3rd International Conference on Electric and Electronics (EEIC 2013)
Transient Stability Analysis of Power System Using Improved Least Square Method
and Energy Function
Dongrong Jiang
School of Electronic Information and Automation
Chongqing University of Technology, CQUT
Chongqing, China
e-mail: drjiang@126.com
Lian Zhang
School of Electronic Information and Automation
Chongqing University of Technology, CQUT
Chongqing, China
e-mail: zh_lian@cqut.edu.cn
Abstract—The wide application of WAMS (Wide Area
Measurement System) provided a new thinking for transient
stability analysis of power system. Based on real-time
measurement information from WAMS, a method to quickly
analysis power system transient stability that does not depend
on network structure, model and parameters of power system
is proposed in this paper. Firstly, the parameters of system are
identified online by improved least squares method on basis of
that the multi-machine power system is equivalently reduction
to one machine infinity bus system. The equilibrium point and
stability margin of equivalent system are rapidly computed,
and then the stability of system is quickly judged. The
simulation results show that the algorithm can accurately
judge transient stability of system and has better anti-
disturbance ability and robustness. The issue of dependence of
transient stability analysis on system model and parameter is
solved. The adverse influences of data lose and measurement
errors in actual measurement of the WAMS on analysis results
are avoided effectively.
Keywords-power system; wide-area measurement system;
parameter identification; energy function; transient stability
analysis
I. INTRODUCTION
Last decade, an urgent requirement about the power
system stability control has been proposed because of
the large area blackout frequently occurred on the global.
Power system transient stability control is the first line of
defense of stable operation [1], with a particularly important
research value, while transient stability analysis is its
foundation. Therefore, it is important to look for a quick and
accurate method for transient stability analysis. There are
some traditional transient stability analysis methods such as
mainly numerical integration and direct method [2], but it is
difficult to coordinate its calculation accuracy and speed.
Artificial intelligence method has the advantages of fast
online calculation and easy to generate inspired rule of
© 2013. The authors - Published by Atlantis Press 337
Zhao Li
School of Electronic Information and Automation
Chongqing University of Technology, CQUT
Chongqing, China
e-mail: 396253977@qq.com
decision, which constituted a fine complementation with
traditional analysis methods. Currently, the main artificial
intelligence methods are support vector machine [3], neural
network [4], machine learning method [5] etc. However, the
calculated amount of artificial intelligence method is huge,
so there are many difficulties in practical application.
WAMS has been widely used in modern power system
for monitoring, operating and controlling [6], which
functions and applications have improved. It not only can
provide technical support to building strong and smart grid,
but also provides a new idea in power system transient
stability analysis, which mainly about trajectory prediction
and estimation of power angle. There are curve
extrapolation [7] and trajectory prediction method based on
system model [8] for the moment. However, the adverse
effect on analysis results was not considered in these
approaches, which because of measurement data loss and
measurement error of WAMS in practical application, and
hindered its popularization of application in engineering. A
method to quickly analysis power system transient stability
that does not depend on network structure, model and
parameters of power system is proposed in this paper. The
basis of this method is that the multi-machine power system
is equivalently reduction to One Machine Infinity Bus
(OMIB) system. The parameters of system are identified
online by Regularized Robust Recursive Least Squares
(R3LS) [9] method. The equilibrium point and stability
margin of equivalent system are rapidly computed, and then
the stability of system is quickly judged. Thus, the accuracy
of stability determination of system is improved greatly.
II. EQUIVALENT REDUCTION OF MULTI-MACHINE POWER
SYSTEM AND PARAMETER IDENTIFICATION
A. Equivalent Reduction of Multi-machine Power System
The calculation speed of complex power system
transient stability analysis can be improved effectively by
using homology equivalence technology, which include
identification and aggregation of homology cluster [10]. The Equation (5) reduction to ARMA as
homology cluster is identified by using composite criterion y ( k ) + a1 y ( k − 1) + a 2 y ( k − 2 ) + L + a na y ( k − na ) =
that consist of angleδ and rotate speedω. The composite
power angle is defined as B1U ( k − 1) + B2 U ( k − 2 ) + L + Bnb U ( k − nb ) + (6)
δcom = δ + kΔω (1) ε ( k ) + e1ε ( k − 1) + e2 ε ( k − 2 ) + L + ene ε ( k − ne )
Here, k=n+1,n+2,...,t; n=max[na,nb,nc]; t is time stamp.
Here, k is setting, and Δωis the variation of rotate speed
per sampling period. The critical cluster S and surplus C. R3LS Algorithm
cluster A could be separated at the biggest clearance of
The vector θ = [ a1,a2 ,Lana ,B1,B2 ,LBnb ,e1,e2 ,Lene ] is
T
abutting composite power angle δcom by descending sort the
the parameter vector to be determined. To specify a
δcom of every generator. prediction error criterion, a virtual linear prediction model is
The power angles curve of complementation cluster are constructed as
aggregated by using CCCOI-RM transform after separated. ∧
We can use two equivalent generators replace the y( k θ) =−a1y( k −1) −a2y( k −2) −L−anay( k −na) +
complementation cluster, and reduction to two machine
1 ( k −1) + B2U( k −2) +L+ BnbU( k − nb) +
system. As discussed in [10] and [11], the two machine BU (7)
system can reduction to OMIB system further and the
⎛ ∧ ⎞ ⎛ ⎞ ∧
e1ε⎜k −1θ( k −1) ⎟ +L+eneε⎜k −ne θ( k −ne) ⎟ =ϕT ( k) θ
mathematical model is constructed as
M eq δeq
''
= α + β cos δeq + γ sin δeq (2) ⎝ ⎠ ⎝ ⎠
Here, M eq = MS M A ( MS + M A ) is equivalent inertial
Where
ϕ( k) ⎡⎣−y( k −1) ,L, −y( k − na) ,U( k −1) ,L,U( k − nb) ,
time constant, and δeq = δS − δ A is equivalent power angle,
and α、β、γ are parameters to be identified. ⎛ ∧ ⎞ ⎛ ∧ ⎞⎤
T
(8)
ε⎜ k −1 θ( k −1) ⎟,L, ε⎜ k − ne θ( k − ne) ⎟⎥
B. Model of Parameter Identification ⎝ ⎠ ⎝ ⎠⎦
In [10], the parameters to be identified of (2) were The prediction error at time k can be defined as
⎛ ∧ ⎞ ∧⎛ ∧ ⎞
ε ⎜ k θ(k)⎟ = y(k) − y⎜ k θ(k)⎟
solved by Recursive Least Square (RLS) method with
(9)
forgetting factor, but the calculation accuracy and anti-
⎝ ⎠ ⎝ ⎠
disturbance ability are not enough ideal. Thus, R3LS Define
algorithm is used to solve it in this paper. But before, (2) ∧
needs reduce to Auto Regressive Moving Average (ARMA).
Define
( )
ε k θ(t) = y(k) − y k θ(t) ( ) (10)
And
δ''eq ( mT) = ⎡⎣δeq ( m + 1) T − 2δeq ( mT) + δeq ( m − 1) T⎤⎦ T2
T
1⎛ −
⎞ ⎛ −
⎞
m = 1,2,L,n − 1 (3) J ( θ( t )) = ⎜ θ( t ) − θ ⎟ ∏( t ) ⎜⎝ θ ( t ) − θ ⎟⎠ +
Here, T is sampling period, and m is number of sampling 2⎝ ⎠
times. (11)
( )
t
According to (3), (2) can be written as ∑ λ ρ ⎡⎣ε k θ ( t ) ⎤⎦
t −k

