a regulated energy marketplace, the utility's central control system knows the
costs of all the generators, so the marginal fuel cost component of
prices is computed by the formulas of this book. However, under a
structure, the regulated T and 0 company's Market Coordinator
not necessarily know the precise operating cost characteristics of the
generators. Hence, the marginal operating cost component of spot prices
be determined empirically by observing how the generation responds to
t prices at various times. Alternately, the Market Coordinator could ask
private generating firms to furnish their operating cost data (confidentially
",';'",r<>t,'",
of course) to facilitate system dispatch. We can see little reason for the privat
firms to say no or to try to play games and provide incorrect data.
Since the Market Coordinator knows the details of the transmission an
distribution systems, the network components of the spot price are determine
directly by the equations in the main text of this book.
There is no generation revenue reconciliation term; i.e. in (2.2.1), I'R (I) = O.
The network revenue reconciliation term IJR.k(t) remains.
Conclusion: Spot price evaluation is mostly but not entirely the same for both t
regulated energy marketplace and the deregulated system of Figure 5.1.1.
System Security
117
'~"'"~"'~.J
close to its desired value (60 or 50 Hz), to keep the sum of tie-line
near the net scheduled interchange levels during normal operation, and to
emergency support when needed. The regulated T and D company pays
generators to accept and respond to AGC signals. The price level is
ned by the marketplace. Of course, it is conceivable for the regulated T
D company to own some limited generation (pumped hydro would be nice)
sole purpose is to accept AGC and emergency response signals.
"~lJl1LlI.~"JlI.
At long last, the deregulated marketplace has resulted in something new: the
for pricing dynamic behavior. Research) is needed before the details of how to do
are clear, but no major obstacles are foreseen.
,vll'v.u"v.Il.
Jpt'.l/{\"It
of Monopoly Power
rll>rl"P ...,five
{\rt.tprm
C':"'~TI'''''''''S.3.
~/kwh
8000
><1000
7000
6000
5000
4000
3000
2000
1000
0.010
140
120
Sat.
160
Sunday
Shows net revenue under the spot price curve for the week January 7 to 13. Only
one unit's net revenues (based on a marginal cost of 2.0Se) is shown for sake of
clarity.
6000r-----------------------------------------------~
Week of October 6 to 12, 1980
5000
4000
3000
2000
1000~~~--~~--~~--~~~--~~--~~--~~--~~
0.0 10
enues
ilable
some
duled
enues
~ firm
n per
later
)st of
20
30 40
50 60
70
80 90 100
120
140
160
Shows net I evenue under the spot price curve lor the week October 6 to t 2 Only
one unit's net revenues (based on a marginal cost of 2.0Se) is shown for sake of
clarity. Total revenue for a year is the sum of weekly revenues for each week the unit
is available.
Economies of scale are a key issue in the argument of whether private firms will
build generators that are socially desirable (economically efficient). Private firms
will tend to build smaller plants that can be put on-line fast, so they can start
to get their money back sooner. If one lives in an ideal world where future costs,
In a regulated energy marketplace, the vertically integrated utility can coordinate its generating and network expansion plans so the network is not overloaded and desirable system dynamics are maintained. One of the potential
weaknesses of the deregulated energy marketplace of Figure 5.1.1 is that such
coordination will not be as close.
When a private generation firm is considering building a new plant, it will pay
the regulated T and 0 company to do load flow and system stability studies to
determine whether the new generation can be expected to see unfavorable
network and/or dynamic quality of supply components of the spot prices due
to weak transmission in the area. If the network is too weak, the regulated T and
D company will decide whether or not it is socially desirable (relative to the
customer's bills) to strengthen the network. Alternatively, the regulated T and
o company might offer to build new lines at the private firm's request and then
charge for their use by using the line-by-line decomposed method of revenue
reconciliation (discussed in Chapter 8) so the private generation firms pay for
the amount of the new lines they use.
A potential obstacle to overall economic efficiency of the deregulated marketplace is that the regulated T and D company will have to forecast what other
types of private generation might be built in a given area in subsequent years.
For example, for a given new generating plant, adding a 130 kV line might be
optimal, but if a second generator were to be added a few years later, building
a 230 kV rather than 130 kV line might be better. This exemplifies the economy
of scale issue for transmission. There is also the potential problem of the
regulated T and D company building lines for a proposed new generation plant
which is subsequently cancelled. These potential problems are still another
reason we do not advocate deregulation as in Figure 5.1.1 until detailed analyses
are done.
I?rirj~(,9!~tinl~
121
in a Deregulated World
tion and planning of any electric power system requires forecasts of future
ill pay
lies to
)rable
!s due
Tand
:0 the
rand
I then
In today's system (or a regulated energy marketplace), all the demand foreand decisions are done by the central utility. Demand and equipment
ty are the critical variables influencing long-term operation (unit com, plant maintenance, and fuel purchase) decisions. For investment
ng, other variables such as future fuel prices, the cost and availability of
labor conditions, and the possible availability of alternative generation
gies are also forecast.
nder the deregulated energy marketplace of Figure 5.1.1, separate forecasts
made by each private generation firm (or purchased from Information
ts) and by the regulated T and D company. Private generation firms
their long-term operating decision on forecasts of future spot price patterns
the appropriate time span. These spot price forecasts replace the demand
unit availability forecasts used in the regulated case. These firms' investment
are also based on a similar spot price behavior foreast in addition to
variables used in the regulated case (fuel prices, capital prices, and so on).
One advantage of the deregulated case is that many different, independent
and subsequent decisions are made. Some private generation compago bankrupt because of their forecasting errors, but others will realize
profits. 7 Thi~ is more desirable than the centralized case, where a single
's forecasti!lg errors can influence all of the generators in a given
. Moreover, price feedback makes spot price patterns easier to predict.
, today we do not know whether the errors in long-term
of spot price behavior will have a larger or smaller impact on operating
investltlent decisions than the forecasting errors of the central regulated
This'is still another reason we are not advocating establishment of the
energy marketplace of Figure 5.1.1 at the present time.
and Long-Term Contracts
122
The deregulated energy marketplace of Figure 5.1.1 still has the T and 0
company, which is subject to regulation. 1i This regulation is done pretty much
as in the present system except that only the rates the T and 0 company charges
to transmit and distribute the energy are regulated (e.g., the revenue reconciliation mUltiplier is set to recover network capital costs). The regulatory commission does not directly set the rates that the users payor that the generators
receive.
123
5.4. A SCENARIO
costs added) without explicitly worrying about generation capital costs. The
regulated T and D company acts like a middleman in the transactions.
Steps II and HI of Table 5.4.1 can remove these differences.
Step II: Mandatory Wheeling
The wheeling rates of Step II use spot prices which incorporate revenue reconciliation for both generation and the network. As a result, a private user, for
example, would not find it advantageous to buy from a private generator or
other utility solely to escape large generation capital costs of his/her own utility
(say resulting from a new nuclear plant entering the rate base). The underlying
logic is that as long as the user's own utility has an obligation to serve the user,
the user has an obligation to help pay for generation capital costs. A similar
logic applies to private generators from which the utility has an obligation to
buy. (See also discussions in Section 9.5).
s.
nsac-
user,
nilar
>n to
result of Step III is a mixed system which combines regulated utilities (with
own customers whom they have to serve) and free agents, users and private
Step IV Ls to wait and see how many users and private generators
to enter the free market and what affect such actions have an overall
s.xstem behavior, etc.
Sometime during this wait-and-see period it could become necessary to introduce the~ynamic pricing discussed in Section 5.2. However, as discussed
, it c~uld also come into existence during Step I, i.e., as part of the
regulated energy marketplace.
If free agent status becomes popular and has desirable effects, the regulatory
commission (or legislature) might take further action such as discouraging (or
preventing) the regulated utility from building more generation. This would
eventually lead to the deregulated marketplace of Figure 5.1.1. The regulatory
commission might even go all the way and have the regulated utility sell off its
existing generation to the highest bidders (which introduces another host of
questions for which answers do not exist today).
The main advantage of the four-step scenario of Table 5.4.1 is that one is able
to play court to deregulation for a long time before committing to a binding
marriage. It is also possible to achieve many of the potential advantages of
deregulation without full implementation, i.e., living with a mixed marriage of
both regulated and deregulated participants.
SECTION 5.5. DISCUSSION OF CHAPTER 5
There are a host of papers and reports for and against deregulation, with
as many definitions of what the term means. Since deregulation is really
side excursion on the main journey of this book, we choose not to provide
detailed guide to the literature. The ideas of this chapter result from a """''''''''6
of some of our earlier deregulation papers. (See the Annotated Bibl,roo,rll',\hv
section on Deregulation and Wheeling). Many other issues in those papers
not covered here.
The deregulation concept of this chapter is based on a supply and demand
marketplace. Most of the other electricity deregulation literature is oriented
only to the supply side i.e., to deregulating generation without altering the way
users buy electricity. We believe that deregulation which considers only the
supply side of the supply-demand equation is dangerous and could have very
negative results.
A second major difference between this chapter and most of the rest of the
deregulation literature lies in our concern that the economics and physical
security of power systems not be destroyed or compromised. Many other
deregulation scenarios bypass issues such as those discussed in Appendices A,
B, and C. If they examine issues such as plant dispatch at all, they propose that
a central purchasing agent will coordinate the deregulated generators, telling
them when and how much to generate. Long term purchase contracts with
variable quantities and fixed prices are advocated. We believe that such arrangements give all the power to the central purchasing agent, and quickly reduce the
"deregulated" generators back to the status of virtual utilities, with little incentive for innovative behavior. Issues such as responsibility for outages and for
spinning reserves would be very complex in this scenario, and the resulting mess
of contingent long term contracts would have few advantages compared with
present regulation. The new proposal for deregulation in the U.K. may provide
some evidence about this, since it seems to have some of these characteristics.
It is also interesting to compare the situation in electricity with that in natural
gas. Wellhead price deregulation replaced years of strict price regulation. It lead
eventually to considerable deregulation of demand, and to the rise of spot
markets for natural gas.
Our earlier papers on deregulation contain further discussion of these issues
and a discussion of partial deregulation. For example, the present regulatory
system could be maintained but with free entry allowed into generation by
anyone willing to be paid the spot price.
ues
ory
by
127
agree. Experimentation is appropriate now, while drastic change is preWe also agree that reforms of pricing methods (spot pricing) will have
bigger beneficial effect than deregulation could, at least in the next one
two decades.
Finally, we have a substantive disagreement with the implicit assumptions in
Joskow/Schmalensee book, and in much of the rest of the deregulation
. The issues is the willingness of private companies to build power
which would sell into a spot market, without a firm "take or pay
1." This is closely related to economies of scale in generation, since it is
likely for small plants than large ones. We expect that plants costing
a few hundrell million dollars will be quite viable, and not suffer
ISe:O'nomles of scale compared with 1000 MW, billion dollar plants. Although
large plants look good on paper, they are inflexible, take a long time to
(for a variety of reasons), have poorer reliability records, and impose
securi~y costs on the rest of the system. [Ford, 1985] In a full spot
tplace, these liabilities would be reflected in the spot price paid to them.
any industries today are characterized by plants costing several hundred
dollars, which are built despite no guarantees about a market for their
and we expect that electricity could/would be the same.
The real resolution of this issue is an empirical question. It can actually be
today, without much deregulation. State utility commissions have the
power to allow a utility to pay spot prices to any private firm that wants to build
a generator and sell to the utility. We hypothesize that if such offers are made
(with appropriate protection e.g. an enforceable promise that the prices will not
be reregulated once the plant is built), private firms will come forward to do this.
See Bohn et al. [1984] for additional discussion.
1. Sales directly from generators to users are not forbidden. However, when spot prices are
properly calculated, they are irrelevant; customers and generators always do at least as well by
dealing only with the T and D company.
2. Precisely defining how many is too many is a difficult task that depends on overall system
characteristics.
128
t pr~sented the basic ideas without burdening the reader with mathematical
Part II presents the explicit concepts and equations which enable
calculatiQn of hourly spot prices and associated energy marketplace transacPart n~tries to prevent mathematical manipulation from obscuring the
concepts. Nevertheless, a lot of complex-appearing equations loaded with
symbols and sub- and superscripts will be found. Some but not all readers
be happy to see such equations replace the handwaving of Part I.
mathematical theory of the most general case could be presented in one
mathematically beautiful step. However, we have chosen a less elegant
,....."'''h which starts with simple cases and proceeds sequentially to more
formulations. This step-by-step approach takes more pages but should
hope) make it easier to understand what is going on. The sequence is as
r
r
6. GENERATION ONLY
o
o
The model's simplicity makes it easier to understand the basic concepts underlying the hourly spot price.
Section 6.1 starts off with the simplest of all situations, which considers only
fuel and variable maintenance costs assuming there is always enough capacity
available. This leads to the "system lambda" component of the hourly spot
price. Sections 6.2 and 6.3 then address the impact of capacity constraints,
which leads to the generation quality of supply components. Two approaches
are presented, a cost function approach in Section 6.2 and a market clearing
approach in Section 6.3. Section 6.4 presents a side discussion on using prices
to "dispatch generation" as well as demand. Section 6.5 concludes with a very
important discussion on the multiple time-period couplings of supply costs and
demand responses. Since such multiple time-period effects greatly complicate
the notation, we usually do not explicitly include them in the notation of this
book. However, in practice they can rarely be ignored.
131
In this section, the hourly spot price is discussed considering only generation
operating costs associated with fuel and variable maintenance and assuming
that there is sufficient generation capacity available so that demand never
exceeds supply. This unrealistic assumption is removed in Sections 6.2 and 6.3.
Define
t: Time index in one-hour steps
1,2, ...
j: Generator Index
j
1,2, ...
k = 1,2, ...
dk(t): Demand of kth customer during hour I (kWh)
d(t) = Lkdk(t): Total demand of all customers during hour t (kWh)
equipment outages.
dk(t) and hence d(t) are random processes in time t because of random
phenomena such as weather.
gj(t) j = 1 ... are functions of the gj.ma.(s) and des) s = T 1 t ... T2 as
determined by economic dispatch and unit commitment logics which the
utility uses to minimize its operating costs. T2 - TI is the time interval
considered by the unit commitment logic.
of this chapter is that predictions for one hour into the future are perfectly
accurate. Chapter 9 discusses the impacts of removing this assumption.
Define
Gj,F[git)]: Fuel cost of jth generator with gj(t) kWh of output during hour t ($)
Gj,M[gj(t)]: Maintenance cost of jth generator with gj(t) kWh of output during
hour t ($)
Gj .FM [gAt)] = Gj.f[gj(t)] + Gj.Mfgj(t)]: Total fuel and maintenance cost of jth
generator ($)
GF[g(t)] = LPj.f[gj(t)]: Total fuel cost of all generators with total output get)
during hour t ($)
GM[g(t)] = LjGj.M[g/t)]: Total maintenance cost during hour t ($)
GfM[g(t)] = Gffg(t)] + GMfg(t): Total fuel and maintenance cost during hour
t ($)
(6.1.1)
A more precise notation for GFM [get)], etc., would be GfM fgj(t) j = I ... ], but
in this chapter all generators are assumed to be aggregated into a single equivalent generator.
Assume the total cost (ignoring capital) of providing electric energy is given
by
T2
GFMtg(S)]
s=T\
a(Total Cost)
Pk(t) =
M, (t)
In this chapter only an aggregate demand d(t) is considered, so there is only one
hourly spot price given by
(I(Total Cost)
p( t)
==
iJd( /)
A key assumption made in this section (and in Sections 6.2 through 6.4) is that
there are no multiple time-period couplings of the costs. Thus g(s) is independent of d(t) t =I s. This yields
(t) == DG FM [get)]
(!d(t)
(6.1.2)
Energy Balance Constraint: Total energy generated during hour t has to equal total
amount used.
For the present case, losses are ignored, so the energy balance constraint yields
get) =
d(t)
(6.1.3)
aG rM [d(t)]
(6.1.4)
ad(t)
Yr(t)
YM(t)
(6.1.5)
aGdg(t)]
og(t)
(6.1.6)
(6.1.7)
Fuel and maintenance components are combined into system lambda, where
A(t)
System Lambda
),(t)
cGFM[g(t)]
iJg(t)
=
)'F(t)
l'M(t)
(6.1.8)
.1(/)
(6.1.9)
A first course in basic economic theory often derives the fact that marginal cost
pricing is optimum in a social welfare sense. We will now repeat the derivation
which proves that (6.1.9) is optimum. It is important to understand the principles in this simple case.
6. Generation only
135
B[d(t)]: Total benefits all customers receive from using d(t) kWh of electrical
energy during hour t.
Social Cost I
(6.1.10)
Hourly Spot Price p(I): Rate charged to customers which minimizes the social
cost assuming optimum customer behavior.
Customers exhibit optimum behavior if d(t) is chosen to maximize the difference
between the customers' benefits and their bills, i.e. to maximize
B[t/(I)]
p(t)d(t)
(6.1.11)
= pe,)
(lB[d(I)]
(6.1.12)
td(t)
Substituting d(l) = d[p(I)] into (6.1.10) yields
Social Cost = C1ML!;(t)] - B[d[fl(t)]]
~
>
(6.1.13)
4~
(6.1.14)
/l,(t): Lagrange multiplier introduced for the energy balance constraint d(t) = get)
and then find the optimum values of both g(l) and p(I). Setting
DQ(t)
ng(l)
()
DCFM[g(t)]
ilg(l)
Setting
oQ(t)
op(t)
}.(t)
d(t),
L'r
l ( )
c t
OB{d(t)J] Od(t) =
Od(l)
Op(t)
or
oB[d(t)J
od(t)
=
).(1)
(6.1.15)
Using (6.1.12),
pet)
= A(t)
(6.1.16)
which is (6.1.9).
The derivation of (6.1.16) differentiates the Lagrangian net) of (6.1.14) with
respect to g(t) and p(t). An alternative approach (that yields the same result) is
to find the optimum get) and d(t) by setting
an(t)
og(l)
an(t) =
Od(t)
to yield (6.1.15). The argument that leads to (6.1.12) can thus be viewed as a way
to introduce prices p(t) in a way which causes the customer to choose a value
for del) which satisfies the social optimality conditions for d(t). We mention this
alternative approach because later on, when the problem formulation gets more
complicated, we will sometimes use it. It is best to understand the basic ideas for
simple problems before the notation starts to get messy.
Solution for p(t):
A[d(t)]
d[p(t)]
A[d[p(t)lJ
(6.1.17)
6. Generation only
137
The solution of (6.1.17) for pet) is usually viewed in the economic literature
as the intersection of the supply and demand curves (see Figure 2.10.1).
Effect of Purchase-Sales From-To Other Utilities
5)
5)
Many utilities purchase and/or sell electric energy from/to neighboring utilities
via transactions ranging from long-term multiyear contracts to hour-by-hour
economy transactions. The cost of purchases and profit of sales are explicitly
included in the GFM [g(t)J cost function. Thus in many situations, },(t) is directly
determined by purchases and/or sales and only indirectly by the utility's own
variable fuel and maintenance costs. This phenomenon is of particular importance for a utility with only hydro generation.
In most subsequent discussions in this book, we talk about A(t) as if it is
determined directly by the utility's own explicit fuel (and other) costs because
it simplifies the discussion. However, at various points, we reintroduce the fact
that purchases and sales cannot be neglected.
s
SECTION 6.2. GENERATION QUALITY OF SUPPLY, I'Qs(t): COST
FUNCTION APPROACH
Section 6.1 considered only fuel and variable maintenance costs. An additional
term due to quality of supply (i.e. ability of generation to meet demand) is now
add~>to the cost function.
CG1lsider a utility with demand d(t) and maximum available generation gmax (t)
duri~g hour t. If d(t) >~ gmax (t), the system will black out. Therefore a utility will
take'lextreme measures to prevent such a condition from occurring Define
gerlt.l'U): Critical Generation Level: When g(t) = d(t) approaches gerit.i,(t), the
utility takes~measures to prevent g(t) from exceeding gerit,y(t).J
geril.l,(t) = gmax(t) - gres(t).
(6.2.1)
gres(t): Operating reserve requirements. Can be a function of gj(t) j = 1 ... and
d(t).
The measures a utility starts to take to prevent get) from exceeding gcrit.y(t) cost
money. Define
GQS[g(t)]: Generation quality of supply costs incurred to provide reliable energy to
customers during hour t; i.e. to prevent g(t) from exceeding gcrit.,(t) ($).
Define
I'Q5(1): Generation quality of supply component of hourly spot price.
Then
I'QS(t)
aGQS[g(t)]
ag(t)
(6.2.2)
prices
Thus for the present case, the hourly spot price p(t) of Section 6.1 expands to
become4
pet)
A(t)
I'Qs(t)
(6.2.3)
o
o
A utility that is interconnected with its neighboring utilities may be able to buy
emergency energy over the transmission tie lines. In some cases, a utility can also
buy additional energy from customers with emergency back-up generators or
cogeneration systems. Hence one way to quantify the GQs[d(t)] costs is in terms
of such purchases.
In actual operation, utilities are often buying and selling energy from each
other even ifno one utility is near its own generation capacity. In such cases, A(t)
includes the effects of such transactions. Hence the YQs(t) resulting from emergency purchases can be considered to be part of A(t) also. We choose to take this
point of view in the rest of this book. Thus in all further developments we
assume that
Costs of emergency purchases are included in A(t) and are not part of I'Qs(t).
When g(t) > gcrit,y(t), a utility can exercise load management to reduce load
instead of, or in addition to, buying emergency power. Ways to achieve such
load management include
o
o
o
o
Interruptible contracts
Direct load control
Reduction in utility house load
Rotating blackouts
Such load management introduces direct costs to the utility, and/or the
customers,
These load management-unserved energy costs can be quantified by
GQs[d(t)] =
f}QS,y(t)uy(t)
(6,2.4)
6. Generation only
to
!.3)
u;.(t) =
0Q5.;-(l):
otherwise
139
This yields
')'Qs(I)
OQs;(I)b, (I)
(6.2.5)
au,.(I)
b'(l)
ag(l)
lly
so
Of
1S
:h
otherwise
(6.2.6)
t)
ris
'e
Assume the utility has gone through its planning process and by combining
The calculation of a value for AQs .y can be done conceptually by use of generation expansion computer programs that look many years into the future. This
approach requires a lot of computation and is subject to the effects of forecasting errors in future load growth, costs, etc. (see Chapter 10). A more pragmatic
approach is to define AQs,y in terms of the annualized capital cost of a peaking
plant such as a gas turbine,
Given AQs,y, it is reasonable to quantify the generation quality of supply costs
by
8760
L GQS[g(t)]
tal
KQs .)": Generation plant addition made to maintain reliability of generation supply
(kW)
Denoting by ,1. incremental changes yields
_
YQs (I) -
aGQsfg(t)] _
ag(t)
-
!l.KQs.'l
AQS.;. !l.g(l)
d(t),
I:!.KQs .;
AQs .; I:!.d(l)
I:!.KQs .;
M(t)
a
8760
L a(l)
1=1
a(I):
Allocation function
d(l) = dmax
otherwise
liN
otherwise
LOLPy(l)
LOLH;.
6. Generation only
141
LOLP),(t)
1=1
Further discussions in this book usually use the loss of load probability
allocation function of (6.2.11), i.e.,
(6.2.12)
The a!location function~' of (6.2.8) through (6.2.12) are reasonable. For those
who til<e mathematical l!1anipulation, we will now present a derivation.
Assllme the generation plant addition Kos.y is chosen to satisfy a constraint of
the form
8760
Cy =
(6.2.13)
1= I
cy[d(t) -
Assume the constraint of (6.2.13) is active (binding). Then to maintain it, incremental changes in d(t) and Kos.y must satisfy
t = 1 ... 8760
(6.2.14)
I'S
Od(t)
1 .... 8760
QS.)'
oc,[d(t)
KQs., 1
=
Od(t)
M(t)
a(t)
IlKQs y
/ld(t)
(6.2.15)
a
aCy[d(t) - KQs .y]
ad(t)
a(t)
8760
L a(t)
(.1
The special cases of (6.2.9), (6.2.10), and (6.2.11) result from (6.2.15) by
choosing
o
dma
I( N(dma.
dma 1
+ ... dmm
N - I - KQs .,
(Installed Capacity)
Assume KQs y is nonzero but small and that K Qs y is perfectly reliable generation
(i.e., is always available when needed) and assume that hourly future demands
are known. s Then
d(()YQS.,
[d(t) _ K .y - gcrit.y(t)]
Qs
(6.2.16)
a(t) = aUE,.(t)
ad(t)
J
o
p[gcrit.l'(t)]dgcrit.y(t)
LOLPy(t)
(6.2.17)
6. Generation only
143
Evaluation of LOLPy(t)
15)
by
The reader should notice that LOLPy(t) is a probability and hence can take
values between 0 and I. If LOLPy(t) is estimated at the beginning of hour t, all
uncertainty affecting the system during the hour is known according to the
assumption of Section 6.1. Hence, if calculated at hour t, the value of LOLPy(t)
as derived above is either 0 or 1, because gent.y(t) g is known and
UEy(t) = del) - gerit.y(l) if del) > gent.y(t), zero otherwise. However, as discussed in Chapter 4, a more pragmatic approach is to use a smoothly varying
LOLP)'(l). Several methods are discussed in Chapter 4, one of which is repeated
here, namely
k[d(t)] - gcnt.;.(t)
LOLP.(t) =
t\g
d(t) >
gent.;. (t)
t\gcrit
otherwise
'i,'
In~Bection
;.(1)
l'Qs(1)
(6.3.1 )
where
on
lds
\6)
(6.3.2)
This is called the market clearing approach because when the demand del)
approaches the critical level gerit.y(t), the value of I'Qs(t) is raised until the market
clears,6 i.e., until demand reduces enough.
Mathematical Derivation
In order to derive (6.3.1) (6.3.2), the hourly spot price p(/) is defined as in Section
(6.3.3)
d(t) and
(6.3.4)
!l(I) =
+ Ite(t)[d(t)
- g(t)]
(6.3.5)
ItQs.,(t)[g(t) - gerl,.y(t)]
As in Section 6.1, /le(t) is determined by finding the optimum get) and is given
by1
Ite(t) =
A(t)
Using d(t)
=
iJO(I)
ap(l)
PQs.,(t)
_ oB[d(t)] =
Od(t)
(6.3.6)
A(t)
(6.3.7)
)'Qs(1)
where the generation quality of supply component I'Qs(t) is the Lagrange multiplier
(6.3.8)
I'QS (t)
= ACri' (I).
(6.3.9)
Assume a linear demand response model (other forms are discussed in Appendix
E).
6. Generation only
d(l) =
d,,(I) {I
f3(t)
[P(I) -
),erit(t)]}
Aedt (I)
145
(6.3.10)
(J(t) is the demand elasticity parameter (a negative number). If do(t) > derit.;.(t),
then it is necessary to set get) = gerit.:.(t) in (6.3.10). This yields
[gmt/(I) - do(t)]A,,,t(l)
(6.3.11)
P(I)do(l)
Therefore
A . (I) [gern.y(l) - do (I)]
,rn
do (t)!3(I)
(6.3.12)
otherwise
NUMERICAL EXAMPLE.
Assume
10/kWh = 100$/MWh
Aerit
do
1000MWh
900MWh
Then:'6.3.12 yields
100(900-1000)
P1000
10
'.1_
- 7f $!MWh
-1.0 /kWh
f3
f!.
LQs/kWh
-0.01
100
-0.1
10
-0.2
-0.5
p.
In Section 6.3, prices were used as a mechanism to cause the customer to choose
the socially optimum value of demand d(t) while the generation get) is specified
by the utility to meet the demand. A somewhat different perspective on what is
going on leads to the conclusion that the generators can also be viewed as
"price-takers" who "self-dispatch" themselves based on prices.
To see this, consider what happens when the Lagrangian of (6.3.5) is optimized directly with respect to g(t) and d(t) (as discussed in Section 6.1). This
yields
an(t)
A(t) - JJ.e(t)
ag(t)
an(t)
ad(t)
aB[d(t)]
ad(t)
JJ.QS,y(t) = 0
JJ.e(t) = 0
OGFM [get)]
A(t)
og(t)
(6.4.4)
pAt)d(t)
JJ.QS,y(t) =
(6.4.5)
aB[d(t)]
ad(t)
+ pAt)
= 0
(6.4.6)
6. Generation only
147
The results (6.4.5) (6.4.6) yield the optimum conditions (6.4.1) (6.4.2), provided
From this point of view, the customers and the generators are making
independent decisions on the values of d(t) and g(t) based on the same prices
(paid by customers and to generators) while the price is adjusted (somehow)
until the energy demand constraint d(t) = g(t) is met. This concept may not
seem to be overly exiciting for the special case being considered here in Chapter
6. However, it becomes more interesting in later chapters when many different
generators are considered and some of them are owned by the customers.
SECTION 6.5. MULTIPLE TIME PERIODS
For most utilities, the marginal cost of generation during any given hour
depends on the generation levels during previous hours and those expected in
futur't hours. Examples of such multiperiod time couplings are startup and
shutdown costs, hydro::with storage, and various types of fuel contracts.
Ins,tead of using (6.1.8), the multiple period definition of }.(t) is given by
(6.5.1)
I)
;)
i)
148
The customer demand response model underlying (6.1.l2) did not have any
multiple-period coupling dependence. In practice, the potential of customers to
reschedule electric power usage from times of high price to times of low price
provides such multiperiod coupling. Examples of residential rescheduling are
thermal storage of either heat (in hot water heaters, buildings, or thermal bricks)
or cold (in buildings or ice). Industrial examples include the variations of the
times at which metal is melted, water is pumped, and very large machines are
used.
As with multiple-period generation costs, multiple-period demand' response
complicates the computations but does not change the basic principles. (Section
9.4 is an example of an analysis with multiple-period coupling).
SECTION 6.6. DISCUSSION OF CHAPTER 6
The random nature of demand, cost, and equipment availability is a basic fact
of life in a spot price based energy marketplace. However, these random
characteristics did not playa major role in this chapter because we assumed the
hourly spot price is specified after the values for demand, costs, etc., for the
given hour are learned - i.e., after the random processes have been realized (or
at least predicted perfectly).
Two of the basic components of the hourly spot price were developed:
6. Generation only
149
at least for inital implementation. Two cost function approaches (cost of load
management/unserved energy and annualized cost of peaking plant) were
presented. It is easier to obtain the necessary input data for the peaking plant
approach.
In Chapter 7 we will combine both the cost and market clearing approach to
quality of supply into a single formulation so that the reader can always choose
which one best meets his/her needs.
HISTORICAL NOTES AND REFERENCES - CHAPTER 6
The basis of this chapter is Caramanis, Bohn and Schweppe [1982], although the
formulation used in this chapter is much simpler to make it more understandable. Outhred and Schweppe [1980] discuss quality of supply components from
a more general point of view, albeit with words, not specific equations.
