Sei sulla pagina 1di 12

906

IEEE Transactions on Power Systems, Vol. 9, No. 2, May 1994

A N OPTIMAL REACTIVE POWER PLANNING STRATEGY AGAINST VOLTAGE COLLAPSE


Venkataramana Ajjarapu Dept. of Elect. & Compt. Eng. Iowa State University Ames, IA Ping Lin Lau Sargent and Lundy Chicago, IL Srinivasu Battula Dept. of Elect. & Compt. Eng. Iowa State University Ames, IA

Abstract
This paper introduces a method of determining the minimum amount of shunt reactive power (VAr) support which indirectly maximizes the real power transfer before voltage collapse is encountered. Using a relaxation strategy that operates with a predictor-corrector/optimization scheme, a voltage stability index that serves as an indirect measure to the closeness of reaching the steady state voltage stability limits is obtained. Sensitivity information that identifies weak buses is also available for locating effective VAr injection sites.

Introduction

In recent years, full utilization of transmission among utilities for economic transfers has made the problem associated with voltage instability or voltage collapse a pressing issue. Numerous articles have been published around the world to discuss and find ways to tackle this pressing problem [l-91. Countries like Japan, France, Canada, and the United States have reported cases of voltage collapse with losses in millions of dollars. To prevent such instances to occur, system operators are looking for tools that can enhance their understanding of where the system is operating with respect to the point of collapse. In literature, this point is often referred as the critical point (steady state voltage instability limit). In addition to the identification of the critical point, system operators are also interested in knowing how much external reactive support is needed and where to locate so that both secure and economical solution can be achieved. Characterized by a progressive decline in voltage magnitudes, steady state voltage instability can occur because of the inability to meet increasing demands for reactive power at certain buses in a stressed power system network. While the process of voltage instability may be triggered by some forms of disturbances resulting in changes

in reactive power demand, the causes of a stressed system are many. The high costs of upgrading and strengthening existing transmission lines to meet increasing energy consumption; the difficulty of acquiring right-of-way; the delay of obtaining licenses to build new transmission lines; the shift of generation pattern due to environmental constraints are just a few reasons that lead to the increase of vulnerability in todays power system network. Since the identification of the critical point indirectly defmes the boundary between the stable and unstable steady state operating region, the search for the critical point is of important value. Unfortunately, conventional power flow analysis was hampered by the ill-conditioning at and near the critical point when the Jacobian matrix of the Newton-Raphson method becomes singular. For this reason double precision computation and anti-divergence algorithms such as the one found in [IO] have been used in an attempt to overcome the numerical instability.

To avert collapse, operators are seeking tools that can enhance their understanding of where the system is operating with respect to the point of collapse (or critical point). In addition, how much shunt VAr supply is required and where it should be located so that both a secure and economical solution can be achieved. To address such needs, this paper describes a method that extends the capability of the continuation power flow [ll]to maximize real power transfer through minimum amount of shunt VAr support. It features a relaxation strategy from which a voltage stability index is available to indicate the closeness to voltage collapse from current system load level. By capturing the sensitivity information obtained from the continuation technique at each load level, weak load buses that have large differential change in voltage magnitude (dV) for a given differential change in load power (dX) are recorded. When load bus voltage falls below permissible limits, an optimization routine will take on these buses as shunt VAr injection sites and minimizes the amount of VAr support upon specified system constraints.
This paper is organized as follows. Section 2 briefly explains basic steps involved in the continuation power flow. Section 3 derives the sensitivities of key components (for e.g. buses, branches, and generators) for a given change in system load. Finally Section 4 proposes an optimal strategy that will maximize the power transfer within a given system while minimizing the necessary amount of shunt reactive power support before voltage collapse is encountered.

93 SM 535-5 PWRS A paper recommended and approved by the IEEE Power System Engineering Committee of the IEEE Power Engineering Society for presentation at the IEEE/PES 1993 Summer Meeting, Vancouver, B.C., Canada, July 18-22, 1993. Manuscript submitted Feb. 3, 1992; made available for printing April 16, 1993. PRINTED IN USA

0885-8950194/$04.00Q 1993 IEEE

907

Continuation power flow

A
/Predictor

Continuation power flow (CPF) is based on locally parameterized continuation technique [12]. The key idea here is to avoid the singularity of the Jacobian by slightly reformulating the power flow equations, and applying the locally parameterized continuation technique [13]. In order to apply the continuation to the power flow problem, the power flow equations must be reformulated to include a load parameter, A. The inclusion of this load parameter X results in the steady-state power flow equations given by F ( c , X) = 0 (1) Here z = k,VJJT. Where 4, represents the vector of bus voltage angles and represents the vector of bus voltage magnitudes. In general the dimension of F will be 2 1 1 , ~ + npu, where npqand tipv are the number of P Q and P V buses respectively. in equation (1) and taking Substituting z = the derivative, we can get the fundamental equation of sensitivity analysis

Critical Point

63 h\

Corrector

bd

k,VJT
dF

Load Figure 1: An illustration of the predictor-corrector scheme used in the CPF

dF d F = -d6+

as

-dV+ dV

-dX

aF

aX

=0

(2)

Once the tangent vector has been found by solving (3), the prediction can be made as follows

Once a base solution (zolX,) of F ( z , X ) = 0 has been found (this is nothing but ordinary power flow which corresponds to X = 0), the continuation problem is to calculate further solutions, ( 2 1 , XI), ( 2 2 , X2), . . . until the target point is reached. The continuation power flow employs a predictor-corrector scheme to find these continuum of power flow solutions, i.e., first a solution will be predicted and then the corrector tries to make this predicted solution converge to the required solution. Each of these steps are described separately here.

where * denotes the predicted solution and a is a scalar that designates the step size.

