Effect of Load Modeling on Voltage Stability

S.Z Zhu, J.H Zheng, L Li*, S.D Shen, G.MLuo** Dept. of Electrical Engineering, Tsinghua University, China * Dept. of Electrical Engineering, Univ. of California, Los Angeles ** Dept. of Applied Mathematics, Tsinghua University, China
Abstract - This paper analyzes the analysis of the relation between the mechanism of voltage stability and load models. Two methods, IV (Instrumental Variable) and MSAR (Minimum Sum of the Absolute Residues), are used to' identify load parameters of nonlinear and linear load model. ESA (Eigenvalue-Structure Analysis) and two sensitivity methods based on ESA are presented to analyze the voltage stability: sensitivity on state variables or control variables and sensitivity of total generated reactive power. Examples are presented by simulation on IEEE-14 buses system, which shows quantitative effects of different load models.

Keywords

voltage stability, load modeling, parameter

identification.
INTRODUCTION

In recent years the voltage stability problem has been paid much more attention primarily due to a number of stability accidents that occurred in some countries. In the present research on voltage stability, although the mechanism of voltage instability is not well understood yet, it is almost sure that load representation can have a significant impact on voltage stability [I]. Therefore, the main purpose of this paper is to discuss the method for load modeling and the effect of load models on voltage stability study.
Now, electrical load modeling has been an important issue in power analysis and control area. Electrical load models are the basis of analysis, programming, operation and calculation of power system. Especially the simulation results of dynamic behavior have been more affected by the selected model in the system calculation. It is known that different load models may lead to different conclusions [2,3]. The studies of the western region of the United States indicated that using a constant impedance load model in small signal stability analyses tends to overestimate the damping by about 25% as compared with a more accurate load representation [4]. In transient stability calculation for the power systems in China, the power output of specific power plants and its economics were affected by 4 6 % depending on the load model used. In the studies on the impact of the loss of a major transmission line feeding a power region, a substantial difference was found on the results of applying dynamic load models and static load models. Some typical models may lead to total wrong conclusions. The load models are also important in voltage stability analysis. Initial efforts using simple load models failed to explain the voltage collapse scenario [5,11]. So it is necessary to study load-modeling problem and to determine practical parameters of load models.

The characteristics of electrical load in a power system are generally represented by non-linear time-variant with a random behavior. The constructions of electrical devices are different and the characteristics of electrical load are not the same, such as constant impedance, constant current, constant power or considering different dynamic-static percentage load models. These load models can be described as polynomial, exponential or difference equation load model. When voltage or current suddenly change, electrical load also change according to its dynamic responds. It will be difficult and complex to set up the system model represented for dynamic and static feature in a large scale of power system. Parameter Identification is the key for load modeling [ 6 ] . Two identification methods, IV (Instrumental Variable) and MSAR (hhnimum Sum of the Absolute Residues), have been applied on modeling. The methods are both fast and advanced. And both of the methods IV & MSAR can be used online. To analyze the voltage stability, two sensitivity methods, which are based on ESA (Eigenvalue-Structure Analysis), are presented: sensitivity on state variables or control variables and sensitivity of whole reactive power. The methods can be used to search for voltage instability area, optimal reactive support locations and optimal load shedding strategy. They can also be used to judge the possible instability model, voltage instability, power angle instability or the combination of both. In this paper, the results from examples simulating on IEEE-14 buses system show that it is better to use load parameters obtained from field than that from conventional load models.
2

ELECTRICAL LOAD MODELS

Some of the mathematical expressions for modeling static and dynamic loads are described as follows.
21 . Static load models

Static load models can be defined as the function of the active or reactive power of load wt the voltage and ih frequency on the load buses at the moment. Usually it is expressed as exponential or polynomial function. Exponential function (eq. 1) is defined as follows.

