Sei sulla pagina 1di 6

Paper accepted for presentation at 2003 IEEE Bologna PowerTech Conference, June 23-26, Bologna, Italy

A Load Shedding Algorithm for Improvement of


Load Margin to Voltage Collaplse
F. M. Echavarren, E. Lobato, L. Rouco, Member IEEE, M. I. Navarrete, R. Casanova and G . L6pez

lines) the load margin to the saddle node bifurcation may be


Abstrad-Voltage stability is concerned with the ability of a too low, and the power system may become close to voltage
power system to maintain acceptable voltages at all buses. A collapse.
measure of the power system voltage stability is the distance to In power system operation planning studies, the System
the saddle node bifurcation of the power flow equations, which is
called the load margin. When the power system load is very high, Operator analyzes the forecasted 3tate of the power system and
andor there exists a large generation-demand imbalance in the performs in advance appropriate control actions to fulfill the
power system areas the load margin to the saddle node security criteria. This control actions consists of voltage
bifurcation may he too low, and the power system may become control resources (transformer taps, shunts reactors and
close to voltage collapse. In case that active and reactive power generation bus voltages) adjustments and modification of the
generation resources in the importing areas are exhausted, generation dispatch (increase and decrease of generation in
corrective load shedding may beeome the last option. This paper
presents an LP-based optimization load shedding algorithm to several units, and connection of off-line ones).
improve the load margin. The objective function consists of In on-line voltage security analysis, when the power system
minimizing the total system demand decrease. First order is getting close to the load margin, the available control
sensitivities of the load margin with respect to the load to be shed actions may have been exhausted or there may not be fast
are considered. The performance ofthe method is illustrated with enough to prevent the system voltage collapse. Therefore
highly loaded scenario of the Spanish power system. emergency load shedding measures may be required. However,
it must be considered as a last resort, so demand reduction
Index Terms-Voltage stability, load shedding, optimization.
must be minimized (load shedding in the importing areas and
I. INTRODUCTION an equal amount of generation reduction in the exporting
areas).
In recent years, voltage stability has become one of the This paper presents a LP-based optimization load shedding
most important problems in the power system operation. algorithm for load margin improvement. First order
Voltage stability is concerned with the ability of a power
sensitivities of the load margin with respect to the load to be
system to maintain acceptable voltages at all buses in the shed are considered. The algoriihm selects both the optimal
system under normal conditions and afler being subjected to a location of generation and demand buses, and their
disturbance [I]. corresponding power reduction. The objective function
Voltage stability is a dynamic phenomena. However, the consists of minimizing the total system demand decrease. The
voltage instability of a power system can be measured
problem constraints consist of 51 target improvement of the
obtaining the distance of the static power flow equations from load margin, the power balance equation and the generation
the initial point of operation to its saddle node bifurcation and demand limits.
point, known as the voltage collapse point. To calculate this The load margin improvement is approximated with a linear
distance, the bus injections are increased in each bus by a
constraint built based on sensitivities with respect to the load
constant rate of initial injections and a scale factor called load to be shed. The main problem of the load margin linear
margin [2]. formulation is the loss of accuracy for large variations of
When a power system load is very high, and/or there exists
generation and demand. To overcome this difficulty, the
a large generation-demand imbalance in the power system algorithm has been designed as an iterative process that
areas (producing large power transfers between exporting and imposes an additional constraint of the total demand reduction
importing areas and high voltage drops in the interconnection in each iteration.
Voltage collapse sensitivity studies have been proposed in
F. M. Echavarren (email: Francisco.Echavarren@iit."pco.es), E. the literature [3], to evaluate the influence of arbitrary
Lobato (*mail: Ennque.Lobata@iit.upco.~) and L. Rauco (email: parameters on the load margin. Several authors concentrate on
Luis.Roueo~iit.upco.ff) re with the School of Engineering of
Universidad Pontificia Cornillas, CIAlberto Aguilera, 23, 28015 Madnd,
the effect of the variation of the load margin with respect to
Spain. topology changes caused by contingencies [4]. In fact,
M. I. Navarrete and R. Casanova me with N J R A , C/Avda. de preventive load shedding algorithms have been proposed in the
Bmselas, 35,28108, Aleobendas, Madrid, Spain. literature [5], taking into account contingencies and improving
G. Idpez is with Red Elbtnca de EspaAa, P. Conde de 10s Gaitanes,
177, La Moralela, 28019 Madrid, Spain. a load margin estimation based on first order sensitivities. This

