2019 Prognostics & System Health Management Conference—Qingdao (PHM-2019 Qingdao)

A Review of Power System Transient Stability

Analysis and Assessment
Huimin Wang Zhaojun Li
School of mechanical and electrical engineering Department of Industrial Engineering and Engineering
University of Electronic Science and Technology of China Management
Chengdu, China Western New England University,
[email protected] Springfield, MA, 01106
[email protected]

Abstract—The failure of power system transient stability is one of criterion, which is proposed by constructing a function directly
the main factors causing catastrophic accidents of power systems. in order to quantitatively measure the power system transient
Therefore, it is of great significance to evaluate the transient stability [5]. Traditional transient stability calculation is carried
stability of a power system. This paper first introduces the out under the condition that the topology, parameters, operation
evaluation methods of power system transient stability, including conditions and disturbance modes of power system are given.
the assessment methods based on time domain simulation, direct The structure of power system network model is presented in
method, artificial intelligence-based methods and the Figure 1.
probabilistic assessment method. The key challenges in power
system transient stability assessment are reviewed and analyzed, Exi E yi
xi
including the stability evaluation of power-electronized power δi Coordinate
system and the main elements of artificial intelligence method transformation
I xi I yi Other synchronizer
equation
used in transient stability assessment. Last, the future research Vti
I di I qi Edi Eqi
directions and conclusions are discussed. Electric
Electromagnetic loop E fi Excitation system power
equation of Load equation
equation network
Keywords- transient stability assessment; power system; power- synchronizer equation
electronized power system; probability assessment; artificial
Pmi
intelligence Rotor motion equation
Equation of prime Other dynamic
mover and governor devices
ωi
I. INTRODUCTION
Figure 1. The model structure of power system whole network
Power system transient stability is that the ability of
generators to continue to operate synchronously after the In Figure 1, Idi and Iqi are rotor current of direct-axis and
system is disturbed [1]. The causes of failure of power system quadrature-axis respectively; Edi and Eqi are rotor voltage of
transient stability include short-circuit fault, sudden direct-axis and quadrature-axis; ¥i is power angle; Exi and Eyi
disconnection of lines or generator, etc. Accurate and fast are stator voltage of direct-axis and quadrature-axis; Ixi and Iyi
transient stability assessment method is important to the are stator current of direct-axis and quadrature-axis; Efi is
security operation of power system. With the gradual excitation voltage; Vti is node voltage; Pmi is input power of
advancement of smart grid construction, long-distance, huge prime mover; wi is angular speed of prime mover.
capacity transmission mode and high-proportion power
electronics, the new risks are introduced in power system [2]. However, previous works focused on the application of
Power shortage accidents and complex cascading failures different methods to establish Lyapunov functions for power
further rise the difficulty of power system stability analysis and systems. With the development of Lyapunov functions, it is
control. recognized that another key problem is to accurately estimate
the stability region after failure of power systems [6]. The
The direct method, time domain simulation method and calculations of energy function of flexible AC transmission
artificial intelligence (AI) method are commonly used for system (FACTS) devices and revised transient energy function
transient stability analysis of traditional power system [3]. of FACTS are formulated in [7].
The time domain simulation method is to solve the With the advantage of wide-area measurement system
differential equations and algebraic equations, which describe (WAMS) technology, the artificial intelligence prediction
the transient process of the system by various numerical method based on WAMS can use real-time measurement data
integration methods. Then the stability is judged according to to train the transient stability classifier online, instead of using
the change of the relative angle between the rotor of the off-line model to simulate various disturbances to obtain data
generator. In each step interval, it is approximated that the rotor [8]. Artificial intelligence generates databases as input of
is in constant acceleration motion [4]. established networks through a large number of off-line
The direct method is mainly based on Lyapunov stability simulations, and uses intelligent algorithms to construct stable
classifiers. Then the stability of the system is evaluated by
training stable classifiers [9]. The method of Artificial