2.2

Parameterization & Corrector

2.1

Predictor

The prediction is made by taking an appropriately sized step in a direction tangent to the solution path. Thus, the first task in the predictor process is to calculate the tangent vector. This tangent vector can be obtained by factorizing equation (2), i.e.,

K6,EuiEAl

[ {i]

On the left side of the equation is a matrix of partial derivatives multiplied by a vector of differentials. The former is the conventional power flow Jacobian augmented by one column (FA), while the later t = [dl, d l , dXIT is the tangent vector being sought. A normalization has to be imposed in order to give 1 a non zero length. For example, one can use

After the prediction is made, the next step is to correct the predicted solution. The best way to present the corrector is to expand on parameterization, which is vital to the process. Parameterization is a mathematical way of identifying each solution on the branch. It is a kind of measure along the branch. As already mentioned, the parameterization scheme used here is called local parameterization. By looking a t a PV curve, one can see that the voltage is continually decreasing as the load nears a maximum. Thus, the voltage magnitude at some particular bus could be changed by small amounts and the solution can be found for each given value of the voltage. Here the load parameter is free to take on any value it needed to satisfy the equations. This is called local parameterization. In local parameterization the original set of equations is augmented by one equation that specifies the value of one of the state variables. In equation form this can be expressed as

where ek is an appropriately dimensioned row vector with all elements equal to zero except the k t h , which equals one. If the index k is chosen properly, letting t k = f l . O impose a nonzero norm on the tangent vector and guarantees that the augmented Jacobian will be nonsingular at the point of maximum possible system load [12]. Thus the tangent vector is determined as the solution of the linear system

where TI is an appropriate value for the kth element of E. Now once a suitable index k and the value of 11 are chosen, a slightly modified N-R power flow method (altered only in that one additional equation and one additional state variable are involved) can be used to solve the set of equations. This provides the corrector needed to modify the predicted solution found in the previous section. More details can be found in [ll]. Figure 1 demonstrates the predictor-corrector process used in the CPF.

908
In the above continuation process, the tangent vector proves useful because it describes the direction of the solution path a t a corrected solution point. A step in the tangent direction is used to estimate the next solution. However if one looks at the elements of the tangent vector as differential changes in bus voltage angles (db,) and magnitudes ( d E ) in response to a differential change in load connectivity ( d X ) , the potential for a meaningful sensitivity analysis becomes apparent. Next section explains how this information can be used to identify critical buses, branches, and generators with respect to voltage stability [14]. about in [11].The weakest bus with respect to voltage stability limit is to find the bus with the largest d%/dPtor,,, where dPiotal is the differential change in active load for the whole system and is given by
n n

= CdX
with the following definitions

K L , = multiplier to designate the rate of load change at bus i as A changes


$i

= power factor angle of load change at bus a'


propriate scaling of X

Sensitivity analysis from the tangent vector

SaBAsE apparent power which is chosen to provide ap=


The weakest bus j is

Here we are interested in the sensitivity of system response with respect to a change in the connected parameter (A). In other words, we derive an expression for the differential change in a scalar valued function h ( x , A), because of differential change in A. Here, l i ( x , X ) is any power system operating constraint such as branch flow, reactive output of a generator, bus voltage magnitude. Using the differential chain rule, we obtain

dh -- dh dx
dX-zdx

+-ah d X

when the weakest bus j, reaches its steady-state voltage stability limit, dX approaches zero, the ratio dVj/CdX will become infinite or equivalently, the ratio CdX/dl/;. tends to zero. The ratio C d X l d V j , which is easier to handle numerically, can be defined as a voltage stability index for the entire system.

(4)

Applying the same chain rule to the power flow equations (1) resulted in ( 2 ) . Equation ( 2 ) can be expressed as an ordinary differential equation of the form

3.2

Branch sensitivities

z=[z] ax

dx

8F

dF

Let us consider a branch ij. Let KLbi and l/;.Lbj be the voltages at buses a' and j respectively, and let yijLBij be the line admittance. Then the losses in the line i j can be derived as Defining this loss expression as function the sensitivity equation as one can get

(5)
11,

For a specific variation of parameter A, the corresponding variation to solution x is calculated by evaluating the Jacobian 6'Fldx. This procedure, as it was proposed in Davidenko [15] fails at critical points where the Jacobian is singular and the inverse does not exist. At this point the solution x is very sensitive to even small perturbation in the parameter. However, by the nonsingular augmentation procedure described in Section 2 , dx/dX is directly given by the tangent vector in the continuation power flow. It can be directly substituted in equation (4) to get the sensitivity of any operating constraint.

+ ( 2 6 - 2ECOS(bi - b j ) ) dl/;. dX

+ ( 2 E V j s i n ( S i - 6 j ) ) - dbi dX
Branch sensitivity indicates how important that particular branch is to voltage stability. Table 1 shows the branch sensitivities obtained near the critical point by considering the Q l o s s e s in the branches for the 30-bus test system. in Figure 2 shows real power vs. total Qlosses the first five highest and the last five lowest participating branches (i.e. the five branches with the highest and the lowest sensitivities) for the 30-bus system. It can be noticed that the slope of the curve which shows the &losses in the first five highest participating branches is steep, compared to the last five lowest participating branches. This tells us that it is the rate at which the &loss in a particular branch is changing which is important, but not the magnitude of the Qloss.This can be observed by looking at the Q l o s s e s and the sensitivities from table 1. For example in table 1, the QloSs the branch 25 is more than the Q l o s s in branch in 3, but branch 3 has higher sensitivity compared to branch

3.1

Bus sensitivities

For bus sensitivities the function h ( x , X) can be either bus voltage magnitude or angle at a particular bus i. For bus voltage magnitude at bus i, from equation (4)

Since the value of dX is the same for each dl/i or d6i in a given tangent vector, the bus sensitivities are nothing but the tangent vector elements themselves. Bus sensitivities indicate how weak a particular bus is near the critical point, and helps in determining the areas close to voltage instability. The higher the value of bus sensitivity, the weaker it is. These bus sensitivities are very similar to the voltage stability index the authors talked

25.