The basic mathematical approach used here (and in Chapter 7) of minimization of a social cost function subject to constraints is classical economics, which
is used by many authors. Our approach differs from most of the other literature
because of the time scale of concern, which, of course, can make a large
difference in practice.
Most of our approaches in Sections 6.2 and 6.3 to quantifying the quality of
supply component are translations of existing ideas on marginal cost pricing to
the one-hour time scale where events are revealed before prices are calculated.
We '}l1le the term "quaHty of supply component" because it seems to be the most
des1$Eptive term available. Other authors use terms such as "capacity compon~nt" or "reliability component."
Mbre detail on the multiple time-period problem of Section 6.4 can be found
in Caramanis, Bohn, and Schweppe [1982]. See also Kaye and Outhred [1988].
NOTES
,t
,t
d
:t
g
Ie
a
1.
Ie
151
152
The power flows over a given line in a network and the losses in that line are
determined by the generation and loads at all buses, not just the buses at either
end of the line.' The line flows are determined by Kirchoff's Laws and depend
on the network's interconnected structure as well as physical parameters such
as impedances.
In general, the solution of Kirchoff's Laws involves both real and reactive
powers combined with voltage magnitudes and angles. This is called an AC load
flow. In most of this chapter we consider a special case (approximation) that
involves only real power flows called a DC load flow. 2 The full AC load flow case
is discussed in general terms in Section 7.9; see also Appendix A.
Generation-Load Definitions
As in Chapter 6 define
g/t): Output of jth generator during hour t (kWh)
dk(t): Demand of kth customer during hour t (kWh)
get)
Ljg/t): Total generation
d(t) = Lkdk(t): Total demand
(7.1.1 )
(7.1.2)
Define
g(t): Vector of all generation
~(t): Vector of all demands
A given bus can contain a generator, a load, or both a generator and a load.
Define as in Chapter 6 3
GFM [get): Total Generation Fuel and Maintenance Costs
GFM [g(t)] = L,GFM,j[g;(t)]
(7,1.3)
GQslg(t)): Generation Quality of Supply Costs
get) ~ gcrit,;,(t): Maximum Total Generation Constraint
gcrit,y(t): Depends on maximum available generation and operating reserve
requirements; see (6,2, I)
g)(t) ~ g),max: Individual Generator Maximum Capacity Constraints
B[~(t)J: Total Customer Benefit
(7.1.4)
Assume that the kth customer acts optimally, i.e. the demand dk(t) is the
function of price Pk (t) defined by
(7.1.5)
153
Network Definitions
, are
ther
,end
iuch
A well-posed network flow problem results by specifying the gj and dk at all but
one bus, which is called the swing bus. 4 We assume the swing bus contains only
a generator and let * denote the value of j which specifies the swing bus
generator. Define
:tive
oad
that
;ase
*(1): Vector of all gj(t) except g*(t), the generator on the swing bus.
Thus the dimension of *(t) is one less than the dimension of (t). Define
Zi(t): Energy flows over line i during hour t (kWh) i = I ...
z(t): Vector of all line flows
y(t): Vector of net bus injections for all buses except the swing bus, i.e. the
If all buses except the swing bus contain both generators and loads,
1.1)
1.2)
~(t)
[*(1) - ~(t)
Z;
z,[,t(t)
""=
(7.1.6)
id.
~(t)
Hy(t~
(7.1.7)
This yields
rve
OZ;(I) _
ildk(t) -
H
-
;k
(7.1.8)
1.4)
the
L;(f)
L(I)
1.5)
L[~(I)
(7.1.9)
When line losses are present, the energy balance constraint used in Chapter
6 becomes
get) =
d(t)
L gj(t)
L dk(t)
d(t)
L L;(t)
i
get)
L(t)
L[::[,[*(t), q:(t)])
d(t)
L[::[ K(t)]J
Network lines have maintenance costs which depend partly on the line flows.
Define
NM,i(t): Maintenance cost for line i resulting from flows during hour t ($)
NM(t): Total Maintenance Cost
= LjNM,i[z;(t)] = NM[~(t)]
Any line i has a limit on how much energy flow it can handle. In practice, this
limit can vary with time duration of flow, outside temperature, direction of flow,
and conditions at other buses. However, for the present discussion, assume the
power system has to be operated such that
!Zi(t)! :;:; Z"m ..
(7.1.12)
The more general case just introduces a lot of extra notation. This line flow
constraint is analogous to the maximum available generation constraints
gil) < g},max (t) of (7.1.3). The physical constraint on line flow is not necessarily
a hard constraint (as in (7.1.12 because a small violation will not instantaneously cause the line to fall down or burst into flames. Instead, overly high
flows cause loss of line life and increase the probability of a line failure immediately or in the future, which decreases the quality of the supply the utility is
furnishing its customers. s Furthermore, line flow constraints may be used to
reflect other system operating constraints such as stability. A cost function
which gets large rapidly as the line flows approach or exceed some level can be
used to capture these effects. Define
NQs,;{t): Quality of supply costs of line i during hour t ($)
NQs(t): Total network quality of supply
= Li NQS,;[Zi (t)] = NQs[~(t)]
(7. 1.I 3)
The difference between line maintenance costs NM,i(t) and quality of supply
costs N Qs )!) is not always sharp. We define them separately because in some
applications, a separate point of view is appropriate.
SECTION 7.2. GENERAL RESULT
The purpose of this section is to derive for the hourly spot price the following
155
equation, which incorporates all of the cost functions and constraints of Section
7.1.
Pk(t) =
A(t)
YQs(t)
tlL.k(t)
Fu~1
tlM.k(t)
tlQS.k(t)
(7.2.1)
and Maintenance
OGF.M/t)
ogj(t)
oGQs[g(t)]
og(t)
!lQs.)
(t)
[A(t)
+ YQs(t)] a;IJi)
[A(t)
YQs(t)J
L oL;(z;(t
;
oz;(t)
oz;(t)
Odk(t)
L aNM.;[Z,(t)] oz;(t)
;
oz;(t)
Odk(t)
't
~W
Odk(t) =
J.l.Qs.~.;(t):
!lQs.~.,
(t)] oz;(t)
Odk(t)
..
Lagrange multiplier arising from the maximum line flow constraints (7.1.12)
+ tlk(t)
+ YQs(t)
(7.2.la)
156
'1M.k (t)
'10S,k (t)
L ~ (t) OZi(t)
i
'
Odk(t)
OzJt) {y(t)Li[Zi(t)]
NM.i[Z;(t)
NQs,;[z,(t)]}
/lQs.~,;( t)
GFM [get)]
GQsfg(t) - B[d(t)
Ile(t)[d(t) - get)]
/lQs.l'(t)[g(t) - gcrit.y(t)]
and differentiating to find the optimum get) and p(t) where d(t)
would have yielded
oGQs[g(t)]
og(t)
YQs(t) =
d[p(t)]. This
/lQs.y(t)
which is the sum of the two separate results. Then by assuming GQS = 0 or
= 0, either of the two Chapter 6 approaches can
be obtained (actually in some applications it could be desirable to keep both
types of quality of supply terms).
This concept of using both a cost function and a constraint for the quality of
supply is used in the following.
Derivation of (7.2.1)
G[(t)
N[::(t)
B[~(t)]
(Customer Benefit)
11c(t)[d(l)
L[::(t)) - get)]
;)
GrM[(t)]
+ Godg(t)]
L f.lmaqJ(t)[g/t)
(Maximum Individual
Generation Constraint)
- gma,/t)
L NM,[z,(t))
+
(Maintenance Costs)
L Nodzj(t)
+ L, f.lOS.'I.,(t)[Zi(t)
1.57
Z,.ma,)
The second step is to set the derivatives of the Lagrangian of (7.2.2) with respect
to
j = I ... and Pk (t) (or dk (I)) k = I ... equal to zero. This yields
(lG[g(r)
(ig/(t)
oN[.:(r)]
ilg/(t)
'~+~+Ilc(t
)[ilL[Z(r)
----I
cg/t)
i')B[~(t)J
M,(t)
(7.2.3)
(7.2.4)
() =
Pk t
Idt
)[1
ilL[Z(t))J
rid, (t)
+~
ilN[z(t)]
rid, (t)
+~
(7.2.5)
Note that if the hourly spot price at a generator busj is defined to be given
by
()
pt
--
IJ
(7.2.6)
such a definition is consistent with (7.2.5) provided the sign of gj(t) is changed
to make it a negative load. Thus (7.2.5) and the basic result (7.2.1) apply to all
buses j and k, and the optimality conditions (7.2.3) (7.2.4) can be neatly
summarized by
158
p(t)
8gj (t)
8B[d(t)]
adk(t) =
Pk(t)
The third step is to evaluate lie (t). Assume without loss of generality that the
generator on the swing bus is on the margin so the * is a value of j that denotes
a marginal generator. Then since [*(t) does not contain g*(t},
8z i (t)
8g*(t)
8z i [g*(t), d(t)]
i = I ...
~=o
8g,,(t)
t) =
8G[g(t)]
8g* (t)
..1.(t) =
..1.(t)
)'Qs(t)
8GQs [g(t)]
8g*(t)
( )
iJos.y t
where the fact that g* (t) is on the margin is used to get IImax.y,* (t)
ing (7.2.10) into (7.2.5) yields
Pk(t) =
..1.(t)
)'Qs(t)
[..1.(t)
8L[z(t)]
l'Qs(t)]--=adk(t)
O. Substitut-
8N[z(t)]
+ ----""-adk(t)
The final step to get (7.2.1) is to insert the definition of N[z(t)] from (7.2.2)
and collect terms.
Note that the Lagrange multipliers IImax,y,) which were introduced to handle
the gj < gmaxJ constraints do not explicitly appear in (7.2.1). Their role in life is
to help determine which generators are at their maximum values.
Although (7.2.1) is a simple-appearing form, a lot of computations can be
required to find numerical values. First it is necessary to solve the optimum
159
The network loss component of the hourly spot price is given by (7.2.1) as
[),(t)
'( )
[A t
Yos
(t)]
yos(t
~
adk(t)
(7.3.1)
The DC load flow approximation theory of Appendix D yields for the losses in
the ith line
(7.3.2)
where R, is a constant that depends on the resistance of the line. More accurate
app~ximations can be, used (such as including the effect of no load losses) but
(7.3.2) is satisfactory in most applications. The total losses
L(t) =
L L,(t)
I
(7.3.3)
:,T(t)Rz(t)
l(t)By(t)
HTRH
2[),(t)
yos(t)]
'"
OZi(r)
Rizi(t) odk(t)
(7.3.5)
160
I1u(l)
~!
(7.3.6)
l'Qs(l)l~!L(t)
kth row of B
The network maintenance component of the hourly spot price is given by (7.2.1)
as
I1M.k(t)
__ "oNM.,[z,(t)] oZ,(t)
L.,
,
oZ,(t)
adk(t)
(7.4.1)
Although maintenance costs associated with a given line depend to some extent
on the energy carried by the line, good mathematical models for this dependence
are not known at the present time (by the authors at least). Therefore we will
just hypothesize a structural form which seems reasonable.
The hypothesized form is
N~.,
NM,,[z,(I)] =
N~.,:
N~.,[z,(t)J
(7.4.2)
N~ ,[z,(t)] =
Iz,(t)1 <
Quadratic in Iz,(t)1
Iz,(t)1
>
Zcril.'
Zcril.'
There is a threshold effect so that below some critical loading level, maintenance is independent of line loading.
Maintenance is proportional to heating (losses) when line loading is above the
critical value.
L I1M.k.,(t)
(7.4.3)
OZj(t)
I1M.k., (t)
BM.ilz,(t)1 Odk(t)
Iz;(t)1 >
Zedl,;
IZj(t)1 <
Zeril.i
161
It is also possible to express 'lM.k(t) in terms of bus injections, but the line flow
form of (7.4.3) is most natural.
Equation (7.4.3) is simple to use. The problem is to specify the constants, 8M ,;
and Zerit,i' Note that in practice, The Nt.i term will often dominate, so care is
needed if historical data are used to try to specify the needed constants.
In general we believe the effect of the maintenance component is usually not
critical. This is the reason it was not discussed in Part I of this book. It is
included in the general theory because it could be important in some applicaSECTION 7.5. NETWORK QUALITY OF SUPPLY: I1Qs.k(t)
The network quality of supply component of the hourly spot prices is given by
(7.2.1) as
(7.5.1 )
In most applications, both network quality of supply terms will not be used at
the same time.
,Pih,
Netw~!$
j,
Assume {lQS.'J.i(t)
o in (7.5.1).
Quadratic in z,
t
IZ,(I)I >
Z,.m.,
Z,.m.,
(7.5.2)
where the value of the cost for large line flows may vary with the line flow itself.
Comparison of (7.5.2) with the network maintenance cost (7.4.2) shows that
they are very similar in appearance. We discuss the two terms separately because
they model two different types of phenomena: maintenance is associated with
normal operating conditions, while quality of supply is associated with times
when the system's capacity is being stretched. Thus even if identical-looking
equations were used, '1M.k(t) and '1QS.k(t) could behave numerically in quite
different ways.
Market Clearing Multiplier: I'QS,,,i(t)
Assume NQS.i(t) = 0 in (7.5.1). Assume the flow in one particular line, iO,
exceeds a hard limit ziO.ma, The flow in line iO can be reduced by simultaneously
redispatching the generators (to get a new line flow pattern) and by customer
response to changed prices.
Consider first the effect of customer response. Assume only line iO is overloaded. The hourly spot price for the kth customer is
Pk(t)
(7.5.3)
oz;o(t)
odk(t)
Customers a long distance (electrically) from the overloaded line iO have a small
(in magnitude) H jOk and hence will not respond. A second difference is that
network quality of supply can be positive or negative.
A third and perhaps even more basic difference between generation and
network market clearing is
The market clearing approach for generation capacity limit always requires customer
response, but customer response is not necessarily needed for network line capacity
limits.
For many (but not all) cases, generation redispatch can be used to remove a line
overload even if no customer responds. 7 Conceptually each generator sees a
price (see (7.2.6,
and redispatches itself. Thus in this case, the market clearing value of J1.QS.~.iO is
determined by the generation costs, not by customer response.
Quality of Supply Costs Versus Market Clearing
The difference between the use of network quality of supply cost functions and
the market clearing approach is not basic. Market clearing involves a hard
constraint, while quality of supply costs are penalty functions.
SECTION 7.6. TWO-BUS EXAMPLE
The general equations of Section 7.2 are now solved for special cases involving
a two-bus system. Hopefully this adds insight to the nature of the equations. A
more general three-bus example is provided in Appendix D.
Swin9 Bus
z_
G"~"mlr
G,"""" 2
92
9,
163
Lood, d
Bus 2
Bus 1
Problem Formulation
Figure 7.6.1 shows a two-bus system with generators at both buses and a load
at Bus 2. Only a single hour is considered, so the time dependence on t is
dropped from the notation. The numbers are chosen so that Bus 1 is the swing
bus.
For this example, assume
o
o
o
No maintenance costs
No network quality of supply costs
No
ge~eration
ThO; the quality of supply components are given by the Lagrange multipliers.
F'\Ir most cases to be discussed, the demand d is assumed to be fixed and
independent of price. This assumption is removed for the last case.
The generator's fuel and maintenance costs are modeled by
Since Bus I is the swing bus, it folIows from Figure 7.6.1 that
o
=
164
GZ
Og2
5 /k Wh =
50 $/MWh
10/kWh =
(baseload)
(peaking)
100$/MWh
R;
0; i.e.,
0.05
g2
The capacity limits, gl.max, g2,max, and Zmax values are quantified in each particular
case considered.
General Equations
A2
+
+
fJos"
lie
fJOS,)'
Ily,max,2
(7.6.1)
-
IIOS"I
(I
2Rz )fJe
(7.6,2)
The Bus I equation is much simpler both because it is the swing bus and because
Generator I is always on the margin. Similarly (7.2.6) yields
Assume gl,maX' g2.maX' gent.y(t) and Zmax are all very large so they impose no
constraints. Thus f1.Qs.y = f1.y,max.2 = f1.Qs.~ = 0 and (7.6.1) becomes
A2
(I
2Rz)fJc
165
gives four equations in the four unknowns, gl' g2, Z, and J1.e' If it is assumed that
g2 > 0 then Ily.max.2 = 0 and solving for g2 yields
or numerically
o
100
100 - 50
- (2)(5)(10 5)50
- 9000MWh
1000
5 10- 5 (IOOOi
1050
50$/M/Wh =
P2 =50[1
~
5/kWh
+ 2(510- )1000]
5
55$/MWh = 5.5/kWh
g2
800MWh
202MWh
5/kWh
P2
5.1 /kWh
which is as expected; i.e. the price at each bus is the marginal cost of generation
at each bus.
Case 2: Line Flow Limit Active
Assume
gl.max, g2.max,
and
IlQs.)'
gcrit.r
Zmax
1000MWh
For Case 2, it turns out that g, > 0, g2 > 0, so both generators are on the
margin. Thus (7.6.1) becomes
=
,,1,1
,,1,2 -
~e
~Q'," =
(I
2Rz)~e
Rz~.x
which is four equations in the four unknowns, g" g2, J.lQs," and J.le.
Solving these equations yields
or numerically
~QS,"
gl
100 - [I
600
5(10-s)(60W =
780 MWh
where of course
g2 =
50$/MWh =
50(1
5/kWh
5
~QS,"
53
47
100$/MWh
1O/kWh
167
is not a numerical accident. The presence of an active line flow limit effectively
decouples the prices at the two buses.
Case 3: Generation Limit Active on Base Loaded Unit gl
Assume
#0'.1'
g2.max,
#1.01.X.2
/los",
gl.max
500 MWh
The values of gl and g2 for this case are obvious by inspection, i.e.
so
or
so
For this case, (7.6.1) and (7.6.2) are not valid, since g2 not gl is the marginal
generator. For this case
P2 =
1O/kWh
Assume
#OS.1'
Zmax
and
/los ., =
gerit.y(t)
500
g2.m.x
500
are large so
The actual values depend on the assumed structural form of d[P2J, and the value
of the demand elasticity, etc.
SECTION 7.7. PRICE DIFFERENCE ACROSS A LINE
Section 7.6. considered two adjacent buses connected by a single line. Here we
consider two adjacent buses, but they are now embedded in a network. Two
buses are adjacent if there is one line that directly connects them in addition to
the many others interconnecting through the rest of the network. The price
difference between these two adjacent buses (i.e. the price difference across
interconnected line) is discussed in this Section.
For simplicity we ignore the network maintenance component and
network quality of suppply cost component; i.e. assume
NM [Zi(/)]
NQS.,[Zi(/)]
~,(I)
Y(I)
+ ~ ~,(I)
Y(I)
iJL,[Z,(/)]
OZi(t)
GZ,(t)
ad.(/)
(Losses)
(Network Quality of Supply)
k =
Define
"
GZi(/)]
- Od (t)
2
169
the DC load flow approximation, the partial derivatives of the line flows with
respect to the bus injections become the constants H21 and H 12 , but they depend
on the rest of the network configuration. However, if we make one further
approximation on the network, (7.7.3) is greatly simplified. The additional
assumption is
Constant Reactance to Resistance Ratios: The ratio of the reactance to the resistance for
every line i in the network is constant, i.e. the same for all lines, i = I ... ,
: Real power flow over the line (from Bus I towards Bus 2) and the
effective resistance of the line that directly connects Buses I and 2
7.7.4) shows that the effect of the losses have been greatly simplified (the
quality of supply term still involves coupling of prices throughout the
network).
If no network quality of supply terms are present (i.e. 11Qs.",i(t) = 0 all 0. then
.7.4) says
price difference L\PI2(r) between two adjacent buses. I and 2. is proportional to the
I power flow ::1,(1) on the line connecting the two buses.
1'01'
The formulation thus far has considered the hourly spot price Pk(t) seen by the
kth customer for using dk(t) during hour t where the generation levels get) are
determined by the utility. We now consider the case where some customers have
their own generators and sell energy back to the utility.
The derivation and discussion of the basic result (7.2.1) implied that del) > 0,
but such an assumption was never used. Furthermore the benefit Bddk(t)] could
actually be negative if desired. Therefore if the kth customer owned a generator
with output gt(!) and had a demand of lik(t), the net demand dk(t) is given by
d,(t)
df(t) - g'f(t)
(7.8.1)
Similarly, if the fuel and maintenance costs of the customer-owned generator are
GFM.dg~(I)l, the net benefits are
(7.8.2)
Using (7.8.1) and (7.8.2) in (7.2.2) leads to the important conclusion that
The hourly spot price Pk(O at bus k is valid even if the kth customer has a generator.
(7.8.3)
This conclusion had already been reached using (7.2.6) but is important enough
to repeat.
If the kth customer has only a generator and no load (dk(t) = 0); then the
customer is paid Pk(t) by the utility for the energy gHt) which is fed into the
network during hour t. The customer determines the value of g'k(t) by minimizing
(7.8.4)
s-
subject to g%(t)
171
~ g~ax.k
i.e., by self-dispatching the generator based on the hourly spot price. Thus gk(t)
is determined by the condition
(7.8.5)
J.i.max.;'.k (t)
if gHt) = g~ax.Y.k
For most applications, the reader can stop reading this section at this point.
However, we will now discuss some fine points which can be important in some
situations.
Nonuniqueness of Pk(t)
e
e
e
Assume for simplicity that the kth customer has only generation and no load.
The use of the Pk(t) of(7.2.1) causes the customers to self-dispatch in a fashion
which minimizes social costs. However, if the generator is at its maximum
output, i.e., gW) = g~ax.k' then any value of Pk(t) which is greater than
I,
:i
r
will )'~eld the same level of generation. Thus the theory, as presented here, does
not necessarily specify a unique buy-back rate for customer-owned generation.
There is a demand-side analogy to this non uniqueness result. Assume the kth
customer has Qnly demand dk(t) and no generation. If the Pk(t) is low enough
that dk(t) is at its maximum possible level (or at least is completely insensitive
to price), then any value of Pk (1) that does not influence dk (t) can be used.
Throughout this book we assume that the buy and buy-back rates for
customer demand and generation are given by the Pk(t) of (7.2.1) because we
believe this to be the "fair and equitable" approach. We discussed the possibility
of using a lower value (when the generation or demand is at its maximum) only
because it exists.
Role of Generation Quality of Supply
In Section 6.4, the concept of the utility generation being self-dispatched was
discussed. This idea is now extended to the present case.
It follows from the mathematics of Section 7.2 that the optimality condition
for the utility gj(t) are satisfied if each utility generator chooses gj(t) to minimize
(7.8.6)
subject to
172
o
g(t)
g(t)
:0:;;
:0:;;
gcri<.;.(t)
L gj(t)
Thus far in this chapter it has been assumed that all lines are individually
modeled. In practice this is rarely true. Aggregation is almost always required
at the distribution level and often it is desirable at the transmission level.
The modeling of the various network components, losses, maintenance and
quality of supply will be discussed for an aggregated network which does not
model individual lines.
Network Losses IIL.k(t)
Equation (7.3.6) gives 11L.k (t) as a function of bus injections. Thus in some sense,
this equation does not involve the explicit calculation of line flows, i.e., the
explicit modeling of individual lines. However, the computation of the "Bmatrix" of (7.3.4) does require manipulation of individual line parameters.
Hence, use of (7.3.6) can be viewed as disaggregated at a planning level but at
least partially aggregated at the operational level.
The buses used in (7.3.6) can be lumped together in various ways to reduce
the level of detail (i.e., the dimensions of the B-matrix). Some of the potential
problems that arise are discussed in Section 4.5. With enough aggregation of
buses, the network loss component becomes independent of customer locations,
i.e. has no k dependence.
Network Maintenance: IIM.k(f)
As discussed earlier, good data on variable network maintenance costs are not
readily available today. Hence even when network losses and network quality
of supply are handled at a disaggregated line-by-Iine level, an aggregated
approach to network maintenance may be desired.
173
If individual line flows are not being modeled, the market clearing approach
cannot be used and it is necessary to resort to the use of cost functions. By
analogy with Section 6.2 on the cost function approach to generation quality of
supply, two approaches are discussed:
o
Following the development of Section 6.2, one possible formula for I1Qs(t) is
eQs.k(t)b~(t)
IIQ5(t) =
(7.9.1)
bn(t) =
otherwise
Th6:problem with implementating (7.9.1) is the difficulty in specifying a reasonablesalue for dent y . One approach is to say that b~(t) = 1 any time any load is
not being served because of a network capacity shortage. This philosophy has,
however, many obvious shortcomings.
Cost of annualiZ'ed network investment
Section 6.2 yielded an equation (6.2.12) for YQs(t). The aggregated network
analogy to (6.2.12) is
(7.9.2)
LOLH~ =
LOLP~(t)
1=/
The difficulty associated with using (7.9.2) lies in the fact that practical models
for computing LOLP~(t) are not readily available today (except for special
cases). Therefore, we will follow the usual procedure when one doesn't have a
model: simply assume a reasonable form. The simplest reasonable form is
174
k[d(t) - d ]
otherwise
cnt,~
LOLP (t) =
"
Substituting (7,9,3) into (7,9,2) shows that the choice of a value of k has no effect
on "Qs(t) - a most fortunate occurrence. Thus to use (7,9,2) it is just necessary
to specify a value for derit,q' In some price studies we have used "engineering
judgment" to specify dcrit,q as some percentage of the maximum demand where
the percentage ranged from 80 to 95 percent.
When (7,9,3) is used, (7,9,2) becomes for d(t) > dcri"q
1JQS ( I)
AQs.,Jd(t) - dcr;,.'!]
-::8~760::::':":'---'---==
f~l
where the summation is only over d(t) > dcrit.q' A simple approximation to
(7,9.4) results when d(t) is viewed as a random variable whose probability
density is flat for dmox ~ dcri,.q' This point of view yields
p: Value of probability density for dm.. > del) > der;. .
I 8~ [de )
d]
[dm., - deri, . ]2 p
8760 ,~
t - er;I.. ~
2
(7,9,7)
175
The preceding discussions have considered only real energy. Reactive energy
flows can also be important since they affect both real line losses and voltage
magnitudes (see Appendix A). In practice, it is sometimes desirable to include
a spot price on reactive energy as well as a constraint on the allowable voltage
magnitudes at each bus. The preceding equations can be generalized by viewing
the energy and prices as complex numbers whose real and imaginary parts
correspond to real and reactive energy respectively.9 When this is done, a
nonlinear AC load flow has to be used to solve the network equations instead
of the linear DC load flow approximation used here.
SECTION 7.11. DISCUSSION OF CHAPTER 7
The basic principles used in this chapter are the same ones used in Chapter 6
where only generation was considered. However, the equations are now much
more complicated in appearance because of Kirchoff's laws, which determine
how line flows and losses come into play.
The biggest effect of the network is that spot prices become spatially dependent, i.e. they depend on where the customer is located on the network (unless
an aggregated network model is used). Under normal operating conditions, the
generation components of the hourly spot price usually dominate the network
components. However, when the network is heavily loaded, the network comp~ents can dominate.
;Jhe network components of the hourly spot price should never be ignored
entirely unless explicit computations have proven them to be negligible.
HISTORICAL NOTES AND REFERENCES - CHAPTER 7
nsns-
pie,
itly
enlest
The basic reference for Chapter 7 is Bohn, Caramanis and Schweppe [1984]. See
also Schwejilpe, Bohn, Caramanis [1985] Caramanis, Bohn, and Schweppe
[1986] and Caramanis, Schweppe and Roukos [1988] which address wheeling.
Dobbs [\983] looks at optimal spot pricing and finite line capacity.
Readers with a power system background in economic dispatch and optimal
load flow (with line flow constraints) will find that the equations of this chapter
look familiar. This is to be expected, since such power system theory motivated
their development. See Appendix B for further discussion on this relationship
to power system operation. Ponrajah [1984] is an example of a very close tie. The
work of Luo, Hill, and Lee [1986] dealing with bus incremental costs is related
to spatial spot pricing, but uses only system lambda and losses.
The network causes prices that vary spatially. Several previous results in the
economic literature have studied how public utility prices should vary over
space. Relevant models include Takayama and Judge [1977] (which was not
directed at electricity), Craven [1974], Dansby [1980], Scherer [1976, 1977], and
Schuler and Hobbs [1981]. All of these models are deterministic and most are
static. Agnew [1977] is stochastic but does not deal with electricity.
Scherer has the best model of electricity line losses and line constraints, and
includes T and D investment options. Scherer's approach is to use a mixed
176
8. REVENUE RECONCILIATION
TR
L L Pk(t) dk(t)
I~
(8.1 )
Define
Total Operating and Capital
Costs Incurred by Utility
t = I ... TR
(8.2)
The capital costs are the embedded costs (interest, debt payment and rate of
return of investment) of existing generation plants and the network as allocated
to the time interval t = I ... T R
Define
(8.3)
177
The goal of revenue reconciliation is to make A(TR ) close to zero in some sense.
Section 8.1 starts out by considering how to modify the hourly spot prices to
achieve revenue reconciliation which combines generation and network costs.
Two approaches are used in Section 8.1: the conventional one of "Ramsey
pricing" and the simpler to use but less elegant weighted least squares. Section
8.2 contains the "buy-back" version of Section 8.1. Later on, Section 8.5 returns
to the problem of Section 8.1 but with reconciliation decomposed into generation and network components or even all the way to individual transmission
Jines. Sections 8.3 and 8.4 present the surcharge-refund and revolving fund
approaches. Section 8.6 briefly discusses the role of fixed charges. Section 8.7
contains an interesting, but not critical side discussion on nonlinear pricing,
Section 8.8 concludes with a discussion on revenue neutrality.
SECTION 8.1. MODIFY SPOT PRICES: AGGREGATE RECONCILIATION
The first approach to revenue reconciliation discussed here modifies the formula
for computing the hourly spot price so that for some fixed value of TR (say one
year), the expected value of the revenue R(TR ) equals the expected value of the
costs C(TR ) where the expectation is an average over all equipment outages, fuel
costs, demand variations, etc. that occur over the time interval t = I ... TR
Define
CO(TR ): Administration, metering, billing, and other fixed but noncapital costs
over t = I ... TR ($)
C)'G(TR): Embedded capital costs of installed generation over t = I ... TR ($)
C~(TR): Embedded capital costs of instaIled network over t = 1 ... TR ($)
NM (TR): Network Maintenance costs over t = I ... TR ($)
GFM(TR): Fuel and generation maintenance costs over t = I ... TR ($)
GFM(TR ) = r.;~1 r.PFMJ{gj(t)]
(8.1.1)
Then
(8.1.2)
Define
{h(t): The hourly spot price developed in Chapter 7 for customer k during hour
A(t)
I'Qs(t)
AL.k(t)
I1M,k(t)
I1QS.k(t)
(8.1.3)
E{C(TR)}
8. Revenue reconciliation
179
This formulation assumes the formula for computing Pk(t) is a priori specified,
Le. before time t = I.