δ eq ( k ) − 2 δ eq ( k − 1 ) + δ eq ( k − 2 ) = ⎡⎣ α T 2 +
(4)
β T 2 cos δ eq ( k − 1 ) + γ T 2 sin δ eq ( k − 1 ) ⎤⎦ M eq
Consider the disturbance ε(k) of system, (4) can be
written as
y ( k ) + a1 y ( k − 1) + a 2 y ( k − 2 ) + L + a na y ( k − na ) =
b1u ( k − 1) + b2 u ( k − 2 ) + L + bnb u ( k − nb ) +
c1v ( k − 1) + c2 v ( k − 2) + L + cnc v ( k − nc ) +
d1w ( k − 1) + d 2 w ( k − 2 ) + L + d nd w ( k − nd ) +
ε ( k ) + e1ε ( k − 1) + e2 ε ( k − 2 ) + L + ene ε ( k − ne )
Define
B1 = [ b1 , c1 , d1 ] ,L , Bbn = [ b bn , ccn , d dn ] ;
−
k =n +1
Here, θ is constant vector and the prior guess of
parameter θ. The matrix ∏ ( t ) is a positive definite matrix.
It describes the confidence level in this initial guess at time t.
Its main function is that it resolves the under-determined
problem associated with a normal LS estimation. For the
regularized LS algorithm, the regularization term helps
skew the estimate towards the initial guess. Note that if the
(5) initial guess is different from the true value, a bias is
introduced into the estimation results. The constant λ is the
forgetting factor. ρ(ε) is a loss function that defines a
measure for the errors. It is used to reduce the potential
effect of no typical data, and that’s the robustness property.
In the normal case defined as