This research is partially supported by the National Science Foundation of

China, No. 71771038.
2019 Prognostics & System Health Management Conference—Qingdao
978-1-7281-0861-2/19/$31.00 ©2019 IEEE (PHM-2019 Qingdao)
978-1-7281-0861-2/19/$31.00 ©2019 IEEE
intelligence is used to develop a load disaggregation approach where ¥ is angle of power energy; w is angular frequency; wN
for bulk supply points based on the substation rms is rated angular frequency; TJ is electromagnetic torque; PT is
measurement in [10]. The optimization problem is formulated, mechanical power and Pe is electromagnetic power.
and the Cuckoo search algorithm is adopted for optimal
designing of power system stabilizer in [11]. The analysis B. Direct Method for Transient Stability
about artificial intelligence optimum plans and improving the
functioning of the power systems economically are made in The direct method is mainly based on Lyapunov stability
[12]. Compared with the time domain method, the artificial criterion. The direct method contains potential energy boundary
intelligence method does not need to establish the mathematical surface (PEBS), relevant unstable equilibrium point (RUEP)
model of the power system. Artificial intelligence method uses and extended equal area criterion (EEAC) [13]. The EEAC
the measured response information to extract the characteristics method refers to that the regulation of excitation can also
that can reflect the physical nature of the system transient promote the transient stability of power systems [14]. A multi-
stability. Then the transient stability assessment is carried out objective optimization method is proposed, which is to model
by establishing the mapping relationship between the transient stability as an objective function rather than an
characteristics and the system stability. inequality constraint in [15]. Only the excitation system with
fast rise of excitation voltage and high peak voltage can have a
Usually, artificial intelligence takes an offline approach to significant effect on improving transient stability. The reason is
obtain feature samples which can accurately characterize the that fast excitation reduces the acceleration area, increases the
inherent mechanism of power system operation under deceleration area and then improves the transient stability of
anticipated accidents. The acquired samples are trained and the system. Good excitation control plays a more important role
learnt repeatedly, and the method of predicting the transient in increasing artificial damping and eliminating the second
stability of power system is constructing the classifier. pendulum or multi-pendulum out-of-step [16]. Basic steps of
Meanwhile, the feature samples are input into the classifier in direct method are shown in Figure 2.
real time. The comparison of the three methods for evaluating
the power system transient stability is shown in Table I. Transient stability analysis
at device or system level

TABLE I. COMPARISONS OF THREE METHODS

Method Information Rapidity Application
Time domain Time response Computation is Main method in industry; Modeling
simulation of system state heavy and time- standards for testing
method consuming other methods
Direct method Estimates of Fast calculation Intensified time domain
attraction speed solution combined with Dynamic polymerization
domain simulation method based
on time domain for on-
line transient stability Reduction of order according to time
analysis scale
Artificial Stability Fast calculation Pre-processing and post-
intelligence indicators speed processing of data
method

What is the form of the simplified system?

II. BASIC ANALYSIS STEPS OF DIFFERENT METHODS
APPLIED TO TRANSIENT STABILITY ANALYSIS

A. Time Domain Simulation Method for Transient Stability Constructing Lyapunov functions
From the previous discussion, the calculation of traditional
transient stability is carried out under the condition that the
topology, parameters, operation conditions and disturbance Estimating attraction domain
modes of power system are given. The time domain simulation
method is to solve the differential equations and algebraic
Figure 2. Basic steps of direct method
equations, which describe the transient process of the system
by various numerical integration methods. The equations are as
follows: C. Probabilistic Assessment for Transient Stability
dδ Some parameters of power system are random due to errors
= ω − ωN = Δω = fδ (δ , Δω ) (1) in measurement, estimation or calculation [17]. The operating
dt conditions and random disturbances are ever-changing, and
deterministic analysis does not consider the possibility of
d Δω ω N
= ( PT − Pe ) = fω (δ , Δω ) (2) various accidents. Transient stability probability analysis is
dt TJ different from deterministic analysis [18]. Different from
deterministic analysis, the transient stability probability
analysis determines the probability indicators according to the

2019 Prognostics & System Health Management Conference—Qingdao

(PHM-2019 Qingdao)
statistical characteristics of the main stochastic factors affecting Network topology Network topology

power system transient stability.