General Structure of Equation for Pk(t)
There are many ways to modify Pk(t) in order to achieve (8.1.4). However, the
following forms result from' all of the approaches to be discussed:
Multiplicative Form:
(8.1.5)
Additive Form:
(8.1.6)
(I
m)p,{l)
(8.1. 7)
(8.1.8)
-=--~--'-'--'--
TK
(8.\.9)
L L E{pdt)d,(I)}
I~
TR
a == E{C(TR )}
2:: L E{p,(I)dk(I)}
/0;;1
(8.1.10)
J..
YQs(l)
l1u(l)
I1M.k(I)
I1QS.k(l)
YR(I)
(8.1.11)
'IR.k (I)
m[A,(I)
YQs(t)]
(8.1.12)
m[tlu(t)
+ tlM.k(t) +
tlQS.k(t)]
The weighted least squares approach to deriving (8. I .5) (8.1.6) starts by defining
Measure of closeness _
of fit of Pk(t) to Pk(t) -
{r. L
1=1
[Pk(t) - I\(t)f}
2Qk(t)
(8.1.15)
Setting the derivative of (8.1.15) with respect to Pk (I) equal to zero yields]
(8.1.16)
which yields4
Pk(t) =
p,(t)
flRQ,(t)dk(t)[1
Ck(t)]
(8.1.17)
(8.1.18)
8. Revenue reconciliation
liB
Table 8.1.1. Weighted Least Squares: Effect of Choice of Weighting Function Q,(t)
IF
{ Q,(t) =
THEN
{ p, (I) =
d,(t)
p,(t)
p,(I)
d,(t)
Choice 2 yields additive form: a,(I) = I'R[I + ",(I)]
. Choice 4 yields Illultiplicative form: 111,() = I'R[I + ,()]
Value of PR varies with choice
E{C[TRJ} -
E{~ f
E{~ f Qk(t)df(t)[1
I\(t)d,(t)}
(8.1.19)
Ck(r))}
T~le 8.1.1 summarizes the equations for Pk(t) that result from (8.1.17) for
four;:aifferent choices of Qk(t).5 The motivation follows from (8.1.14), since the
valu~ of Pk(t) is closest to Pk(t) for customers k and/or times t when Qk(t) is
small~ Thus, for example, Choice 2 of Table 8.1.1 results in hourly spot prices
Pk(t) that are closest to Pk (t) when dk (t) is large so most revenue reconciliation
is done for cus~omers k and/or times t with small demands. 6 Such arguments can
be used to justify Choices I to 4 of Table 8.1.1. Note however that different
personal judgments can yield choices not in Table 8.1.1. For example, one might
argue that Qk (I) = dk (t) should be used so the fit of Pk (t) to Pk (t) is closest when
the demand is small.
There are three important points to note on the role of the elasticity Ck(t) in
Table 8.1.1. First, if ck(l) is a constant, independent of customer k or time I, (i.e.
if ek(l) = c), then the value of C does not have to be known since C always
appears with IlR and (l + c) can be factored out of the denominator of (8.1.19).
Second, if ek(l) = c, then for Choices 4 and 2 of Table 8.7.1,
mk(t)
m = tiR[1
ak(t)
a = tiR[1
+ cl
+ el
where of course the value of IlR changes with the choice of Qk(I). Third, since
the magnitude of Ck(t) is usually small relative to one, the actual value of Ck(t)
is not very important for any of the choices.
n.
J82
It is also important to note that the sign of Ji.R can change from year to year
(assuming TR = 1 year). For example Ji.R can go from negative to positive when
a nuclear power plant "comes on-line" and enters the rate base.
Throughout this book we assume Pk (I) > 0, since the case of a negative p(t) is
too rare to justify the use of the more complex notation of (8.1.20).
Ramsey Pricing Derivation
"Ramsey pricing" is a term often used in economics to denote the prices Pk(t)
which minimize "social cost" subject to the revenue reconciliation constraint of
(8.1.4). It is also known as "second-best pricing".
In Section 7.2, the following Lagrangian was developed to find the Pk(t) which
minimized social cost (see (7.2.2:
a(t) =
Gf(t)]
N[~(t)] - B[q:(t)]
Ile(t)[d(t)
L[~(t)] - g(t)]
For the present problem with expectations and a time interval I to TR , define
another Lagrangian by adding the revenue constraint of (8.1.4):
(8.1.21)
Setting the derivative with respect to Pk(t) equal to zero yields (after manipulation as in Section 7.2 and remembering that the Pk(t) of Chapter 7 is now called
Pk(t
(8.1.22)
8. Revenue reconciliation
Ul3
Table 8.1.2. Ramsey Pricing: Effect of Choice of Model For Derivative ad,(I)/apdt)
CHOICE OF
RESPONSE
FUNCTION
(Linear)
2
(Exponential)
3
(Log)
4
(Power)
0)
IF
ad,(t) =
ap, (I)
{THEN
p,(t) =
ct, (I)
lii,(I) - J.iRd,(t)ct,-I(t)](1
[Pk(t) - 11Rctkl(t)](l
ct,(fl/p, (I)
ddt)
ct,(I)-()
fl, I
(I
(I
11R)-1
J.iR)-1
Y'
_
Pk(l)
IlR
J.iRdk(t)
~
J.iR
ct~(t))
-I
p,(t)
, ~,(I) is a negative quantity which does not depend on tI, (I) or 1', (I),
, For (,hoice 4 ~,(I) = ", (I),
, Choice 4 yields multiplicative form: "', (I) = [\ + i'R + I'Ri", (t)]-',
is
')
If
(8.1.23)
capital investment etc. Care is needed in using the results since myopic applications of(8.1.23) run the risk of getting outside the range of validity of the model.
To illustrate, consider an industrial customer (the kth customer) who uses
one MWh during hour t at full capacity and receives a benefit of 50 /kWh from
the electric energy. Assume that during hour t, this customer has orders to fill
which require operation at full capacity, i.e. at I MWh. Then this customer will
behave as follows:
dk(t)
Hence Odk(t)/OPk(t) = when Pk(t) < 50 /kW. If,uR > 0, the result of(8.1.23)
is that the short-run Ramsey pricing says that this customer should pay 50/
kWh independent of Pk(t) provided Pk(t) < 50/kWh. A long-run model that
looked over many years and had a long-range price elasticity that reflected the
fact that charging 50 /kWh might cause the customer to move would yield more
reasonable results at the expense of even more difficult-to-implement equations.
Some of these "problems" associated with (8.1.23) go away if it is assumed
that the derivative is constant for all k in some customer class.
An interesting possibility results if the derivative is assumed to be constant for
all k (or all k in some customer class) but assumed to vary with aggregate
demand. It can be argued that elasticity goes down (in magnitude) with increasing demand. This yields a time-varying m(t) that increases the variability of the
hourly spot prices.
SECTION 8.2. BVY -BACK RATES
P,.buy(t) ddt)
Pk.seli
(8.2.1)
ddt) < 0
(8.2.2)
(8.2.3)
8. Revenue reconciliation
~a
1115
leI.
ses
(\ +
(\ +
)m
m"I\)Pk(t)
(8.2.4)
mhlly)l\(t)
(8.2.5)
fill
rill
3)
at
he
Ire
I1lscll
lS.
ed
or
te
lS-
he
and
~d
ve
11/,,11
le
~y
(8.2.6)
m)Pk (I)
Pk."I\(I)
(1
PUUy(t)
(\ - I1l)Pk(l)
(8.2.7)
2)
Pk.,,1\ (I)
(Selling to Customer)
PUlIy(l)
3)
~o
+ I1lIPk.sel\(t)1
+ I1llih.huy(t)1
(8.2.8)
where Pk.blly(t) is now a negative number (whose magnitude equals the Pk.buy(t)
of (8.2.7. The form (8.2.8) is more "symmetric" then (8.2.7).
We will use (S.2.7) (S.2.S) in subsequent developments. Thus readers who are
happy with hand-waving arguments can skip the rest of this section. However,
for those who like to see mathematical gymnastics, we will define some general
fomulas and then show how (8.2.7) (8.2.8) result as a special case. The mathematics is also useful in understanding what is going on and provides some
alternative equations which have potential use in certain applications.
Weighted Least Squares Derivation
holds whether elk(t) is positive or negative. However, care in the choice of Qk(t)
is required because it is necessary that Qk(t) > O. Thus, for example, Choice 4
of Table 8.1.1, should be replaced by
P. (I)
(8.2.10)
Qk (I) = Id (/)1
k
+1
-\
If the assumption is made that lek (t)1 = lei for all k and t, independent of
whether buying or selling, then (8.2.11) yields (8.2.4) (8.2.5) where (because c is
negative for usage and positive for generation)
mscll
mbuy
III -
1<:1}
i-lR (I
If the effect of
1<:1)
lei is ignored,
which is (8.2.7). Similar types of conclusions result from other choices of Q,(t).
Ramsey Pricing Derivation
The Ramsey pricing approach of Section 8.1 (minimization of social cost subject
to reconciliation constraint) yielded (8.1.23), i.e.
(8.2.12)
This is directly applicable for both buying and selling, since no assumption on
the sign of elk (I) was made.
In order to be more explicit, consider Choice 4 of Table 8. 1.2 which yields
8. Revenue reconciliation
187
r,
al
e-
(8.2.13)
Ie
(usage)
ek(t) ~ 0
(generation)
Sk(t)
})
r)
4
I)
As discussed in Section 8.1, the short-run character of our Ramsey formulation leads to the need for care in its strict application. A similar situation arises
in the present buy-back case. To illustrate, assume J..tR > 0 and consider a
customer-owned generator with maximum capacity of I MW and fuel costs of
IO/kWh, independent of amount generated. This means the customer will
behave as follows:
Customer generation
IMW
Customer generation
Hence the customer's elasticity ek (t) is "zero" when the buy-back price is greater
than IO/kWh. The result is that (8.2.13) yields 10
I)
. Pk(t)
p,(l)
IO/kWh
no
6k(l) =
ii;(t)[1
liR
+ ~~
sgn (d,(I)))
(8.2.14)
1+
lI1,cll
[I +
[I +
fIR
fiR
+ lei
fiR -
fiR
-
lei
JI
JI
(8.2.15)
+ J..tR'
In Sections 8.1 and 8.2 we specified a priori (before time = I) formulas for
hourly spot prices which yield reconciliation in an expected value sense over the
intervalt = I ... TR The "surcharge-refund" approach to be discussed here is
a posteriori.
For simplicity assume dk(t) ~ 0. Buy-back rates follow in a similar fashion.
The basic approach is to
o
o
Define
t.k(TR ):
t.k(TR )
TR
Llk(TR )
= m
L dk(t)Pk(t)
1=1
TR
L d(t)Pk(t)
t.(T)
,~I
Bill for t
I ...
TR
lR
L dk(t)Pk(t)(l
1=1
TR ($).
+ m)
n.
Hence this approach has yielded an effective spot price, (l + m)Pk(t), which is
identical to the multiplicative form of the modified spot price (8.1.7) of Section
8.1, except that in (8.3.2), m is specified a posteriori after time TR using Ll(TR)
while in (8.1.7) m is specified a priori before time 1 using E{Ll(TR)}'
An alternative approach is to make Llk(TR) proportional to the kth customer's
energy usage; i.e.
(8.3.3)
ss
eto
a.is
Ig
This yields an effective hourly spot price of Pk(l) + a, which is the additive form
of the modified spot price of Section 8.1, the only difference being the time at
which the constant a is specified.
Analogous ways can be found to allocate Ll(TR) among customers to yield
most of the formulas of Section 8.1 and Section 8.2.
A Priori vs. A Posteriori
:h
I)
2)
time interval TR , which is usually one year. We now consider the "revolving
fund" approach, which can be viewed as the result of assuming TR is extremely
large."
Set Pk(t) = Pk(t). Define as in (8.3)
~(T)
C(T) - R(T)
(8.4.1)
where now T is a running time index with one-year steps. The revolving fund
(referred to with a variety of different terms in the literature) works as follows.
o
If ~(T) < 0 (i.e. utility is received too much revenue), put the excess into a
"revolving fund" which is invested in the capital market.
If ~(T) > 0 (i.e. utility is receiving too little revenue), remove the funds from
the revolving fund (if it is positive) or if necessary, borrow the money from
the open capital market.
With this approach, the customers see Pk(t) = Pk(t) while simultaneously the
utility neither makes excess profits nor incurs inappropriate financial losses.
Define
T
RF(T) =
L ~(s)
(8.4.2)
s= I
In practice, (8.4.2) would be modified to reflect the interest received (paid) when
RF(T) is positive (negative).
The equations of Sections 8.1 through 8.3 keep ~(T) close to zero by modifying the hourly spot price. With the revolving fund approach, the hourly spot
price cost is not modified either a priori or a posteriori. Instead, the utility
generation and network investment policy is used as the mechanism to try to
keep RF(T) close to zero over a period of many years.
The basic fact motivating the revolving fund is
If the utility has an "optimum" generation and network system, revenue reconciliation
is automatic since marginal cost prices exactly cover the operating and capital costs (see
Chapter 10).
8. Revenue reconciliation
'191
Sectf(;n 8.1 used a single revenue reconciliation constraint that contained all
gener~tion and network costs. An obvious variation is to introduce multiple
revenue reconciliation constraints such as to
Achieve revenue reconciliation separately for generation and the network by using two
constraints,
'!
or
Achieve revenue reconciliation separately for each generator and each line by introducing a separate constraint for each generator and each line.
It can be argued that capital costs are "more equitably" distributed among
the customers when such decomposed revenue reconciliation is done for different parts of the system. For example, the capital costs of a major new
generating plant or transmission line would then be borne most by the customer
who makes "most use" of the new facility.
There are, however, disadvantages to the decomposed revenue reconciliation
approach. If the goal of revenue reconciliation is to keep Pk(t) as close as
possible to Pk(t), then the addition of extra constraints helps defeat this goal,
since they will increase (or at least not decrease) the "distance" between Pk(t)
and pdt). Thus there is a tradeoff between equity (customers bear "fair" share
of costs) and economic efficiency (Pk(t) close to Pk(t.13
192
Decomposition of Revenue
L [yet)
,=1
yet)
A(t)
YR (t)] get)
(8.5.1)
YQs(t)
R,,(TR) =
L L ["k (t)
,=
1 k
YR (t)L(t)
(8.5.2)
because since
get)
L gj(t)
L(t)
g(t) - Ldk(t)
(Total generation),
(Total Losses),
it follows that
(8.5.3)
L [yet)
YR (t)] gj(t)
1=1
and
R~.;(TR):
R",,(TR ) =
L L ["k,;(t)
1=1
- L;(t)[y(t)
"k.;(t)
"R.k.;(t)] dk(t)
"l,k,;(t)
YR (t)J
+ "M.k,;(t) +
"QS.k.;(t)
where
Li(t): losses on line i
1JL.k.j
[).(t)
+ YQs(t)]
aLj(t)
adk(t)
etc.
Using such equations, it is possible to compare the revenues "obtained" from
different components of the system to the costs of those components. This can
be useful in helping to guide future investment decisions even if aggregate
revenue reconciliation is done.
Separate Generation and Network Reconciliation
erR ):
C).
Total costs associated with generation
C./TR ): Total costs associated with the network
The problem 1's to define the revenue reconciliation components i'R (t) and YJR.k(t)
of the spot price so that
E{C\.(TR )}
E{R\.(TR )}
(8.5.7)
E{C./(T~)}
E{R,/(TR )}
(8.5.8)
)'Qs(t)] = my)'(t)
(8.5.9)
so that
(8.5.10)
The use of the absolute value of 11k (t) is motivated by the form of the buy or sell
equation of (8.2.8) and the fact that I1k(t) can be positive or negative.
Substituting (8.5.9) into (8.5.7) and (8.5.8) yields the conditions
TR
+ my)
(I
I~
E{r(t)g(t)}
(8.5.11)
TR
I L E{[lJk(t)
I-I
m~llJk(t)lJdk(t)}
TR
- (I
my)
E{r(t)L(t)}
(8.5.12)
I~I
m~.
Note that
m~
depends
(Amount ReceiVCd)
(
Amount Paid to
)
from Customers
Gencrating Department
I I p,(t) ddt)
I
(I
1Il;.>(t)g(t)
Lkdk(t)
Net)
( Right Hand Side of (8.5.12) )
without Expectation Operator
( Revenue =
If the values of fJ.? and m" are not equal, then the use of the aggregate
reconciliation procedure of Section 8.1 can be viewed as causing the Generation
Department to "subsidize" the Network Department or vice versa. Since my and
mil can have different signs, the Generating Department can overrecover while
the Network Department is underrecovering.
8. Revenue reconciliation
,ell
II)
195
The developments of Section 8.1 assumed the total costs C(TR ) of (8.1.2) are to
be recovered (in some sense) by the hourly spot price Pk(t) where C(TR ) included
CO(TR ), the administration, metering, etc. costs. In practice it can be more
desirable to recover some of these COCTR ) costs such as metering and billing in
terms of a fixed charge so that
MonthlY)"
Bill
=
Pk(t)dk(t)
't'
.
FIxed
Charge
12)
Ids
ity
red
Irk
ice
It'S
(t)
It'S
Define
Rk(t): Revenue received by utility from customer k using dk(t) during hour t. ($)
In earlier discussions, the revenue was constrained to have the "linear" structural form of
ate
ion
md
lile
(8.7.1)
n(TR )
[GfM[g(t)] -
IlR[C(TR) -
Bk[dk(t)] - Ile(t)[g(t) -
(~I
dk(t)]]
TR
I I
t= 1
Rk(t)]
(8.7.2)
(8.7.3)
Setting
).(/) = aGFM[g(/)]
ag(/)
(8.7.4)
(8.7.5)
Rk O
).(t)dk(t)
(8.7.6)
8. Revenue reconciliation
197
(8.7.7)
C(TR ) = L,LkRk(t)
2)
is
to
Thus the Rk(t) of (8.7.6) is the revenue which minimizes social cost subject to
both energy balance and revenue reconciliation constraints.
The constants R k O in (8.7.6) can be chosen in any way that satisfies (8.7.7)
provided only that they satisfy the condition
t = 1 ... TR
(8.7.8)
Define
R(t): Total revenue obtained by utility during hour t ($)
R(~= LkRk(t)
Ro +
A(t) d(t)
(8.7.9)
I)
(1
m)A(t) d(t)
(8.7.10)
Figure (8.7.1) plots R(t) of (8.7.9) and (8.7.10) versus d(t) assuming ~ and mare
greater than zero, i.e. underrecovery in the absence of revenue reconciliation.
Figure (8.7.1) provides motivation for using a "nonlinear" pricing scheme
that yields a R(t) vs. d(t) curve which looks more like the ideal R(t) of (8.7.9).
One approach is to hypothesize a nonlinear polynomial
(8.7.11)
Total
Revenues
R(t)
($)
Ideal Revenue:
Slope = A(t)
where the Pk.1 (t), Pk.2(t) etc. are found by repeating the derivation of Section 8.1.
A more explicit nonlinear pricing function motivated by Figure 8.7.1 is
ex
>
(8.7.12)
For a fixed value of r:t. (based, for instance, on judgment), a value for Po can be
found. Using (8.7.12), revenue reconciliation impacts more on small demand
levels. Obviously one can also hypothesize nonlinear pricing structures where
revenue reconciliation impacts more on large demand levels, etc. Equations
(8.7.11) and (8.7.12) are only two examples of the many possible nonlinear
pricing structures.
It should be noted that the developments of Section 8.1 and 8.2 started with
the hypothesized "linear" form of (8.7.1), but can yield nonlinear prices. Table
8.1.1 and 8.1.2 showed that some choices of the weighting function QR (t) (for
weighted least squares) and the derivative od(t)/op(t) (for Ramsey) yielded Pk(t)
which are functions of dk(t), i.e., nonlinear pricing structures.
We do not pursue these nonlinear pricing ideas further because they complicate the analysis without any proven real-world advantages. The resulting
formulas generally depend strongly on the hypothesized customer benefit
function Bddk(t). However it is important to emphasize that
The basic principles of hourly spot pricing and the energy marketplace apply to nonlinear
as well as linear pricing structures.
8. Revenue reconciliation
199
The preceding sections have discussed revenue reconciliation for a full implementation of a spot price based energy marketplace. During the initial testing
and implementation phase, it may be expedient to do revenue reconciliation by
the concept of "revenue neutrality" where
A spot price sequence Pk (t), t = 1 ... 8760 is said to be revenue neutral if it recovers the
same annual revenue that would be recovered under a present rale, assuming no change
in the kth customer's behavior, i.e., no response to price charges.
Define
= 1 ... 8760: Spot price sequence expected next year with no revenue
reconciliation.
Pkft:> = (I + m)Pk(t): Revenue neutral spot price sequence.
Pk(t), 1
~-
It
+ m)R2
Bill, - (I
(8.8.1)
8760
R2
1=
df(t)ih(t)
I
~-I
R2
(8.8.2)
I
k-l
[Billk - (I + m)R2]2
Qk
(8.8.3)
200 . . n. Tneofyof
L (Bill k)R2!Qk
1
(8.8.4)
An Extension
L [1
met, a)p,(t)dZ(t)
+ ....
~:
ao
a" sin
~)
wt
a2 cos
wf
+ .,.
where the fundamental frequency is 24 hours, one year, or whatever does a good
job.
We must emphasize that revenue neutrality as it is being discussedhere is not
a fundamental approach to revenue reconciliation. It is merely an expedient
which may prove useful during intitial testing of spot pricing.
SECTION 8.9. DISCUSSION OF CHAPTER 8
~)
8. Revenue reconciliation
20t
Many of the aggregate reconciliation ideas of Sections 8.1 through 8.4 are based
on the extension of existing ideas into the time scale of hourly spot prices. The
weighted least squares concept in Section 8.1 appears to be new. The disaggregated reconciliation ideas of Section 8.5 are based on Schweppe, Bohn and
Caramanis [1985].
The general problem of efficiency constraining prices to meet a budget con-
202
straint has been vigorously debated in the economic literature. Hotelling's [1938]
article considered the problem of financing public works such as bridges where
the marginal cost of crossings is usually trivial. His answer to the pricing
problem was to set prices through taxes which (ostensibly) would not distort
consumption decisions, such as income taxes or inheritance taxes. Coase [1946,
1970] argued that from a broad public policy perspective, user support was an
important market test for efficient allocation of resources, and thus fees should
cover the total cost of the enterprise. He suggested the use of multipart tariffs
(such as declining block rates or a fixed fee plus a commodity charge) as an
alternative to government subsidies. Vickrey [1955] stressed that a misallocation
of resources can result if marginal cost pricing principles are not followed.
Baumol and Bradford [1970] proposed optimal departures from marginal cost
pricing with a generalization of Ramsey's [1927] rule. A much discussed special
result of their analysis is the "inverse elasticity rule":
If the cross elasticities of demand between the commodities in question are zero, then the
percentage deviations in price from marginal costs should vary in inverse proportion with
the own price elasticity of demand.
If cross-elasticities are not zero, a somewhat analogous rule still holds. More
recently, proposals for nonlinear pricing or multipart tariffs (see e.g., Willing
[I978]) have been suggested to be Pareto superior to the Baumol and Bradford
rules.
Peddie et al. [\983] address revenue reconciliation with variable elasticities
directly in a spot pricing context (although called dynamic pricing).
NOTES
I. E( C( T R )} in (8.IA) is assumcd to be a given number because this corresponds to the way revenue
reconciliation is done for most utilities. An alternative approach is to consider thc fuel cost
component GFM (TR ) of C( T R ) to be a variable which depends on the g,(t). which in turn depend
on the Ii, (I). which in turn depend on the (I, (t). This approach leads to a slightly different set
of equations for p, (I). If a reader prefers this approach, it is relatively straightforward to modify
the equations appropriately.
2. Remember that Ii, (I) depends on pdl) so that, in general, (8.1.9) and (8.1.10) are not explicit
equations.
3. We ignore cross-elasticities and assume d, (I) docs not depend on p, (T) T #- I. We also ignore
the fact that I', (I) depends on d, (I) which in tUI'll depends on p,(I). The extra terms resulting
from such dependence could be carried along if desired.
4. The expectation operator is used only in evaluating JIlt.
S. If Q, (I) = Q, independent of time. demand, or price, then (8.1.17) and (8.1.19) show that p'(l)
does not depend on Q.
6. This argument is valid if demand response evidence indicates that demand is elastic during peak
periods and nonelastic during off-peak periods. Hence prices deviating from optimal spot prices
during ofT-peak periods do not affect customer behavior appreciably.
7. As in (8.1.16), we ignore cross-elasticities.
8. In the future, when a lot of customer response data become available, this statement may
change.
9. This is the "inverse elasticity" rule that results from Ramsey pricing.
8. Revenue reconciliation
203
10. A more elegant mathematical argument which leads to the same result would introduce a
Lagrange multiplier to handle the explicit constraint of 1 MW on the customer's generation
capacity.
11. An interesting academic exercise is to go to the other extreme and assume TR = I hour in
Section 8.1. This effectively destroys any resemblance to marginal cost pricing.
12. It should be noted on the other hand that overcapacity generally implies low spot prices. Hence
if spot prices are left unchanged, they would reinforce load growth and hence increase the chance
that revenues will catch up with revolving fund deficits.
13. Other problems with the separate revenue reconciliation approach applied to individual generators or lines can arise from old equipment whose capital costs are already "written off."
14. The mathematics of weighted least squares can be used to yield (8.5.9) and (8.5.10) in a relatively
straightforward fashion. However, we have not developed a corresponding Ramsey pricing
methodology.
~"
Q!tapters
6 through 8 presented the theory underlying the hourly spot price
defified as follows:
Hourly Spot Price: Marginal value of energy ($/kWh) for the next hour computed at the
beginning of the hour assuming complete knowledge of the operating conditions, costs,
etc., that will ex,ist during the hour.
Here in Chapter 9, we discuss how to compute spot price based rates - i.e.,
transactions which are based on the hourly spot price.
Sections 9.1 through 9.4 discuss transactions with different time scales and
types of contracts between the utility and its customers:
o
206
Sections 9.5 and 9.6 return to an hour-by-hour time scale and discuss
wheeling rates, i.e., the theory of how to charge for transmission services.
Section 9.5 discusses spot wheeling rates that two nonadjacent utilities ought to
pay for trading electric power using the transmission networks of other interconnected utilities. These wheeling rates depend on the hourly spot prices
prevailing at the boundaries of each utility. Section 9.6 considers wheeling
between a user and a private generator or a user and another utility. The key
issue here is how revenue reconciliation for generation is handled.
The only ideas developed in Chapter 10 from Chapter 9 come from Section
9.1, and they are not essential to the main theme of Chapter 10. Therefore
readers can take one "giant step forward" to Chapter 10 if the concepts of this
chapter are nol of immediate interest.
SECTION 9.1. PREDETERMINED PRICE-ONLY TRANSACTIONS
The framework used for the discussion is a marketplace where some customers
see hourly spot prices while the remaining see predetermined rates.
Costs and benefits of generating and consuming electric power at a specific
moment in time are assumed to be independent of past and future generation
and consumption levels in order to simplify the derivation (see Section 6.5).
Revenue reconciliation as discussed in Chapter 8 is not included here. It can be
added if desired at the expense of notational complexity. For simplicity, we
discuss only the case of customers buying energy from the utility. However, the
results generalize to the case of customer-owned generation that is being bought
by the utility.
Problem Formulation and Notation
For a 24-hour update (see Chapter 3), r might be 4 PM while t varies 24 times
with one-hour steps starting at 2 AM the next morning. However, to simplify
metering, a given implementation might keep Pk(tlr) constant over some range
of t. A three-hour example is-
In general symbols'
- Pk(tlr)
<X
t(r)o
207
I ct::. I(r): Is mathematical notation for all the hours I belonging to the set t(r).
JSS
es.
to
For the three-hour example, the set t(r) contains 2 AM, 3 AM, and 4 AM and
T(r) = 3.
To simplify the appearance of subsequent equations, the t and t notation is
often suppress and we define
~r
:es
ng
ey
rA = PAUl,): r, is the value of the spot price based rate predetermined at time, and valid
for I ct::. I(r).
rs
1c
m
The results generalize readily to the case where different classes of customers see
predetermined rates with different update times (as determined by r) and period
definitions (as defined by the set t(r.
General Result for Predetermined Rates
i).
Tbihourly spot prices, pdt), and predetermined rates, r k , seen by each of the
tw~classes of marketplace customers are shown later in the section to be given
)e
by'~
,e
(9.1.1)
le
It
L
lFI(1)
E { {ik(l) -"-.-
L
In
es
fy
~e
I f 1(1)
(""
E {(I(~f'(
tl}
(9.1.2)
( , J...
A more accurate (and classy) notation would use E {Pk(t)!r} instead of just
E {Pk (t)}, etc.
When the set 1(1') contains only one hour I, (9.1.2) can be written
--a,:;E{ild!(.[)}
+ ----"--;----:-"--<orA
(9.1.3)
Cov{a, b}
The basic result (9.1.2) is complicated by the presence of the derivative which
depends on the choice of a customer response model.
If the derivative iJdf(t)/iJrk is constant in time and deterministic, then (9.1.2)
becomes
p:
Elasticity Parameter
ro)
Assume P and ro are deterministic and constant in time. Substituting the exponential or power response models into (9.1.2) yields
E{Pk(t)dk(t)}
rX;f(r)
(9.1.4)
while the linear and log response models yield a similar result with the nominal
dO.k(t) replacing dk (t).2 When the set t(r) is one hour, (9.1.4) becomes
Thus the predetermined rate rk can be greater than or less than the expected
value E {Pk (I)} because the covariance can be negative or positive for a specific
customer.
It should be noted that since dk(/) depends on Pk(/), socially optimal marketplace behavior of hourly spot price customers can be achieved by appropriately
selecting either dk(/) or Pk(t). The same is not necessarily true with df(t) since
the predetermined rate rk can be constant over many hours. This limitation in
the degrees of freedom results in reduced social welfare, which should be
compared to the savings in metering, communications and transactions costs.
Derivation of General Result (9.1.1) and (9.1.2)
The definitions of Chapter 7 are used, except the demands are decomposed into
the two classes of customers: those that see hourly spot prices, and those that
see predetermined rates. Define (in addition to the definitions of Chapter 7)
df(t): Demand of kth customer seeing predetermined rates (kWh)
dP(/) = "f.kdt(/)
dP(/): Vector of all dk(t)
B[~P(t)]: Total Customer Benefit for the customer class seeing predetermined
rates.