909

Branch no.
24 30 32 23 3 25 22 21 33 34 16-19 21-22 23-24 16-17 2-3 16-21 15-16 14-15 25-26 26-27 2.9336 2.7159 0.9558 0.7863 1.0473 1.3171 0.7611 0.1217 0.5242

Sensitivity
1.0000

'
~

0.5924 0.5249 0.2377 0.1839 0.1784 0.1541 0.1349 0.1033 0.0807

2.3443 1.843

1.0000 0.5318

Table 2: Generator sensitivities near the critical point for the 30 bus system indicates which generators are important in maintaining voltage stability near the critical point i.e. generators with high sensitivity are important in maintaining voltage stability of the system. For example, with the outage of generator 29 one could transfer less power than with the outage of generator 25. Bus sensitivities can be used to identify the buses for possible reactive compensation. Branch sensitivities can be used to identify critical contingencies and generator sensitivities can be used to identify the generators to share the increase in the load. In the next section bus sensitivities are used to place the capacitors to increase the power transfer.

Table 1: Branch sensitivities near the critical point for the 30 hus system
p losssa p.u.
I

~ u s 1 5 1 ~ s

I
I200
11.00
10.00

i z s' l i i

9.00 8.00
/

7.00
6.00 5.00 4.00 3.00 200
1.w

Formulation of the optimal strategy

As described before, by comparing the sensitivity information of each bus to CdX, the buses that are most susceptible to voltage collapse can easily be identified. To make use of this intelligence, the next approach will be to minimize the necessary amount of shunt VAr support when bus voltages fall below operating limits. This lends itself into a nonlinearly constrained minimization problem and can be expressed as
minimize [Total injection a t selected weak buses]
I I
I I

00 .0

60.00

80.00

100.00

1a.00

Powcrp.~.

subject to 1. continuation power flow equations,

Figure 2: Real power vs. Qlosses curve - 30 bus syscein

3.3

Generator reactive power sensitivities


11.

2. all bus voltages within specified operating limits,


3. allowable shunt VAr supply at each injection site.
In mathematical form,

For this the reactive power at a generator can be defined


as the function i.e.
L ~ A ~I (

ll(z, A) = Q

( L , S A , A ~ E ~ ~ ~ ( $ J ~ )QT, )

where QL,, is the original reactive load and KL,, SaBAsE as defined before. Therefore the sensitivity are equation becomes

qi,

Table 2 shows the generator sensitivities calculated for the same 30-bus test system near the critical point. It should be noted that there are total nine generators in the 30-bus system, and near the critical case only two generators are participating, whereas the other seven generators already hit their Qlimits. Generator sensitivities

Sw = the set of selected weak load buses


SL = the set of P Q load buses

910
Maximum system real pwr. transfer in p.u. 106.855 118.255 Reactive injection in p.u. cap./ind. Weak load bus locations before volt. collapse 12,7,8,5,11 12.7.8.5.11 Lowest bus voltage in p.u.

Method

PCs

O.OOO/O.OOO
-4.990/0.000

0.914 0.950

I
LOAD

Pcos
\-corrector

Table 3: PCS, PCOS results from the 30-bus system

qej

= additional capacitive VAr at bus j = additional inductive VAr at bus j

E igure 3: Illustration of the predictor-corrector optimizat,ion scheme

yr3

P G ~ , G ~ active and reactive generation at bus i Q = P L ~ , L ~ active and reactive load at bus i Q = PT(,QT* = active and reactive injection at bus i
At the programminglevel, the best place to implement the problem statement is to interface the predictor-corrector scheme (PCS) of the CPF with an optimization routine after each corrector step. This approach is called predictorcorrector optimization scheme (PCOS). Figure 3 illustrates this process with a P-V curve. The corresponding flow chart is provided in Figure 4. A sequential quadratic programming (SQP) algorithm [16] is adopted to solve for the optimal solution. Clearly, the above objective can easily be converted to minimize the cost of VAr injection by adding a fixed installation cost and a variable unit cost for capacitive and inductive sources. Nonetheless, one should study the feasibility of those suggested injection sites prior to actual planning. Since a power system has to operate within voltage limits, the optimization routine will not be activated unless voltage violation is detected in any one of the load buses.

SOLVE BASE CASE


POWERFLOW

SPECIFY TIIE NUMBER OF


WEAK

~ u mR a

REACTNE INJECTION

1.