0-7803-6420-1/00/$10.00(c) 2000 IEEE

395

3 IDENTIFICATl[ON METHODS

Error function is defined as following:

N
N

-. 2

In the procedure of parameters identification, sometimes it will happen that the identified results are different under the same type of load model and the same combining. It is called parameters instability. The way is called unbiased uniform estimation in the case of colored noise. Both W a n d MSAR are good methods for solving the problem.
3 1 Instrumental Variable (IV) .

Where N is the number of sampling points, Wm(i) is the i" measured value of active or reactive power, Wc(i) is the ih calculated value of active or reactive power on Ui, fi by equation. Xp. Xq are the vectors of estimated parameters. This estimation problem is transferred as a minimum optimal problem, which means to search a group of parameters to make error Ew minimum.

Difference equation is linear to the estimated parameters, which can be denoted as the least square (LS) format. With the consideration of measurement error and stochastic fluctuation of loads, and with the assumption of unknown statistical characteristics of error, we have:
A y ( k ) = p T (k);+

X p =IP0.ai,a2 1 X q = IQ s b l . b 2 1 0

(3)

E(k)

(7)

Polynomial function is defined as follows:

jI = f O [ r p .v . . + b P ( y ) L

](1+k Af)

IQ
2.2

Where Ay@) is the increment of active or reactive power, (pT(k)is row vector for increment or input and output variable, U is the column vector for all the estimated parameters, E(k)
(4)

P V
= QoIoq
2 Cy)
0

P
0
+ c

0 v

is correlated noise with unknown statistical characteristics. If least square method is directly applied to Equation (7), the biased estimation of 0 will be obtained [7]. Since IV method needn't know the composition of noise in advance and can be expressed as recursion formula, it is feasible for this case. The general algorithm of IVmethod is:

.+

V b (-)

Y V

K 1 + k Af) 4

Because P equals to PO as V equals to VO,we get the following equations: a,,+b,+c,= 1, %+b,+c,= 1 Dynamic Load Models

Dynamic load models can be used to represent the function relation between the active or reactive power of load and the voltage and frequency of load bus at the moment or several times intervals before the moment. Mathematics models have many parameters that are difficult to be identified. Input/output models consider the load as a "system". The input variables are the voltage U and frequency f on load bus, the output variables are all active and reactive power of the load. K order difference equations, which are PQ decoupled, are used to describe dynamic characteristics of load in order to improve the precision of models [8][9]. The nonlinear feature of load can be represented by the square of voltage. The linear feature of load can be represented by one order function of the voltage.
AW ( n ) = C AiAW ( n 1=1

Where instrumental variable t(k) can be selected as &k)= (p (k-ko);ko is positive integer no more than the order of noise to obtain unbiased uniform estimation for parameters. For on-line identification, equation (8) can be transformed to iterate formula:
;& (

+ 1) = ;(k) + L(k + l ) [ A y ( k + I ) - p T ( k + L);(k)]

/o\

P(k+ I) = P(k)-

P (k)&k

+ 1)J

( k + 1)P(k)

1+ J ( k + l ) P ( k ~ ( + I ) k

I )

k + C EiAU (n
i=O

i)

(5)

Initial matrix P(1)=(I k ) I , 6 (I)=h', where E is positive real number as little as possible, 6 is real vector as little as possible, I is identity matrix. In order to avoid the sensitivc problem of recursive T method for initial matris P( I), we I' use LS method first before starting W method to improve calculation results, that is, {(k+ 1) is replaced by (p (k+ I) for about 50 steps, then IVmethod comes into use.
3 2 Minimum Sum of the Absolute Residues (MSAR) .

'

+ C C i A / ( n - i)+
i=O

1=O

Di(AU(n

i))'

When the data are processed, each inputloutput variable uses increment, the identified parameters can be represented by a vector: X=[A 1, ...AbB, ... ,Bk ,Ch ... ,CbDa ... ,D/c] (6) Error Function is the same as equation (2).

