Optimal Load Shedding Strategy in Power Systems with Distributed
Ding Xu, Student Member, IEEE
Department of Electrical and Computer Engineering
Clemson University, SC29634, USA
Abstract: An optimal load shedding strategy for power systems with
multiple distributed generation is presented. According to
distributed generation’s role in emergency state, they are classified
as the centraI generation type, non-dispatchable type and storage
type. The dynamic and static models of each type following a major
disturbance are developed. Based on the analysis of major
disturbance in grid system, interconnection or distributed generation
system, the load shedding is formulated as an optimization problem
subject to system, operation and security constraints. To handle the
differential equationsthat representthe dynamics of different types
of generators, a discretization technique is used to transfer
differential equations into algebraic equations. The final non-linear
optimization problem is then solved by mathematical programming.
The method is first tested on a small system. Then a 30-bus
distribution system with multiple distributed generation is adopted
to test the proposed method for a more practical application.
Keywords: Distributed generation, Load shedding, Optimization,
Dynamic model
I. INTRODUCTION
The primary function of power systems is to supply
electricity to its customers. However, whe,n the system itself
is in emergency state, it may shed partial loads to ensure the
power supply to important loads, as the last resort to maintain
system integrity. Load shedding in bulk power system has
been studied many years [1 ]-[6]. Basically, these methods
are either based on the simplified dynamic model that treat all
the generators in the system as one equivalent generator, or
based on steady state models. In [7], the optimal load
shedding problem was treated in a different way, The
dynamic model was used and they were not cohered to one
equivalent generator. ‘The problem was then formulated as an
optimization tlmction with system, operation and security
constraints. In solving this problem, the discrezation
technique was used to transfer the differential equation to
algebraic equations.
Load shedding in the power systems with distributed
generation is studied here. Distributed generation is small-
scale power generation that is ‘usually connected to the
distribution system. It could be combustion engine, small
hydro, photovoltaic arrays, fiel cells, microturbines or battery
storage [8]. For utility, distributed generation can help to
maintain system stability [9], provide the spinning reserve
[10] and reduce the transmission and distribution cost. For
end user, the distributed generation can provide power supply
flexibility and improve power quality. For society, the
renewable energy can reduce the emission from the
traditional power plants.
0-7803-6674-3/00/$10.00
0-7803-6672-7/01/$10.00
(C)(C)
2000
2001
IEEE
IEEE 788
Generation
Adly A. Girgis, Fellow, IEEE
Either utility or end user can install distributed generation.
Both of them will play a great role in restructured power
industry. Deregulation separates the power generation,
transmission and distribution. It is hoped that the competition
would reduce power generation cost. However, there will be
electricity price fluctuation in new environment. The
transmission and distribution cost may also be shown clearly
in the electricity bill. To avoid the effect of price fluctuation
and T& D cost, installing distributed generation is a better
comprise for end user. It is also predicted there will be more
distributed generation in utility systems. After deregulation,
the utility distribution companies may no longer own the
traditional power plants. And the competition makes them to
provide the best service to customer. Distributed generation is
good choice for utility distribution company to improve
power quality, power supply flexibility and optimize the
distribution system. In the future, not only the industrial
customers, but also the utility distribution company will
install distributed generation,
More distributed generation in the power systems will
change the structure of distribution system and will introduce.
the new problems. One of the problems is the optimal load
shedding. There are some researches for end-users have been
finished in this field. In [11], the power system in a large
pump station is simulated by computer program. The studies
show the load shedding scheme can maintain the power
supply to important load during disturbance. In [12], the
islanding and load shedding design for captive power plants
is presented. The main duty of load shedding is to maintain
the power balance afler islanding. In [13], the simulation
method is used to design the load shedding in an integrated
process plant. [14] also describe a simulation based load
shedding design process for an integrated steehnaking
cogeneration facility. For the utility system, load shedding
with multiple distributed generation is a new issue. However,
it will become an important subject in restructured power
system.
This paper will fnt study the emergency state in the power