SELECT LOA0 PhRAyETER


CONTINUATION

PARMdFlFB

I-[

SOLVE M TANCEm R

NEXT CONTINUATION

4.1

Numerical results

PREDICTTIE SOLUTION WITH THE CURRENT CONWATION

1
I

PERFORM OPTIMIUTION

Figure 4: Flow chart of the predictor-corrector optimization scheme

The validity of the proposed method is demonstrated on the New England 30-bus system, which is commonly used in voltage stability research. The initial system real load is 50.465 p.u. Table 3 demonstrates that the maximum real power transfer for this system can be improved from 106.855 to 118.255 p.u. with a total reactive (capacitive) injection of 4.99 p.u. Here only the first five weak buses are considered for reactive injection. With the same system, the effect of increasing the specified number of injection buses to the real power transfer capability is studied. From Table 4 it was found that the real power transfer actually decreases as the number of injection buses increases. This can be explained as follows. As the number of injection sites increases, the voltage profile across the system will be improved. Correspondingly, the incident of voltage violations will reduce and the demand for reactive power support necessary to rectify the voltage violations will decrease also. As a result, the transfer of real power which is enhanced by the support of reactive power decreases. It is also found that the weakest buses tend to cluster around the generator that has recently reached its maximum reactive limit. According to eouation 6. however. the sensitivitv information tends

911

Specified no. of inj . buses at each load level


1 3

Maximum

system real pwr. transfer in p.u.


125.757 119.642 118.255 114.992 112.509

Total no. of cap. inj. sites


3

Required
total cap.

Identified
weak buses for cap. inj.

inj . in p.u.
-8.547 -5.616 -4.990 -3.085 -2.099

before vol. collapse


28 12,7,5 12,7,8,5,11 21,15,24,16, 12,14,13 21,15,24,16, 12,14,13,4,17

5 7
9

1
I

1
I

7 10 12
14

I
I

Table 4: Effect.of increasing the number of injection buses to real power transfer with the 30-bus system

Real power transfer in p.u. before sudden voltage collapse 63.0812 68.8484 74.5445 80.1396 85.5849 90.9081 100.5802 104.9210 108.8679 110.3960

Voltage stability index 0.1300 0.1285 0.1269 0.1230 0.1203 0.1174 0.0957 0.0918 0.0665 -0.0061

System real power transfer in p.u. 90.908 109.063 110.904 111.950 112.962 113.940 114.574 115.279 118.255 119.556

Voltage stability index 0.1174 0.0663 0.0087 0.0085 0.0083 0.0081 0.0080 0.0614 0.0578 -0.0020

Table 6: Voltage stability index under RS in the 30-bus system where the system will experience voltage collapse, it is of particular interest to determine how much reactive power supply is required and where it should be located so that the power system can be operated at maximum reliability and economy. A CPF (continuation power flow) was first developed to overcome the ill-conditioning near the critical point; where the Jacobian matrix of the Newton-Raphson method becomes singular. By applying a locally parameterized continuation technique, the C P F initially starts with a known power flow solution and then enters a PCS (predicty-corrector scheme) to find subsequenb path-dependent solutions above ana below the critical point for a given load increase scenario. Interfacing with the PCS, this paper attempts to.develop a methodology that meets the system planners needs. It first recognizes the sensitivity information derived from CPF as a reliable source for identifying weak buses that are prone to voltage collapse. Then, buses within this set of weak buses are further selected to provide remedial action against voltage collapse. In this research, the idea of shunt reactive power injection at selected weak load buses is proposed. To achieve an economical solution at a given load level, the proposed method is formulated as a nonlinear constrained optimization problem with an objective that minimizes the amount of shunt reactive power injection. This approach is called PCOS (predictor-corrector optimization scheme). Not only is the voltage collapse process delayed by the injection of reactive power, but the real power transfer capability of the system is improved also. Furthermore, bus voltages are ensured within permissible voltage operating limits. This makes the proposed strategy specially attractive for those utilities interested in economic transfer but hindered by the steady-state voltage instability problem. Unfortunately, the voltage stability index derived from the CPF fails to provide a good measure to voltage collapse. One possible explanation would be the continual improvement of the voltage profile to meet specified voltage limit constraints in optimization. These voltage corrections may indirectly generate flat tangent vector profiles in the state variable space from which the index is derived. As a result, a RS (relaxation scheme) is suggested to improve the performance of the voltage stabil-

Table 5: Voltage stability index under PCOS using.the 30-bus system to be contaminated by the rather flat profile of the tangent vector once the PCOS is in progress; the index becomes less indicative to the projection of potential collapse. This is evident from Table 5. In view of this, a relaxation strategy (RS) is suggested which involves an alternative execution between the PCS and PCOS after the detection of voltage violation. If there is no violation the PCS alone will be executed. The progress of voltage stability index with RS scheme is shown in Table 6. The index decreases monotonically from 0.1300 to 0.0665. It can be concluded that the results under RS are generally more conservative than that of the PCOS. Because RS provides a real power transfer level at which a small drop of reactive power support will lead to voltage collapse, the value of real power transfer is expected to be smaller. However, the degree of reduction depends upon the characteristics and the initial state of the system. In the case of the 30-bus system, the maximum real power transfer under PCS and RS are found to be 106.855 and 108.868 P.u., respectively. Though the net increase is only 2.013 P.u., it is a well known fact that the New England 30-bus system is initially stressed system.

Discussion & Conclusions

To prevent voltage collapse from occurring, system operators and planners are looking for analytical tools that can enhance their understanding of where the system is actually operating with respect to the point of collapse - critical point. In addition to knowing the load level

912

ity index. With the RS, the performance of the voltage stability index does improve significantly. The maximum load level obtained from RS actually represents the load level beyond which a sudden drop of reactive support in the system may lead to sudden voltage collapse. Since no system operates in perfect condition, this indirectly provides a built-in security margin to forewarn possible sudden voltage collapse. Finally, it is important to realize that every system is unique and it is possible to encounter network constrained problems before maximum load levels can be reached. This further implies that a good knowledge of the system topology and behavior at various operating conditions is essential to the operation of maximum reliability and economy. In terms of contributions, this paper:
0

[7] Chiang, H. D., I. Dobson, R. J. Thomas, J . S. Thorp, and F. A. Lazhar. On Voltage Collapse in Power Systems. Proceedings of the 1989 IEEE Power Industry Computer Application Conference (PICA). [8] Alvarado, F. L., and T . H. Jung. Direct Detection of Voltage Collapse Conditions. Proceedzngs: Bulk

Power System Voltage Phenomena- Voltage Stability and Security, Potosi, Missouri, Jan. 1989.
[9] Sauer, P. W., C. Rajagopalan, M. A. Pai, and A. Verghese. Critical Modes and Voltage Instability in Power Systems. 1986 IEEE ISCAS 3 (May 1986): 1019-1022.