The linear equation with the estiiniled parameters is assumed as follows:

Where xli=[x,b

x2b

... , x,,,klrTis an input column vector, yk is an

0-7803-6420-1/00/$10.00 (c) 2000 IEEE

396

output column vector, &[O1,81, ....e,,,lT is a column vector for all the estimated parameters, ek is the error. The rule function is defined as:

obtained by distant technologic with adverse iterative theory. The calculation burden is decreased and the calculation speed is fast. The studied system scale is not limited. The study results of 14 buses are discussed.
4.1 ESA method

By comparing this rule function with LS objective function, we can convert this question and solve it with Weighted Least Square, and find the solution with Variable-weighted Least Square iteration for multiple times. This is called Variable-weighted Slack Least Square. If the residue of solution to MSAR is e,@), we have:
(12)

The power system load f o equations are: lw

Considering static voltage characteristics, we take deviation from equation (14) and ingest analysis on work point, and then we can get the follow.

Where e&is the residue of equation (10). However the rule of Weight Least Square is:

I;:[

That is:

[:; :; 3 ; ;

[I

(15)

Let b,= lIIeu(k)l, we obtain JU(O)=J,(O). In another word, if we select the weight as l/leu(k)l, we can get the solution with the rule of MSAR. Therefore, the residual key problem is how to get leu(k)l. We can solve this problem with concept of Slack Algorithm as follows: Get e&) by LS method first, and replace eu(k) by eu(k). Next, let el(k) = e&), extract the variable weight l/[e,(k)l. Then work e2(k) out with Variable-weighted LS. e,(k) approaches eLA(k)more closely than e,(@. So we obtain the weight llle2(k)l. Afterward we find the solution to Variable-weighted LS again. Repeating the steps above, we have ti (LS), o" ( l ) , e' (2), i (L) and get e^ (LA) when L ---*CO.
4

AY = JAX (16) J is the Jacob Matrix with polar coordinates form. AY is the variation of inject vector and .AX is the deviation of state vector.

If J is of non-zero & non-singularity characteristic root A I . A 2... A and corresponding right distinction vector UI U2...U,,, , left distinction vector V I , V2...Vm. relational expression between value of determinant with characteristic root will be follow:

VOLTAGE STABILITY

As we know, mathematical models have important effect on calculation results. We have achieved more conclusions [ 101 in transient stability studies in power system. According to researching results, the mathematical models, especially load model and its parameters, also have important effect on the calculation of voltage stability [l2]. In this paper the ESA method is provided for the study of voltage stability. The decision of steady stability limit is singularity value of Jacob Matrix of load flow and the least pattern of characteristic root in the Jacob Matrix will be the criterion of margin indicated of stability in the power system. The ESA method not only provided criterion of abundance indicated of stability in the power system but also directly provided sensitivities for steady variable and control variable. And an algorithm that is used to calculate sensitivity of all-round reactive power was provided. The criterion of abundance indicated of stability and sensitivities will be

It will be a zero characteristic root as Jacob matrix J is singularity after limit of steady stability amved. The matrix changes to singularity when normal operation condition transfers to limit of steady stability. First the least pattern of characteristic root in the Jacob Matrix must change monotonously to zero. Now we can choose the least pattern of characteristic root as a criterion of abundance indicated of stability in the power system. Both AY and AX are expressed for right characteristic root respectively.

By relational expression:

There is:
V~TAY =

vir2a , U i
IS1

= a i

Vi'AX

V I Z 1.1

f piu, = p ,

(21)

Because of Ju , = i i u ,
Then:
a .
I

(22) (23)

A 1. p 1.