U ( k − 1) = ⎡⎣ u ( k − 1) , v ( k − 1) , w ( k − 1) ⎤⎦ ,L ,
T

U ( k − nb ) = ⎡⎣ u ( k − nb ) , v ( k − nc ) , w ( k − nd ) ⎤⎦
T

338
 ⎧ 1 2 b) If ΔV ' < −ε , the power system transient is instable.
⎪⎪ 2 ε , if ε ≤ 3σ
c) If 0 < ΔV ' < ε , the power system transient is critically
ρ (ε) = ⎨ (12)
stable.
⎪ 1 ( 3σ )2 , if ε > 3σ
⎪⎩ 2 d) If −ε < ΔV ' < 0 , the power system transient is critically
Here, the variable σ is the standard deviation of theε. instable.
Then, a regularized weighted LS estimation of the θ can
IV. SIMULATION RESULTS AND ANALYSIS
be defined as
∧ An IEEE39 model is used in this paper to comparative
θ ( t ) = arg min ⎡⎣ J ( θ ( t ) ) ⎤⎦ t ≥ n + 1 (13) analysis the transient stability results which using the
method above, BPA simulation and the online RLS method
III. TRANSIENT STABILITY ANALYSES with forgetting factor in [10]. In BPA simulation, generators
There are two equilibrium points s and u in the use the classical model and load use the integrated dynamic
equivalent system (2) after fault. Equation (2) can be written load model. The online RLS method which contains
as by being equal to zero forgetting factor and the R3LS algorithm are realized by
MATLAB software. WAMS measurement data and fault
α + β cos δeq + γ sin δeq = 0 (14) Critical Clearance Time (CCT) are obtained by BPA
Two radicals δeqs and δequ can be obtained with leading simulation which the sampling period is 0.05s. Analysis
results of the three methods are shown in TABLE Ⅰ .
the parameters from using R3LS algorithm in the equation
The ’∗’is the point of fault and all the faults are three-phase
above. The equilibrium point s corresponds to δeqs is stable grounded short circuit.
equilibrium point, and the equilibrium point u corresponds
to δequ is instable equilibrium point. According to δeqs TABELI. ANALYSIS RESULTS OF TRANSIENT STABILITY

and δequ , the transient state energy and critical energy of Fault
Fault Fault R3LS RLS
BPA
CCT clearance Energy Energy
system can be obtained. location Judgment Judgment result
(s) time (s) margin margin
The transient state energy of system at the fault
clearance moment is 0.10 2.013 Stable 2.516 Stable Stable
1 0.15 0.632 Stable 0.839 Stable Stable
Vcl = Vke + Vpe = M eq ωeqcl
2
−
2 (15) 0.20 0.259
Critically
0.314 Stable
Critically