Asrwme a bus k may have at most one generator and one customer from each
"'" classes. As in Chapter 7, assume for simplicity that the swing bus has
of th~wo
only"!t generator. Thus the individual line flows are given by
d(t)
dP(t)
L[~(t)l
(9.1.6)
L
fOCI{r)
E{!l(t)}
prices
where
~P(l);
nct)
i.e.,
O(t) =
is the
nCt)
G[,[(t)]
N[::(t)]
- B[~(t)J
- B[~P(t)]
/1,(t){d(t)
Since generation dispatch and hourly spot decisions are made on-line whereas
predetermined rates are set in advance, the first order optimality conditions are
aO(t)
(9.1.7)
=
(9.1.8)
aN[z(t)]
adk(t)
[aL[z(t)]
)le(t) 1 + - a(d
)
k
(9.1.9)
L
IEIV)
E{[aN[Z(t) - aB[df(t)]
Mf(t)
Mf(t)
)l (t)
c
(I +
aL[Z(t)])] Me(t)}
adf(t)
ark
L E { [Pk(t)
Me(t)}
- r.) -iJ-
IE/(r)
rk,
(9.1.10)
The relationship between predetermined price-only rates and hourly spot prices
has been derived without modeling rationing costs. In practice, predetermined
rate customers will sometimes be rationed by the utility on occasions of generation or network capacity shortfalls.
This is a reasonable conclusion. After the hourly spot price reaches a certain
level, it is socially better to drop a predetermined rate customer than to raise
further the price seen by hourly spot price customers.
Derivation of (9.1.11)
Actual consumption level of predetermined rate customer k after rationing. If there is no rationing, ~k(l) = df(t)
~(t)
Lk~k(l).
df!) -
~k(l).
(9.1.12)
L 'E{n(t)}
IE/(r)
net) = G[~{t)l
N[~(t)]
B[~(t)
B[~P(t)
11c(t)[d(t)
- g(t)
~(l)
L[:.(I)J
(Energy Balance)
+ L Rk(df(t)
k
- 6 k (t
Considering that generation dispatch, hourly spot prices and rationing (if
needed) are on-line decisions while predetermined rate setting is an off-line
decision,
on(t)
ogj(t)
on(t)
OPk (t)
on(t)
06 k (t)
(9.1.14)
(9.1.15)
I~) E {o~:t)}
(9.1.16)
(9.1.17)
Equations (9. J. 14) and (9.1.15) yield (9.1.1). Equations (9.1.16) and (9.1.1 7)
yield, after collecting terms using (9. I. 9),
pdt)
oRk
a6 (t) - IlR.k(t) =
k
~E {[[pdt) +
aRk
(9.1.18)
a6 k (t)
oB(df(t
odf(t)
(9.1.19)
No rationing when
(9.1.20)
o
213
Thus there is always a ceiling to the hourly spot price when a relatively large
segment of marketplace customers see predetermined rates. The ceiling is the
marginal rationing cost to the predetermined rate customers. When rationing is
done through a nondiscriminating procedure such as rotating blackouts or
brownouts (as opposed to a priority list in order of increasing rationing costs),
the spot price ceiling might be the average customer cost of unserved energy.
In Chapters 3 and 6, several different ways to specify the quality of supply
components of the hourly spot price were discussed, two of which were
Market Clearing: Change the hourly spot price until customers respond enough
so that rationing is not required.
Cost of Unserved Energy: Use quality of supply cost function based on cost of
unserved energy.
The ceiling demonstrates a relationship between the market clearing and the
cost of unserved energy approaches.
SECTION 9.2. PRICE-QUANTITY TRANSACTIONS
T~price-only
transactions of Chapters 6 through 8 and Section 9.1 are fixedpriCe-variable quantity contracts wherein the utility specifies a fixed price (for
thenext hour in Chapters 6 through 8 or for hours, days, months, etc., in Section
9.I}and the customer can buy any amount of energy at the quoted price. In this
section (and in Section 9.3) we explore other types of contracts which involve
some specification of quantities as well as prices.
Motivation for Price-Quantity Transactions
2n
The formulation of the hourly spot price has so far considered generation and
consumption costs and benefits, transmission line flows, and energy balance,
assuming that all relevant uncertainty is known at the time of price setting.
System security control introduces additional considerations, some of which can
be summarized in the following two issues: 3
o
2i5
The above two issues are systemwide. The only contingency planning requirement in Chapter 6 was an operating reserve requirements constraint. Utilityowned generators were assumed (see discussion in Section 7.8) to be dispatched
subject to the operating reserves constraint while other participant individual
decisions ignored it. This assumption is now removed by treating operating
reserve requirements as an additional commodity. Marketplace participants,
consumers and non-utility-owned generators can also contribute towards
meeting system security planning and control requirements. Therefore their
costs in preparing to supply resources (contingency planning requirements
participation) and their costs in actually supplying them, if needed, should be
distinguished and modeled. For example, a generator that contributes to operating reserve requirements (a common contingency planning method) incurs costs
for maintaining an operating state, and possibly opportunity costs for not
generating at a higher capacity so that it will be capable to make its reserved
capacity available if needed for security control. The same generator will incur
additional fuel and maintenance costs if it is called upon to actually utilize its
spinning reserves for security control implementation. Marketplace participants
are also responsible for the level of resources needed for effective contingency
planning and control implementation. For example, a large generator imposes
a higher burden on operating reserve requirements than a smaller generator.
The major results that follow when security control uncertainty is modeled
are~iscussed next.
Prii;., with security control modeling
The units of the above quantities are energy. A more precise characterization
. might be capacity available for a certain duration. For notational simplicity, we
assume that security control implementation requires capacity for a fixed a
priori known duration, and hence energy units are appropriate.
The model used in previous chapters is augmented by the addition of two new
constraints,
l>,.p
Sp
(9.2.1)
S,
(9.2.2)
L 5,.,
k
and also the inclusion of the relevant costs of participants for supplying Sk.p and
The constraint g(t) ~ gcrit,y(t) imposed in earlier formulations will always
be met if inequality (9.2.1) is satisfied. The Lagrange multiplier /-lQSs (of the
earlier formulation) is zero when enough generating capacity to meet load and
operating reserves is available. The Lagrange mUltiplier /-lp will be zero much less
often since operating reserves must be always secured.
The reader should note that gcrit.y(t) which was defined as a constant in Section
6.2 (i.e., gCrit.y(t) = gmax(t) - gres(t is actually dependent on individual generation levels. Indeed gres(t) in (6.2.1) is equivalent to Sp of this section which
depends on generation levels. For example, gres(t) may be considered to equal
the output of the largest generator at time t. In the simpler formulation of
Chapter 6, gres(t) was assumed to be a constant and hence
Sk.c'
for all k was used in deriving (6.3.6). This simplifying assumption is removed
here and Sp is generally considered to depend on the gk levels.
After the usual formation of the Lagrangian and optimization with respect to
4
Sk.p and Sk,c' the following results are obtained.
For a generator:
(9.2.3)
For a demand:
(9.2.4)
where
Ih(t): Normal operating spot price as in Chapter 7
Pp: Contingency planning spot price; equals /-lp
Pc: Security control implementation spot price; equals /-lc
Skc
otherwise
To see how security control prices work, consider a customer (load) who
pledges all of his/her demand. Thuss
If Skc
0,
If Skc
Skp'
Therefore if
dk =
dk =
Skp
ad
ys
he
Ild
:ss
[Pk(t) - PpjSkp
a-
lal
of
(9.2.5)
[- Pp - PcjSkp
)n
:h
217
The contingency planning spot price Pp (which equals the Lagrange multiplier
J1.p) has an intuitive interpretation when the utility provides alJ required operating reserves by spinning standby generators. In this instance Pp is the incremental cost incurred by the marginal spinning generator k in supplying the last kWh
of operating reserves Skp' In equation form,
~d
to
3)
4)
Depending on the characteristics of the marginal contributor to system operating reserve requirements, Sp, Pp may be quite small and at times may even be
zero.
The result (9.2.3) (9.2.4) has a form which is similar to the familiar results of
previous chapters where the various additive components of the optimal spot
Pfice correspond to modeling a specific marginal cost imposed by an increment
the participant's consumption or generation level. In this case, each participant is charged/paid for its incremental impact on system contingency
planning reserve requirements and system security control implementation
requirements. It should be noted that most loads do not affect Sp and Sc. Thus,
except for very large loads,
if
However, for large generation the last two terms of (9.2.3) can be important.
For example, if gk is the largest generator's output at time t, security control
contingency planning could require Sp = gk> which results in
10
thus implying that the kth generator should be "charged" at the rate of pp.
Similarly, if generator k is suddenly forced to disconnect from the grid causing
the need for Sc = gk> it should be charged at the rate of Pc' The spot price
218
modification of (9.2.3) internalizes security control costs and can thus be viewed
as providing the proper incentives for generators to behave in a way that
minimizes system security related costs.
Implementation of security control prices
The general security control approach involves a price reduction Pp for pledges
and a postpayment Pc for actual control. The resulting bill for demand is
given by (9.2.5). Two simpler cases are
Skp
Prepayment Only:
o
(9.2.6)
Pho
Postpayment Only:
o
Peo: Postpayment only control implementation price.
(9.2.7)
:d
It
Values of Pbo and Pco can be chosen by matching the expected values of the bill
of (9.2.5) to those of (9.2.6) and (9.2.7).
To illustrate how this can be done, define
1'l
.e
= 0,
Then the expected bill, E{Billd, for the general case (9.2.5) is given by
:t
e
e
(I - n)[ - Pp - .okclskp
skp[nPk(t) - Pp - (I -
:-
(9.2.8)
n)hcl
.okC: Expected value of Pc over all values exceeding the reservation price of customer k
,0
For the prepayment only case, the expected bill from (9.2.6) is given by
(9.2.9)
v
Equating(9.2.8) and (9.2.9) yields
r
Pbo =
I
- [pp
+ (\ - n)/\cl
(9.2.10)
the postpayment only case, the expected bill from (9.2.7) is given by
(9.2.11)
When considering the results (9.2.10) and (9.2.12), remember that for most
situations the probability n of no control will usually be close to one.
Many present-day interruptible contracts are examples of prepayment only,
where
Interruptible Rate = Pk(t) -
Pbo
Hence a consistent setting of such rates should follow the procedure outlined
above if the security control philosophy is being used.
A new pricing proposal (Chao and Wilson [\985]) related to post payment
only is interruption insurance, in which the utility pays customers who are
interrupted. One difference is that the proposed insurance payment rates differ
across customers, while in the energy marketplace proposed here, the same Pc
applies to all customers. The two proposals are alike, however, in that they both
cut off exactly those customers who are most willing to be interrupted.
Reduced Transactions Costs Motivated Price-Quantity
Noting that
we define further
rk,O
Pk(tlqk = 0)
rk,l
221
Postpayment Only: Charge the interruptible customer the expected spot price
rk but offer a payment when interrupting at the rate i'k.l so the expected bill
is
k O
so the
The pricing structures for the above prepayment only and postpayment only
approaches are identical to the corresponding security control structure of
(9.2.6) and (9.2.7). However, the underlying logic for computing the price is
basically different since in the present case the price-quantity transactions are
really just a way to implement a price-only transaction.
Multiple Levels of Reliability
The basic ideas can be extended in many ways to yield quite sophisticated
utility-customer transactions. An interesting extension involving varying levels
of reliability is the following. A utility could offer its customers a choice in the
level of reliability they want to buy so that customers who pay a lower rate are
lo~r in the priority ordering and thus get interrupted more frequently.
Direct Load Control
Assume tnat for a particular day, the utility knows a priori that the customer's
needs can be represented by the constraine
24
L d(t)
dreq
1=1
where dreq may depend on day of week and/or outside temperatures (for space
conditioning). Then the utility will obviously control the device's d(t) to meet
the constraint while minimizing
ZZ2
where Pk(t) is the utility's marginal cost of providing energy to the customer.
The flat rate the utility offers the customer is based on the expected values of the
Pk(t) and the optimum d(t); t = I ... over say a one-month interval.
Such direct utility control of water heating and/or thermal storage can be
viewed as a utility-provided tactical control service (See Section 4.3).
Discussion of Price-Quantity Transactions
Price-quantity transactions are based on the behavior of the hourly spot prices.
Many price-quantity transactions are very close to or equivalent to special types
of price-only transactions.
In general we question whether there are actually significant transactions
cost reductions from such transactions, compared with just putting the
customer onto a standard price-only spot price, perhaps with only two price
levels. The metering, communications, and customer response costs will be
close, or even lower for the spot price since customers do not have to pre-select
and communicate the quantity information. Hence for large (non-residential)
customers, we will be surprised if transactions cost savings outweight the
reduced benefits of price-quantity transactions.
However, price-quantity transactions based on the utility system's needs for
security control are basically different than price-only transactions, and fill a
unique role of fast and accurate response. A large practical difference between
security control motivated price-quantity transactions and present interruptible
contracts or interruption insurance proposals [Chao and Wilson, 1985] is the
duration of advance commitment required from participants. Most customers
(and generators) have time-varying and stochastic willingness to provide
reserves, as well as stochastic time-varying overall demand. For example, an air
conditioning load can provide more reserves on a hot summer day while it can
provide no reserves during the winter. Industrial consumers such as electric
furnace factories are affected by the overall business cycle, specific orders in
hand, shift changes and the physical production cycles of their equipment. It is
therefore inefficient to force them to precommit to a fixed level of interruption
for a year as is the practice with many interruptible contracts. Hence, the
reserves market should be frequently resolved for the appropriate group of
participants. A reasonable time scale is roughly one order of magnitude longer
than the duration of reserve use. For example, "spinning" type reserves may be
beneficially reauctioned each hour or possibly every few minutes. Of course such
auctions would be conducted computer to computer, using standard response
logic set by the customers to place their bids.
SECTION 9.3. LONG-TERM CONTRACTS
223
contract between the utility and its customers. The price-quantity transactions
of Section 9.2 combine price- and quantity-based transactions (such as interruptible) where these quantity transactions operate within a price-only update
cycle.
Here in Section 9.3, we combine price-only transactions with a fixed-pricefixed-quantity contract that works on a slower time scale than the corresponding price-only update cycle. These long-term fixed-price-fixed-quantity
contracts are motivated by customer desires to know in advance what their
energy bills are going to be.
Two basic implementation approaches are
Utility Long-Term Contracts: A fixed-price-fixed-quantity long-term contract
between the utility and one of its customers.
Futures Market: Fixed-price-fixed-quantity futures market contracts between
buyers and sellers as negotiated by an independent energy broker.
A futures market will evolve only if there is sufficient demand for its existence.
Basics of Fixed-Price-Fixed-Quantity Transactions
Assume the fixed-price-fixed-quantity contracts are being combined with onehour update spot prices. Extensions to other predetermined price-only transac,tions such as 24-hour update (and to price-quantity) is conceptually straightfqJ;.ward.
The bill for energy used during hour t by the kth customer with a fixed-pricefixed-quantity long-term contract is
Bill, (I) = Pk (I)[d, (I) - du(t)] + dk.r(t)fk(t)($)
(9.3.1)
The values of dk,r(t) and/k(t) are specified hours, days, weeks, months or years
before hour t .
Equation (9.3.1) illustrates the two key features that result when a price-only
one-hour update is combined with a fixed-price-fixed-quantity contract:
o
If the actual demand equals the quantity fixed in the contract, i.e., if
dk(t) = dk,r(t), then the bill is independent of Pk(t), i.e. depends only onlk(t).
The customer's decision at hour t on how much energy to use (i.e., the value
of dk(t)) is based solely on the hourly spot price and is independent of either
dk/(t) or Ik(t).
To illustrate the second feature, let Bk[dk(t)] denote the customer's benefit
function so at hour t the customer chooses dk(t) to maximize
Bddk(t)] - BilIk(t). This yields a dk(t) given by
or negative.
The Billk(t) can be negative if, for example, dk(t) = 0 andh(t) < Pk(t).
For the futures market-energy broker implementation, there are various
possible ways the cash could flow. One approach is
o
o
Another approach is
" Customer pays Utility Pk(t)dk(t)
o Customer pays to or receives from Broker (h(t) Pk(t dk;(t)
The following subsections address the issues of how a customer chooses a
value of dk,f(t) and how the utility or energy broker specifies a value for h(t).
Customer Choice of dk,r(t)
Leth(tlr) be the priceh(t) as specified at the much earlier hour r. The customer
at hour r specifies the value of dk,f(t) given knowledge of h(tlr). Two cases of
interest are
Case I:hUlr) > E{Pk(t)lr}
Case II:h(tlr) < E{Pk(t)lr}
In Case I, where hUlr) is greater then the expected value of the spot price
(given all information available at time r), the customer loses money on the
average. Hence the customer chooses dk,f(t) > 0 only if the customer receives a
benefit from having a prespecified price - i.e., if the customer is willing to pay
a premium for the insurance provided by the fixed-price-fixed-quantity
contract.
In case II, where!k(f\r) is less than E{Pk(t!r)}, the customer makes money on
the average. The expected profit increases as dk.,(t) increases. In this case, the
customer chooses a value for dk./(t) based both on the benefit of a prespecified
price and on the customer's willingness to gamble. For example, if a customer
chooses a very large dk./t) and when hour t finally comes, Pk(t) < };,(t\,), the
customer will have to pay a lot of money. Case II might occur when utility
long-term contracts are affected by revenue reconciliation, as will be discussed
shortly.
Energy Broker Specification of fk(tl,)
o
o
o
The utility can act just like an independent energy broker where fees are either
set at costs or are set higher with the profits used to reduce the bills of regular
customers.
Maximize profit
The utility can act as the energy seller who is part of a futures market and simply
set
On the average, the utility always makes a profit which can be used to reduce
the bills of its regular customers.
The utility can set the socially optimum h(t) by repeating the derivation of
Section 8.1 where the revenue reconciliation constraint of (8.1.4),
is modified to become
(9.3.3)
Either the weighted least squares or Ramsey pricing approach can be used to
find the optimum Pk(t) and/k(t) simultaneously. The Ramsey approach yields
h (t) that are very sensitive to the benefit function chosen to model the value to
the customer of prespecified prices and the method of modeling the customer's
willingness to gamble - i.e., risk aversion.
Use of long-term contracts to achieve revenue reconciliation
E{,~lf (j~(t) -
I\(t)dk./(t)l}
(9.3.4)
where Pk (I) is the unreconciled spot price of Chapter 7 which is now seen by all
of the utility's regular customers. This should work (in theory) if the utility
overrecovers when charging Pk(t), since then
(i.e., Case II as discussed above), and one expects that enough customers would
be willing to make a profit (on the average) to enable the condition (9.3.4) to be
met. If the utility underrecovers when charging p(l), the condition (9.3.4) is
achievable only if the benefits of the insurance are large enough to that the
resulting dk./(/) are large enough.
Discussion of Long-Term Contracts
227
quantity contract is sound. However we have not yet fully explored the many
issues associated with its implementation, e.g. utility incentives to manipulate
forward prices, and the exact terms of such transactions.
SECTION 9.4. SPECIAL CUSTOMER CONTRACTS
eLR'
The basic argument for special customer incentive (or disincentive) rates is that
the effect of the long-term capital stock elasticity eLR can dominate the shortterm operating elasticity BS R '
We will now do some mathematical gymnastics which quantify the above
ideas. Only readers with a special interest in the subject should proceed with the
remainder of this section. We do not address the subject again at any later part
of the book.
228
where
Pk(t): Hourly spot prices without revenue reconciliation
Pk(t): Spot price with the Ramsey approach to revenue reconciliation
Ji.R: Lagrange multiplier adjusted to yield the desired annual revenue
Ji.R > 0: Implies excess capacity (i.e., underrecovery)
Ji.R < 0: Implies shortage of capacity (i.e., overrecovery)
where
fh: Vector of the Pk(t) t
Vector of the Pk(t) t
dk : Vector of the dk(t) t
Pk:
=
=
=
I ... TR
I ... TR
I ... TR
Dk = (odk/opd: A TR by TR matrix
T R : Time Tnterval for revenue reconciliation
Derivation of (9.4.2)
To keep the book from having too many equations, the derivation of (9.4.2)
ignores all uncertainties, network effects, capacity constraints, etc. Thus, we
return to the case of Section 6.1 except that the cross-elasticities and Ramsey
revenue reconciliation are included.
The resulting Lagrangian is
!l(TR)
+ Il{ Cr
f efc!.kJ
(9.4.3)
TR
I ... TR
I ... TR
229
1 ... TR
0,
ofJ.
(9.4.4)
(9.4.5)
(9.4.5~ields
(9.4.6)
.~
= B,k
Assume
dk(t) =
(9.4.7)
KkUk(t)
o~
Uk(t) ~ I
Further assume
(9.4.8)
Kk([h)
.2..
(e PR): Time average of all prices, t :=
TR - T
1 ... T R
Thus the kth customer's decision on what values of Kk to choose depend on all
of the prices, i.e. Pb over the revenue reconciliation time intervaLS The utilization decision at time t depends only on the price Pk(t) at hour t. There are no
cross-elasticities with respect to utilization.
Assume further that
(9.4.9)
I
Short-run ealsticity
(9.4.10)
231.
= - 0.01
New Customer: k = 2
SLR.2
= - 0.5
sR.2
= -
0.01
Then
Old Customer Pays:
o
o
Line losses
Redispatch of generators
Transmission line flow constraints
Other power system security issues 9
Recovery of embedded capital costs; e.g. revenue reconciliation
There are many types of wheeling terms and conditions in use today and
others are being proposed. Wheeling can be a controversial subject especially
when it is associated with terms like "mandatory", "free access", etc. Discussions on its advantages can get very lively and we have been engaged in such
discussions; as one example see Schwepe [1988]. However, the discussions here
neither survey nor compare all the approaches and do not address the pros and
cons of wheeling. Consider only the wheeling rates that result from the theory
of spot price based energy marketplace.
There are various possible types of wheeling such as
Utility to Utility: Utility S is selling to the buying Utility B where part of all of the energy
flows over the wheeling utility's lines (there may be several wheeling utilities).
Utility to Private User: Utility S is selling to a Private User B located within the wheeling
utility's service territory.
Private Generator to Utility: Private Generator S located within the wheeling utility's
service territory is selling to Utility B.
Private Generator to Private User: Private Generator S is selling to a Private User B
where both are located in the wheeling utility's service territory.
As will be discussed, revenue reconciliation can cause major differences in the
wheeling rates for these different types of wheeling.
Unreconciled rates: General
Define
Wss(t): Energy seller is selling to buyer during hour t (kWh); i.e., amount of
energy being wheeled.
Cw(t): Costs incurred by wheeling Utility during hour t ($).
wss(t): Spot wheeling rate during hour t if revenue reconciliation is ignored
($jkWh).
The spot wheeling rate is the marginal impact of wheeling on the wheeling
utility's costs:
(9.5.1)
~
oW + '[,1,()I +
()l~
oW
YQS I
(9.5.2)
where
NI~(t)]
L[~(t)l
The details of the derivation can be found in Caramanis, Roukos, and Schweppe
(1988). They are not repeated here because the reader is undoubtedly becoming
tired of such manipulations and because (9.5.2) is a reasonable result. The terms
ilW
[A(t)
(I)) ilL[z(I))
YQs
il W
Under some existing wheeling terms and conditions, the wheeling utility
subtracts its extra losses from the amount of energy delivered to the buyer; in
such a case the second term of (9.5.2) is obviously not appropriate.
It is important to note that (9.5.2) can yield a negative wheeling rate! As a
simple example, assume the wheeling energy flow is in the opposite direction
from the rest of the energy flows on the wheeling utility's network. This yields
a negative oL[z(t)/aw. However, it can be shown that the wheeling utility's net
benefits (wheeling revenue minus change in costs) are never negative.
The basic result (9.5.2) holds for all types of wheeling: utility to utility, utility
to private user, etc. However, for utility to utility wheeling, implementation of
(9.5.2) requires a lot of book-keeping because the partial derivative with respect
to W implies simultaneously changing the net scheduled interchange of Utilities
Band S and then modeling the effects of their AGC, etc. (see Appendix B).
Computer programs to do this book-keeping are available (see Caramanis,
RoU\kos, Schweppe (1988 but the details are not covered in this book. Instead
we w11l go into more interesting issues associated with revenue reconciliation. To
sim'i,lify the algebra, only the special case of bus to bus wheeling is considered.
Bus to Bus Wheeling
Bus to bus wheeling occurs when the seller provides all the energy at bus Sand
the buyer receives all the energy at bus B, where both buses are in the wheeling
utility's service territory. This case applies directly to the Private GeneratorPrivate User type of Wheeling and to the other three types if only one or two
of the wheeling utility's tie lines to external utilities are affected (which could be
a reasonable approximation to many cases). However even if the explicit bus to
bus equations cannot be applied, the basic principles to be discussed are valid.
For bus to bus wheeling (9.5.2) becomes
(9.5.3)
or equivalently
(9.5.4)
where the Band S subscripts obviously refer to the buying and selling buses.
234
Revenue Reconciliation
After reading Chapter 8, the reader should suspect that there are many ways to
do revenue reconciliation for wheeling. We will only present some of the main
results by using the simple, constant multiplier approach. Define
(Uss(t): Wheeling rate including revenue reconciliation.
Class 0: A reasonable approach for Class 0 is to use the form of (9.6.1) with
m)[y(l)
flu(t)
(I
f!s(t)
(I - 111)[;'(1)
+ IJs(I)]
+ 17s(l)]
which yields
(I)ns(l)
(l)us{t)
m[2i'(t)
Ils(t)
IJs{t)]
(9.5.6)
Class ON: A reasonable approach for Class ON when the wheeling utility has
no obligation to serve only the buyer is to use the form of (9.5.3) with
f's(t)
/'(1)
lJu(l)
f's(t)
)'(1)
IJs(t) - m/lJs(t)/
m(y(l)
IJs(t)]
which yields
(9.5.7)
Comparison of Classes
The main difference between Class 0 (9.5.6) and Class ON (9.5.7) is the factor
of 2 which multiplies y(t). The main difference between Class N (9.5.5) and
Classes 0 and ON is that Class N has no yet) term (except as it enters into the
loss components of the I'/(t).
The impact of the generation marginal costs y(t) can be very large. If no
network quality of supply components are present, the network component I'/(t)
contains only loss components, so, for example,
aL(t)
'1e(t) =
yet) adB(t)'
; to
ain
d)
ith
.6)
The hourly spot prices developed in Chapters 6, 7, and 8 provide the basis for
a wide variety of spot price based transactions. This chapter presented the
resulting theory for predetermined price-only transactions, price-quantity transactions, long-term contracts, special rates for special customers, and wheeling
rates. The techniques used can also be applied in other types of situations such
as assignments of power pool reserves. The wide range of possible spot price
236
based transactions illustrate the power (or it is energy?) of a spot price based
energy marketplace.
Many of the results of this chapter bear on politically sensitive issues (i.e.
disputes) of valuing and compensating various non-standard generation and
load management schemes. For example, one argument against large generators
per se and nuclear units in particular is that their size forces the utility or power
pool to carry more reserves, both total reserves and operating reserves. Pricequantity rates for system security control provide a way to measure this penalty,
if any. Another example is the time pattern of solar and wind generators. The
hourly spot price gives a precise way to measure the actual value of energy
generated. Many similar questions surround cogenerators, thermal storage
systems, etc.
We would be happier to see spot pricing used not just to evaluate the value
of existing generation and load management equipments but to actually change
day to day behavior of the units to increase their social value. One is pie splitting
(at least once the units are built), while the other is pie enlarging. We believe that
many existing units can have their social value (and private profitability)
enhanced in this way.
NOTES
I. Since revenue reconciliation is ignored here, we could use the notation of Chapter 8, and write
p, (tiT). However, we choose to use the Pk(tlr) notation of Chapter 7. In Chapters 2 and 3, Pk(t)
is associated with a one-hour update spot price. In actual implementation, as discussed in
Chapter 4, a one-hour update is a predetermined rate which is specified, for example, five
minutes before the start of the hour. Thus, to be exactly correct, all price-only transactions are
in the category of predetermined rates.
2. Using (9.1.4), the expected revenues received by the utility are the same for /'k and Pk(t),
assuming dk(/) does not change.
3. These by no means exhaust the issues of "dynamics pricing" which can be argued to include the
dynamics of individual generators. For investigation of such details see Caramanis, Bohn,
Schweppe [1987J.
4. For details of derivation see Caramanis, Bohn, Schweppe [1987J.
5. We assume here for simplicity that if interruption is needed it will last for the whole pricing
period. Extension to fractional period interruptions is, of course, possible.
6. That is, constraint (9.2.2) will be met after the value of S is revealed by setting Sk.c = Sk.p with
k values ranging over the required subset of participants in order of increasing contingency
prices. Once (9.2.2) is met, Pc will be set equal to the highest reservation price reached and all
s'.c corresponding to higher reservation prices will be set to zero.
7. This simple constraint assumes the hot water or thermal storage capacity is very large. In
practice, the logic usually has to be more complicated to account for finite storage and
time-varying (within a day) customer needs.
8. In practice, capital stock decisions would consider prices over several revenue reconciliation
intervals, but this assumption simplifies the manipulation of the equations.
9. Much more detail can be found in Caramanis, Schweppe, and Bohn [1986J and Schweppe, Bohn
and Caramanis [1985J.
10. The term "tie line coefficient" is our own; there seems to be no standard terminology.
II. More detailed discussions can be found in Schweppe, Bohn and Caramanis [1985J.
Lie
ge
19
at
y)
ite
(I)
in
.ve
Ire
f).
he
n.
th
~y
III
In
Id
*1ourly spot prices and spot price based rates were derived in Chapters 6
through 9 by considering short-term issues, i.e., capital stock was treated as
fixed. This chapter examines investment decisions in the spot price based marketplace and their relationship to the statistical behavior of spot prices. Optimal
investment conditions from the point of view of society's cost minimization are
derived and compared to optimal investment conditions from the point of view
of individual marketplace participants. The use of spot price notation leads to
very straightforward mathematical results, and permits us to treat generation,
transmission, and end use technology in a single framework.
This chapter limits discussion of investment conditions to the case of ideal
revenue reconciliation which does not affect participant behavior.
Investment decisions are evaluated on the basis of the price duration curve,
a concept analogous to the load duration curve. Statistical hourly spot price
behavior information is shown to be necessary and sufficient in evaluating
investment conditions in most cases.
Section 10.1 presents the overall problem formulation while Sections 10.2,
10.3 and 10.4 discuss the specification of generation, customer, and network
investments. Section 10.5 discusses revenue reconciliation for an optimum
system. This chapter concludes with a discussion in Section 10.6 on short-run
versus long-run marginal cost pricing.
This chapter concentrates on mathematical conditions for optimality. Real
world power system planning (as discussed in Appendix C, in particular, Section
C.I) involves many concerns, beyond mathematical optimality of social welfare.