Extends the capability of the continuation power flow to include a methodology to identify critical buses, branches, and generators with respect to voltage stability. Extends the capability of continuation power flow to capacitor placement problem. Explore the possibility of achieving maximum real power transfer while minimizing the amount of shunt reactive support to a given system. Preserves the performance of voltage stability index using a relaxation scheme. Establishes a built-in security margin to guard against sudden voltage collapse.

LO] Iwamoto, S., Y. Tamura. A Load Flow Calculation Method for 111-conditioned Power Systems. IEEE Transactions on PAS-97, no. 5 (Sept./Oct. 1978):
1586-1599. 111 Ajjarapu, V., and C. Christy. The Continuation Power Flow: A Tool for Steady-State Voltage Stability Analysis. IEEE Transactions on Power Systems, vol. PWRS-7, (Feb 1992): 416-423.

[12] Rheinboldt, W. C. and J . V. Burkhardt, A Locally Parameterized Continuation Process, ACM Trans. on Mathematical Software 9, no.2, (June 1983): 215235.
[13] V. Ajjarapu, and C. Christy, The Application of Locally Parameterized Continuation Technique to the Study of Steady-State Voltage Stability, Proceedings of 21 st Annual North American Power Symposium, Rolla, Missouri, October 1989. [14] Ajjarapu, V., and S. Battula. Sensitivity Based Continuation Power Flow. Proceedings of the

References
[l] Proceedings: Bulk Power System Voltage Phenomena - Voltage Stability and Security, EPRI EL-6183, Project 2473-21 , Electric Power Research Institute, January 1989.
[2] Tiranuchit, A., and R. J . Thomas. VAR Support and Voltage Instabilities in Electric Power Networks. Proceedings of the North American Power Symposium, October 13-14, 1986, Ithaca, NY: 21-29.

twenty-fourth annual North American Power Symposium, Reno, Nevada, Oct. 5-6, 1992.
[15] Davidenko, D. On A New Method Of Numerically Integrating A System Of Nonlinear Equations , Dokl, Akad, Nauk, USSR, 88, (1953):601-604. [16] Numerical Algorithm Group (NAG), Fortran Library Manual, Mark 13, vo1.4, 1988.

BIOGRAPHY
Venkataramana Ajarapu, received his Ph.D. degree in Electrical Engineering from the University of Waterloo, Ontario, Canada, in 1986. Currently, he is an associate professor in the Department of Electrical and Computer Engineering at Iowa State University, Ames, Iowa. His present research is in the area of reactive power planning, voltage stability analysis, and power system security.

[3] Galiana, F. D. Load Flow Feasibility and the Voltage Collapse Problem. Proceedings of the 23rd Conference on Decision Control, Las Vegas, Nevada, December 1984: 485-487. [4] Tamura, Y., K. Iba, and S. Iwamoto. Relationship Between Voltage Instability and Multiple Load Flow Solutions in Electric Power Systems. IEEE Transactions on Power Apparatus and Systems, vol. PAS102, (May 1983): 1115-1125. [5] Kersel, P., and H. Glavitsch. Estimating the Voltage Stability of a Power System. IEEE Transactions on Power Delivery, PWRD-1, no.3, (July 1986): 346354. [6] Ajjarapu, V. Identification of Steady-State Voltage Stability in Power Systems. International Journal of Energy Systems, vol. 11, no. 1, (1991): 43-46.

Ping Lin Lau, received his Bachelor of Science degree in


May 1989, and his Master of Science degree in November 1991, from Iowa State University. Currently he is working with the Sargent and Lundy Company in Chicago, Illinois.

Srinivasu Battula, received his B. Tech degree in Electrical Engineering from J.N.T.U. College of Engineering, India, in 1989. He received his Master of Science degree in December 1992 from Iowa State University and is continuing towards his Ph.D. program at the same University.

913

Discussion Empros Power Systems Control Minneapolis MN:The authors are to be congratulated for their paper which presents a method for planning VARs for improved security to voltage collapse. In the formulation (section 4 of the paper), the authors minimize the total reactive injections at specified buses so that a power flow solution exists with all load buses voltages within the specified ranges. The following questions and comments arise: Essentially the authors have used voltage magnitude as an index of security to collapse. Since voltage magnitude alone is not a good indicator, several indices of the margin to collapse have been proposed [2,5,A]. The VAR planning for voltage collapse security enhancement problem should then involve planning VARs to improve a margin to collapse (defined by a chosen index) to an acceptable level. Such is the nature of the work reported in [B]. The planning problem is usually to minimize the total costs of installation of the VAR equipment. Minimizing the total costs is closely tied to minimizing the number of locations where the VARs are installed. The bus sensitivities as computed in the paper will provide a list of good candidate locations. How does one choose the optimum number and which ones?

0.0. Obadina,

In the planning process one should consider several base and contingency cases. How would such a consideration fit into the proposed method?
In closing, I would like to congratulate the authors for a well-written paper.