( α + β cos δeq + γ sin δeq ) dδeq

δeqcl stable stable
∫δ eqs
25∗-26 0.21
0.21 0.104
Critically
stable
0.147
Critically Critically
stable stable
1 Critically Critically Critically
Here, Vke = M eq ωeqcl
2
is the kinetic energy and 0.22 -0.194
instable
0.078
stable instable
2
( α + β cos δ + γ sin δeq ) dδeq is the potential
δeqcl 0.25 -0.872 Instable -0.721 Instable Instable
Vpe = − ∫ eq
δeqs 0.15 2.537 Stable 2.782 Stable Stable
energy of system. ωeqcl is the angular velocity and δeqcl is 0.20 0.931 Stable 1.251 Stable Stable
the power angle of the equivalent system at the fault Critically Critically
0.25 0.271 0.418 Stable
clearance moment. stable stable
3∗-4 0.26 Critically Critically
The critical energy of system is 0.26 0.194
stable
0.303 Stable
stable
( α + β cos δ + γ sin δeq ) dδeq
δequ
Vcr = − ∫ eq (16) 0.27 -0.071
Critically
0.046
Critically Critically
δeqs instable stable instable
The energy margin of system is 0.30 -1.192 Instable -1.027 Instable Instable
ΔV = Vcr − Vcl (17)
Equation (17) can be re-formatted as The correctness of transient stability simulation results
ΔV ' = ( Vcr − Vcl ) Vcl
of above method is verified, because the results are
(18)
consistent with BPA simulation. Nearby the critical stability
According to the La Salle invariance principle and the point of the system, the results of RLS with forgetting factor
second Lyapunov theorem, transient stability of system can sometimes are false. However, the method above can
be judged after obtaining the energy margin. We set a accurately determine system transient stability after clearing
threshold value ε(ε≥0) of transient stability determination. fault, because R3LS algorithm leads a positive definite
The different values can be set at the instance of accuracy, matrix in the objective function, which makes the estimated
and in the normal case we set it 0.3. Judge transient stability value regularization and to be close to actual value. Thus,
as follow the accuracy of parameter identification is superior to
a) If ΔV ' > ε , the power system transient is stable. ordinary least squares method.
In order to verify that the algorithm has strong