237
The load, generation, and network related terms defined in Chapters 6 through
9 are also used here. The reader should thus review the definitions of $/t), get),
~(t), ~*(t), ("(t), d(t), cJ.(t), ~P(t), dP(t), (iP(t), ~(t), Lf~(t)], G[~(t)J, N[~(t)J, B[(i(t)J,
B[dP(t)] and other related terms.
Additional quantities are defined here to allow inclusion of investment terms
into the model. It is assumed that the capital stock associated with each
generator, customer or line is represented by a scalar and is not subject to
indivisibilities. Whereas the scalar assumption is made for notational simplicity
only, the indivisibility assumption is needed for a differential calculus based
description of first-order optimality conditions.
Using subscripts to indicate generation, demand and network related quantities, as usual, define
K;.i: Capital stock of generator j which is in place at the beginning of the period
under consideration and expected to be retired at the end of that period.
KD,k: Capital stock of hourly spot price participant (customer) k in place at the
beginning of the period under consideration and expected to be retired at the
end of that period.
Kg.k: Similar to above for predetermined rate participants.
K~J: Similar to above for network line.
K y , K D , Kg, K,,: Vectors with elements the quantities defined above.
I;(K-;} The mvestment cost ($) associated with capital stock KyJ utilized by
generator j.
Ik (KO,k), Ik (Kg,d, ~(K~,;): Defined similarly to the above for hourly spot price
participant k, predetermined rate participant k, and network line i respectively.
[(Ii,)
r.j~(KYJ)
The short- and long-term Lagrangians are defined next. They represent the net
hourly and whole period costs to society respectively.
The short-term Lagrangian is identical to that defined in Section 9.1 of
Chapter 9.
!let) = Gf(t)]
N{:,(t)
- B(q,(t)1
- B(q,P(t)
/-le(t)[d(t)
+ dP(t) +
L[~(t)J - g(t)J
(10.1.1)
h
),
I,
s
h
nLT
I(K y )
+ I(KD ) + I(Kb) +
I(K~)
-
L E{n(t)}
(10.1.2)
1 ... 0
where
T: The retirement time of capital stock invested and put into operation at time
zero.
E: The expectation operator over random variables to be realized at time t.
Expectations are evaluated with probability density functions that represent
the knowledge at time zero of the behavior of random variables at time t i.e., a conditional expectation.
Conventional (not spot price based) investment conditions are usually
obtained from a formulation which involves a reliability constraint such as an
allowable loss of load probability (or hours) or expected unserved energy. Our
development here assumes that the generation and network quality of supply
terms of the spot price are calculated by the "market clearing approach." Thus,
th~re is no unserved energy in the sense of anyone being "blacked out";
cystomers simply see very high prices instead and reduce demand as required.
Hence, no explicit reliability constraint is included. Since the market clearing
al'proach is being used, there are no generation or network quality of supply
costs; i.e.,
GQS[g(t)]
NQs!.~(t)
in the definitions of G(t) and N(t) of (10.1.1). The Gos and Nos terms could be
included, if desired, at the expense of more complicated-looking equations.
Two implicit assumptions in the above long-term Lagrangian formulation are
o
o
These assumptions are made only for notational simplicity. Existing capital
stocks with varying retirement rates can be modeled without major structural
changes in the formulation. Also, discounting to yield a more conventional
calculation of the present value of costs can be introduced in the summation sign
of (l 0.1.2) without changing the problem formulation in any qualitative sense.
Equation (10.i.2) is sufficient to present the basic ideas without a lot of technical
notational complexity. However, the reader must be warned that
In any actual calculation using the ideas of the chapter, the equations must be modified
appropria tely.1
an LT
oK;",
an LT
aKD,k
an LT
aKf,.k
an
LT
aK'I,'
for alii
(Generation)
( 10.1.3)
for all k
( 10.1.4)
for all k
(10.1.5)
= 0
for all i
(Network)
( 10.1.6)
Utility Generation
Developing the rest of (\0.2.1) holding g,(t) constant (and remembering that
GQs{g(t)] = 0) yields
.
- flmax.)./(I)
ogma.J(t)}
oK.
(10.2.2)
rJ
(10.2.4)
so
(10.2.6)
L E{(j(t)}
(10.2.7)
e(t) =
J1Qs.y(t)
plant is available
J1ma'.YJ(t)
.I
If the plant is either turned off or not at peak capacity (gj(t) < gmax), then
If the plant is at full capacity, then the optimality condition (7.2.7), the definition
of Gf(t)] of (7.2.2) (remembering that GQS = 0), and use of (10.2.5) yield
Ai
tIQS.i(t)
J1max';'j(t) =
Pj(t)
(10.2.8)
(10.2.9)
pj(t) - Aj
g/t) =
gmaxJ
IIQs.i(t) = YQs(t)
gj(t) <
gmaxj
Assume that the variable fuel and maintenance costs of generator j are constant.
Then if the effect of flQs.y(t) = YQs(t) is ignored, a good approximation to
(10.2.9) is
Investment Cost of Incremental kW
(Aj)
* (a;) * (Hours)
(10.2.10)
where
A,:
Area under hourly spot price duration curve from Aj to infinity as in Figure 10.2.1
a,: Probability that the incremental generating capacity will be available for generation during any future hour
Hours: The total number of hours in the service life of the incremental investment<
This underscores the significance of hourly spot price duration curves in evaluating incremental investments. 3
Quality of Supply and Generation Investment Costs
One of the methods for calculating the generation quality of supply spot rpice
components that was discussed in Section 6.2 was the,"Annualized Cost of
$/kWh
8760
Time
\~.,.,'
K/.,A Qs.;
L E{YQs(t)}
YQs(t) =
LE{a;(r)}
r
(10.2.11)
244
for any ay(t), a fact which can be verified by substitution of YQS(t) in (10.2.11).
The ay(t) obtained in Section 6.2 was
Thus the Section 6.2 result is one of many possible ways to quantify YQS (t),
Customer Owned Generation
If a customer-owned generator k sells its output to the grid at the hourly spot
price and self-dispatches itself according to the price Pk(t), the individual generator's optimal investment problem is
minimize
K,.k
1[Ky,.]
"
E'T
{gk(t)Pk(t)
- GFMfgk(t)J}
subject to
:0::;;
gk(t)
:0::;;
gmax.k(t)
L: E[Ck(t)]
o
(10.2.12)
otherwise
Comparing the customer investment condition (10.2.12) with the corresponding utility investment condition (10.2.9) shows that the only difference lies in the
effect of the generation quality of supply benefit !lQS,y(t) seen by the utility. This
makes sense because the utility is responsible for quality of supply.
If the effect of the !lQs)t) = YQs(t) term is neglected, then individual profit
maximizing behavior coincides with the utility's optimal behavior under the
mild assumption that the same future price duration curve information is
available to the private and utility planner. 4 Finally, if system security control
requirements are modeled as commodities to which all market participants can
contribute (see Section 9.2), rather than as a constraint on utility-owned generation, utility- and customer-owned generation investment conditions coincide
completely. This happens because security control or quality of supply requirements are properly internalized in the price system and are accessible to all
marketplace participants alike.
z:: E{ _
adk(t)
(\0.3.1)
OJ[KOk ]
oKD .k
(10.3.2)
(10.3.3)
E'h!,lation (10.3.3) states that the optimal investment level is such that incremental investment costs equal the present value of the expected incremental
electricity savings over the life of the investment. Equation (l0.3.3) clearly
coinCides with individual consumer profit maximizing behavior.
Equation (10.3.3) shows that forecasts of the properties of the hourly spot
price are necessary to calculate investment decisions. The price duration curve
information discussed above in conjunction with generating capacity expansion
investments may not be sufficient in describing the necessary information
needed to evaluate incremental investments for customers in all cases.
Consumers with the ability to respond to hourly spot prices using some type of
storage (thermal, product, etc.) may benefit from additional information describing the time series nature of the spot prices. Thus, typical daily spot price
trajectory forecasts may be necessary. See Bohn (1982] for an analysis of how
to calculate the incremental value of additional storage.
The investment conditions for generators, lines and customers given in this
chapter ignore the effect of revenue reconciliation. It can be argued that this is
reasonable for utility investment in new generators and lines. However, in
practice, customers would consider future revenue reconciliation if it modifies
the hourly spot price.
Predetermined Rate Participants
246
n.
'7
aL[y(t)]]
adf(t)
-It ( t ) -- - + _aB-.::..d!....;df:...;.(t...:.!)]}
adk(t)
aKg.k
aKb.k
( 10.3.4)
Substituting the following relationships derived in Section 9.1 (see (9.1.8) and
(9.1.3)
Pk(t) =
aN{ y(t)]
adfCt)
aBkfdf(t)]
=
adf(t)
Ite(t)
aLl yet)]
Pe(t) adf(t)
rk
yields
a/[Kb.d =
-aKP
D.k
"E{[" ( )] adf(t)
~
"
Pk t
'VD
I
0 no b.k
(lBk[df(t)J}
oKP
D.k
(10.3.5)
L E{OBddf(t)]}
I
aKg.k
(10.3.6)
(10.3.7)
247
t/,/
l.max ] }
az13K.
(10.4.1)
'1,1
Equation (10.4.\) states that the network investment in line i should be undertaken to the level that renders the cost of an incremental investment equal to the
present value of the stream of expected benefits through losses reductions,
maintenance cost reductions and alleviation of line flow constraints.
Appendix D on the DC load flow shows that if hourly spot prices are paid to
generators and charged to consumers, the system will receive a net income which
can be considered to be the income of the network. Disregarding NM , the
network income is shown in Appendix D (see Section D.2) to be during hour t
t1e(t)L[y(t)J
L PQs.".,(t)z,(t)
i
"'>f
uK".,
aK".,
Section D.2 of Appendix D shows how these quantities can be computed under
the DC load flow approximation. The impact on system losses of an incremental
investmentS in line i is shown to be proportional to the square of the power
flowing over line i, while the impact on line flows is proportional to the power
flowing over line i. Appendix D (see also Section 7.7) also shows that the power
flowing over line i is proportional to the difference of spot prices across the ends
of line i (assuming a constant reactance to resistance ratio on a\llines). Thus the
optimal investment condition for line i can be evaluated in terms of the behavior
of spot prices at the ends of that line.
248
In Sections 10.2, 10.3, and lOA, we showed that socially optimal investment
conditions are characterized by the following rule: Investments are made until
the last MW of investment capacity "earns" an expected stream of new income
whose present value equals the incremental cost of investment. Assuming that
there are no economies or diseconomies of scale (i.e., the investment cost per
MW of capacity is invariant of the total installed capacity), optimal investment
conditions quarantee that each unit of capital (generation, network or consumption) will make precisely enough to cover its investments costs ifpaid according
to optimal spot prices. Therefore, revenue reconciliation would be automatically
achieved.
In real systems, a number of issues may result in imperfect revenue recovery.
Two are
o
Nonexistence of spot price based revenues/costs and hence historical investment levels that are far from socially optimal levels.
It is also necessary to worry about intertemporal effects of present generration-consumption levels on future costs. These effects were assumed nonexistent in Chapters 6 through 8. They can be taken into consideration as discussed
in Caramanis (1982]. However, the intertemporal effects modeled in there are
fairly short-term, extending over the unit commitment time scale (days to
weeks). Longer-term intertemporal coupling extending over periods of years is
very cumbersome to model, since it requires long-term marginal cost models of
the form discussed in the next section.
SECTION 10.6. LONG-VERSUS SHORT-RUN MARGINAL COST PRICING
Arguments on short-run versus long-run marginal cost pricing are often encountered in the electricity pricing literature. This section addresses the shortversus long-run issue in the context of a spot price based marketplace.
In the economic literature, short-run marginal cost usually refers to the cost
of incremental production, including variable costs such as fuel and raw
materials but not including the cost of capital or other fixed (sunk cost) factors
of production. Long-run marginal cost usually refers to the costs of incremental
production over a long period, including both variable and fixed costs.
In most power system costing literature, short-run marginal cost is associated
with a production cost model whereas long-run marginal cost is associated with
incremental revenue requirements calculated by a capacity expansion planning
code.
liy
'y.
me
.'lat
,er
mt
IPng
;t-
r-
s~d
:e
o
is
,f
can cause a major jump (up or down) in a long-run marginal cost price
computed for a deterministic world which ignores future uncertainty or priceresponsive future demand. History has shown that such events are a fact of life.
Energy Marketplace Long-Run Marginal Cost Pricing
A spot price based energy marketplace can provide long-run marginal cost
prices which are selected to maximize the long-term social welfare Lagrangian
Q LT of (10. \ .2) with the conditional expectation operator reflecting future uncertainty at the time of the selection. Assuming that capital stock is an explicit
function of future demands and denoting for simplicity different types of capital
by one variable, we have the following necessary maximization condition:
p(t)
under the condition that all marketplace participants are under optimal spot
prices and that socially optimal investment decisions as defined in earlier
sections of this chapter are in place. Indeed, as shown in Chapters 6 and 7,
optimal spot prices are selected such that
Od(t) n(t) =
(10.6.2)
Equations (10.6.2) and (10.6.3) imply the necessary conditions of (10.6.1), and
hence long-run marginal cost coincides with the optimal spot price.
Before discussing the conditions under which PLT(t) and pet) may diverge, the
reader should note the difficulties involved in actually arriving at the optimal
investment conditions which insure that (10.6.3) holds. Optimal investments
require evaluation of future uncertainties as described by the conditional expectation operator of (10.1.2). Although in principle we can write the necessary
conditions, their implementation is hard as is proper accounting of uncertainties
in the context of present-day long-run marginal costs models. However, the
reader ought to note that the adjustment to future uncertainty realizations,
through the selection of closed-loop type optimal spot prices, mitigates the
impact of the uncertainty and hence the impact of shortcomings in modeling it.
Another difficulty in implementing spot price based long-run marginal costs
rests on the unavailability of proper demand models that capture the interdependence of intertemporal demands. In principle one needs to use benefit
functions which depend on a whole time stream of demands rather than demand
at a single point in time; i.e., one needs to evaluate
where
Hi~orical investment decisions which have been guided by average cost pricing
con-siderations guarantee that current capital stock levels are not at their socially
optimal levels. In addition, the stated difficulties in dealing with uncertain
intertemporal demand dependencies, noncoincidence of social and private
optimal investments for predetermined price participants, etc., indicate that
future capital stock levels will "almost never" reach and hold their socially
optimal levels. We can thus say that the following relationship holds:
PLT(r)
p(f)
The reader should note that pet), the spot price of Chapters 6 through 8, does
provide information about the future which is associated with long-run
marginal costs. For example, when generating capacity is in shortage, spot
prices will tend to be frequently high since capacity constraints will also be
frequently active. This behavior of pet) will signal to consumers that increased
demand will require additional generation investments.
We wish at this point to conjecture that the behavior of PLT(t) is likely to be
very similar to that of pet) and hence the magnitude of "other terms" in (10.6.4)
will be small relative to pet).
The motivation for the conjecture is our belief (neither proven nor tested) that
a change in demand at hour t has very limited impact on future capital and
operating costs (many years in the future) for two reasons:
~",.
II'.
meory or
1I0urWspolprices
Hopefully, someone will someday develop a good demand model which can
either verify or disprove this belief.
SECTION 10.1. DISCUSSION OF CHAPTER 10
253
the optimal spot price formulas. Thus future uncertainties on load growth, fuel
costs and other important unknown variables are substantially smoothed
through spot price driven short-term reoptimization that cushions the impacts
of investment planning shortfalls or longfalls, changes in environmental
standards, new fuel discoveries, etc.
A general discussion of short-versus long-run marginal cost pricing was also
provided in this chapter. The basic spot price equations developed in Chapters
6 through 9 are primarily (but not entirely) evaluated in a short-run context.
Some of the options for computing the generation and network quality of
supply components involved, use the annualized cost of peaking generation
plants and the marginal cost of expanding the network. These approaches can
be viewed as pragmatic approaches to long-run marginal cost pricing. However,
a spot price energy marketplace could be based on pure long-run marginal costs
if desired, provided the massive uncertainties associated with expansion
planning are included in the long-run marginal costing formulation, valid
demand models are used, etc. - i.e., that the formulation of (10.6.1) is used.
Establishment of an energy marketplace based on present-day approaches to
long-run marginal cost pricing is not recommended.
The reader must be careful not to confuse the mathematical optimality
conditions of this Chapter with real world power system planning as discussed
in Appendix C; particularly Section C.I.
HIS]ORICAL NOTES AND REFERENCES - CHAPTER 10
The' basis of Sections 10.1 through 10.4 is Caramanis [1982]. Other material
listed in the annotated bibliography also develops investment conditions for
particular parts of the marketplace. Other authors have developed optimality
conditions for generation, but have not used the spot price formulation (which
gives very clean statements of the optimality conditions). See for example the
discussion at the end of Chapter 3. As usual, an exception is Vickrey, who
provided a lucid discussion of the issues and the optimal investment conditions
under responsive pricing.
There has been very little previous work on optimal conditions for transmission investments, although of course utilities have developed transmission
plans by the method of simulating "System behavior with and without a particular proposed line. (See Appendix C.) To be sure, transmission lines are one
topic where the assumption of continuously variable investment sizes can fail,
since the addition of a new line connecting two widely separated (electrically)
buses can lead to a discontinuous shift in power flows. The results of Section
10.4 are most accurate for the case of strengthening an existing line (or examining the size of a proposed line),
The situation on customer end use technology is somewhat different. Since
utilities have long been interested in buying and installing load management
equipment for cuistomers, there are a number of methods of evaluating the
value of such investments. For example the entire second half of Talukdar and
254
Ac~on,
Jan Paul and Bridger Mitchell, "Evaluating Time-of-Day Electricity Rates for Residential
Customers," in Mitchell & Kleindorfer, eds., pp. 247-273, 1980.
AgQcw, Carson E. "The Theory of Congestion Tolls", Journal of Regional Science, Vol. 17, No.3,
pp. 381-393.
Anderson, Ronald, W. The Industrial Organization of Futures Markets, D.e. Heath. 1984.
Baumol, William J. and Bradford "Optimal Departures from Marginal Cost Pricing", American
Economic Review, Vol. 60, June 1970, pp 265-283.
Bergen, Arthur R. Power Systems Analysis, Prentice Hall, 1986.
Berger, Arthur "Pricing Mechanisms in Coupled Dynamic Systems", PhD thesis, Harvard University, May 1983.
Berger, Arthur W. and F.e. Schweppe, "Real Time Pricing to Assist in Load Frequency Control",
submitted to IEEE Transactions on Power Systems, 1995.
Berrie, T. "Interactive Load Control, Parts 1 to 6", Electric Review, 1981-1982.
Bertsekas, Dimitri, Dynamic Programming: Deterministic and Stochastic Models, Prentice-Hall,
1987.
Blackmon, B. Glenn, Jr., "A Futures Market for Electricity - Benefits and Feasibility," Discussion
Paper Series E-85-07, Energfy & Environmental Policy Center, John F. Kennedy School of
Government, Harvard University, July 1985.
Bohn, Roger E., "A Theoretical Analysis of Customer Response to Rapidly Changing Electricity
Prices," August 1980, Revised January 1981, MIT Energy Lab paper MIT-EL-081-001 WP.
Bohn, Roger E., "Spot Pricing of Public Utility Services," PhD dissertation, MIT, 1982. Also
available as MIT Energy Lab Technical Report MIT-EL 82-031.
Bohn, R.E. "Procedure for Electricity Auctions with Optional Spot Pricing", unpublished memo,
November 1985.
Bohn, Roger E., M.e. Caramanis, and F.e. Schweppe, "Optimal Spot Pricing of Electricity:
Theory," Energy Lab Working Paper MITEL-08100SWP, 1981.
Bohn, Roger E., Michael C. Caramanis and Fred C. Schweppe, "Optimal Pricing of Public Utility
Services Sold Through Networks," Working Paper HBS 8331, Graduate School of Business
Administration, Harvard University, corrected January 1983.
255
256
References
Bohn, R,E" B,W, Golub, R.D, Tabors, and F.C. Schweppe, "Deregulating the Generation of
Electricity Through the Creation of Spot Markets for Bulk Power", The Energy Journal, Vol.
5. No, 2,1984 pp 71-91.
Bohn, R,. F. Schweppe, and M, Caramanis, "Using Spot Pricing to Coordinate Deregulated
Utilities, Customers, and Generators", Chapter 13 in Plummer et ai, editors, 1983.
Boiteux, M. "La tarification des demandes en point: application de la theorie de la vente au CaUl
marginal", Revue Generale de /'electricite, Vol. 58, 1949, pp 321-340,
Boiteux, M" "On the Management of Public Monopolies Subject to Budgetary Constraints,"
English translation of his 1956 article, Journal of Economic Theory, 1971, Vol. 3, pp. 219-240,
comment vol. 3, p, 341.
Brown, G. Jr. and M. B. Johnson, "Public Utility Pricing and Output Under Risk", American
Economic Review, Vol. 59, March 1969, pp 119-129,
Caramanis, M. "Analysis of Non-Dispatchable Options in the Generation Expansion Plan", IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-102, No, 7, July 1983 pp. 2098-2103.
Caramanis, M. "Production Costing of Interconnected Electrical Utilities Under Spot Pricing",
Large Scale Systems: Theory and Applications, Vol. 9, 1985, pp 1-18.
Caramanis, Michael, "Investment Decisions and Long-Term Planning Under Electrical Spot
Pricing," IEEE Transactions on Power Apparatlls and Systems, Vol. PAS-IOI, No. 12, Dec. 1982,
pp 4640-4648,
Caramanis, M. C, "Electricity Generation Expansion Planning in the Eighties: Requirements and
Available Analysis Tools," Energy Models and Studies 1983, pp 541-562.
Caramanis, M. C, R. E. Bohn, and F, C Schweppe "Spot Pricing of Electricity: Practice and
Theory", IEEE Transactions on Power Apparatus and Systems, Vol. PAS-IOI, No.9, September
1982, pp 3234-3245.
Caramanis, Michael C, R. E. Bohn, and F. C Schweppe "The Costs of Wheeling and Optimal
Rates", IEEE Transactions on Power Systems, Vol. PWRS-I, No, I, February 1986, pp 63-73.
Caramanis, M. C, R, E. Bohn and F. C Schweppe "System security control and optimal pricing
of electricity" Electricity Power & Energy Systems, Vol. 9 No.4, October 1987, pp 217-224.
Caramanis, M. C, N. Roukos, and F, C Schweppe, "WRATES: A Tool for Evaluating the
Marginal Cost of Wheeling", submitted ot IEEE Transactions on Power Systems, 1988.
Chao, Hung-po, "Peak Load Pricing and Capacity Planning with Demand and Supply Uncertainty," The Beff Journal of Economics, Vol. 14, No. I (Sprin 1983), pp, 179-190.
Chao, Hung-po and Robert Wilson, "Priority Service: Pricing, Investment, and Market Organization," American Economic Review, Vol. 77, No.5, December 1987, pp. 819-916.
Coase, Ronald H., "The Marginal Cost Controversy", Economica Vol. 13, 1946, pp 169-182.
Coase, Ronald H" "The Theory of Public Utility Pricing and Its Application", The Beff Journal of
Economics and Management Science, Vol. 1, spring 1970, pp 123-128,
Coate, Thomas M" "Some Strategies for Reducing Electricity Costs at MIT", MIT bachelor's
thesis, May 1979,
Cohen, Linda, "A Spot Market for Electricity: Preliminary Analysis of the Florida Energy Broker,"
Rand Report N-1817-DOE, February 1982,
Constantopolous, Panos "Computer Assisted Control of Electricity Usage by Consumers", PhD
Thesis, MIT, May 1983,
Constantopoulos, p" R. Larson, and F, Schweppe, "Decision Models for Electric Load Management by Consumers Facing a Variable Price of Electricity", in Energy Models and Studies, B.
Lev (editor), North Holland Publishing Company, 1983.
Constantopoulos, P., F. C Schweppe, and R. C. Larson, "ESTIA: A Real Time Consumer Control
Scheme for Space Conditioning Usage under Spot Electricity Pricing", manuscript, 1987.
Craven, John, "On the Choice of Optimal Time Periods for a Surplus Maximizing Utility SUbject
to Fluctuating Demand," The Bell Journal of Economics and Management SCience, Vol. 2,1971,
pp. 495-502.
Craven, John, "Space-Time Pricing for Public Utilities", The Bell Journal of Economics and
Management Science, Vol. 5, 1974, pp 187-195.
Crew, M. A. and P. R. Kleindorfer, "Peak Load Pricing with a Diverse Technology," The Bell
Journal of Economics, spring 1976, pp. 207-231.
Crew, Michael A. and Paul R. Kleindorfer, Public Utility Economics, N. Y.: St. Martins Press, 1979.
Crew, Michael A. editor, Issues in Public-Utility Pricing and Regulation, D.C. Heath and Company,
Lexington Books, 1980.
Dansby, Robert E. "Spatial Considerations in Public Utility Pricing", in M. Crew, editor, 1980.
References
257
Daryanian, Bahman [1986], "The Definition and Application of an Optimal Response Algorithm
for Electricity Consumers and Small Power Producers Subject to Spot Prices", MIT Master's
thesis.
Dobbs, Charles "Optimal Spot Price Theory and Finite Line Capacity", MIT BA thesis, May 1983.
Elgerd, One I., Basic Electric Power Engineering, Addison Wesley Publishing Company, 1977.
Elgerd, OUe 1.. Electric Energy Systems Theory: An Introduction, Second edition, McGraw-Hill
Book Company. 1982. (TKI45.E487).
Ellis, Randall P., "Electric Utility Pricing and Investment Decisions Under Uncertainty." Ph.D
Dissertation, MIT. 198\.
Enders. Robert H. "Power System Load Modeling for Frequency Mediated Control". MS thesis,
MIT, February 1985.
Faruqui, Ahmad and Robert Maiko, "Load Shape Modifications Associated with Time of Use
Rates for Electricity". in Hotelling, ed. Challenges for Public Utility Regulation for the 1980s,
MSU Press, 1981.
Fitzgerald, A. E., Charles Kingsley Jr.. and Stephen D. Umans, Electric Machinery, fourth edition,
1983.
Flory, John E. and O. L. Parker, "Communication and Metering Equipment for Electricity Spot
Pricing", IEEE Transactions on Power Apparatus and Systems. Vol. PAS-104, No.6, June 1985.
pp 1363-1369.
Ford, Andrew "Short-Lead-Time Technologies as a Defense Against Demand Uncertainty", in
Plummer, Oatman, and Gupta, 1985, pp 115-136.
Garcia, E. V. and J. E. Runnels, "The Utility Perspective of Spot Pricing", IEEE Transactions on
Power Apparatus and Systems, Vol. PAS-104, No.6, June 1985, pp 1391-1393.
Gellerson, Mark W. and Shawna P. Grosskopf, "Public Utility Pricing, Investment, and Reliability
under Uncertainty: a Review", Public Finance Quarterly, Vol. 8, 1980, pp 477-492.
Golub. Bennett, and Leonard Hyman "Financial Problems in the Transition to Deregulation". in
Plummer, editor, 1983.
Golub, B., R. Tabors, R. Bohn, F. Schweppe, "Deregulating the Electric Utility Industry: Discussion
and a Proposed Approach," MIT Energy Lab Working Paper MIT-EL 81-043, \98\.
Gol1'!b. Bennett, R. Tabors, R. Bohn. and F. Schweppe "An Approach for Deregulating the
~eneration of Electricity", ch. 5 in Plummer, editor, 1983.
Hasfings, B. F. "The Detroit Edison Second Generation Load Management System", IEEE
J'ransactions on Power Apparatus and Systems, Vol. PAS-104 No.3, March 1985, pp 550-53.
Henney, Alex "Privastise Power: restructuring the electricity supply industry," Centre for Policy
Studies, study no. 83, London, 1987.
Hiaasen, Eric M. "Spot Wheeling Rates for Electricity", MIT MS thesis, August 1984.
Hotelling, H. "The General Welfare in Relation to Problems of Taxation and of Railway and Utility
Rates". Econometrica, Vol. 6, July 1938, pp. 242-269.
Hyman, Leonard S. America's Electric Utilities: Past. Preselll, and Future. second edition, Public
Utilities Reports, Inc., Arlington Virginia, 1985.
ICS. "The ICS Project: TranstexT Participant Development Program on Innovative Pricing for
Electric Utilities", unpublished prospectus, Atlanta, August 1984.
Joskow, Paul and Richard Schmalensee, Marketsfor Power: an Analysis of Electric Utility Deregulation, MIT Press, 1983.
Kaye, R. J. and H. R. Outhred [1988], "A theory of Electricity Tariff Design for Optimal Operation
and Investment", submitted to IEEE Transactions on Power Systems, 1988.
Kelly, K., S. Henderson and P. Nagler, Some Economic Principies for Pricing Wheeled Power,
Report of the National Regulatory Research Institute, Columbus Ohio, 1987.
Kepner, John and M. Reinbergs, "Pricing Policies for Reliability and Investment in Electricity
Supply as an Alternative to Traditional Reserve Margins and Shortage Cost Estimations,"
unpublished manuscript. Approx 1980.
Littlechild, S. C. "Spot Pricing in the Electricity Industry: an appraisal of the MIT model",
presented at a conference in honor of A. Charnes, October 1987.
Littlechild, Stephen c., "A State Preference Approach to Public Utility Pricing and Output Under
Risk", The Bell Journal of Economics and Management Science, Vol. 3, 1972, pp. 340-345.
Luh, Peter B., Yu-Chi Ho, and Ramal Muralidharam, "Load Adaptive Pricing: An Emerging Tool
for Electric Utilities," IEEE Transactions on Automatic Control, April 1982.
Luo, J. S., E. F. Hill, and T. H. Lee "Bus Incremental Costs and Economic Dispatch", IEEE
Transactions on Power Systems, Vol. PWRS-I, No. I, February 1986.
258
References
Maiko, J. Robert and Ahmad Faruqui, "Implementing Time-of-Day Pricing of Electricity: Some
Current Challenges and Activities," in Crew, 1980.
Manichaikul, Yongyut, "Industrial Electric Load Modeling", Ph.D. Thesis, MIT, 1978. (Also
Electric Power Systems Engineering Lab Report 54.)
Manichaikul, Y. and F. C. Schweppe, "Physically Based Industrial Electric Load", IEEE Transactions on Power Apparatus and Systems, July/August 1979, pp 1439-1445.
Manichaikul, Y. and F. C. Schweppe, "Physical/Economic Analysis of Industrial Demand," IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-99, No.2, March/April 1980, pp.
582-588.
Marchand, M. G., "Pricing Power Supplied on an Interruptible Basis," European Economic
Review, July 1974, pp. 263-274.
MIT Energy Laboratory, New Electric Utility Management and Control Systems: Proceedings of
Conference, MIT Center for Energy Policy Research and Electric Power Systems Engineering
Laboratory, Report EL79-024, June 1979.