REFERENCES
(A) 0. Obadina and G. J. Berg,"Determination of Voltage Stability Limit in Mutimachine Power Systems", IEEE Trans. on Power Systems,Vol. 3,NOV. 1988,pp 15451554. (B) 0. Obadina and G. J. Berg,"VAR Planning for Power Systems Security" IEEE Trans. on Power Systems,Vol. 4,May 1989,pp 677-686.
Manuscript received August 9, 1993.

K.N. Srivastava, S.C. Srivastava, P.K.Kalra (Indian Institute of Technology, Kanpur 208 016, INDIA): The authors are to be commended for proposing an optimal reactive power planning

strategy against voltage collapse, a phenomenon experienced by several utilities all over the world during the last decade. The discussors would like to have authors' opinion on the following: (1) The authors have based their study on the singularity of the load flow Jacobian which typically corresponds to the saddle node bifurcation of the power system. However , the presence of saddle node bifurcation does not preclude the presence of any other kind of bifurcation such as , period doubling Hopf bifurcation bifurcation, cyclic fold bifurcation etc in the power system. Voltage collapse can take place even when Jacobian is not singular [I,21 .The discussors would like to know how the authors propose to implement their scheme under these conditions. (2) It is not only the magnitude but also the pattern of the load that matters in estimating the proximity of operating point to the steady state stability limit defined by the surface det{J}=O , as demonstrated by [3]. In addition, the Dobson et a1 characteristic of the load has the profound effect on the system stability. The discussors would like to know the reasons for considering only magnitude and neglecting the pattern and characteristics of the load in the study. (3) What is the (numerical) criterion/ threshold for treating to be maximum in eq (6) for deciding j bus as the weakest bus ? In our opinion, this will be system dependent. (4) The traditional approach such as capacitors may prove successful for dealing with voltage problems caused by insufficient reactive power. However, if the low voltage problem is an indication of operating too close to the steady state stability limit of the transmission system, the traditional approach may not be appropriate. With ample amount of reactive power support, voltage can be maintained within acceptable limits over a wide range of loadings. In fact the voltage may never decline below acceptable limit even though the system approaches its steady state stability limit. Thus, only voltage magnitude violation may not act as a foolproof indicator but angular separation may also required to be monitored. References: [ l ] H. Wang, E. H. Abed and A. M. H. HamdanI1lIs voltage collapse triggered by the boundary crisis of a strange attractor ?I!, Proc. 1992, American Control Conference, Chicago, June 1992. [2] K. N. Srivastava, S. C. Srivastava and P. K. Kalra, IIChaotic oscillation in power system under disturbancesf1, Accepted in.Int. Conference on Advances in Power System Control, Operation h Management, Hong Kong Dec 7-10,1993. [ 3 ] I. Dobson and L. Lu,"New methods for computing a closest bifurcation and worst case load power margin for voltage collapsell, paper # 9 2 SM 587-6 PWRS, I E E E PES Summar meeting Seattle, WA, July 1992.

I$&[

Manuscript received August 20, 1993.

914
M. K. Pal: The paper is interesting. We have, however, some difficulty in following the predictor part of the continuation algorithm as described in this as well as the authors' two other papers (References 11 and 13 of the paper). The authors state that in equation (3) the vector e, has all elements equal to zero except the kth, which equals one. Then it follows from (3) that the magnitude of the kth element of t is also equal to one. Is this rational except at the start of the solution process when the kth element of ekmay be the last element. Selecting any of the voltage magnitudes or angles as the continuation parameter seems unreasonable, since the step change in any of these is not likely to be anywhere near unity. The use of the scaling factor is not likely to correct this problem either. Some explanation from the authors through a numerical illustrationusing a simple two-bus system will be appreciated. The authors' method, as illustrated in Figure 1, detects the passing of the critical point rather than locating it exactly. Since there can be an infinite number of critical points, depending on how one chooses to model the distribution of loads and generations, trying to locate the exact critical point is pointless anyway. However, a simpler way of approximately locating the critical point might be to convert a part of the load at the weak buses (which can be easily detected from the sensitivity information derived from the power flow Jacobian while the system is still some distance away from the critical point) to equivalent admittances. The test for the critical point would be easy. If the load served decreases with increase in admittance, the critical point has been passed. Converting part of the load to constant admittancewill also improve convergence in general. This will also eliminate the problem of the power flow Jacobian being singular at the critical point. Sensitivity information derived from the power flow Jacobian can then be used for the voltage stability index as well as for selecting buses for reactive support in the optimization routine, in more-or-less the same way as done in the paper. The scheme seems to work well in very small test systems. Have the authors tried this approach in their test systems? We have some comments on the section discussing numerical results. The statement ".. the real power transfer actually decreases as the number of injectionbuses increases" may be misleading. We note from Table 4 that the total reactive support has decreased with the increase in the number of injection buses. What is actually happening is that although there is no voltage violation, the overall voltage level remains lower when the number of injection buses is higher. The fallacy will be removed by raising the minimum voltage limit and bringing it closer to the maximum value. The argument presented in favor of the relaxation strategy does not appear to be sound. The index seems to be as indicative with the RS as without it, provided the results are properly interpreted. In Table 5, all of the power transfer levels above 109 p.u. represent operation close to the turning point on the P-V curve, and can be considered the limit in the absence of additional support that is provided in the optimization. * Our main concern is, however, the authors' repeatedly referring to the critical point as the point of "voltage collapse," implying that reaching a critical point will inevitably lead to voltage collapse. This is of course not true. Voltage instability and collapse would depend on the load characteristics [A-B]. Depending on the speed of response of the load dynamics, and the distance of the actual load delivery point (i.e. the utilization bus) from the high-voltage bus where the load is modeled, the instability and collapse point may come well before the critical point based on a steady-state formulation. On the other hand, for certain load types, specifically static loads, one can continue loading the system after the critical point is reached without any danger of instability and collapse, althoughthe demanded load will not be served.