Considering the power system, tcr and tcl are assumed to be Load distribution Load distribution (random)

the critical and actual clearing times of the fault respectively.

Fault location Fault location (random)
The principle of calculating the probability indicators of
transient instability is shown in Figure 3 [19,20].
Fault type Fault type (random)

Fault-clearing time Fault-clearing time

(random)

Fault simulation
Fault simulation

Calculating system limit

Is the sample size
parameters
sufficient?

Calculating the probability

indicators of transient stability

Figure 4. Comparison of flow charts of deterministic and probabilistic power

Figure 3. The principle of calculating the probability indicators of transient system transient stability analysis
instability
But the calculation of probabilistic transient stability
If and only if tcl>tcr, the system will be transient unstable. analysis is usually very large. Probabilistic analysis of transient
The fault clearing time tcl is a random variable, and the system stability of power system involves the probability of the system
transient stability assessment can be expressed by the state, which depends not only on the location and type of the
probability of system instability when the fault occurs. fault, but also on the relay protection settings, and the system
+∞ state before the fault. The probabilistic transient stability
Prl ( I ) = Pr ( tcl > tcr ) = ³ f ( tcl ) dtcl (3) analysis procedures are shown in Figure 5.
tcr

In the formula, I refer to the event leading to transient Usually, disturbance accident simulation is that using
instability of the system and the probability density function of probability model to simulate disturbance accident, disturbance
the fault clearing time is tcl. If tcr and probability density includes location, type and other information.
function of the fault are known, it is easy to obtain the Determining network topology and
generation mode
probability indicators of transient instability of the system
under this fault condition.
Monte Carl simulated load level
In summary, the method of probabilistic transient stability
analysis of power system is divided into analytical method and
Monte Carlo method [21]. In order to assessing the probability Random analog circuit, fault type and
fault clearing time
of power system stability, the conditional probability theory is
used. The influence of probability distribution of random
factors is mainly considered. Comparison of flow charts of Transient stability simulation

deterministic and probabilistic of transient stability analysis of

power system is shown in Figure 4.
Is it stable?
The analytical method uses conditional probability theory
in statistics to evaluate the stability probability of the system.
The probability indicators of transient stability are determined Revision of evaluation indicators
according to the statistical characteristics of the factors that will
affect the power system transient stability. Probabilistic
transient stability analysis makes up for the limitation of Is the sample sufficient?
deterministic method in transient stability analysis, which is an
important breakthrough and supplement to deterministic
method. Revision of evaluation indicators