339
robustness, Gaussian random white noise is added into in this paper. It was shown that this method could accurately
WAMS data of BPA simulation in this paper. The simulation determine system transient stability by comparative analysis
results are shown in TABLEⅡ. In order to simulate atypical between BPA and MATLAB program simulation. Moreover,
measurement data because of short-term communication this method could still ensure the accuracy of judgment with
failure of WAMS, the data values of two sampling periods WAMS measurement data containing noise or occurring
of node 16 are adjusted to zero. The simulation results are short-term communication failure, and it has better
shown in TABLEⅢ. robustness, which could avoid the adverse effect of data loss
and measurement error effectively. The parameters of this
TABELII. ANALYSIS RESULTS OF TRANSIENT STABILITY WITH ADDING method are obtained with the real-time measurement data or
NOISE TO MEASURED DATA calculating directly, without prior knowledge of network
Fault R3LS RLS
structure and parameters. Thus, this method can adapt to the
Fault Fault complex system model, reflect the actual operation of the
clearance Energy Energy
location CCT (s) system accurately, and can be used for real-time transient
time (s) margin Judgment margin Judgment
stability analysis.
0.10 2.253 Stable 2.371 Stable
0.15 0.571 Stable 0.492 Stable REFERENCES
Critically [1] Lu Qiang, and Mei Shengwei, “Important basic researches on power
0.20 0.237 0.302 Stable systems facing 21 century,” Natural Science Development, vol. 10, no.
stable
25∗-26 0.21 Critically Critically 10, pp. 870-876, Oct. 2000.
0.21 0.092 0.114
stable stable [2] WU Wei, TANG Yong, SUN Huadong, and XU Shiyun, “A Survey on
Critically Critically Research of Power System Transient Stability Based on Wide-Area
0.22 -0.184 0.091 Measurement Information,” Power System Technology, vol. 36, no. 9,
instable stable
pp. 81-87, Sep. 2012.
0.25 -0.931 Instable -0.825 Instable
[3] Wu Qiong, Yang Yihan, and Liu Wenying, “Electric power system
transient stability on-line prediction based on least squares support
TABELIII. ANALYSIS RESULTS OF TRANSIENT STABILITY WITH vector machine,” Proceedings of the CSEE, vol. 27, no. 25, pp. 38-43,
ADJUSTING DATA OF TWO SAMPLE PERIODS AROUND FAULT OF NODE 16 TO
Sep. 2007.
ZERO
[4] A.G Bahbah, and A.A Girgis, “New method for generators angles and
Fault R3LS RLS angular velocities prediction for transient stability assessment of
Fault Fault multimachine power systems using recurrent artificial neural
clearance Energy Energy network,” IEEE Transactions on Power Systems, vol. 19, no. 2, pp.
location CCT (s) Judgment Judgment
time (s) margin margin 1015-1022, Feb. 2004.
0.10 2.126 Stable 1.824 Stable [5] Ye Shengyong, Wang Xiaoru, Liu Zhigang, and Qian Qingquan ,
“Power system transient stability assessment based on severely
0.15 0.681 Stable 0.789 Stable disturbed generator attributes and machine learning method,”
Proceedings of the CSEE, vol. 31, no. 1, pp. 46-51, Jan. 2011.
Critically Critically
0.20 0.233 -0.047 [6] Adamiak M, Apostolov A, Begovic M, Henville C, Martin K, et al,
stable instable
25∗-26 0.21 Critically Critically “Wide Area Protection-Technology and Infrastructures,” IEEE
0.21 0.093 -0.083 Transactions on Power Delivery, vol. 21, no. 2, pp. 601-609, Feb.
stable instable
Critically Critically 2006.
0.22 -0.169 -0.159 [7] DENG Hui, ZHAO Jinquan, LIU Yongjun, and WU Xiaochen, “A
instable instable
novel post-fault rotor-angle trajectory prediction method based on
0.25 -0.721 Instable -0.624 Instable
modified gray Verhulst model,” Power System Protection and Control,
vol. 10, no. 9, pp. 18-24, May. 2012.
In TABLEⅡ and Ⅲ, nearby the critical stability point [8] Yujing Wang, and Jilai Yu, “Real Time Transient Stability Prediction
of the system, the results of RLS with forgetting factor of Multi-Machine System Based on Wide Area Measurement,” IEEE
sometimes are false, too. But the judgment of R3LS is Trans on Power Delivery, pp. 1078-1084, 2009.
accurate. Because the loss function of measurement error is [9] Ning Zhou, Daniel J. Trudnowski, John W. Pierre, and William A.
added to the objective function of R3LS algorithm, which Mittelstadt, “Electromechanical Mode Online Estimation Using
Regularized Robust RLS Methods,” IEEE Transactions on Power
can reduce the potential adverse effects of atypical Systems, vol. 23, no. 4, pp. 1670-1681, Nov. 2008.
measurement data. Simulation results show that this method [10] CHU Yunlong, and XIE Huan, “A Method to Fast Predict Power
has better robustness. System Transient Stability Based on Concave-Convex of Phase
Diagram and System Identification Theory,” Power System
V. SUMMARY Technology, vol. 32, no.5, pp. 59-65, Mar. 2008.
A new method was constructed by using the WAMS [11] WANG Xiaoming, LIU Dichen, WU Jun, HUANG Yong, YUN Lei,
et al, “Energy Function-Based Power System Transient Stability
measurement data, which could analysis power system Analysis,” Power System Technology, vol. 35, no. 8, pp. 114-118,
transient stability accurately and quickly after clearing fault Aug. 2011.

340