system with distributed generations and formulate the load
shedding problem as an optimization problem. Then the
characteristic of different distributed generation is discussed
and models are developed. Afler describing the solution
approach, the test results on a small system and a bigger
system are given.
 II. EMERGENCY STATE AND LOAD SHEDDING
FORMULATION
A power system with distributed generation may have
following structure.
I
II
I
I
I
o-j pK&’D3
LOAD1 LOAD2
Fig. 1 Typicrd power systems with multiple distributed generation
The distributed generation may spread around
distribution system that is connected to a grid system.
When ‘tiere is a major disturbance, the power system with
distributed generation may enter the emergency state. Load
shedding is one of the emergency control actions. It is needed
in three cases. First, when there is a disturbance in the grid
system, the system operator may request the distribution
utility or industrial customer to shed load to maintain the
system integrity. Sometimes, the distributed generation
system has agreement with the grid system to curtail power in
emergency state [16]. Second, when there is a major
disturbance in distributed generation system itself, for
example, one or more distributed generation is lost, the load
shedding is needed to ensure the power consumption within
certain limit. Third, when there is a major disturbance in the
interconnection, the power systems with distributed
generation may be disconnected horn the grid system. To
maintain the power balance and. system stability, the load
shedding may be conducted in the power system with
distributed generation. All the three cases can be divided into
two categories. In the first category, the distributed
generation system is still connected to the grid system.
Usually, the tiequency in such system is maintained by the
grid system. In the second category, the distributed
generation system is separated ffom the grid system. It
becomes ishmding system. The dynamic model is needed in
analyzing such islanding system.
Basically, there are two objectives for load shedding. The
fust one is to maintain the power supply to the important
0-7803-6674-3/00/$10.00
0-7803-6672-7/01/$10.00
(C)(C)
2000
2001
IEEE
IEEE
loads, The second one is to minimize the cost of load
shedding. In the past, when distributed generation system is
in emergency state, most loads may be shed, Only important
loads are supplied by the nearby distributed generation. This
method may maintain the power supply to important loads,
However, it can not utilize the maximum capacity of
distributed generation. Besides, if there are multiple
distributed generation in the distribution system, it may be
difficult to define nearby generation. In this paper, we
formulate the load shedding problem in the power system
with distributed generation as an optimization problem. Its
objective fimction is to minimize cost of load shedding afier
disturbance while maintain power supply to important load.
In formulating this problem, we first establish the objective
function that is to minimize the cost of total load shedding. It
can be written as
1=1
(1)
where m is the number of loads to be shed. Ci is
the cost of each load.
This objective function should be subject to system,
security and operation constraints. These constraints may also
exist in the bulk power system. However, for the power
system with distributed generation, there may be some
special requirements. Firstly, to ensure the uninterrupted
the
power supply to certain load is the primary function of some
distributed- generation. So there may be some important load
can not be shed at any time. This will add following
constraint
(2)
where P/iis the remaining load (after shedding) at bus i
~fiin is the amount of load that cart not be shed at bus i
Secondly, the distributed generation exists in the
distribution system. Unlike the meshed transmission system,
the distribution system is usually radial system. Because the
restriction of protection system, the reverse power may not be
allowed in some lines. So besides the overload constraints,
there are constraints on power flow direction for certain lines.
It can be written as,
~j>o (3)
<J is the active power flow from bus i to bus j
Thirdly, there are some special characteristics for different
distributed generation. We will discuss them in detail in next
section.
Some other constraints are also included in formulating
this problem. Load flow equations are the system constraints.
It maybe written as,
789
 be slightly affected by the system frequency and voltage. It
~l~n~vnlcos(oin + an -di) - Pgi - P,, = o (4) may be written as
fl=l
where Y is the admittance matrix Pg,= F’(f, Vgi) (lo)
V is the bus voltage
The third kind of distributed generation is storage
~ is the bus angle
equipment. Some storage equipment can be dispatched in the
emergency state. However it has different characteristic.
The frequency, voltage and line flow are also required to
Usually, the storage equipment has certain energy E. If it
be in certain range. It may be written as,
release power fast, the energy would be consumed quickly.
This constraint can be written as,