The critical point represents a network limit based on a specific load-generation pattern. Although this limit has important implications in system planning, in general, it has nothing to do with voltage instability and collapse [B-C]. Only in specific situations, when a number of conditionsare satisfied simultaneously, a critical point will also be a voltage collapse point [B]. [A]
[B]

[C]

M.K. Pal, "Voltage Stability Conditions Considering Load Characteristics," IEEE Trans. on Power Systems, Vo1.7, No.1, pp.243-249, Feb. 1992. M.K. Pal, "Voltage Stability: Analysis Needs, Modelling Requirement and Modelling Adequacy," IEE Proc. C, V01.140, No.4, pp.279-286, July 1993. M.K. Pal, Discussion of "An Investigation of Voltage Instability Problem," by N, Yorino, et al. IEEE Trans. on Power Systems, Vo1.7, No.2, pp.600-611, May 1992.

Manuscript received August 27, 1993.

V . AJJARAPU and S. BATTULA, Iowa State TJniversity, Ames, Iowa: We wish to thank the discussors or tlirir genuine interest in the subject of our paper, and we very much appreciate their thoughtful comments. We will respond to each of the them separately.

M. K. Pal: In the predictor part of the algorithm, selecting any of the voltage magnitudes or angles as the continuation parameter is not unreasonable. If X is the continuation parameter all through the process, problem arises because a solution doesn't exist above some maximum possible value of A. (This can be visualized on a P - V curve). At this point any one of the state variables is effectively used as the parameter to be varied. Here the load parameter would be free to take on any value it needed to satisfy the equations. This principle is called the locally parameterized continuation. Once the continuation parameter is selected (A, or K , or S i ) , the corresponding tangent vector component (dX, or d x , or d 6 i ) will be f l . But while predicting the next solution we are taking a step size U (see the equation in section 2.1 of the paper), which will be typically of the order of say 0.02 (the step size can be large if we are far from the critical point). Thus the change in that particular variable is not unity but equal to the step size U . T h e continuation process is demonstrated here with a simple example. Consider the following simple example with a single unknown x f(x, A) = x2 - 33: A = 0 (1)

The Jacobian is

[x af = [(2x - 3) a]
af

1 1

Let t,he base solution (x0,Xo) be ( 3 , O ) . Then the series of solutions (21, XI), (22,A2), . . . can be found using predictor-corrector continuation as below:

915

Continuation step I:
Predictor
To start with, let X be the continuation parameter. Calculate the tangent vector as below: (here the index k is equal to 2 ) .

Corrector iteration:

A x = -0.00004075

and

AA = 0

-[lr
*dx=

:][:;I=[;]
- and d X = 1
3
-1
; : ] + U [

Now, max{Ax. AX} < E . So stop the corrector iterations. After the first continuation step, the point ( X I , XI) is equal to (2.8228757,O.S). Repeat the entire process until we reach the critical point. For this example, the critical point is (1A. 2.25).

Predict the next solution as

[;:I=[
where
U

:;]

Thus the predicted solution

is a scalar designating step size (say 0.5). (21 X I ) becomes (2.8333,0.5). ,

Corrector Correct the predicted solution by solving:

X versus x curve for the example will be as shown in the Figure 1. For the predicted solutions at points (1) and (21, we can choose X as the continuation parameter (i.e. fix X at that particular value) and converge on to the curve with corrector iterations. But at point (3), X can no more be a continuatjon parameter, as there is no solution for that value of A. At this point (i.e. when we are close to the critical point), we use local parameterization technique and choose z as the continuation parameter and solve for the system. This can be clearly observed in the example we considered. We know the solution as (1.5,2.25). Consider the augmented Jacobian of the continuation process

*-[

2 6666

'0

1 1][:;]=[
and

0.0277768

0
AX = 0.

* Ax = -0.0104165

At x = 1.5, det { Jaug}= 0. So the method diverges near the critical point, if A is the continuation parameter. Suppose we choose x as the continuation parameter instead, then

Repeat these corrector iterations, until reasonable accuracy is obtained (say E = 0.0001). Now, max{Ac,AA) > E . So update T I 9c and repeat the corrector iteration.

.
I

Det {Jaug} # 0. Then we can solve for the system. In this example, because we have only one variable 'z', we choose x as the continuation parameter. But for power flow equations, x is a vector of bus voltage magnitudes (LG's) and angles (6i's). There we can choose any of these state variables as the continuation parameter, especially near the critical point. This can be illustrated with the simple two bus example, whose single line diagram is as shown in the Figure 2.

11

j0.l

0.995/10.

I
10 .
AI

I
(l.O+jO) PU

fixed
Figure 1: X versus
2

fixed
curve Figure 2: Simple two bus example (base case)

4 2

916

C P F was run for this simple example. Near the critical point, the tangent vector calculation in the predictor step is
5.00
-5.00

6.67
0.00

-5.00
1.00

5.00 0.00

] [ y; ] [ ]
=
0 ,

dSz

0.0

Authors in general agree with the discusser that the critical point as calculated from the power flow formulation does not tell the whole story about voltage collapse. To fully understand the voltage collapse we have to consider system dynamics which includes load dynamics. The importance of this aspect is explained in [All.