0-7803-7967-5/03/$17.00 02003 IEEE


paper focus on a corrective algorithm, using the sensitivity demand. To calculate the distance to the saddle node
factors of the load margin with respect to load shedding, and bihcation in the direction given by the parameterized vector
.improving the real load margin iteratively Those sensitivities Fa, the load margin 1 is introduced. Then, the augmented
are calculated from the Lagrangian multipliers of the non- active and reactive power flow equations can be expressed as
linear optimization problem which solution is the system follows:
saddle node bifurcation [6].
The paper is organized as follows. Section I1 provides an
F(B,V)-F,-l.Fa = O (4
The saddle node bifurcation is obtained solving a non-linear
overview of the algorithm. Section 111 explains how the load
optimization problem [ 6 ] : The objective function is to
margin sensitivities can be obtained from the Lagrangian
maximize 1 and the constraints are the augmented power flow
multipliers. Section 1V describes the mathematical formulation
equations:
of the proposed algorithm. Section V illustrates the
max /z
performance of the proposed algorithm applied to an actual
s.t. F ( B , V ) - 8 - - I . F A = E = O (3)
scenario of the Spanish power system. Finally, Section VI
contains the conclusions of the paper. where E is the right-hand-side of the equality constraints.
Let p be the lagrangian multipliers corresponding to the
11. OVERVIEW OF THE ALGORITHM equality constraints. Then the Lagrangian of the problem is
Fig. I depicts a flowchart of the proposed algorithm. The built as follows:
algorithm starts from the initial saddle node bifurcation L = 1-pT.[F(B, V) - F, - -I.Fa] (4)
solution of the base case. The algorithm has been designed as The optimality conditions are obtained derivating the
an iterative algorithm, where limits on the demand reduction
Lagrangian with respect to the load margin, the state variables
have been imposed in each iteration to avoid the loss of
x ={e,V ) and the Lagrange multipliers:
accuracy caused by the linear approximation of the load
margin. While the desired security load margin is not reached, V A L=l+pT.F, =O (5)
a new iteration is performed updating the sensitivities and
VJ = -F:,,LI =O (6)
solving a new LP optimization problem, obtaining the optimal
location of both generation and demand buses (and their V A L= -F(B, V ) + F, +l.FA = 0 (7)
corresponding power reduction) and calculating the new load where Fx is the Jacobian matrix of the power flow equations.
margin. Section 111 details the mathematical expressions of the Therefore, in the saddle node bifurcation, the augmented
sensitivities, and Section N contains the LP-optimization power flow equations (7) must be satisfied, and also there
problem of each iteration to obtain the minimum load exists a non-empty vector (5), p , that multiplied by Jacobian
shedding. matrix is equal to zero (6), so the Jacobian matrix of the power
flow equations F, is singular.
lNlTl4L SADDLE NODE BIFURCATION SOLUTION
The Lagrangian multipliers p are the sensitivities of the
objective function with respect to the right-hand-side of the
equality constraints [7]:

1 - CALCULATDN OF THE SENSKlVlTlES Assuming a proportional relation between the initial bus
injections F, and the parameterized change in bus injections

I SOLVE LP OPTIMIZATION PROBLEM


I FA, let 7 be the proportional ratio between them:
Fa = T F ~ (9)