AP@At< E
t=l
(11)

where P@is the release power at time t.

HI. CHARACTERISTIC AND MODEL OF
DISTRIBUTED GENERATION
As we mentioned in last section, the load shedding is
needed in two cases. In the first case, the power system with
distributed generation is still connected to the grid system.
The system frequency is primarily decided by the grid
system. In this time flame, only the steady state model of
distributed generation is required. In the second case, the
distributed generation system is disconnected from the grid
system. The system frequency is decided by the power
balance in the islanding system. The dynamic model is
required to analyze the load shedding in such case. So two
models for each distributed generation are developed in this
paper.
E is the total available energy in the storage equipment.
Another consideration is the reactive power output of
distributed generation that is more diversified than the
traditional generation. Some renewable energy based
generation, such as wind generation, does not generate
reactive power. There are three methods to handle reactive
power for these generations. They may obtain reactive power
from the system, by fixed amount of capacitor, or by
controlled capacitor with fixed power factor. So there are two
kinds of reactive output constraints for such distributed
generation.
Q. = Qf. or Q. = W’g) (14

For the steady state model, there are three kinds of The dynamic models of the distributed generation are more
distributed generation. The fwst kind of distributed complicated. For the distributed generator that equipped with
generations may include gas turbine, combustion engines and governor, a similar model for the central generation may be
hydro generation. Those generations are similar to the central used. Their diagram is shown in figure 2 and can be written
generation. They can be dispatched in emergency state. as
There may be two constraints for this kind distributed
generation. That is the output and the ramp rate. It must be
pointed out that minimum output of some generation is an (14)
important constraint because of the cogeneration. They must
where Pm,is the prime mover input power
generate certain power to ensure the heat supply. The ramp
rate exists because the generator needs certain time to P. is the prime mover output power
increase its output. These constraints can be written as,
P=is the power in generator

For the wind turbine, there is no governor control. Its

~git ~ Apgi ,,mit
(9) dynamical model can be represented as two connect
dynamical components [ 15], [17]. Its mathematical equation
Where APgi, is the increasing output tlom moment t-1 to would be,
moment t.
●
-( DG-D,)uG +( DG-D, )m, +~T _~pG
to~ =
The second kind of distributed generations include some MG MG MG m MG
renewable energy, such as wind turbine, photovoltaic array. . (15)
These generations can not be dispatched in emergency state.
Their outputs are mainly decided by the input energy, such
wind or solar thermal. Some type of generation output may

0-7803-6674-3/00/$10.00
0-7803-6672-7/01/$10.00
(C)(C)
2000
2001
IEEE
IEEE 790
 x,+l=xi +At*dxldt (17]
APm, APm
1 Aco The dx/dt is treated as an independent variable in the
Ref. +
z optimization problem. So only the algebraic equations is
— involved in the entire problem. Similar to the first case, the
problem can be solved as a non-linear optimization problem,

AP v. EXPAMPLE FOR SAMLL SYSTEM

The proposed method is first tested on small system as

1
shown in figure 1. Generator 1, 2 and 3 generate 6MW,
F 10MW and 2MW. Generator 2 is non-dispatchable generator.
Load 1, 2 and 3 consume 8MW, 6MW and 8MW. The
distributed generation system obtain 4MW from the grid
system. In the first case, the power systems with distributed
Fig 2. Generator Model generation lose its 10MW generation and there is contract
between the distributed generation system and grid system
Iv. SOLUTION APPROACH that limits the power transmission to 6MW. In this case the
static model is needed, We also assume the 2MW DSG is
As we mentioned before, there are two kinds of load non-dispatchable generator. Using the optimization algorithm
shedding problem. ln the first case the distributed generation to conduct load shedding, we obtain following results
system is still connected to the grid system. In this case, the
generator is using static model. However, the generator ramp Table 1
rate and storage constraints need to be included in this model.
Although there is no differential equation involved in this
optimization problem, the discretization is still needed
because of these time-related equations. In solving this
problem, the load shedding is divided into several steps. The
system, ,operation and security constraints should be satisfied
in each step. Then the time related equations, such as the
In the second case, there is a disturbance in the grid
generator ramp and storage feature can be represented as the
system. The distributed generation system is required to cut
additional constraints. So the entire problem become a typical
the total electricity consumer tlom the grid system to 2 MW.
non-linear optimization problem. The mathematical
The distributed generation system needs to conduct load
programming can be used to solve it.
shedding. This is still a static load shedding. Using the
developed algorithm, we can obtain,
In the second case, the distributed generation system is
disconnected from the grid system. In this case the frequency Table 2
is changed because of power unbalance, Here brings two
problems, fwst one is how to estimate the power unbalance in
such system, If we neglect the oscillations between different
distributed generator, at the power change and frequency
have this relation,

(16)
This equation can help us estiinate the power unbalance.
From measurement, we can obtain the frequency and its rate
of change. Using these two parameters, total power unbalance
in the system can be estimated. These unbalances are divided
among the dispatchable generators according to their inertia.
This value can be used as the initial value of each generator.
In the third case, there is a disturbance in the
interconnection. For example, one line is open in the double
line connection. The power limit in the interconnection
becomes 3MW. This is still a static case. Using the proposed
method, we obtain,
Table 3

The second one is the dynamic models that is needed in the

islanding system. The dynamic models is represented by
differential equations. To handle the differential equation in
the problem, we also divide the load shedding into several
steps with time. The differential equations can be discretized
by following formulaj

0-7803-6674-3/00/$10.00
0-7803-6672-7/01/$10.00
(C)(C)
2000
2001
IEEE
IEEE 791
 In the fourth case, we simulate the disconnection of the
grid system and distributed generation system, In this case,
the generators are modeled as the traditional generator with
dynamical model. So the load shedding is divided into several
steps.