=$

[ ]; [ ]
=
-1.00 -1.00 0.00

K.N. Srivastava. S. C. Srivastava,and P.K.Kalra: We address the questions in the same order, as follows:
(1) We agree that the singularity of the load flow Jacobian typically corresponds to the saddle node bifurcation of the power system. The entire study we carried out in this paper is based on steady-state analysis. Hopf bifurcation, period doubling bifurcation, cyclic fold bifurcation etc. occur when we consider system dynamics. Our analysis in this paper is just based on power flow equations. The aforementioned dynamic bifurcations are observed by the authors in [Al,A2]. (2) We also agree that it is not only the magnitude, but also the pattern of load that matters in estimating the steady-state limit defined by the surface det {.J} = 0. The continuation power flow ( C P F ) has the option of changing the load pattern in whichever way one wants. i.e., we can specify the bus numbers where one wants to increase the load and also the factor by which the load is to be increased at a particular bus. Also, the C P F is capable of handling nonlinear loads. Nonlinear loads based on EPRIs LOADSYN [A31 are successfully incorporated in CPF. This study defined the concept of load connectivity and the confusion about the critical point with respect to PV curves is clarified. More details about this can be found in [X4].

The value of dX becoming zero indicates that we are a t the critical point. Here either of the angle or voltage a t bus 2 can be the continuation parameter. The value of 1.0 a t ( 1-3)th element of the augmented Jacobian indicates that we choose angle at bus 2 as the continuation parameter. If we choose X as the continuation parameter a t this step the augmented Jacobian will be

Javg =

j.00 -5.00 0.00

-5.00 5.00 0.00

6.67

0.00 1.00

which is singular. Even if we are not exactly at the critical point but close to it, the augmented .Jacobian will be very close to singular. Thus, any of the bus voltage magnitudes or angles can be the continuation parameter in the process, especially near the critical point. For this example, Pcrzltcal is 5.0 p.u. and Vcrtttcal is 0.707L - 45. The method proposed is well conditioned for not only constant power type load representation, but also nonlinear loads. It is close to Newton-Raphson type power flow method and can be easily implemented. The method proposed by discusser, i.e. converting part of the load to constant load impedance may improve the convergence of the power flow. However, the critical point depends on the type of load model we use. If the load is other than constant impedance, the results may not be accurate. We have not tried this approach. The real power transfer decreases as the increase in number of injection buses is valid with respect to maintaining voltages within the limits. Since we are minimizing the reactive power, the optimization tries to inject enough reactive power to satisfy lower limits of voltages. The in- crease in lower limit increases the required reactive power injection and indirectly results in increase in real power transfer. The main reason behind the relaxation strategy is to predict the change in the sign of index for slight drop in reactive power support. Table 5 indicates for a fast continuous injection of reactive power the power transfer can be increased to 118.255. But table 6 gives the przctical power transfer of 108.8679. Since after this power transfer, slight change in reactive power leads to an opcLrat,irig point beyond the turning point.

( 2 ) When it comes to spotting the weakest bus, it is helpful to think in terms of a P - V curve, or considering the more general situation, X - V curve. Heuristically speaking, the weakest bus is the one that is closest to the turning point or knee of the curve since that is where stability is lost. Equivalently, a weak bus is one that has a large ratio of differential change in voltage to differential change in connected load. The weakest bus is then the one with the largest ratio, where CdX gives the differential change in active load for the whole system. Actually, the formula for the index is not system dependent. The value of the index depends on the operating point. Accordingly, the weak buses which are calculated from this index change from base case to critical case.

(4)When the weakest bus j, calculated as above, reaches its steady-state voltage stability limit, dX becomes zero, i.e. dX is zero a t the critical point and is negative beyond the critical point. Thus, the change in sign of the dX component from tangent vector reveals whether or not the critical point has been passed. Also, it was mentioned in the paper that the ratio is defined as the voltage dV, stability index. The criterion we used in the optimization

917

routine is the voltage magnitude violation in conjunction with the above index.

Oladiran 0. Obadina: We address the questions in the


same order, as follows: Once again, we didnt use voltage magnitude alone with optimization process. This was explained in our response to K. N. Srivastava, S. C. Srivastava. and P. K. Kalra. The objective function proposed in the paper can be converted to minimize the cost of VAR injection by adding a fixed installation cost and variable unit cost for capacitive and inductive sources. Location in some aspect deprnds on the physical constraints of locating the reactive power devices. Selecting optimum number depends on several factors and may lead to multi-objective programming. For example in Table 4 by increasing the number of locations, the amount of reactive power required to rectify the voltage violation decreased. However, increase in number decreased the power transfer. 0 We agree that in planning process, one should consider several base and contingency cases. In the C P F , each continuation step can be (approximately) considered as a base case. For planning against contingencies, the methodology proposed can easily be modified with increase in CPU time. To achieve this objective, first the contingencies which are critical with respect to the voltage
0

stability can be identified from the branch sensitivities, which can be calculated from the tangent vector (how to calculate these sensitivities are explained in section 3.2 of the paper). Then we can re-run the C P F for each critical contingency identified.

References
[All
V. Ajjarapu, and B. Lee. Bifurcation Theory and its Application t o Nonlinear Dynarnical Phenomena in an Electrical Power System. IEEE Transactions on Power Systems, vol. PWRS-7, pp. 424-431, Feb. 1992. B. Lee, and V. Ajjarapu. Period-doubling Route to Chaos in an Electrical Power System. To be Published an the Proceedings of IEE, Part-C

[A21

~ ~ 3 1 General Electric Company. Load Modeling for Power Flow and Transient Stability Computer Studies. EPRI Publication, EL-5003, vol. 1-4,
Jan. 1987.

~441

V. Ajjarapu, and S. Battula. Effect of Load Modeling on Steady-State Voltage Stability. To be Published in Electric Machinrs and

Power Systems Journal


Manuscript received October 12, 1993.

Potrebbero piacerti anche