Moretti, Didier Rene, "Pricing Policies for Power System Security Control," Master's Thesis, MIT,
1983.
Munasighe, Mohan, The. Economics of Power System Reliability and Planning, Johns Hopkins
University Press, 1979.
Nguyen, D. T., "The Problem of Peak Loads and Inventories," Bell ]ou/'//al of Economics, Vol. 7,
spring 1976, pp. 197-206.
O'Rourke, Paul "A Physically Based Model of the Space Conditioning Load Under Spot Pricing",
MIT Master's thesis, Operations Research Center, January 1982.
O'Rourke, Paul and Fred C. Schweppe, "Space Conditioning Load Under Spot or Time of Day
Pricing," IEEE Transactions 011 Power Apparatlls alld Systems, Vol. PAS-102, May 1983.
Oren, Shmuel, Stephen Smith, Robert Wilson and Hung-po Chao, "Priority Service: Unbundling
the Quality Attributes of Electric Power", EPRI report EA-4851, Palo Alto, 1986.
Oren, Shmuel, Steven Smith, and Robert Wilson, "Capacity Pricing," Econometrica, Vol. 53, No.
3 (May, 1985), pp. 545-566.
Outhred, Hugh and F. C. Schweppe, "Quality of Supply Pricing for Electric Power Systems",
Conference Paper, IEEE Summer Power Meetings 1980, No A-80-084-4.
Peddie, R. A., G. Frewer, and A. Goulcher "The Application of Economic Theory Utilising New
Technology for the Benefit of the Customer", Discussion Paper, South Eastern Electricity
Board, U.K., 1983.
Plummer, James, Terry Ferrar, and William Hughes, editors, Electric Power Strategic Issues, Public
Utilities Reports, Arlington Virginia, 1983.
Plummer, James L., Eugene Oatman, and Pradeep K. Gupta, Strlltegic Management and Planning
[or lectric Utilities, Prentice Hall, 1985.
Po,irajah, R. A. "Derivation and Applications of Optimum Incremental Costs", M.S. Thesis,
McGill University, 1984.
Ramsey, F., "A Contribution to the Theory of Taxation ", Economic 10u/'//al, Vol. 37, March 1927,
pp 47-61.
RAND, "Real Time Pricing and the Future of the Electricity Industry," Unpublished memo by
Derek McKay, October, 1979.
Ray, D. J. "A Cost Analysis of Wheeling in the Electric Utility Industry", PhD thesis, University
of Wisconsin-Madison, 1987.
Ruane, Michael F., "Physically Based Load Modeling", MIT PhD thesis, June 1980.
Sanghvi, Arun "Economic Costs of Electricity Supply Interruptions", Energy Economics, July 1982,
pp 180-198.
Scherer, Charles R. Estimaling Electric Power System Margillal Costs, North Holland Publishing
Co., 1977.
Schuler, Richard E. and Benjamin Hobbs, "Spatial Competition -- Applications in the Generation
of Electricity", manuscript, April 1981.
Schweppe, F. c., "Power Systems 2000," IEEE Spectrum, July 1978.
Schweppe, F. c., "A Rapid Deployment Demand Side Management Tool for Reduction of Summer
Peaks", unpublished manuscript, November 1987.
Schweppe, F. c., "Mandatory Wheeling: A Framework for Discussion", submitted to IEEE
Transactions on Power Systems, 1988.
Schweppe, F. C., R. E. Bohn, and M. C. Caramanis, "Wheeling Rates: an Engineering-Economic
Foundation", MIT Technical Report TR 85-005, September, 1985,
References
259
'fhe following is an annotated list of reports and papers written at MIT and
subsequently at Harvard and Boston University on spot pricing and related
isslles. Many of these were discussed in the Historical Notes and References at
the end of each chapter, but this list provides a self-contained summary. Spot
pricing is one part of a larger overall approach to electric power systems called
Homeostatic Control, so there are some references to Homeostatic Control.
The list is separated into three areas:
" Spot Pricing
" Customer Demand Response Modeling
Deregulation and Wheeling
SPOT PRICING
"Power Systems 2000," by Fred Schweppe, IEEE Spectrum, Vol. 15, No.7, July
1978, 6 pages. An informal discussion of how a decentralized state-of-the-art
power system could work using spot pricing.
"New Electric Utility Management and Control Systems: Proceedings of Conference," by the Homeostatic Control Study Group, June 1979, 200 pages,
MIT-EL 79-024. Discussion papers and audience reaction from a conference on
homeostatic control.
261
262
Annotated bibliography
Bibliography
263
264
Annotated bibliography
Some of the following do not explicitly address spot pricing but contribute to
the development of a solid foundation for the physically based, end use
modeling. Most papers use a control theory approach in which customers
exogenously specify control rules and objective functions. Others use a different
formulation, based purely on profit maximumization by industrial customers.
Electric Load Modeling, by James Woodard, Garland Press, New York, 1979,
350 pages. Provides a deterministic framework for physicaIly based end use
modeling of electrical demand with emphasis on the residential sector.
"Physically Based Industrial Electric Load," by Y. Manichaikul and F.
Schweppe, IEEE Trans. on Power Apparatus and Systems, Vol. 98, No.4,
July/Aug. 1979, 7 pages. Application of the stochastic framewo~k to the industrial sector by individual case studies of seven specific customers.
"Physical/Economic Analysis of Industrial Demand," by Y. Manichaikul and
F.e. Schweppe, IEEE Transactions on Power Apparatus and Systems, Vol.
PAS-99, No.2, March/April 1980, 7 pages. Models loads as random processes.
Does not explicitly model effects of changes in electric rates, but suggests how
to estimate these effects. Based on seven specific case studies.
"Physically Based Load Modeling," by M. Ruane, MIT Ph.D. Thesis, 1980.
Provides a detailed stochastic framework for modeling residential loads.
"A Theoretical Analysis of Customer Response to Rapidly Changing Electricity
Prices," by R. Bohn, 1980, revised January 1981 MIT-EL 81-00IWP, 150 pages.
A series of models of electricity use, emphasizing the response of profit-maximizing customers to spot prices. Derives the increase in customer profits from
various forms of spot pricing. Some discussion of actual case studies, but no
real-life numerical examples. Later incorporated into Bohn thesis.
"Physically Based Modeling of Cold Load Pickup," by S. Ihara and F.
Schweppe, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100, No.9,
September 1981,9 pages. Application of stochastic structures to modeling load
level response to short interruptions.
"A Micro-Processor Based Energy Usage, Rescheduler for Electric Utility Load
Leveling," by J. Wilber, Bachelor's thesis, Massachusetts Institute of Technology, June 198 I. Learn hot water heater usage patterns and then control under
spot prices.
"Electric Load Management by Consumers Facing a Variable Price of Electricity" by P. Constantopoulos, R. Larson and F. Schweppe, 1982 Joint National
Meeting, San Diego, California, October 26,1982,30 pages. See also "Decisions
Models for Electric Load Management by Consumers Facing a Variable price
of Electricity," by P. Constantopouios, R. Larson, and F. Schweppe, in Studies
in Management Science and Systems: Energy Models and Studies, edited by
Bibliography
265
Burton Dean and Benjamin Lev, North Holland Publishing Company, New
York, Vol. 9, pp. 273-292, 1983. General discussion of a theoretical framework
for analyzing customer response to spot prices.
"Computer Assisted Control of Electricity Usage by Consumers," by P. Constantopoulos, MIT Ph.D. Thesis, 1984. Theory of above plus detailed numerical
example of optimization of space conditioning under spot pricing.
"Space Conditioning Load Under Spot or Time of Day Pricing," by P.
O'Rourke and F. Schweppe, IEEE PAS, May 1983, Vol. 102, pp. 1294-1301,
9 pages. Develops simple-to-use formulas to evaluate savings-discomfort
tradeoff for space conditioning under spot pricing.
"Stochastic Modeling and Parameter Estimation of Residential Electric
Loads," by D. Wang, MIT Ph.D. Thesis, 1984. Use of semi-Markov Model with
extensive application to heat pump data from one house.
"Spot Pricing of Public Utility Services," by R. Bohn, Chapters 4 and 5. Models
large industrial customers as profit maximizers with explicit storage and rescheduling options. Solves for optimal behavior under weekly update and 24 hours
update spot prices. Later extended in thesis by B. Daryanian.
D~EGULATION
AND WHEELING
NeIther deregulation nor wheeling is a necessary part of spot pricing, and in fact
they are much less important than core applications of spot pricing. However,
both topics are getting a lot of attention (at least in the U.S. and U.K.), and spot
pricing has a lot of relevance to them.
There are two types of deregulation/wheeling proposals in the following
articles. One is incremental: how can small amounts of competition (or
wheeling) be allowed, on a voluntary basis, within a conventional regulatory (or
public ownership) system. The other is complete: how could a completely
deregulated system function. (The analogy for wheeling is mandatory wheeling
to/from customers.) Chapter 5 emphasized the complete approach, while many
of the following look at incremental issues. (E.g. PURPA qualifying facilities
could be allowed to select spot pricing as the basis of avoided cost calculations,
giving them an incentive to self-dispatch.)
"The Costs of Wheeling and Optimal Wheeling Rates," by Michael C. Caramanis, R.E. Bohn, and F.C. Schweppe, IEEE Transactions on Power Systems,
Vol. PWRS-I, No. I, February 1986, pp. 63-73. The theory of wheeling,
involving regulated utilities with simple examples. Non-intuitive nature of some
wheeling rates.
"Wheeling Rates: an Engineering-Economic Foundation", by Fred C.
Schweppe, R.E. Bohn, and M.C. Caramanis, MIT Technical Report TR 85-005,
September, 1985. Major expansion of the above, covering revenue reconcilia-
tion for wheeling, and more complex cases such as private generators. Mandatory wheeling and related issues.
"Comments on 'Regulation of Electricity Sales - For Resale and Transmission
Services (FERC docket 85-17-000 Phase II),," by F.C. Schweppe, R.E. Bohn,
R.D. Tabors, and M.e. Caramanis, October, 1985. Comments on how FERC
might regulate bulk utility sales and wheeling to encourage optimal prices (spot
price based).
"WRATES: A Tool for Evaluating the Marginal Cost of Wheeling," by M.C.
Caramanis, N. Roukos, and F.C. Schweppe, submitted to IEEE Transactions on
Power Systems, 1988. Summarizes a PC based model for calculating wheeling
rates in multi-bus systems. WRA TES implements the theory of the preceeding
articles.
"Mandatory Wheeling: A Framework for Discussion," by F.e. Schweppe,
submitted to IEEE Transactions PS, \988. Does not emphasize spot pricing.
Instead looks at the many paradoxes and definitional issues surrounding
wheeling, such as "paper wheeling" and the possibility that certain wheeling
transactions can reduce social welfare.
"Marginal Cost Pricing for Mandatory Wheeling," by F.e. Schweppe, unpublished manuscript, 1988. A summary of wheeling issues and equations for a
general audience.
The following papers discuss various forms of deregulation:
"Deregulating the Electric Utility Industry," by R. Bohn, R. Tabors, B. Golub,
and F. Schweppe, MIT Energy Lab Technical Report MIT-EL 82-003, 1982. An
analysis of a deregulated system based on spot pricing. Looks at a number of
possible market structures, with mixtures of regulation and deregulation for
generators, users, and transmission system.
"Deregulating the Generation of Electricity Through the Creation of Spot
Markets for Bulk Power," R. Bohn, B.W. Golub, R.D. Tabors, and F.C.
Schweppe, The Energy Journal, Vol. 5, No.2, 1984, pp. 71-91. A revised version
of the above working paper.
"An Approach for Deregulating the Generation of Electricity," by Bennet
Golub, R. Tabors, R. Bohn, and F. Schweppe, ch. 5 in Plummer, editor, 1983.
Emphasizes complete deregulation.
"Using Spot Pricing to Coordinate Partially Deregulated Utilities, Customers,
and Generators", R. Bohn, F. Schweppe, and M. Caramanis, Chapter 13 in
Plummer et ai, editors, 1983. (Note: the authorship of this article was mistakenly
labeled on the way to the publisher.) Looks at a partial disregulation scenario,
similar to today's system, and why spot pricing should be used as the basis of
transactions by deregulated marketplace participants dealing with regulated
utilities. E.g. why sales by PURPA facilities should be based on spot prices.
APPENDICES
PREFACE TO APPENDICES
~
--r:his appendix summarizes some of the key ideas in the analysis of network
flows and power system dynamics. It also discusses various local controllers that
are on the power plants and scattered throughout the network. Appendix B
discusses centralized control and operation.
A comment on notation: In the main text of this book, "t" denotes discrete
time, usually with one-hour increments. In this appendix, t denotes a continuous
time variable. Similarly in the main text, real power is measured by energy
(kWh) and denoted by y, d, and g as appropriate. In this appendix, real power
is denoted by p or P with units of kW.
SECTION A.I. NETWORK FLOWS
Single-Phase AC Power
The instantaneous electrical power flow pet} at any point in an electrical network
is given by the product of the voltage vet) and the current i(t}. Power system
voltages and currents vary sinusoidally with time (ignoring harmonics). Thus
for some point on a single phase system,
V(I)
i(t)
IX)
where
269
Vmax
Voltage amplitude
Imax
Current amplitude
IX
v(t)i(t)
VmaJmax
+ IX)
(A.l.l )
(A. 1.2)
where by definition
P: Real Power
=
VI cos (IX)
Q: Reactive Power
VI sin (IX)
V
Vmax/Ji
Imax/Ji
The real power component of (A. 1.2) (i.e., P[I - cos (2wt)]) has a time-average
value equal to P while the reactive power component (i.e., Q sin (2(01) has a zero
time average.
Three-Phase AC Power
Single-phase power has a pulsating characteristic. For medium and large loads
and generators, pulsating power is undesirable. This is avoided by using a
polyphase system, usually three-phase.
The relationships among voltages and currents at any point on a three-phase,
balanced electric network are
Va (t)
V sin (WI)
ia (t)
I sin (wt
Vb(t)
ib(t)
+ IX - 21t/3)
V sin (wt + 21t/3)
I sin (WI + IX + 21t/3)
vc(t)
ic(t)
I sin (wt
IX)
Appendix A 271
where "a", "b", and "c" denote the three phases. The total power flow at that
point is the sum of the powers due to each phase, so
(A.I.3)
(A.I.4)
3Vlsin(a)
(A. 1.5)
The fact that the three-phase pet) of (A. 1.4) has no Q sin (wt) component as in
(A.I.2) does not mean there is no three-phase reactive power. After all, each of
the three lines is carrying a pet) that looks like (A.I.2). Reactive power is
sometimes called "imaginary" power. I
Transmission Line Modeling
'7'>
Consider a balanced three-phase transmission line between two buses (or nodes)
of a network. Assume at Bus} and Bus k the "a" phase voltages are respectively
VaJ(t)
Va.k(t)
+ b)
V. sin (wt + 15 k )
with phases "b" and "c" shifted in place by 120 and 240.
A new variable,
8): Voltage phase angle at Bus j,
has been introduced to denote the fact that the phases of the voltage sine waves
vary with location on the network. The voltage magnitudes and angles vary
continuously between buses.
Define
~:
~k:
or -)
Then
(A. 1.7)
272 Appendices
where the summation is over all buses connected to Busj. Qj and Qjk are defined
similarly.
Assume line i connects Bus j to Bus k. The equation that relates ~k to the
voltage magnitudes Vj, /ik and voltage phases bj , bk and the characteristics of line
i is
~k
G,v} -
+ n, V; /Ik
n;
G,V;~COS(JJ -
sin (t5j
t5d
15 k )
(A. 1.8)
-x,
R; +
X;2
x,:
Reactance of Line i 2
Network reactances, resistances, real power, reactive power, and voltage are
usually expressed in a per unit system wherein a reference voltage and usually
MWh level are specified and all variables are normalized with respect to these
references. In a per unit system, the voltage magnitude V is usually within the
range 0.95 to 1.05.
AC Load Flow
The purpose of an AC load flow computer program is to calculate the real and
reactive powers flowing through the transmission lines of a given network for
some specified bus conditions, such as real and reactive power or voltage
magnitude and real power. The network's structure and parameters (e.g., resistances and reactances) yield a set of 2Nb - I (Nb = number of buses) simultaneous nonlinear equations which have to be solved, e.g.,
Appendix A
273
(A.I.9)
~ == [1i';, ... VNb,Dj, ... DNbf
~:
Note: The value of Pb k = .. is not included in u and the value of 15k , k = .. is not
included in ~. The bus .. is called the swing bus (Or reference bus)
The value of x is to be found, given u. The line flows are then easily computed
from x using (A.1.8) (x is sometimescalled the state vector). The real power at
the sWIng bus k = .. is-not specified in order not to overspecify the problem. The
power assigned to the swing bus is the difference between the total load power
and the calculated line losses minus the total power at all the generator buses
except the swing bus. Generally, the angle at the swing bus is arbitrarily set to
zero, since the absolute values of the voltage angles [)j are not important.
AC load flows are solved by iterative techniques. Two solution methods are
"Gauss-Seidel" and "Newton-Raphson."
Gauss-Seidel
This method first assumes initial values for the voltages (magnitude and angle)
at each bus. Then the voltage at the first bus is calculated to match the specified
powers at that bus, keeping the voltages at all other buses at their intial values.
Then the voltage at the second bus is calculated, keeping the voltage at Bus I
at the value just calculated. This procedure is repeated for each bus until
convergence. This method does not converge fast, but is very simple and usually
works.
Newton-Raphson
This method makes corrections to the initital voltage values assigned to the
buses while taking into account the interactions with other buses. Define
!!.)
(If"(.>:, u)
-'=-----
ox
{(x,)(:5.. - ~,)
for x =
~+ I
b'
.
Ian matrIX
is given by solving
(A.1.l0)
274 Appendices
This iteration continues until x j + I ;;:; Xi> which means that !(x j , u) ;;:; O.
Solving the linear system of (A. I. IOns not trivial, since
order can be very
high (e.g., 500 to 2000). Fortunately the Jacobian matrix is very sparse (i.e., it
has a lot of zero entries). Special numerical techniques are used to solve these
sparse matrices.
its
An AC load flow works with four basic types of variables: real powers, reactive
powers, voltage magnitudes, and voltage angles. The nature of the physical
system is such that there is strong coupling between
Real power P and voltage angle J
and between
Reactive power Q and voltage magnitude V
with much weaker coupling between, say, voltage angle and reactive power. This
property is often used to decouple the AC load flow logic into two separate
subproblems, which then iterate with each other.
DC Load Flow
An approximation called the DC load flow is used in this book instead of the
full AC load flow. This approximation yields a linear set of equations relating
real power P to voltage angle <5, i.e., it is "one half" of the decoupJed AC load
flow. It is developed in detail In Appendix D.4
Optimum Load Flow
An optimum load flow is a computer program that tries to find the set of bus
power injections, voltage magnitudes, etc., that minimize some criteria subject
to constraints which include the condition (A.1.9). For example, the criteria
might be to minimize losses where in addition to (A.l.9) the real power flow
through line i is not allowed to exceed a prespecified value. Optimum load flows
can be important in system operation relative to economics and security (see
Appendix B).
Time Variations
Thus far we have been considering the magnitudes and phases of the sinusoids
to be constant in time. In practice, concern is usually with the time variations
of the amplitudes and phases of the sine waves. Let V(t), P(t), and (j(t) denote
these time variations. V(t) can be viewed as an "amplitude modulation" of the
sinusoidal voltage, etc.
Appendix A 275
X: location of Fault
C2
R2
C3
R3
C4
R4
Relays are extensively used for the protection of transmission lines and transformers. A relay contains the logic that decides to open or close a circuit breaker
if certain locally measured conditions are met. Two commonly used network
relays are overcurrent relays and impedance relays.
To illustrate the sophistication of network relaying, consider the three bus
network of Figure A.2.1. Four circuit breakers Cl to C4 are shown. Each has
an associated impedance (also called distance) relay, R I to R4, which detects the
presence of a "fault" and estimates its location by measuring the voltage and
current at the relay's location. The 'x' on the transmission line of Figure A.2.1
represents a fault, which in this case is a short circuit due to a lightning strike
which established an ionized electrical path for current flow to ground (or
between phases). This path is sustained if the potential difference between the
line and ground is high enough.
The following is one possible sequence of events:
At
At
At
At
t
t
t
t
276 Appendices
At t = 0.5 sec + 8 cycles, relays R3 and R4 reopen the circuit breakers if the
fault is still there.
The circuit breakers Cl and C2 did not trip because their impedance relays RI
and R2 determined that the fault was not within their zone of protection. In this
example, the fault is in the zone of protection of R3 and R4, but not R I and R2.
However, ifC3 and C4 fail to open for some reason, R2 will trip circuit breaker
C2 after a preset time delay.
Network Controllers
Relays are used to protect the power plants as well as the network. They are set
to "turn off the plant" when they sense a problem which could damage the
plant.
Because of the massive capital investment associated with any given power
plant, these relays are set very conservatively; i.e., it is much better to "cry wolf'
than to damage the plant. It is the job of the central controllers discussed in
Appendix B to make sure that the sudden loss of anyone power plant does not
affect the service being provided the customers.
Power Plant Controllers
The power plant operators are the most important controllers. A modern power
plant control room has a vast array of displays and switches for the operators'
use. Digital computer-driven display and diagnostic systems are playing an ever
increasing role.
There are also many automatic control loops within a power plant. To
illustrate, consider a fossil steam power plant.
The voltage regulator controls the excitor in order to maintain output voltage
magnitude at the desired set point.
The governor controls steam flow into the turbine so that the frequency does
not drop "too much" as load increases (as will be discussed in Appendix B, the
local power plant governors do not attempt to maintain exactly 60 Hz).
Boilers have extensive automatic firing control loops to maintain pressure and
temperature within acceptable limits while providing the needed steam flow to
the turbine.
The best way to really appreciate what a power plant is and how it is
controlled is to visit one.
Appendix A 277
The number of differential equations used to model the boiler of a steam power
plant and its controllers varies from two to 200. Time constants for boiler
transients range from seconds to 20 minutes.
A turbine is often represented by two to four differential equations whose
time constants range from one to 15 seconds.
The generator is often approximated by a set of two to five differential
equations with time constants from 0.01 to 0.1 seconds.
A key equation of motion for the jth generator is given by Newton's second
law to be
=
HJ. dJ;(t)
dt
~:
(A.3.1)
J;(I): Frequency of generated power which is close to being proportional to the speed
J(t)
(\(D:
Load dynamics are not really understood and so are usually modeled algebraically, i.e., transients are ignored. Typical models are constant impedance,
constant power, frequency sensitive, and voltage sensitive.
SECTION AA. POWER SYSTEM DYNAMICS
Three types of power systems dynamics with different time scales of c('ncern are
discussed.
278 Appendices
Transient Stability
If a short circuit occurs on a transmission line, the protective relays "clear" the
fault within cycles, as discussed in Section A.2. During this time, the abnormal
conditions cause mechanical transients in the generators which are governed by
the swing equation (A.3. I). If thejth generator is close to the fault, then as long
as the short circuit exists, its electrical power output is zero (or very small),
because it is trying to supply a load with zero impedance. Since the mechanical
power input to the jth generator cannot change within cycles, the right side of
the swing equation is positive. This implies a positive acceleration, i.e., the jth
generator starts to speed up relative to the rest of the system. If the fault is not
cleared in time, the generator's speed (frequency) increases so much that it
"pulls out of step," i.e., loses synchronization with the rest of the system.
Transient stability is usually studied by numerical integration of the nonlinear
swing equations plus other differential equations of the generator-voltage regulator models. An AC load flow is done at each time step to evaluate the effects
of network coupling between the generators and loads. Boiler and turbine
dynamics are often ignored. For large interconnected systems (say more than
tOo generators), such numerical integrations can run much "slower than real
time" even with powerful digital computers.
Dynamic Stability
Large interconnected power systems with relatively weak transmission links can
exhibit small-amplitude, low-frequency (one- to ten-second period) sustained
oscillations. This is called the dynamic stability problem. It is usually studied
and analyzed by linearizing the nonlinear differential equations such as used in
transient stability studies about an operating point and then doing eigen valueeigen vector analysis. In general, dynamic stability involves slower dynamics
than transient stability. Turbine dynamics may be included while boiler
dynamics are usually ignored.
Long-Term Dynamics
Long-term dynamics looks at transients that are much slower than either
transient or dynamic stability. To illustrate, consider a two-generator system
where
At t = 0 -, Both generators supply the load
At t = 0, Generator 2 is tripped off due to some fault, then
Appendix A
279
Many good books cover the modeling of three phase AC networks, load flows,
relaying, generation modeling and transient stability at various levels of detail.
Examples of different types include Fitzgerald, Kingsley and Umans [1983];
Elgerd [1982] and Bergen [1986]. A good source for material on boiler-turbine
modeling, dynamic stability and long-term dynamics (also on load flow, etc.) is
the IEEE Transactions on Power Apparatus and Systems. s The December issue
of each year contains a detailed index of papers published that year.
NOTES
I. The term "imaginary" is motivated by the common use of complex numbers to represent power
where Q(I) is the imaginary part. Readers who are new to the area should refrain from the mistake
of associating the term "imaginary" 'to "it doesn't exist"
2. These resistances and reactances are equivalent circuit values resulting from the steady state
solution of a partial differential equation.
3. If (A.I.8) is obvious to a reader, that reader is wasting his/her time by reading this appendix.
280
Appendices
4. The term "DC load flow" arose because the linear relationship between!:.. and .Q is analogous to
the relationship between current and voltage in a direct current network which contains only
resistors. "DC analog" circuits were used to solve for line flows in the days before large digital
computers were available.
5. A cycle is a measure of time equal to 1/60 second assuming the power system operates at 60 Hz.
In many parts of the world, power systems operate at 50 Hz. However, since the authors live in
the U.S.A., a 60-Hz system is assumed in this book.
6. The ionized path from the line to ground should go away in 0.5 seconds.
7. In switching studies, network dynamics can be important but such issues are not addressed here.
8. Starting in 1986, Power Apparatus and Systems has been replaced by three publications, IEEE
Transactions on Power Delivery, Energy Conversion and Power Systems.
A key input to the system economics and security functions (to be discussed in
subsequent sections) is a forecast of what future demand will be, say hour by
hour for the next week, or day by day for next month or year.
Diversity
281
282 Appendices
0k(t): Demand for electricity during hour t for the jth usage device (air con-
ditioner, motor, lighting, etc.) of the kth billing entity (customer) (kWh).
m(t): Vector of meteorological variables during hour t such as temperature and
-humidity.
A key assumption is
Elemental Independence: At hour t and for a given value of '!!.(t), the :!Jk(t)'S are
statistically independent over j and k.
(B.l.2)
There are special cases which violate this assumption, but relative to the present
level of discussion, elemental independence is a reasonable assumption.
The demand of the kth customer is
dk(t) =
I0d t)
j
The conditional expected value E {d(t)lm(t)} of the total demand at hour t for
given weather conditions m(t) is givenby the sum of the mean values of its
components; i.e.,
I I Var{0k(t)I'!!.(t)}
k
Var{d(t)I'!!.(t)} =
Nd E {0k(t)I'!!.(t}}
Nd Var{0dt}I'!!.(t}}
The diversity of total demand can be measured by the ratio of the standard
deviation (square root of variance) u{d(t)lm(t)} to the mean value. Assuming
that the order of magnitude of individual conditional means and standard
deviations is close to one,
Appendix 13283
)n-
.nd
are
.2)
mt
1/2
(f{d(t)l~(t)}
""
E{d(t)l~(t)} ""
1
(
Nd)
(B.I.3)
Since Nd for any reasonable sized utility is very large, (8.1.3) says that
accurate short-range load forecasts are possible if one is able to forecast
E {d(t)\m(t)} - i.e., the dependence of the total system demand on weather and
of course time of day, day of week, etc. 1 The key and very important point of
this discussion on diversity is that
The randomness introduced by the independent variations of the individual usage devices
can be ignored when considering total demand behavior.
One single model structure for the conditional mean demand during hour t is
or
its
;e
Real-world complications have the Fourier series coefficients varying with day
of week and season of year and a weather-dependent component that includes
"heat build-up" dynamics, modeled for example by ARMA techniques. 2
Many different types of short-range load forecast models are in use and not
all "lookalike" (B.IA). However (B.IA) illustrates the basic ideas.
In practice, the biggest source of error in short-range load forecasts is the
effect of errors in the weather forecasts.
SECTION 8.2. SYSTEM ECONOMICS
d
.g
284
Appendices
H
(Btu/hr)
(a)
Pmln
H
Pmax
P(MW)
Pm""
P(MW)
Pmax
P(MW)
(b)
(Btu/kWh)
Pmrn
aH
iiP
Prated
(e)
(Btu/kWh)
t
Figure B.2. I. Curves relating efficiency and costs.
Figure B.2.l shows typical curves relating these quantities with variations in
output power; B.2.! a shows the input-output relation of the plant, B.2.l b is the
"heat rate" (HIP) where the heat rate is the inverse of efficiency, and B.2.l c is
the incremental heat rate. The horizontal axis in the three curves is the actual
power going into the grid. The total electrical power out of the generator is
about 5% higher than P. This extra 5% is used to run the power plant itself
Appendix B 285
of
oP =
oH ($jkWh)
oP
C-
The curves of Figure 8.2.1 are smooth functions. In practice, such curves can
be much less well-behaved. One example is the "valve" point loading" issue
associated with many fossil steam power plants. Incremental heat rate curves for
such plants can look like "saw tooth" functions.
Economic Dispatch
The economic dispatch problem is to find the particular output levels for each
available generator that minimize the total fuel costs while meeting all of the
loads plus line losses. Because of network losses, less efficient generators ($/
kWh) located close to the loads may be used more than more efficient generators
located far from the loads. Typically, economic dispatch optimizations are
recalculated every five to ten minutes with a linear extrapolation (based on a
very short-term load forecast) used in between times. "Raise and lower pulses"
are sent to some generators every two to 20 seconds by the AGC system (see
Section BA).
T\;le equations for economic dispatch are closely related to the spot price
equations developed in the main text of this book.)
Unit Commitment
The unit commitment problem considers longer time scales - for example, hour
by hour for one day or one week. Not all of the generators are needed at certain
times of the day. Unit commitment specifies the daily on/off schedule of generators.
A simple example is illustrated in Figure 8.2.2 for a system with three
generators each rated at 100 MW. If the operating costs of the power plants
increases from Generator # 1 to Generator # 3, the most economical way to
operate the system is as shown in Figure B.2.2.
The simple picture of Figure 8.2.2 becomes more complicated when real-life
constraints regarding each generator are considered, such as
Minimum Up Times: The generator must run for a minimum time.
Minimum Down Times: If shut off, the generator must remain in that state for
a minimum time.