Table 4
29
I
Step 1 Step 2 Step 3
Load 1 0.08 0.0796 0,0703
Load 2 0.0521 0.0504 0.0501
Load 3 0.0625 0.0616 0.0704

If the generator 1 is a storage equipment, and it releases

power at 5MW, then we obtain I
Table 5

If the releasing speed change to 7MW, them we have

Table 6
Fig 330 bus system

Step 1 Step 2 Step 3

Load 1 0.08 0.08 0.0736 Table 7
Load 2 0.0551 0.0541 0.0533
Load 3 0.0654 0.0649 0.0733 step 1 step 2 step 3
Load 2 0.522 0.4818 0.397
Compare the above results, we found the releasing speed of Load 4 0.936 0.8368 0.731
storage affects the load shedding schemes.
Load 9 0.189 0.1799 0.148
Load 11 0.336 0.3101 0.271
VI. EXPAMPLE FOR 30-BUS DISTRIBUTION
Load 12 0.657 0.5742 0.473
SYSTEM
Load 13 0.783 0.7454 0.651
,. A. AA
To test the possible application for a bigger distribution ‘ 14 “ ‘--
u. (ZY U.!305J 0.608

=i=l
system with multiple distributed generation, the proposed m- 0477 0.4193 0.366
method is also tested on a 30 bus distribution system [15]. - ___49 0.5226 0.512
This is a typical distribution system with radial structure. In .,.17 tI 0477
-. . ..- 04403
-. 0.385
this system, we assume there are distributed generation at bus 0.432 0.3776 0.348
10, 17, 24, 29 with different characteristics. Before the
disturbance, the grid system supply 4MW to the system with
distributed generation. When there is a major disturbance in
—-—- —-
t, –– J-.
1=
I
0.672
0.495
mfiaaln
0.6397 0.559
0.4569, I n
AnAnt
-. MR
. .-
- ---

the interconnection, it may cause the disconnection between

.-———— I
the grid system and distributed generation system. This nsd 92 1 CM71 17
equals to the system loss about 1/3 of its generation. To make
r, “.””

sure the power supply the important load, we use the 1

I n KAal
,

nbmm[ - A~
proposed method to decide the load shedding. Because of the
loss of interconnection, the distributed generation system
becomes a islanding system. We use the dynamic system
model to formulate the optimization problem. The load lLoad 29 I 0.81
shedding is conducted in three steps. Table 7 is the final Load 30 I 0.8[ , ., -.. .-1
result of this method. This case shows the proposed method
can be used for the relatively larger system