Take (23) and (20) instead into (18), get

0-7803-6420-1/00/$10.00 (c) 2000 IEEE

397

4.3 Sensitivity method based on ESA

2 a;lu
8.1

, v ,= ~

-I

is

called

spectrum

decomposition.
4.2 ESA of Voltage stability

The relationships between V,and Uiwith characteristic root AI are: J,U, = A,U (35)
J I V , = A,V,

(36)

Derivation xk and multiplication

, ' F

arrangement as follows:
(37)

In order to analyze the problem of voltage stability, suppose AP=O, (15) is simplify:
Matrix form :

Here: JR is call deflation form of Jacob matrix. If

an,= I aA, ax ay

' then

The singularity of JpQ, corresponds to instability of power angle and the singularity of JR corresponds to instability of system voltage. We define difference with involving factor so that we can express effect of various units in study of voltage stability. These are three involving factors (1) Bus involving factor

Equation (37) and equation (39) give out the sensibility of AI to state variable and control variable respectively. According to the information of the sensibility, we can analyze the reasons of losing stability in power system and control the control state including the power of load nod under the given constrain condition to increase the static stability margin AmIn of system.
4.3.2 Sensibility of Whole Reactive Power

Margin function of voltage stability I// is defined by (14)

ntm

v/
Pk;=uk;v;k

-1

= Q, =

, = s t + l )-I

,E

Y , Y , ( G , sin

e, - B ,

cos

s,)(QO)

is involving factor of k" bus.

(2) Transmission line J involving factor

S is sensitivity of Qgto Y. Matrix S is composed of active power load, active power generation and reactive power load.

(3) Generator involving factor

We can get frustrates of ameliorate stability by sensitivity factor ordering.

The Procedures of ESA method are as follows: Stepl: To decompose for factor list of (J - SE) gtep2: Tdterative and calculate A,,,;,, and Umin Ly ( t ) = (t), z = (1,l ,..., 1y
J

qF

(30)
LU.

if lmax(b@+'))- max(uXk')l<E, stop iteration, and get:

4.3.4 Effect of load model for static state stability

( 0 )

uw
z

( t . 1 )

= y

(t)

(&.I)

( & * I ) ,

max @/

(1.1))

Exponential model and polynomial model are shown as equation (1) and (2) The Jacob Matrix for analysis of static state stability is as follow:
(43)

Ignoring the effect of frequency, let kp=kg=O, then substitute this into equation(43), we get the diagonal elements of L matrix as foIlows:

0-7803-6420-1/00/$10.00 (c) 2000 IEEE

398

,,o Sensibility
0.8

!
L

The eigenvalues of J will be the center of a circle of diagonal by Gerschgrin Law. Loii is Jacob Matrix taking no account the effects of static load model. It is negative. Both a, and 6, are positive. So LA is positive. Have considered static load model, we got the increased absolute value of diagonal elements in L matrix. And the absolute value of the least eigenvalue also is increased.
5

0.6 0.4

0.2

0.0-

Fi&e 4: Sensibility of Whole Reactive Power to Reactive Power (Z = c)

111I
10

11

12

Bus

14

APPLICATION AND RESULT

The results show us that the index of involving factor and sensibility of 14th buses are the highest one in IEEE-14 buses system, therefore the 14th bus is a weak bus in this system. In

In this chapter, examples of stability calculating to IEEE-14 buses system are presented. The following results are satisfied. Case 1: Constant resistance model and a- = 0.0177 Table 1: Generator Involving Factor of IEEE-14 buses
Order 1 2
3

biggest ones in the system: It narrates that the mode of losing stability in the system is the instability of power-angle and voltage coupling. Case 2: Constant current model and I Involving Factor 10 . =0.3253

GEN NAME 01 06
08

PFACTOR 1.0000 0.1633 -0.002 2

03

-0.0012

0.8

I 5I
BUS1 BUSJ
R

02

-0.0010

I
Involvin

0.6

0.4

02 .

0.0

9 1 0 1 1 1 2 1 3 1 4

Bus

1
0.8

Figure 3: Bus Involving Factor of IEEE-14 buses system (I = c)

Sensibility

0.6

0.4

1 .o

Involving Factor
-:
!