I ' NEW SADDLE NODE BIFURCATIONSOLUTION


is applied in bus i ,
In that case, if a power redispatch WO,
the right-hand side of its Corresponding power flow equation
E, will be modified as follows:
Fig. I . Flowchart of the iterative optimization load shedding algorithm
AE~= .(I + a.7j) (10)
111. LINEAR APPROXlMnTION OF THE LOAD MARGIN Using(lO)in(l):
Let us start with the power flow equations:
F(B,V)-F, = O (1)
where F, is the initial bus injection. Thus, the sensitivity S, of the load margin 1 with respect
Let FA represents the parameterized change in bus to the initial bus injection F,, of the bus i :
injections due to variations in generation dispatch and load
PG; : Lower limit of generation in bus i .
POo, : Initial active demand in bus i
Therefore, the linear approximation of the load margin can S O , : Sensitivity of the load margin R with respect to
be expressed as follows: the initial active demand in bus i
a = n* + Cs,.q, (13) QD,, : Initial reactive demand in bus i
S O , : Sensitivity of the load margin 1 with respect to
where 1 ' is the initial value of the load margin, and MO, the
the initial reactive demand in bus i
bus injection redispatch in the i -th bus.
C Unfeasibility cost of the residual variable
In order to generalize these expressions, let us separate F,
1' : Initial value of the loatd margin
and F, into two terms corresponding in both cases to
4, : Security value of the load margin to be reached
generation and demand
MDR : The maximum demand redispatch allowed in each
F, = FG, - FDo (14) iteration
F, = FG, -FDA (15) 2) Variables
Hence, there exists a proportional ratio qG between initial APG,, : Active generation variation in bus i
bus generation FG, and the parameterized change in bus APD,, : Active demand variation in bus i
generation FG, : AQD,,, : Reactive demand variation in bus i
FG, = qG.FG, (16) H,: is the Residual variable included to avoid
and a proportional ratio qD between initial bus demand FDo infeasibilities
and the parameterized change in bus demand FD, : B. Objective function
FD, = qD.FDo (17) The main objective function is to minimize the total active
Thus, two different sensitivities of the load margin with demand reduction.
respect to both initial bus generation and demand may be
formulated
Also a penalized residual variable H A is included in the
objective function with an unfeasibility cnst c >> 0 to prevent
the unfeasibility of the optimizatian problem.
CH, (22)
So the complete obiective funci:ion will be:
Now the linear approximation of the load margin can be
expressed as follows:
n = a*+ CSG, CSD,
.AFG~,+ ,AFD~~ (20)
(23)

C. Constraints
where A' is the initial value of the load margin, and MG,,
The following constraints have been considered
and AFDo, the generation and demand redispatch,
The linear approximation of the load margin, bounded
respectively. by a security value 4., A residual variable has been
added to avoid infeasibility.
W . MATHEMATICAL FORMULATION
OF THE PROBLEM
Total load shedding bounded by the maximum demand
This section details the mathematical formulation of each redispatch MDR , to avoLi the loss of accuracy. The
iteration of the load shedding algorithm. Firstly the notation is value of MDR must keep a balance between the loss of
introduced. Then the objective function and the constraints are accuracy and an excessive number of iterations
provided The power factor after the redispatch must be the same
A . Notation as the original in every load bus
The load shedding algorithm is formulated in term of both Generation-demandbalanct: equation
active and reactive power parameters and variables. Therefore, Limits of both active and reactive demand redispatch
the following notation has been adopted Limits of generation redispatch
I ) Parameters The residual variable H , must be a positive variable
PG,, : Initial active generation in bus i These constraints have been respectively formulated as
S G , : Sensitivity of the load margin 1 with respect to follows:
the initial active generation in bus i
described in Section 111. The optimization has been performed
~SC;APC,+
, CSD,,.APD0,
with a security value for the load margin of ?,=IS%, and a
+~ S .AQD,,
D + H~, 2 4, -a' (24) maximum total load shedding of 500 MW. In each iteration,
the maximum demand redispatch was MDR =25 MW.
The maximum relative gap hetween the post-optimization
- C A P D o l i MDR (25) load margin and its linear approximation:

QD0;APD,, -PD,;AQDo1 = O (26) a(%)


= 100,-1p";aI (31)
C A P G O -, C A P D O , = O
(27) was rM=l.5%.
TABLE I summarizes both the initial and final active power
-PO,, i @Do, i 0
(28) generation and demand in each area of the Spanish power
-QD,, i AQO,, i 0 system.
PG; - PG,, i APG,, _C 0 (29) TmLE I
ACT^ POWER GENERATIONA N D DEMAND tN EACH AREA OF THE SPANISH
H, 20 (30) POWERSYSTEM

V. RESULTS initial fmal reduction


The performance of the algorithm is illustrated considering AREA GEN DEM GEN DEM GEN DEM
a highly loaded hourly scenario of the operation of the Spanish INORTHW. 9175 3515 8846 3515 329 0
power system. The power system model includes 2 NORTH 3790 3814 3740 3814 50 0
representations of the French, Portuguese and Moroccan
3 EAST 6240 6112 6240 6161 0 II
systems. The total load ofthe Spanish system is 29108 MW.
The security criteria of the Spanish power system requires 4CENTER 7593 10526 7472 10090 121 436
that branch power flows and bus voltages are within their 5 SOUTH 3309 4481 3309 4428 0 53
limits. In addition, the system must he far enough to the critical TOTAL 30107 29108 29607 28608 500 500
loading condition in order to prevent voltage collapse [SI.