0-7803-6674-3/00/$10.00
0-7803-6672-7/01/$10.00
(C)(C)
2000
2001
IEEE
IEEE 792
 VII. CONCLUSIONS
The optimal load shedding strategy in power system with
distributed generation was studied. According to its role in
the emergency state, the distributed generation is divided into
central generation type, non-dispatchable type and storage
~pe. Then the dynamic and static model for load shedding is
developed. The load shedding problem is formulated as an
optimization problem. The proposed discretization method
solved the problem successfully. The test results show the
proposed method may be used in the distribution system with
multiple distributed generation.
There are two potential applications for this method.
The fwst one can be used in the planning stage to study the
system for emergency plan. For this application, the
calculation speed is not very important. So the more detailed
system can be used. And the more cases need to be prepared.
The second application is for real time application. In this
application, this method can be good helper for system
operator to decide load shedding. With today’s powerti.d
computer, this method can be successfully applied to the
increasing sized system in real time.
VIII. REFERENCES
[1] P.M. Anderson, M. Mirheydar, “An adaptive method for setting
undefiequency load shedding relays,” IEEE Transactions on
Power Systems, vol. 7, no. 2, May 1992, pp. 647-655
[2] Y, Haievi, D. Kottick, “Optimization of load shedding system,”
IEEE Transactions on Energy Conversion, vol. 8, no. 2, June
1993, pp. 207-213
[3] A.A. Girgis, W.L. Peterson, “Adaptive estimation of power
system frequency and rste of change of for calculating sudden
power system overloads,” IEEE Transactions on Power Delivery,
vol. 5, no. 2, April 1990, pp. 585-591
[4] L.P. Hajdu, J. Peschon, W.F. Tinney, D.S. Piercy, “Optimum
load-shedding policy for power systems,” IEEE Transactions on
Power Apparstus and Systems, vol. 87, no.3 March 1968, pp.
784-795
[5] K.A. Paianiswamy, J. Sharrn& “Optimum load shedding taking
into account of voltage and frequency characteristics of loads,”
IEEE Transactions on Power Apparatus and Systems, vol. 104,
no. 6, June 1985, pp. 1342-1348
[6] M.A. Mostafaj M.E. E1-Hawary, G.A.N. Mbsmsdu, M.M.
Mansour, K.M. E1-Nagar, A.M. E1-Arrrbaty, “A computational
comparison of steady state load shedding approaches in electric
power systems,” IEEE Tmnsaqtions on Power Systems, vol. 12,
no. 1, February 1997, pp. 30-37
[7] Ding Xu and Adly. A. Girgis, “optimal Load Shedding Using
Dynamic Modeling,” IASTED Conference, Power and Energy
Systems (PES), Marbell& Spain, September 19-22,2000
0-7803-6674-3/00/$10.00
0-7803-6672-7/01/$10.00
(C)(C)
2000
2001
IEEE
IEEE
[8] John Doughlas, “Power Delivery in the 21st Century,” EPRI
Journal, summer 1999
[9] M.K Donnelly, J.E. Dagle, D.J. Trudnowski, and G,J. Rogers,
“Impacts of the Distributed Utility on Transmission System
Stability,” IEEE Transactionson Power Systems, vol. 11, N0,2,
May 1996, pp. 741-747
[10] P.J, Okane, B. Fox and D.J. Morrow, “Impacts of embedded
generation on emergency reserve”, IEE Proc- gener. Transm.
Distrib. Vol. 146, No, 2, March 1999, pp. 159-163
[11] S.R. Shilling, “Electrical transient and underfrequency load
shedding anatysis for a large pump station”, IEEE Transactions
on Industry Application, vol. 33, No. 1, Jarr/Feb 1997, pp. 194-201
[12] K. Ragarnarri and UK Hambarde, “Islarrding and load shedding
schemes for captive power plants,” IEEE Transactions on Power
Delive~, vol. 14, No. 3, July 1999, pp. 805-809
[13] G. S. Grewat, J,W. Konowalec and M.M. Hakim, “Optimization
of load shedding scheme in an Integrated Process Plant,” IEEE
Transactions on Industry Application, vol. 35, No. 4, July/August
1999, pp. 959-967
[14] C.T Hsu, C.S. Chen and J.K Chen, “The load shedding scheme
design for Integrated steelmaking cogeneration facility,” IEEE
Transactions on Industry Application, vol. 33, No.3, May/June
1997, pp. 586-592
[15] J. Cardell , Integrating small scale distributed generation into a
deregulated Market: Control strategies and price feedback.
Massachusetts Institute of Technology, Ph.D. thesis, 1995.
[16] G.J. Nolan, V.J. Puccio, C.W. Cathoun, “ Standby power
generation under utility curtailment contract agreements”, IEEE
Transactions on Industry Application, Vol. 33, No. 6,
November/December 1997
[17] R. Chedid, N. LaWhite and M. Hit, “ A comparative analysis of
dynamic model for performance calculation of grid -connected
wind turbine generators”, Wind Engineering, VOI 17. No.4. 1994
pp 168-182
Ix. BIOGRAPHIES
Ding Xu, is a Ph.D student at Clemson University, Clemson, South Carolina.
He obtained his B.S and M.S. horn Tsirrghua University, Beijing,China.
Adly A. Girgis, is a fellow of the IEEE. He received the B.S. (with
distinction first class honors) and M.S. degrees in Electrical Engineering
from Assuit University, Egypt. He received the Ph.D. degree in Electrical
Engineering from Iowa State University. He is currently Duke Power
Distinguished Professor of Power Engineering in the Electrical and
Computer Engineering Department, Clemson University, and the Director of
Clemson University Electric Power Resewch Association (CUEPRA). His
present research interests are rerd-time computer applications in power
system control, instrumentation and protection, signal processing, and
Kalmarr filtering applications in power systems
793