Startup Costs: It takes fuel to heat up a cold boiler.
Ramp Rates: It takes time to go from zero to full load.
Crew Availability: If a plant has two or more generators, the operators may be
able to start only one at a time.
286 Appendices
Demendr----------------------------------------------,
(MW)
200~------~--------------~~------~----------_i
Plants
Plants
1.2
1.2
100r-----------~------_,~----------r_----------~
Plant 1
4am
4pm
Time (Hours)
There are also system-wide constraints such as transmission line capacity limits
and the need to carry operating reserves (see Section B.3).
Taking all the real-world constraints into account, unit commitment becomes
a very complicated problem. Note that economic dispatch is a subproblem of
unit commitment; i.e., in theory, for each possible combination of generators
that can supply the load, an economic dispatch must be run.
Two of many approaches to the unit commitment problem are discussed.
Priority Lists
Given a set of power plants and their operating costs, the generators with
cheapest operating costs are first committed as much as possible. The effects of
the constraints are then incorporated. This heuristic method requires a lot of
insight about a particular system. System operators with experience and a good
knowledge of the system can be very effective. Alternatively, heuristic optimization logics are incorporated into a computer program.
Dynamic Programming
Appendix B 287
Fuel Purchases
Most utilities purchase some or all of their fuel (e.g., oil and coal) on the open
market. This leads to another type of optimization problem. Issues related to
determining fuel contracts and purchases include prices of different suppliers,
transportation costs, storage capabilities, purchasing conditions, etc. Fuel
contracts can be signed for' time spans of months and years with possible
provisions for small weekly adjustments. Linear programming can be an effective tool for optimizing fuel purchases.
Hydrothermal Systems
288 Appendices
Power plants have to be removed from the line for routine maintenance (two to
four weeks per year for a large fossil power plant). This leads to maintenance
scheduling which looks a year in advance to schedule maintenance, taking into
account seasonal demand variations, availability of maintenance crews, etc. A
maintenance schedule determined in January for the rest of the year may be
changed completely in April if a major power plant is forced out in April and
requires extensive emergency maintenance. This can affect maintenance on
other plants (as well as the forced-out plant).
Maintenance scheduling optimization is of the integer programming type.
However, it is usually done using heuristic optimization logics and/or the
judgment of experienced operators.
Maintenance scheduling can be combined with nuclear refueling decision
logics (for a utility with nuclear power plants, that is). Nuclear refueling opti
mization involves complex nonlinear relationships covering long-term fuel cycle
costs (which can span several years). For many nuclear power plants, maximum
capacity is reduced at the end of a fuel cycle to get the most energy out of the
fuel.
SECTION B.3. SYSTEM SECURITY
Section A.2 discussed local relay logics which act to avoid damage to the
equipment. This section discusses system level procedures used to protect the
system in the event of a failure. As an example, assume a heavily loaded
transmission line is automatically withdrawn from the system by local protective
devices. If the overall system is not prepared for such an event, other lines may
become overloaded, and their overload relays may trip them out of the system
as well. This could cascade into a blackout.
System Monitoring
Voltage magnitudes, power flows (real and reactive), circuit breaker status (i.e.,
open or closed) are monitored by measuring instruments scattered through the
system. These measurements are sent to the central control system using the
real-time communication system.
The information received is processed by digital computers and presented to
the operators via display monitors. The computer compares the incoming
measurements with previous ones and upper and lower operating limits. It
warns operators in case of irregularities in the data or measurements that lie
outside of safe operating regions.
State Estimation
Appendix B 289
give the estimate of the state variables which minimize the effects of the measurement and modeling errors.
The most common criterion in state estimation consists of minimizing the
weighted sum of squares of the differences between the measurements themselves and the values of the measurements as computed from the estimated state
variables. A state estimation program can be viewed as a type of AC load flow.
Many similar computational methods are used, e.g., Newton's method, decoupling techniques, sparse matrix methods, etc.
System monitoring can be done using the estimated network conditions
rather than the raw measurements. For example, once the state variables have
been estimated, it is easy to compute all the line flows and to test whether they
and all the voltage magnitudes lie within safe operating regions even if they were
not directly measured.
Corrective Control Actions
290 Appendices
a line outage within the utility depend on the status of the network
generation patterns of the other interconnected utilities. In the case of
dependence, the neighboring utilities may share real-time information on
state of their respective systems. Otherwise external equivalent models are
(developed from off-line studies or identified from measurements).
Operating Reserves
The term "operating reserves" is used in this book to denote the geller,ato,r
reserve the utility has to maintain to prevent blackouts (or major tr"'rtll,"n"'"
and/or tie-line flow deviations) in case of the sudden loss of some generation
tie-line support.
To understand the operating reserves problem, consider again the three-plant
example of Figure B.2.2. If Generator # I fails at 4 AM, a blackout will occur
since no reserve generation is available to take over. Operating reserves is one
way to avoid this. If Generator # 2 shares the load with Generator # 1, but is
not at its maximum output. Failure of Generator # I at 4AM might not cause
a blackout, if the operating reserve of Generator # 2 is large enough and can
react fast enough (see discussion of long-term dynamics in Section A.4).
Operating reserve is sometimes associated with spinning reserve. However,
the term "spinning reserve" actually refers to generators which are connected
(synchronized) to the network but which are not operated at their maximum
output levels. In practice, utilities may also maintain other types of operatingfast-acting reserve such as fast-start gas turbines, pumped and regular hydro,
etc. that can be brought up to full power in less than ten minutes. Operating
reserve can also be "carried" by loads which can be rapidly interrupted by utility
signals or FAPER (see Appendix F) action.
System Dynamics
Appendix B 291
Consider an isolated utility whose power plants are tied together and to loads
by a network that is not electrically interconnected to other utilities. In this case
the role of the AGC is to
Keep frequency close to the desired 60 Hz (or 50 Hz if appropriate).
As discussed in Section A.2, each power plant has a governors which uses
locally measured electrical frequency to increase energy output when frequency
goes down (i.e., when the mechanical power driving all the turbines is less than
the power delivered to the loads and losses). These governors are built with a
"droop" characteristic. Thus, if frequency is initially 60 Hz, then after an
increase in load, generation increases to meet the new demand but the resulting
frequency is less than 60 Hz. This droop characteristic is needed to prevent the
local, independent governors from fighting each other.
For an isolated utility, the AGC readjusts the set points of the local governors
to l'lring frequency back to the desired level. Raise and lower pulses may be sent
to the local power plant governors6 ever two to ten seconds.
Choice of a particular power plant's share in any needed total energy output
change is determined by the economic dispatch logic of Section B.2. Thus, in
addition to maintaining frequency, the AGC also tries to keep the generation
levels as close to the optimum economic dispatch as possible.
:t
"
Interconnected Operation
,-
Life gets more interesting when several independent utilities are electrically
interconnected. Because of short-term economy transactions and longer-term
contracts (see Section B.5), each utility specifies its net scheduled interchange;
which is the total amount of power that is to flow out (in) along the tie lines
connecting the utility to its neighbors.
Consider two Utilities, A and B. The role of Utility A's AGC system is
e
f
1
e
~
1
a
o
In normal conditions, to maintain the sum of the power flowing out (in) over
all of Utility A's tie lines close to Utility A's scheduled net interchange. Thus
if Utility A's generation is greater than its total load plus losses plus its
scheduled net interchage, Utility A's AGC reduces Utility A's total generation.
To maintan frequency close to the desired 60 (or 50) Hz.
Under emergency conditions when Utility B has lost a major power plant(s)
due to local relay actions, to increase Utility A's generation to provide
292
Appendices
B(f(t) -/oj
Pn(t) -
Psch
where
Pn (t)
P"h
J(t)
/0
Desired frequency
Each utility raises or lowers its overall generation proportional to the time
integral of its own ACE(t). We will not go through the analysis here, but the
overall result is the desired behavior.
The beauty of this control logic is that each interconnected utility's AGC
system uses only measurements made on its own system. The only overall
interconnected system coordination needed is to make sure each utility'S net
scheduled interchange is correct - Le., that they all sum to zero.
The AGC systems make no attempt to control individual tie-line flows (unless
there is only one tie line). An AGC system controls only the sum of the tie-line
flows.
Control of Time
The AGC systems keep the overall system frequency close to nominal but time,
as measured by the integral of frequency, can drift. Thus one utility is assigned
the task of comparing the integral of frequency to a time standard and sending
time correction signals to the other utilities say once or twice a day. In normal
operation, USA utilities try to keep the difference between the integral of
frequency and true time to within 3 seconds (usually it is much closer but in rare
cases it can be much worse). In general, electric clocks tend to run a little slow
during the day and a little fast at night. 7
SECTION B.S. INTERCONNECTED SYSTEMS
Appendix B 293
There are various degrees of interconnected cooperation. Two extreme cases are
discussed; independent operation and power pools.
Independent Operation
If Utility A's marginal operating cost AA ($/kWh) is greater than Utility B's AB
for the next hour, Utility A may purchase energy from Utility B instead of
generating the energy itself. The price is often based on a split-the-difference
rule; i.e., the sale price is (AA + AB)/2. Such economy transactions take place
each hour and are made by telephone calls between the system operators. Utility
A may be buying from Utility B while simultaneously selling to Utility C.
Contracts
me
the
:IC
'all
let
"(!R'
IfUtility A wants to buy from Utility B but the energy has to flow at least partly
through Utility C, then Utility C may charge a wheeling rate. In today's system,
wheeling rates often bear very little relationship to the actual marginal costs of
wheeling as discussed in Sections 9.5.
~ss
ne
te,
ed
19
al
of
re
w
Power Pool
A simple power pool uses a single central control system that determines how
energy is to be dispatched from all the utility members' generators to minimize
the total operating cost of all the utilities in the pool. This enables centralized
economic dispatch, unit commitment, and maintenance scheduling. It also
allows for central control of operating reserves and system security.
Power pool operation requires the use of some mechanism to balance the
books, i.e. to transfer funds between the pool members so they payor are paid
for energy obtained from or sent to other pool members. One approach involves
variations on the split-the-difference formulas used for independent operation.
A more sophisticated approach uses the concept of an "own load dispatch."
With this approach, the power pool central office determines, say each week,
how much each utility ought to receive or pay for the energy transactions
performed during the week by evaluating
o
The cost of running Utility A in a way that it would meet its own load without
purchases from or sales to the pool.
The actual costs Utility A has incurred.
294 Appendices
These numbers determine the amount of money that Utility A receives or pays.
Power pool operation does not stop members from having separate long-term
contract arrangements among themselves. For example, Utility A may agree to
sell Utility B the output from a given plant for a period of one year. Then the
two utilities simply inform the power pool office of the arrangement, so that the
capacity of A is decreased and the capacity of B is increased by the same
amount. System operation is not affected.s
Power pools often have free-flowing tie lines. In this case, the utility members
do not have individual AGC systems and there is one AGC system for the entire
pool.
Power pools engage in purchases and sales with other power pools or independent utilities just as if the power pool was a single utility. Utilities within
the pool may also make purchases and sales agreements with utilities outside the
pool.
FURTHER READING
The material discussed in this Appendix (like that of Appendix A) can be found
in many places. One of many good books is Wood and Wollenberg [1985]. The
IEEE Transactions on Power Apparatus and Systems is a good source for
detailed papers.9
NOTES
I. Actually other exogenous variables such as an industrial strike, a World Series baseball game, etc.
can also effect total demand, but we restrict discussion to weather and time effects.
2. ARMA means auto regressive, moving average.
3. Economic dispatch programs provide values of system lambda A. However, some care is needed
in translating the economic dispatch A into the Aof this book as the economic dispath Amay not
include purchase and sales, effects of running gas turbines, and unit commitment effects, all of
which are included in the A of this book.
4. Also called load frequency control (LFC).
5. Governor action may not exist on large base loaded units.
6. In practice, only certain power plants are usually under AGC. Base load units may not see AGC
signals. Neither do most gas turbine peaking plants.
7. This makes the working hours for most people a little longer than they should be; unless they use
their own watches.
8. Such long-term contracts between pool members (and also between independent utilities) provide
motivation for the marketplace long-term fixed price-fixed quantity contracts discussed in the
main text of this book. They serve a useful long-term purpose without affecting the efficiency of
system operation.
9. As of 1986. IEEE PAS is replaced by these three publications: IEEE Transactions on Power
Delivery, Energy Conversion. and Power Systems.
The title of this section can be viewed as a succinct .summary of what power
system planning is all about. Independent of whether the utility is trying to plan
for future generation, network changes, or load management, there are always
two dominant issues:
o
There is no unique scalar criterion to define what is the optimum (best) thing
to do - i.e., multiple attributes.
There is massive uncertainty about what the future will hold.
295
296
Appendices
Multiple Attributes
An example of the multiple attribute (criteria) problem is the desire to simultaneously minimize all of the following:
o
o
o
o
o
In many cases, none of the above criteria are consistent in the sense that a plan
which minimizes one does not minimize any of the others.
Some approaches to this multiple attribute decision making problem are now
discussed, considering both single and mUltiple decision makers.
Constrained Optimization
The idea of simply minimizing a weighted, linear sum of the individual attributes
is usually not satisfactory. However, for the case of a single decision maker,
there are procedures to develop a nonlinear utility or preference function (the
word utility here is not electrical in character) by talking with the individual to
learn and then model his/her preferences. The resulting utility function is then
the desired scalar cost function which can be minimized. This approach is not
directly applicable to the multiple decision maker case because utility functions
of different individuals are usually different. For example, it would be surprising
if a utility chief executive officer, the head of the regulatory commission and a
customer completely agreed on anyone utility function. Various fancy multiple
decision making methodologies based on questioning, voting, etc., have been
published, but their applicability to most power system planning problems is not
obvious.
Sensitivity Analysis
The first two approaches used minimization of a scalar cost function with
possible constraints. Given the resulting, optimum plan, sensitivity analysis can
Appendix C
297
Present
Worth
of
Costs
$
2
G)
0 0
0)
(0
9
3
5
loss of load
Probability
= Inferior Plan
be ""done by perturbing the optimum plan in various directions to see how the
various attributes change. If a particular perturbation yields a small change in
the scalar cost function with a major improvement in one or several of the other
attributes, the perturbed plan might be decreed to be better than the optimum
plan. Alternatively, sensitivity analysis can be done on the numerical values of
the constraints to see how the optimum plan changes. Even the definition of the
scalar criteria can be perturbed.
Tradeoff Analysis
298 Appendices
can be argued further than Plans 2 and 5 should be discarded because, for
example, Plan 2 has much higher costs with only a small improvement in
reliability. A curve drawn through Plans I, 9, and 3 is called the knee of the
tradeoff curve. The tradeoff curve approach does not yield a single answer; e.g.,
the choice between Plans I, 9, and 3 still has to be made. With this tradeoff
analysis approach, the final decision (e.g., the choice between Plans 1,9 and 3)
is left to the final decision maker or makers. After all, they are being paid to
make such decisions. Tradeoff analysis makes it possible for the utimate
decision maker(s) to concentrate on the real issues.
Uncertainty and Risk
Utility planning tries to look many years (10 to 30) into the future. Unfortunately, it is very difficult to accurately predict future events over such time scales.
Examples of major sources of uncertainty include
o
o
o
o
o
o
o
o
Load growth
Cost of capital
Capital cost of new equipment
Development of new technologies
Cost, availability of fuel
Taxes
Treatment by regulatory commission
Environmental standards
Salaries of university professors committed to power system research
Utility system planners can make forecasts of the future behavior of all of the
above, but the general rule is
The Forecast Is Always Wrong!
Hence, the uncertainty associated with the forecasts has to be factored into the
decision making process.
Uncertainty implies risk and the existence of risk introduces new attributes.
For example, assume a constrained optimization approach is chosen and the
present worth of future costs (with constraints) is to be minimized. One plan
may minimize the expected value of the cost but may have very high costs for
certain possible futures. A decision maker who is risk adverse could choose a
plan with a small spread in costs even though its expected value is larger.
Some general approaches for dealing with uncertainty are now discussed.
Use Probability Measures
Appendix C 299
Unfortunately, the specification of the input probabilities is often a very difficult task.
Use Bounds
As when dealing with multiple attributes, tradeoff analysis can be viewed as the
extension of sensitivity analysis for uncertainties into a more formal framework.
One basic idea is to discard plans which are inferior to another plan for all values
of the uncertainty (if only bounds are used) or with some high probability (if
probability measures are used). The concept of the tradeoff curve and the knee
of a tradeoff curve can still be used, although the simple picture of Figure C.l.l
is lost.
l.
Three different characteristics of future loads are often forecasted on a year-byyear basis:
n
r
a
a
300 Appendices
Exponential Models
historical data plotted on log paper. Different growth rates are usually obtained
for total energy and peak demand. The load shape is either assumed to be
constant or adjusted to match both the energy and peak demand forecasts.
Econometric Models
The exponential growth approach ignores the many exogenous variables that
affect load growth such as the state of the economy, demographic trends, etc.
These factors can be introduced by assuming, for example,
d(t): peak demand during year t l (MW)
d(t) = kiai/; [ei(t)]
ei(t): Exogenous variables such as state of economy, etc. in year t
/; [ej(t)]: Pre specified function of dt)
a j : Coefficients which are estimated from historical data
Section 8.1 of Appendix B used an end use modeling framework based on the
demands of individual appliances of individual customers to discuss diversity.
Such an end use framework can be extended into the long-range forecasting
arena to yield a model for the residential sector that depends on
o
o
so that hour-by-hour load shapes can be developed for any future year. Econometric models can then by used to see how electric rates, gas rates, customer
income, etc., effect the customers' (aggregated) appliance holdings. Related end
use models for industrial and commercial customers can also be developed
(although they usually have a different structure).
Appendix C
30t
The big disadvantage of these end use models is that they are hard to develop.
Their big advantage is that they make it possible to ask much more detailed
"what if' questions. For example, the effect of introducing more efficient
refrigerators can be studied.
Comparison of Load Forecasting Models
A production cost model is the heart of all planning studies that involve
generation and/or loads. An annual production cost model looks at a given
future year, with a given load shape, and mix of generator plants, and then
o
E~aluates the fuel consumption and costs needed to meet the load under a
cost minimizing dispatch strategy.
Sometimes evaluates some measure of the generators' ability to meet the load
at all times; e.g. loss of load probability or expected unserved energy.
One key function that distinguishes different types of production cost models
is the approach used to model the effect of generation outages (forced and
maintenance). Another distinguishing feature is whether the load is modeled
chronologically (e.g., hour by hour) or by a load duration curve (number of
hours per year, for example, that the load exceeds a given.levelV
Five types of production models are summarized in Table C.3.1. There are
many variations on these basic ideas and only the main points of each are
discussed. 3
Chronological Simulation
A basic approach starts with hourly demand data for 8760 hours (i.e. a year).
Maintenance scheduling is then simulated to determine which plants are available on a week-by-week basis. Then for each day (or week), a unit commitment
logic is simulated to see what the fuel costs are. Random forced outages of the
generators are handled by derating. For example, a 1000 MW plant with a
probability of 0.1 of being forced out is treated as a perfectly reliable 1000
(0.9) = 900 MW plant.
Method of
Handling
Generation
Outages
Type of Load
Representation
Derating
Chronological
Probabilistic
Derating
Chronological
Load Duration Curve
Probabilistic
Probabilistic
(aggregated)
This approach does not yield useful information on the probability of being
able to meet the load at all times.
Variations on this basic approach work with representative days instead of all
365. For example, a typical weekday, a typical weekend, a peak day, a peak
weekend, for four seasons might be chosen to yield 4 x 4 = 16 representative
days.
Monte Carlo Simulation
A simple version of this approach combines an annual load duration curve with
derating of the generators for both maintenance and forced outages. To illustrate the idea, assume there are three generators whose characteristics are
Generator A
Generator B
Generator C
Peak
Capacity
(MW)
1000
500
100
Outage Rate
(Maintenance
and Forced)
0.2
0.05
0
Derated
Capacity
(MW)
800
475
100
Operating
Cost
(jkWh)
3
5
10
Appendix C
303
Demand (MW)
1400
1275 + 100
= 1375
Ie
'e
800~-----------------~~
Generator A
II
k
8760
Hours
Figure C.3.1 shows how these three plants would be dispatched against an
annual load duration curve with a peak demand of 1400MW. The total fuel
costs are computed by calculating the areas under the curve, multiplied by the
operating costs.
In practice, subperiod load duration curves are often used instead of an
annual load duration curve so, for example, the effects of maintenance scheduling can be handled more directly than by simple derating. These subperiods are
sometimes developed as daily load duration curves for representative days (as
in the chronological and Monte Carlo simulations).
This merit order approach does not yield useful information on the probability of being able to meet demand. The area in the small upper shaded triangle
(from 1375 to 1400) of Figure C.3.1 is not a meaningful measure of how much
demand will not be served.
Probabilistic Simulation
This approach is a generalization of the merit order approach which works with
load duration curves but explicitly handles the random nature of forced outages.
The basic idea relative to Figure C.3.1 is to start with Generator A and to
dispatch it first, assuming it has a 90% probability of being available. Probability convolution is then used to compute a new equivalent load duration curve
(which accounts for the 10% probability of Generator A's outage) that is seen
by Generators Band C. The convolution process is then repeated for Generator
3114
Appendices
::I~
500
I
500
Oemand (MW)
Demand
(MH)
(I~W)
2000
Demand (MW)
2000
Demand
(b)
2000
J~
1.0.~\
....
0.0
. ..
500
(a)
2000
(c)
.............
500
(d)
Figure C.3.2. Graphic illustration of convolution. (a) Loading of a 500 MW plant with 90%
availability. Expected energy = capacity x availability x hours = 75600 MWHjweek; (b) If
the first plant fails, the second plant sees the original customer curve. This event has a
probability of 0.10; (c) If the first plant operates the second plant does not see the first 500 MW
of customer demand. This event has a probability of 0.90; (d) The equivalent load curve for the
second plant is the sum of the two curves weighted by their respective probabilities.
B, etc. This approach yields useful infonnation on both fuel costs and the loss
of load probability and/or the expected unserved energy. Figure C.3.2 elaborates the process of using equivalent load duration curves to model forced outages
of previously loaded units.
There are many variations on the method (other than going to subperiods).
Frequency and duration models can be used to incorporate the length of time
a forced outage can be expected to last. Numerical techniques (such as the
method of cumulants) can be used to reduce the computer time requirements
associated with multiple convolutions.
Appendix C
30S
Some production cost models allow the incorporation of the effects of nondispatchable generation (e.g. wind and solar), load management, time-of-use rates,
etc. This can be done with varying degrees of success with all of the five
approaches, but the details of how are beyond the scope of this discussion.
An obvious question is: Which approach is best? Unfortunately, the answer
is: It depends on the nature of the system, the issues of concern, and the amount
of computer time available. For example, when doing a generation expansion
study for 30 years (as to be discussed in Section C.S), computer time may be a
dominant factor and accuracy of less concern because of the many other
uncertainties influencing long-term decision making. Alternately, when comparing, say, a particular load management scheme with adding a particular
combined cycle gas turbine, the accuracy of the production cost model may
decide the answer. A Monte Carlo simulation has the potential to yield the most
accurate answers if enough computer time is available.
It is an interesting and frustrating experience to try to compare the results of
two different production cost programs run for the same utility.
SECTION CA. FINANCIAL MODELS
A financial model computes a utility's cash flows, i.e., input revenue and output
costs for fuel, maintenance, personnel, loans, taxes, etc. It also worries about the
stock holders (if a privately owned utility) and indices like the debt-to-equity
ratio.
For a given future expansion plan, the financial model is run each year to see
how financially healthy the utility would be. For example, building a large
nuclear power plant might cause a utility major cash flow problems during
construction and result in a large rate shock (Le. jump in rates) when the plant
comes on-line and enters the rate base. Furthermore, the amount of utility debt
can influence the interest rate that has to be paid on loans (i.e. the cost of capital)
which, in turn, influences the choice of future construction decisions.
Financial models are, conceptually, just simple bookkeeping. However, they
306 Appendices
can get very complex in practice. Because of the different ways taxes, depreciation, etc. are handled, a financial model developed for one utility may not be
applicable to a different utility.
SECTION C.S. GENERATION EXPANSION PROGRAMS
For each plan of interest (a particular combination of options), the utility uses
production cost, financial, environmental, etc., models to evaluate the diverse
attributes.
Unfortunately, the number of possible future plans can get extremely large.
For example, there are many possible types of new plants; they come in different
sizes, and they can be built in different combinations in different years. Combining all these options with the many possible load management options, etc.,
leads to an almost uncountable number of possible plans.
Generation expansion programs are tools which allow the computer to search
over at least some of the many possible futures to reduce the number of plans
that have to be considered in detail. A typical generation expansion program is
designed to choose automatically between plans involving new plant types, sizes,
and timing of construction.
Several difficult approaches are discussed.
Target Mix, Heuristic Search
One standard approach has the user specify a desired Target Mix of generation
plants at some future time, say 20 years from now. A possible Target Mix might
be 20% nuclear, 40% base load coal, 30% combined cycle, and 10% peaking
plant. The computer program then starts with the existing generation mix and
builds new plants each year (taking into account construction time, etc.) as
needed to meet demand with some specified measure of reliability. Heuristic
logics are used to move the generation mix towards the Target Mix. A production cost model is used each year to evaluate the fuel costs and reliability levels.
A financial model might be built into the code or run separately. At the end of
the study period, the various attributes of the reSUlting plan are summarized, i.e.
present worth of all capital and fuel, tons of coal burned, etc. Relative to the
discussion of Section C.I, the user-specified Target Mix can be one of the
options to be considered. The program would be rerun many times for various
Appendix C
307
Target Mixes and of course for different values of the input assumptions on
costs, load growth, etc.
Optimization Logic
"'OVO',
'N.:ppenal1::es
Then
(C.S.I)
To illustrate the use of (C.S.2), assume there are three possible plant types and
it is desired to determine how much of each to build.
Let
Type
I
2
3
Nuclear
Coal
Gas Turbine
C)Ky
Aj
$/kW
1200
600
$jkWh
.03
.05
.10
200
Figure C.S.la plots TC)Ky of (C.5.2) versus h to give three straight lines. The
intersection of these lines are then projected down onto the load duration curve
of Figure C.S.l b and the desired capacity of each plant type is determined. The
total fuel costs are computed from the areas under the curve as discussed in the
merit order approach to production costing. In practice, derating to handle
outages is used to modify the Ky.
The number of approximations inherent in this screening curve logic is large.
However, the idea can save a lot of effort if used before going into a big
computer program. Sometimes in this world, one hour of human thinking is
worth more than many hours of main frame computer number crunching. For
example, if the nuclear-coal intersection in Figure C.5.la occurred for
h > 8760, the nuclear plant should not even be considered.
SECTION C.6. NETWORK EXPANSION PROGRAMS
Appendix C
TC
309
r-----------------------------------,
8760 hours. h
We have discussed some of the basic models and procedures used in power
system planning. In practice, there are feedback coupling which have to be
considered, as illustrated by Figure C.7.1. For example, load growth drives
generation plans, which drive network plans, both of which drive financial
310
Appendices
r
Load Management, Conservation
Load Model
Generation
Expansion
Environmental
Impact
Model
I
I
Network
Expansion
Financial Model
Rate Model
I
I
issues, which determine rates, which effect load growth. In practice, the complexity of the overall problem places a premium on breaking it up into pieces
so load experts can work on load modeling, financial experts can work on
financial modeling, etc. The trick is then to put it all back together to get a
reasonable plan for the future.
Combining the feedback coupling of Figure C. 7.1 with the multiple attributes
and massive uncertainties discussed in Section C.I shows why
Power System Planning is an Art, not a Science!
FURTHER READING
Appendix C
3H
discussed in more detail in an MIT Summer course, which has a more complete
bibliography.
Merrill and Schweppe [l984J discuss initial developments in the tradeoff
approach to multiple attribute decision making under uncertainty. Caramanis
[1983] discusses generation expansion logics with emphasis on EGEAS. Munasinghe [1979] is a good book on generation expansion logics. Hyman [1985] is
a good starting point for financial issues. These references just scratch the
surface of the economic literature on power system planning. We assume that
most readers of this book will already be somewhat familiar with that literature.
A good source is Crew and Kleindorfer [1979] but many others exist.
NOTES
I. A change in notation, since usually in this book I is an hourly index.
2. A turned-around probability distribution.
3. The names associated with these different model types can be confusing. We try to follow the most
common usage although there are variations in the literature.
4. The difficulty of specifying these equivalent outage states makes this approach the least popular
of all five.
5. We are biased towards EGEAS since the first three authors of this book were instrumental in its
development [Caramanis, Tabors and Schweppe 1982].
This appendix presents the DC load flow model, which provides approximate
but simple relationships between generation and demand levels at the buses and
real power flows through the lines. These relationships yield explicit formulas
giving the marginal impact on network losses or specific line flows from incremental changes in demand or generation at some bus of the network.
The DC load flow provides an approximate solution for a network carrying
AC (alternating current) power. The term "DC" comes from an old method of
computing a solution using an "analog computer" built out of resistors and
batteries where direct currents were measured.
SECTION D.I. GENERAL RELATIONSHIPS
is given by
(D.l.I)
G,
R; + X/
313
314 Appendices
x;
fli =
R~
+ x/
R i : Line resistance
The equation (0.1.1) can be greatly simplified by making a series of assumptions which are often valid. Assume (>I - >2) is small in magnitude so
cos (<5 1
sin (<5 1
<5 2 )
<5 2 )
(0.1.2)
(<5 1
<5 2 )
I and
(0.1.3)
i.e., the approximation of (0.1.3) yields a lossless line since the power flowing
into one end ZI2 equals the power flowing out the other end - Z21.
Losses on a Line
Zl2
(0.1.4)
Z21
(D. J .3) cannot be used since it is a lossless line model. Therefore, we return to
(D.I.l), and substitute into (0.1.4) to yield
(0.1.5)
I, I/;
I,
(0.1.6)
Zl2
of (0.1.3) into
Appendix D
OJ
(22
[z;]
315
(D. I. 7)
x;
>
Z\2
Zj.
Finally, assuming
R;
yields
(D. 1.8)
Define
e:
(D.I.IO)
316 Appendices
(;iT!;i) -I E.
!i.E.
(0.1.12)
!;i(t!:D- 1
l!.T!!.l!.
(0.1.14)
Most of the approximations introduced so far are quite good for transmission
lines. We now introduce another approximation which is not as good but which
yields equations that are much simpler. This new assumption is
A constant X to R ratio (reactance to resistance ratio) on all lines; i.e., R; =
Il.,
all i.