Fig. 2. A m division of the Spanish power syslem

The Spanish power system and the external systems are


divided into eight areas. Fig. 2 displays the approximate
situation of each area. Northwest and north areas (area 1 and
2) are usually exporting areas, and east, center and south areas
(area 3, 4 and 5) are importing areas. Areas 6, 7 and 8
correspond to the areas representing the Portuguese, Moroccan
and French systems respectively. The worst voltage problems
are located in importing areas. An efficient load shedding
algorithm should reduce their load, and hence reducing the
power transfers between the exporting and importing areas.
The case considered had an initial load margin of
390'
1'=4.6%, obtained with the non-linear optimization algorithm 2 4 0 0 29& 3&0 30500 31600 31503 3 d X l
Demand Evolution (Mw)
t
- - +
370/ 29600 29500 30000 30500 31600 31500 3 2 I 0 100 200 300 400 500
Demand Evolution (Mw) had Sheddmg (Mw)
Fig. 3. Nose curves of pilot buses. Fig 5 Pilot buses voltages denvates wth respect to the load margin evolution
dunng the load shedding
AAer the redispatch, the load margin increases to a final
value of 10.7%. It should be noted that the security target VI. CONCLUSIONS
value of 15% was not achieved because the maximum demand This paper has proposed an optimization algorithm to
threshold allowed (500 MW) was reached before. determine the amount and location of the minimum load
Fig. 4 shows the load margin evolution, as the load is shed. shedding to improve the load margin to voltage collapse.
The algorithm is based on I,P optimization. First order
sensitivities of the load margin with respect to the load to be
shed have been considered. The objective function consists of
minimizing the total system demand decrease. The problem
constraints consist of a target improvement of the load margin,
the power balance equation and the generation and demand
limits
To overcome the loss 01' accuracy of the linear
approximation of the load margin, the algorithm has been
designed as an iterative process that imposes an additional
constraint ofthe total demand reduction in each iteration
The performance of the algorithm is illustrated considering
an hourly scenario of the operation of the Spanish power
4' I system
100 200 300 400 500
h ad Shedding (Mw)
Load margin evolution during the load shedding. VII. REFER.ENCES
P.Kundur, PowerSystem Stabi/i@mzdCon@o/Mc. Graw-Hill, 1994.
V. Ajarapu and C. Chnstie, "The Continuation Power Flow: A Tool to
The load margin is the most used indicator to measure the Study Steady State Voltage Stability," IEEE Transactions on Power
distance to the voltage collapse. However, the derivates of the Systems, vol. 7, pp. 416-423, Februiuy 1992.
pilot buses voltages with respect to the load margin are also S. Greene, I. Dobson, and F. L. Alvarado, "Sensitivity of the Loading
useful indicators [9]. Margin to Voltage Collapse with respect to Arbitrary P m e t e r s , " IEEE
Transactions on Power Systems, YOI 12, pp. 262-272, Feb 1997.
Fig. 5 illustrates the evolution of the derivates of the pilot S. Greene, I. Dobson, and F. L. Alvarado, "Contingency Ranking for
bus voltages with respect to the load margin as the demand Voltage Collapse via Sensitivities fiom a Single Nose Curve," presented
decreases. It shows that the buses which exhibit the lowest at IEEWES 1997 Summer Meeting,, Berlin, Germany, July 1997.
F. Capitanescu and T. V. Cutsem, '"Preventive Control of Voltage
derivates at the base case undergo the greatest growth during Security Margins: A Multicontingency Sensitivity-Based Approach,"
the load shedding process. IEEE Transactions on Power Systems, vol. 17, pp. 358-364, May 2002.
0. 0. Obadina and G. J. Berg, "Determination of Voltage Stability
Limit in Multimachine Power Systems," IEEE Transactions on Power
Systems, vol. 3, pp. 1545-1554,Nov 1988.
M. Minow, Mathematical Programming: Theov and Algorithms: John
Wiley and Sons, 1983.
BOE, "PO 1.1: Cnterios de funcion;unientoy segwidad," 30 de Julio de
1998 (not english version avaliable :yet).
A. C. 2. de S o w C. A. CaRimes, and \'. H. Quintana, "New
Techniques to Speed Up Voltage Collapse Computations using Tangent
VectOin? IEEE Tr'amactions on Power Systems, vol. 12, pp. 1380-
1387, August 1997 Ross Casanova Lebrga was born in Madrid,
Spain, in 1961. She received the maSter in
VlII. BIOGRAPHIES Artificial Intelligencekom the Universidad
Politbnica de Madrid in 1988. Since 1988 she
has been working for Indra Sistemas S.A. as an
Francisco M. Ecbavsrreo Cerezo was born in analyst programmer in the Business Intelligent
Madrid, Spain, in 1977. He received the degree of Department. She has participated in the
Electrical Engineer from Univenidad Pontificia development of a number ofprojects for the
Cornillas in 2001. Since June 2001, he is a t energy sector
Research Assistant at the Institute de
Investigaci6n TecnolO@ea,Universidad Pontificia
Cornillas. His areas of interest include modeling
analysis and Simulation of power systems and the
voltages instabiliv phenomena.

Eorique Lobato Mlgueloz was barn in Burgas,


Spain, in 1974. He received the degree of
Electrical Engineer in I998 and the PhD degree in
2002, from Univenidad Pontificia Camillas,
Madrid, Spain. Since June 1998, he is a
Researcher at the lnstituto de Investigaei6n
Tecnolbgica, Universidad Pontifica Cornillas. His
areas of interest include analysis, planning,
operation and economics in electric power
systems. He has participated in several research
projects for different firms related with the energy
industry

Luis Rouea Rodrfguez(Student Member 1989,


Member 1991) obtained the Elemical Engineer
d e w and the Ph.D degree from Univenidad
Politbnica de Madrid in 1985 and 1990. He is
Associate Professor ofthe S c h n l of Engineering
ofUnivenidad Pmtiticia Comillan. His areas of
interest are modeling, analysis, simulation and
identification of electric power systems. He has
been visiting researcher at Ontario Hydro, MIT
and ABB Power Systems

M*lsabcl Navarmlc Ferahdcz was born in


Albacete, Spain in 1960. She received her degree
in Physics h m the Univenidsd Complutense de
Madrid in 1982. Since 1983 she has been working
for Indra Sislemas S.A. as B softwareengineer.
She belongs to the Business Intelligent
Department. She has supervised the development
of a number of projects for the energy Sector.

Potrebbero piacerti anche