:-BUS
= c)

0.8

Figure 4: Sensibility of Whole Reactive Power IO Reactive Power (I

0.4

02

0.0

1 0 1 1 1 2 1 3 1 4

Bus

Figure 1: Bus Involving Faclor of IEEE-14 buses system (Z = c)

Comparing Fig 1 with Fig 3, Fig 2 with Fig 4, we can obtain different results. The involving factor of bus with constant resistance is the first and the involving factor of bus with constant current is the 9th. All of sensibility on the condition of the constant current will decrease more than under the condition of the constant resistance. It shows that the load model is very important in calculation of voltage stability. The limit of voltage stability-has been calculated on the conditions of five kinds of load models.

0-7803-6420-1/00/$10.00 (c) 2000 IEEE

399

CONCLUSION

A point

M
P=c
I=C

0.33107 0.6667
C

0.83929 0.19998
--Cc

Z=c Pol
exp

0.31574

I 0.58637 I

0.84711 0.26827

0.44673 0.57006 0.31485 0.4911

B point m a V
-c-

-c point
m
V

Both of parameters identification method, IV and MUR, can be applied to online identification in the power system. These methods are effective and advantageous. The ESA of voltage static stability is studied in theory and margin of stability is obtained quickly by rarefaction access technique on algorithm. It can be a useful tool in calculation in the power system. Two kinds of sensitivity methods are used in this paper. These methods are effective. The diagnosis capacity in stability problem study can be enhanced. Parameters and models of load have important effects on the study of voltage stability. The calculation results of voltage stability with constant resistance are more conservative or exceeded than the results obtained with load parameters from real field test.

0.84884 0.8549 0.82731 0.8235

0.4598 0.62081 0.3155 0.41182

0.83438 0.83053 0.83732 0.85

Table 3 shows relation of absorbed power of load and voltage on bus. In the table; m=kv2, v is voltage of bus. It corresponds to A point that det(Js) equals to zero. The singularity of JN corresponds to B point. The inflection point of P-V curve corresponds to C point. According to Table 3 and Fig 5,6, we can conclude that the limit of stability based on B,C is conservative or excess and the best criterion is based on A on which load model is considered.

0.8

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

2.4

2.8

Figure 5 : P-V Curve of Constant Resistance Load

Figure 6: P-V Curve of Static Polynomial Load

RENCE [I1 Wen-Shiow Kao, Chiang-Tsang Huang, Chiew-Yau Chiou, Dynamic Load Modeling in Taipower System Stability Studies, IEEE PWRS Vol. 10. N0.2 May 1995, p 907-914 12] CIGRE study Committee 38,Load Modeling and Dynamics, ELECTRA, 1990, pp124-141 i3] IEEE Task Force on Load Representation for Dynamics Performance, Load Representation for Dynamic Performance analysis IEEE PWRS, 1993, 8(2): pp472-482 i4] Y, Mansour, Application of Engineering Analysis to the Western North American Power System, IEEE Pub, 909292-3, Engineering Analysis and Frequency Domain Methods for System Dynamic Performance, 1990. Deuse J, Stabbe M.,Dynamic Simulation of Voltage collapse, IEEE Tans on PWRS, 1993 8(3), ~~894-904 P.Ju., E. Handschin, D. Karlsson, Nonlinear Dynamic Load Modeling: Model and Parameter Estimation IEEE PWRS. Vol.11 No.4 PP1659 1697 Fang Chon@, Xiao Deyun, Process Identification, Tsinghua University Press, 1988 K.Srinivasan,et al., Load Response Coefficients Monitoring System: Theory and Field Experience, IEEE Trans. On Pas, Vol. PAS-100, No.8, Aug.1981, pp.38 18-3827. T.Dovan, et al., A Microcomputer Based On-Line Identification Approach to Power System Dynamic Load Modeling, IEEE Transactions on Power Systems, Vol. PWRS-2, No.3, Aug. 1987, pp.529-536 S.Z. Zhu, S.D. Shen and H.L. Chen,The direct parameter iden~cation approach for continuous-time model of power system transient stability studies IEEE International Conference 93, Hong Kong Dobson I, Chiang H.D., Towards a theory of Voltage Collapse in Electric Power System, systems and control letters 1989(13) pp25-262