Assuming ~ ~ R;, so
fl, =
IX
IlR;-1
or in matrix form,
Appendix D
317
(0.\.16)
H
Discussion on the Approximations and Assumptions
The assumptions made through (0.1.14) are often used in actual utility operations and planning for transmission lines. The additional assumption postulating a constant resistance-reactance ratio for all transmission lines may lead to
larger approximation errors than are tolerable. However, the results thus
obtained are useful in increasing one's intuitive understanding of spatial
network pricing and can be probably used in long-term planning studies where
general behavior modeling is important but high accuracy is not required.
All of the OC load flow assumptions can become more inaccurate for lower
voltage, sub-transmission and distribution lines and as the line loading increases. Thus when dealing with certain actual operational conditions, it may be
necessary to use more accurate approximations or the full AC load flow
equations.
The choice of the swing bus may have a nonzero (although small) impact on
the numerical results. This is due to the approximations resulting in (0.1.3),
which effectively assume a uniform power flow (and hence no losses) at every
point of the line joining Buses I and 2. Depending on where the swing bus is
placed, the flows on anyone line mayor may not include losses on other lines
thus resulting in slightly different numerical results. Although these differences
are likely to be negligible in most real applications, the reader should be aware
of them and not be surprised when they arise in simple textbook type examples.
SECTION D.2. A POTPOURRI OF RESULTS
The OC load approximation of Section 0.1 yields many useful results of various
types which are exploited in the main text. Some of these results are derived
below.
Effect of Swing Bus
Consider (0.1.12) and (0.1.14). Since J:. does not contain the swing bus, i.e. does
not contain y*, it follows that
(0.2.1)
(0.2.2)
Define
318 Appendices
Disregarding NQS and GQs terms, the Chapter 7 spot price equations yield
fl.,
P*
aL) + -aaz
(-e - -a
2::
2:: -
IIQS,/
(0.2.3)
'
fl.e
where !:!:.QS,,/ is the column vector of Lagrange multipliers. Using (0.1.12) and
(0.1.14)
(0.2.4)
Define further
lle = :ie.
From (0.2.6) and (0.2.7) it is seen that the price difference across each line,
disregarding !:!:..s.,1' is proportional to the voltage angle difference across each line
or the real power flow through the line.
Impact on Incremental Line Admittance Changes on Overall Network Losses and Flows
-(/By)
an; - --
(0.2.8)
Appendix D 319
o!.jon
oz
an,
so
or using (0.1.11)
~
an;
=
=
Ce,(I)1
--
!-
<5 2 )
~:i(:iT~:i)-1
(0.2.10)
(0.2.11)
Since C and n; are exogenous and Zj is readily calculable, the iJzjiJn j term is easily
quantifiable for practical studies.
Net Revenues of Transmission and Distribution Network
Define
s: Total revenue received by T and 0 network assuming all users pay the T and 0
network, and all generators are paid by the T and 0 network at the optimal spot price.
Since p is the vector of prices and y is the vector of all injections at all buses
except-the swing bus, it follows that
320 Appendices
Bus 1
Bus 3
(0.2.13)
where elements of yare negative for net consumption of electricity and positive
for net generation at any bus.
Using (D.2.S), (D.2.13) becomes
(0.2.14)
+ y*
(0.2.15)
which implies that S > O. When line flows are not limiting, i.e., !!:.Qs.~ = 0, then
the T and D net revenue is proportional to total network losses.
SECTION D.3. THREE BUS EXAMPLE
Consider a three-bus utility with generators at all three buses, demand at two
buses only and interconnected with a single tie line through which it inputs
WMWh as in Figure D.3.!. Assume demands, dl and d2 , are completely
inelastic, i.e. insensitive to the price of electricity. Assume that the marginal cost
Appendix D 321
where a{a~, with $jMWh units and aj with MWh units are specified parameters.
If Pj(t) is the optimal spot price at time t seen by generator j, then optimal spot
price theory and economic dispatch imply that the generation level g; will be
selected to satisfy
subject to 0 ::;; gj ::;; gl.max' Assuming for simplicity that gj.max is not binding, we
have for j = I, 2, 3
aHin [Pi - aj] - In an
if PI > aj
aj
otherwise
(D.3.1)
Assume for simplicity that NMl = Nos., = O. Following Chapter 7 we have the
optimal spot price relationships for j = ), 2, 3:
iJL(t)
p/I)
aZ,(I)
(D.3.2)
In addition to the six equations specified by (0.3.)) and (0.3.2), we have the
energy balance equation accounting for imports W, net injections gj - dj and
losses L,
J
L [g;
- djl
+ W
(D.3.3)
i= 1
I, 2, 3,
(D.3.4)
The above ten relationships must be solved simultaneously to yield values for
Pi' gj' 1', !lOS.I,.; at time t for j = 1,2,3 and i = 1,2,3. To do that, however, the
quantities Land Z; must be expressed in terms of gj' ~, tie line power flows W,
and the characteristics of the transmission network. Here is where the DC load
flow relationships are used. The constant network characteristics are included
in the reduced network incidence matrix A and the individual line resistances
and reactances.
To construct i we select arbitrarily positive directions of transmission line
322 Appendices
o
To model the swing bus (or equivalently to remove the singularity which is
introduced because the overall energy balance relationship is included in the bus
specific power flow conservation relationships), we form the reduced network
incidence matrix A by dropping from A the column that corresponds to the
swing bus. Selecting Bus 3 as the swing bus, we obtain the reduced network
incidence matrix
1 -I]
-I
-I
:J
We now assume a constant resistance to reactance ratio for all lines so the loss
matrix B and the transfer admittance matrix H are given by (0.1.15) and
(D.I.16f After the necessary matrix multiplication and inversion operations,
Appendix D 323
1'11'2
!i
II'
I)
+ I' 2I') )
-I',r2
[
-r)(I',
1',1')
-1'2(r,
1'2)
1'))
-r,r)
The Band H matrices are now completely determined in terms of the network
interconnections, the swing bus location and the line resistances. Therefore the
terms o:::,(t)/og,(t) in (0.3.2) are now quantifiable as the corresponding element
of the H matrix, H;k, and the oL/ogj terms can be obtained by explicitly
considering the partial derivatives of total losses given by yT B Y or ZT R z. Indeed,
noting that l is the vector of power injected at each bus except-for theswing bus,
oL
og,
aL
iJg1
= 2B12 (gl
el2)
2B I2 (g, - d,)
In order to proceed with the numerical example, assume that the line resistances all equal RI = R2 = R) = 4 X 10-- 5 MWh- l , so 1'1 = /'2 = r) = 25000
MWh, which yields
2.666
B
1-333
1.333]
2.666
; !i
[333
=
-333
-333]
666
-666
- 333
To illustrate the impact of line overloads on spot prices, we first assume that
the line power carrying capability is unlimited so the constraints 0.3.4 are not
binding. This implies tha J1QS.,/.i = 0 for i = 1, 2, 3 and that we have to solve
simultaneously only seven equations (0.3.1), (0.3.2) for j = 1,2,3 and (0.3.3)
to determine generation levels and spot prices at each bus and y(t). Because of
the swing bus selection at Bus 3 (0.3.2), '13(t) = yet).
The reader can see by inspection that the resulting simultaneous equations
can be solved by iterating on y(t) and calculating generation and spot price levels
324
Appendices
PI
P2
p) = 1'(1)
g)
1032.5
0
1266.8MWh
Total Losses
99.3 MWh
Line Flows
ZI
Generation
levels
gl
g2
1010MWh
1189MWh
178MWh
Z2
ZJ
until the energy balance equation is satisfied within some tolerance. Assume
price inelastic demands of 200 MWh at Bus I and 2200 MWh at Bus 2 and
imports through the tie line at Bus 3 of 200 MW. Consider further that the
exponential generation marginal cost relationships have parameters value.
50
aj
500
100
a~
10
a~
200
(most expensive)
30
a~
a~
500
(least expensive)
The iterative solution yields the results shown in Table 0.3.1. The reader can
verify the above results by substituting in the appropriate relationships. For
example
gl
and
Table D.3.2. Results with a Flow Constraint on Line 2
Spot Prices
PI
P2
PJ = )'(1)
g,
g2
g)
1154.ISMWh
221.S MWh
902.0MWh
Total Losses
7S.3MWh
Line Flows
ZI
977.4MWh
1000MWh
23.3MWh
Z2
z)
Appendix D
L =
+ 2.666( -
325
2200)2
99.3
We now remove the assumption that line flow capacities are not constraining
and impose a maximum flow limitation of 1000 MWh on Line 2 but maintain
the high capacity assumption for Lines I and 3. Since Z2 = 1189 MWh for
fJ.QS,q.2(t) = 0 it follows that we now have to determine the value of fJ.QS.q.2(t)
which results in Z2 = 1000 MWh. This implies an iteration over both y(t) and
fJ.QS,II.2(t) until the energy balance equation (0.3.3) as well as the power flow
constraint 22 = 1000 MWh are met. Carrying out this exercise we obtain radically different spot prices which reflect the nonzero network quality of supply
component. The results are reported in Table 0.3.2.
NOTES
l. The G, and nJ symbols here are obviously not related to the costs G" etc., of the main text. In
(A.l.8) the real power flow on the line was denoted by Pjk' Here we revert to the notation of the
main text and use z to denote line flows.
Various chapters of the main text have used models for customer benefits and
price response. Some general structural forms for benefit functions and price
response are presented here. A completely different approach, designed to help
individual large customers optimize their response to spot prices, is presented in
Bohn [1982, Chapter 4].
SECTION E.1. SINGLE PERIOD
Since Bk(dk(t is usual\y a nonlinear function in dk(t), the use of (E. 1.2) is an
approximation in that the sum of benefits E[d(t)] depends on the individual dk(t)
k = 1,2, ... , not just d(t).
It is often assumed that customer demand dk(t) depends on the price or rate
the customer pays for electricity. Define
327
328 Appendices
rk (t): Rate kth customer pays for electricity during hour t ($jkWh).
(E. 1.3)
where (E.I.5) assumes that the rate rk(t) is exogenously specified or that the
number of individual customers is sufficiently large that the value of anyone
dk(t) does not affect rk (I). The aggregate model analogy to (E. 1.5) is
GB[d(t) _
Od(t)
-
()
r t
(E.I.6)
The benefit function most often used in the main text (for an aggregate model)
is the
Quadratic Benefit Function:
B[d(t)] =
Bo(t)
ro(t)[d(t) - do(t)]
{I + d;th(~do~Ot~t)}
(E. 1.7)
From (E. I. 7)
GB[d(t) =
Od(t)
r (t)
0
{I +
d(t) - do(t)}
P(t)do(t)
(E.I.8)
so
GB[d(t)] =
Od(t)
ro(t)
Appendix E 329
do(t)
{I +
p(t)[r(t) - ro(t)]}
ro(t)
(E. 1.9)
so that
d[r(t)] =
do(t)
From (E.I.9),
Pt)
ro(t) ad[r(t)]
do (t) or( t)
(E. I. 10)
which means
pet) =
e(t)
where
C:(l): price elasticity of aggregate demand at time I
c: I)
(
r(l) od[r(l)]
del) or(l)
(E.I.IJ)
Obviously there are many other possible structural forms for customer benefits.
Table E.l.l summarizes four such structures: linear response (as discussed
above); power; exponential; and log. It is important to note that
A second-order Taylor Series expansion of the power, log and exponential benefit
functions B[d(t)] about d(t) = do(t) yields the quadratic benefit function,
and
A first-order Taylor Series expansion of the response function d[r(t)] resulting from the
power, log, and exponential functions yields the linear response functions obtained from
the quadratic benefit function.
Hence if d(t) ~ do(t), all four structures are equivalent. However, in actual
applications, the assumption d(t) ~ do(t) may not be valid, in which case a
choice of explicit structures is required. For the power structure {J(t) = S(l) for
all ret) and d(f). Hence the power structure is often called the "constant elasticity" structure. I
Benefit
B[d(l)]
iJBld(t)}
M(t)
--;:;;<I)
Response
dl,(I)]
clo(t)
od[r(I)}
ar(l)
Linear
Power
Bo(1) +
ro(1)[d(t) - d o(1)]
Bo(1)
{I
d(l) - do(1)}
2P(t)d(l)
d(l)
do(t)
+ P(t)do(t)
{I +
do(t)P(t)
--;;w
Exponential
{( d(t)
do(t)
ro(l)d(t)
+ P 1(1)
Y-1(n _I}
Bo(l)
(
\I
P(1) [In
Log
ro(1)d(t)
( d(l)
do(t)
Bo(l)
III
lexp
ro(l)do(l)P(t)
[d(t) - do(t) ]
P(t)do(l)
-
[ Tlln
I I [I
{ fl
{I + In.L}
clo(t)P(I)
d(t)
do(t)
r(l)
ro(/)
+ P(t)
d(I)]
n do(t)
ro(t)
d(I)P(t)
d(t)P(I)
--;(i)
r(l)
PU)
1M
ex [d(l) - do(t)]
P
P(t)do(t)
P(t)
ro(t)
,(I)
ro(t)
d(t)
od[r(/)]
7r(i)
1M
{ 1+ P(/) [r(l) - ro(l)]}
'0(1)
'o(t)
r(l)
P(t)
[I + In J
P(I)
r(l)
'0(1)
I)}
Appendix E
331
The "best" structure in Table E.l.l is not known. However, it should be noted
that since {3(t) is negative, d(t) can go negative for the Linear Response and Log
Response models when ret) ~ ro(t). Thus, special care in their use is required.
The values assumed by do(t), ro(t), Bo(t), and {3(t) will often vary depending
on the context of the discussions. As discussed before, d(t) is viewed as a random
process. One set of assump(ions used in the main text is:
"0(1): Deterministic function of time
Disaggregated Customers
The benefit function and response function have been discussed for aggregate
customer demand d(t). Definition of structures for individual customer benefit
and response functions follow in a completely analogous function. For example,
for the kth customer, the linear response function of (E.I.9) becomes
(E.1.12)
which implies that all of the variables that determine the kth customer demand
for electric energy during hour I can vary with the customer index k.
SECTION E.2. MULTIPLE PERIOD
The structures of Section E.I considered only a single period of time; i.e.,
demand at hour t depended only on price at hour t. In practice, demand
rescheduling from hours of high price to hours of low price will occur. Thus a
mUltiple time period model structure is also required.
We now assume that demand at time t depends on pricE:s at times 1, 2, ... t,
... Twhere I to T defines an appropriate cycle. We consider a 24 hour cycle, i.e.,
T = 24. Extensions to different cycle lengths shorter or longer are of course
possible. The multiple period linear and power response models are derived
below. Multiple period exponential and log models can be developed similarly
by the interested reader.
Linear Demand Model
The linear demand model is a linear relationship between demand at time i and
prices at times 1,2, ... 24. It is obtained from the self and cross price elasticities
of demand at the nominal (original) prices and demands. We define:
d(i): Input nominal demand at time i; i = \,2, ... 24
332 Appendices
r
l
constant for
, ,
I,)
= 1,2",,24
~(i) +
d(i)
~(')
24
L:
1=1
(E,2,1)
P )
1,2",,24
The benefit function consistent with the above linear demand relationship is a
quadratic function which is obtained by integrating the demand relationship,
Indeed, the vector form of the linear demand relationship can be written as
~ =
~o
+ f[p - l]
(E,2,2)
a x
dOC')
EO[' ,]_1_
I,)
lU)
Solving the above for f!.. and integrating between ~o and ~ we get:
[Change in Consumer Surplus when prices change from original (nominal)
(E,2,3)
The power model is derived from constant price elasticities of demand (own and
cross) and nominal price and demand levels.
Appendix E
333
The basic assumption is that the elasticity (rather than the slope as in the
linear model) is constant. That is
. .
ad(i) p(J)
E[I,}] = op(J) d(i) = Constant for i,j, = \,2, ... 24
(E.2.4)
which implies
(E.2.S)
where d(i), d(i), p0(i), p(j) are defined as in the linear model.
Equilibrium Demand
Given a simple multiple period demand response model as described above and
a supply price relationship which assigns a spot price to a given generation level,
it is possible to find the equilibrium price and demand levels. This requires,
however, the simultaneous solution of a system of equations since demand levels
at each time period are coupled. The task of obtaining this solution is made
more difficult by the fact that we normally do not have an analytic representation of the supply price relationship. A simple iterative approach for finding the
equilibrium price and demand trajectories over all time periods in the cycle is
described below.
Step 1: Given input vectors pO, dO set do Jd = do;.pCq = po
Step 2: Using supply price reJatio-nship for each element of do Jd find corresponding vector of supply spot prices p' .
Step 3: Set d OCW = d oJd + a(d - dOld ) where the elements of the d vector are
obtained from (E-:-2.1) or (E.2.S)setting the p(j) equal to elements of vector
p' determined in Step 2 above. The parameter a is a scalar input between 0
and I appropriately selected to avoid divergence. Typical values of a that
work in most cases are between 0.3 and 0.6.
Step 4: Check for convergence by comparing d OCW and do 1d If sufficiently
different, set do 1d = d OCW and return to Step 2. When convergence is reached
i:..Of:W and e.cqUil
f!..' and finish.
set dequil
NOTES
I. Since (E.l.6) defines demand as a function of rate, i.e. d[r(t), it follows that
iJd[r(t)]
iJr(t)
iJ 2 B[d(tJr'
iJ 2 d(t)
F.FAPER
Throughout most of this book, the shortest time interval of interest has been
one hour. The Chapter 9 discussions on price-quantity transactions for security
control and operating reserve considered shorter time intervals down to
minutes, and perhaps seconds. This Appendix discusses a different approach to
the use of loads as fast-acting (seconds to minutes) operating reserves. This
approach is based on the microprocessor controller called a Frequency
Adaptive Power Energy Rescheduler or F APER. A F APER provides fast
responses without any signaling from the utility and without the customer
realizing what is happening.
SECTION F.l. BASIC CONCEPTS
_ _ _~!l4(,",,"~....
Appendices
Turns the heater off when the temperature T(l) at time I (continuous time) is
greater than the specified upper limit T max , i.e. off when T(l) > Tmax.
Turns the heater on when T(l) is less than a specific lower limit Tmin , i.e., on
when T(l) < T min .
A F APER continues to exercise the above two conventional control laws; i.e.,
it
o
o
If heater is on, turns it off if frequency is less than desired and T min < T(t).
If heater is off, turns it on if frequency is greater than desired and
T(t) < Tmax
Since the control logics of the conventional control dominate, the customer
does not "know" that control action is taking place.
Energy consumption (over the long term) is not reduced when frequency
drops. The power that has been used is simply rescheduled to a later time,
where the time shift depends where the temperature T(l) lies between Tmin and
Tmax
The FAPER concept makes sense only if many FAPERS are placed on many
energy type loads. For a single heater, no control action results if the heater is
off when frequency drops. Similarly, for a single heater, the length of time before
it comes back on is "random." Hence the utility cannot know in advance what
effect a single FAPER will have. However, when dealing with a large number
of F APER controlled loads, diversity and probability theory conceptually
enable one to predict the overall effect of the FAPER control. This diversity
issue is the same one discussed in Section B.l. A utility can forecast total
demand over short intervals even though the utility cannot forecast the behavior
of an individual customer.
SECTION F.2. FAPER DESIGNS
Appendix F
Temperature
337
I
I
I
I
;1""------I
Tm..
___. 1. v~----
Tupp[f]
T1ow[1J
Tmin
Frequency
Consider an on-off FAPER of the type discussed in Section F.I. Assume again
a heater is being used to maintain some desired temperature range. One practical
FAPER control logic is
o
o
Turn heater off when T(l) is above Tupp[f(t)]; i.e., T(l) > Tupp[.f(l)]
Turn heater on when T(l) is less than T;ow[f(t)]; i.e., T(t) < T;ow[f(t)]
where the T;ow and Tupp control functions are given in Figure F.2.1. This gives
a smoother, more desirable response than the simple logic discussed in Section
F.1.
The "slopes" of the T;ow and Tupp functions are design parameters. Specification of the reference frequency Jo(t) will be discussed later.
A Continuous F APER
Not all energy type load controllers are of the on-off type. Some employ
continuous controllers. For example, assume (again for a heater) that the
conventional controller acts as follows:
u(t): Power to heater at time t (kW)
u(t)
T(l)
>
Tma ,
K [Tm., - T(l) ]
Tm .. - T min
T min
Tmin>
T(t)
338 Appendices
o
u(t)
K [Tm .. - T(t)]
Tm - Tmin
T(t) > Tm
B[f(t) - /o(t)]
Both the on-off and continuous F APER designs depend on a reference freIn a simple-minded FAPER,!o(t) = 60 Hz (or 50 Hz depending on
the country). However, as discussed in Section B.4, the system AGC does not
keep the system frequency J(t) exactly at 60 Hz all the time, even under normal
operating conditions. It tends to be a little bit below 60 Hz during the day and
above 60 Hz at night. Therefore a more sophisticated design has a reference
frequency!o(t) that tracks the normal frequency variation. This can be done by
putting the measured J(t) through a low pass filter (time constant of hours) to
get !oCt).
quency!o(t).
There are many possible variations on the basic concepts. Two are discussed as
examples.
Operations to Reduce Normal Power Plant Control
The F APER is based on the use of locally measured frequency to exploit the
short-term rescheduling potential associated with energy type loads. It is
possible to replace the locally measured frequency with a more sophisticated
and coordinated control signal generated at the utility's central control center.
Such a signal could be transmitted for example, by radio.
Such central utility control has the advantage of enabling direct coordination
of load control with other utility control functions. The concept of rescheduling
the power flows to energy type loads (and such that the customer doesn't know
it is happening) can be maintained.
The potential disadvantage of such central utility control is that the commun-
Appendix F
339
ication costs might override the benefits of the central action. Another potential
disadvantage lies in the inherent time delays associated with long-distance
communication.
SECTION FA. ANALYSIS OF MULTIPLE FAPER RESPONSE
As discussed in Section F.I, F APERs make sense only when there are many of
them on diverse loads. A key question is
What are the statistical-probabilistic properties of the aggregate responses of many
F APERs in a closed-loop system which includes the effect of load change on frequency? .
The F APER concept has a lot of potential value and could become important
in the power system of the future. However, the F APER concept is like the
340 Appendices
A utility contemplating the introduction of spot price based rates may benefit
from a quick estimation of how the spot prices are expected to behave, now or
in the future. This appendix presents a typical approach to such a study.
A typical study consists of estimates of 24-hour trajectories of the utility's
marginal costs for selected days, complemented by an aggregated characterization of the statistical behavior of hourly costs over large periods of time (a
month or a year). The 24-hour trajectory information is useful when obtained
for typical and peak load days in each season. The statistical behavior of hourly
costs (or hourly spot prices) can be summarized by a cost duration curve that
is analogous to the familiar load duration curve. This appendix presents an
example of the necessary data and how they can be translated to variable cost
trajectories and duration curves. The reader should note that the level of
complexity of the study presented here is intentionally simple in order to allow
implementation on a personal computer using widely available spread sheet
accounting software.
Discussions in the remainder of this appendix will consider a simplified form
of the optimal spot price which ignores network maintenance costs. In particular, the spot price at hour t without revenue reconciliation is considered to
be
341
342 Appendices
A(I)
p(t)
+
+
2Bd(I)[).(t)
IJQS(t)
)'Qs(t)
)'QS]
[Network Losses)
[Network Quality of Supply]
where
).(1) = system lambda at hour f
d(f) = demand level at hour f
= Network losses coefficient consistent with a two-bus model with all generation and load located at each of the two buses
Using
p(l)
multiplicati~e
(I
m)
pet)
In this section we will discuss how a 24-hour trajectory of spot prices can be
estimated. In practice spot price trajectories will vary even if the demand
trajectory stays constant. However, a useful description of all possible price
trajectories is provided by their "conditional expectation," that is, the expected
value of hourly spot prices conditional upon the demand level and other
seasonal parameters (e.g., planned generator maintenance). Estimation of the
conditional expectation of the various spot price components discussed in
Section G.I can be obtained as described below. For notational simplicity we
denote by l(t) a function of d(f), i.e., l[d(l)] is denoted as simply f(t).
System Lambda let)
There are many ways to calculate the conditional expectation of system lambda.
Many available generation planning software tools provide marginal cost information that includes the expected value of the incremental variable cost of
meeting the last kWh of a given load level. This is obtained by finding the
probability that each generator will be the marginal generator given that
demand is at a known level, and then obtaining the average variable cost of all
generators weighted by these probabilities. A plot of expected system lambda
versus load can be obtained by repeating the above exercise for various load
levels. The plot should exhibit a monotonically increasing relationship.
A simpler way to obtain the system lambda versus load curve is the following:
o
Appendix G
343
48
46
44
42
J:
~
"-
!
0
"E
D
.3
40
38
CI
36
34
32
30
IIIIJ
29
26
fIJ
[jIPODIll OIDID 0l1li
24
OCCO 0
Din l1li
22
7
11
13
15
(Ti'>ouoond.)
Load "'~walts
Network losses Bd 2 (t) and marginal network losses 2Bd(t) can be obtained from
the loss coefficient B. The B coefficient can be calibrated from information on
losses at peak load or average losses. For example if losses at peak are known
to be 11 % and peak load is dp then B can be obtained from the relationship
344 Appendices
0.07,-------------------------,
o
0.06
0.01S
o
0.04
o
0.03
0.02
0.01
Load (WW)
Bd~/[dp
Bd~l =
0.11
Load duration curve information can be used in place of the 8760 hourly load
data by appropriate frequency weighted treatment of the various load duration
curve load levels.
Generation Quality of Supply I'Qs(t)
Estimation of the "lQs(t) term using say (6.2.5) or (6.2.12) requires input on
generation-induced loss of load probability (LOLP (/)J at various load levels.
This information is usually available at utility planning departments and can be
obtained from generation outage tables or convolution of generating unit
outage probabilities. It can be often fitted analytically with an exponential
relationship. Figure G.2.2 presents an example of an LOLP1,(t) versus load d(t)
curve. Given the LOLP,.(t) relationship, we have from (6.2.5) or (6.2.12)
where C, is either the average cost to consumers per kWh of unmet demand or
Appendix G
345
alternatively the annualized cost per kW of peaking load capacity divided by the
annual expected loss of load hours due to generation shortfalls (LOLH,).
Typical values of customer cost of unserved energy are $l/kWh, and of annualized capacity costs $30-80/kW.
Network Quality of Supply '1Qs(t)
otherwise
Customer response will undoubtedly have a significant impact on price trajectories and price duration curves. Although the exact behavior of customer
response may be hard to model, the direction of the impact can be investigated
using the multiple period demand response models of Appendix E. The following iterative procedure can be employed.
346 Appendices
Step 0: Input an original demand trajectory which yields a duration curve with
the same shape as the annual load duration curve.
Step 1: Derive spot price trajectories as described in Section G.2 or G.3 using
the demand trajectory of step 0 or step 5.
Step 2: Derive revenue reconciliation multipliers using the spot price trajectory
from step I.
Step 3: Derive a spot price versus generation level relationshp using revenue
reconciliation multipliers of Step 2.
Step 4: Using the supply curve of Step 3 and demand response models of
Appendix E.2, derive the demand trajectory that incorporates customer
response.
Step 5: Update demand trajectory and return to Step I. Repeat until updated
demand trajectory converges to a constant trajectory.
The differentiation of an expression (usually a Lagrangian) involving expectations has arisen a number of times in Chapters 9 and 10, as for example in the
context of predetermined rates or investment optimality conditions. In every
occasion, we interchanged the order of the expectation and the derivative
operators while manipulating the necessary optimality conditions to meaningful
forms. No justification for this interchange was given in the text since it holds
for rather general conditions. The purpose of this appendix is to elaborate on
these conditions.
Consider a function g of two variables x and 11' where 11' is a random variable
with probability density functionfw. It follows that
(H.I)
ex
(1
= -;(IX
f
1l1.R
Using Leibnitz's rule, as long as the functionfw(w) does not depend on x and
347
348 Appendices
the set Rw does not depend on x either, we can interchange the order of
integration and differentiation to obtain
:x
f (g(x, w)/w(w
dw =
f (L g(x, W)j~(W) dw
ox g(x, w)
INDEX
California. xiv
Capacity component, 34, 149
Capacity credits, 7S
CAPITAL, xviii
Capital stock, (See also Revenue reconciliation) 78, 238
Generation capital, 238
Transmission capital, 16.238
Peaking plant capital cost method of
setting quality of supply price. 39.
41-42 (example). 139ff
Chronological simulation, 301-302, 305
Closed loop vs open loop feedback, 103, 106,
249, 250, 252. 309
Communication to customer. 4, 16, 17-18,
84-86, 90, 112-113. 288. 238, 339
Competition, 114, 117ff, 175
Computer, uses of, 92. 109 (footnote)
Conditional expectation, 102, 207, 250. 282.
342, 345
Construction Work in Progress, 191
Contingency analysis, 116, 289, 290
Contingency planning, 113,214-215,289,290
Control: see Power system control
Corrective control action, 289
COSTS, 22 .
Annualized capital costs, 41-42 (example),
349
:352
Index
Index 353
POWER SYSTEM
Control, 22, 269, 274ff, 281, 289, 292,338
Dispatch: see Dispatch, central economic
Dynamics, 80, 269, 277, 280, 290, 291
Operations, 22, 57, 103,275,281-294
Planning, 23,106,113, 139ff,290, 295-311,
305,306-310,311
Power system security control: See
Security control
Predetermined price participant, (See also Rationing), 237, 240, 244, 245
Prepayment, 218, 221
Price discrimination, 123, 187
Price duration curve, 47-49, 237, 242, 243,
244, 345
Price forecasts: See Forecasts, price
Price trajectories, 7, 14-15, 24, 45ff, 47-49,
115,341-345
PRICES, TYPES OF
24 hours update, 12, 12-13, 18,58,63, 114,
206
Contingency prices: See Prices, security
control
Dynamic pricing, 52, 116, 202, 236
Fixed-price-fixed-quantity, 11,66-67,223225
Hourly, 7, 10, 51, 58,63, 80
Mandatory vs. optional prices, 68, 90
Marginal cost prices: See Costs, marginal
Nonlinear prices, 44, 73, 195-198,202
Predetermined price, 206-210, 240, 245
Price only, 10, 58
Price-quantity, xvi, 10, 17,63,63-66,80,
114, 213-222
Real time prices, 27, 52
Security control price, 215-219
Market clearing price, 39 (example)
Spatial price, 35, 168-170
Prices, non uniqueness of, 171
Prices, Predictability of, 7,10,331-333,341345
Priority lists, 80, 286
Probabilistic simulation, 47-49, 303-304
Production cost model, 98, 149 (footnote),
301-305,307,345
Pumped storage, 287, 290, 307
Purchase/sale, 101,293,294
PURPA (See also: Costs, avoided), 75
Quality of supply, (See also Spot Price Components), 38ff 40 (example), 51, 9293, 137, I 39ff, 149, 239, 242, 289,
344, 345
354
Index
Index
355