Figure 5. Probabilistic transient stability analysis procedures

In the process of Monte Carlo simulation, some uncertain

parameter models can be obtained from historical data or

2019 Prognostics & System Health Management Conference—Qingdao

(PHM-2019 Qingdao)
assumed to be a probability distribution function. The electronic converter in power supply side, transmission
probabilistic transient stability assessment includes state network and load side of power system is getting higher and
sampling, transient stability simulation and transient instability higher, and the power level of power electronic converter is
indicators calculation. Probability assessment of transient also rising. Since the 1950s, power semiconductor technology
stability of power system based on Monte Carlo method is has made great progress. Converters are formed to connect to
shown in Figure 6. power systems as voltage or current sources [23]. The power-
Probabilistic model Probability model System stability
electronized power system is presented in Figure 8.
of perturbation of protection action analysis
Compared with traditional AC power system, the
characteristic of power-electronized power system is that the
Disturbance accident Perturbation time series Disturbance dynamic
simulation simulation simulation system topology changes with the switching action of power
electronic devices. The whole system is time-varying (non-
No autonomous), and there is interaction of multi-time scale
End of simulation?
control. In addition, the power electronic converter itself has
Yes structural nonlinearity and complexity, and there are over-
Calculating the modulation and limiting phenomena when using pulse width
probability of transient
stability
modulation. So that the power electronics dominated power
system has saturation nonlinearity. These characteristics bring
Figure 6. The assessment of transient stability of power system based on difficulties to power system stability analysis [24]. The stability
Monte Carlo method of power electronic dominated power system was determined
by time domain method, impedance analysis method and
D. Artificial Intelligence Method generalized short circuit ratio previously. At present, the small
signal stability analysis of power-electronized power system
Artificial intelligence method is applied to on-line transient has achieved preliminary results. Small signal stability can
stability analysis. Input feature selection and evaluation model ensure the asymptotic stability of the equilibrium point, which
are the key points in the research of assessment of transient is a necessary step in device design. However, considering
stability of power system based on artificial intelligence. small signal stability alone, the boundary of the stability region
In the construction of assessment of transient stability can not to be determined and the stability margin of the
strategy based on artificial intelligence, the stable discriminant equilibrium point can not to be judged. Therefore, it is needed
input characteristic is composed of some combined operational to analyze the transient stability of power electronics
variables reflecting the dynamic response during system dominated power system.
failure. Subsequently, artificial intelligent technologies are
applied to establish the relationship of the input characteristics B. Artificial Intelligence Method for Transient Stability
and the stable state of transient stability characteristics. In the Assessment of Power Electronics Dominated Power
process of modeling, choosing appropriate input features is the System
key to design, and the electrical value should be converted in Time domain simulation method is considered to be the
center of inertial (COI) frame. A large number of studies have most mature and reliable method, which is also applicable to
applied feature transformation algorithms such as injection power electronic power system. However, the simulation
principal component analysis to reduce the input feature method has the disadvantages of large amount of calculation,
dimension and improve learning efficiency [22]. The basic long simulation time and impossible to simulate all operation
steps of obtaining input characteristic variables are shown in states.
Figure 7.
Another main method of transient stability analysis which
can be used in power-electronized power systems is that direct
method based on modern differential dynamic system. The
direct method determines the transient stability of the system
by comparing the transient energy with the critical energy of
the power system at the time of fault clearing.
Figure 7. The basic steps of obtaining input characteristic variables However, conventional algorithms of stability analysis and
the stability control systems is unable to evaluate the power
system operation under smaller and smaller stability margin.
III. TRANSIENT STABILITY ASSESSMENT OF POWER Thus, the AI method are complementary to the traditional
ELECTRONIC DOMINATED POWER SYSTEM transient stability analysis method. Artificial intelligence
method is applied to on-line transient stability analysis.
A. Power Electronics Dominated Power System
With the rapid development of new semiconductor
materials and control technology, the penetration of power

2019 Prognostics & System Health Management Conference—Qingdao

(PHM-2019 Qingdao)
 DC load and power supply
Power generation side after power electronization

Energy storage station

AC load

Wind farm

Distributed power supply

Photovoltaic station

Power transmission side after

power electronization Dynamic compensation device
Hydropower station

Load
Nuclear power plant

Electric vehicle

Thermal power station

Load after power electronization

Figure 8. Power electronics dominated power system

Artificial intelligence method is mainly used in

combination with wide area monitoring system, which IV. DISCUSSION ON FUTURE TRENDS
undertakes the functions of pretreatment and post-processing. The artificial intelligence method can realize the
Transient stability assessment is a mapping from characteristic identification and decision-making of fast determination of
information to stability category. Its basis is to determine the
classification step of decision attribute. Eight attribute variables Power system
are listed in the Table II. The class variables are stable and
unstable, which are expressed by 1 and 0 respectively.

TABLE II. COMPOSITION OF ATTRIBUTE VALUE OF INPUT VECTOR

Constructing input
Construction of input characteristic based on
characteristic based on
Number Attribute variable training data set
power system history data

X1 Maximum initial acceleration of generator

X2 Maximum rotor kinetic energy of generator during fault clearing
X3 Initial rotor angle of generator with maximum acceleration Feature extraction to Power system transient
X4 When the fault is cleared, the angle of the generator is given from higher-order stability assessment
features model
X5 When the fault is cleared, the kinetic energy of the generator is
obtained
X6 Energy regulation of the system
X7 Minimum initial acceleration of generator Training to get a
Judgment of power
classifier for transient
X8 Mean square error of initial acceleration for all generators stability assessment
system transient stability

Off-line training On-line application

There are many characteristics reflecting transient stability.

One of the purposes of classifier design is to select Figure 9. The assessment of transient stability of power system
representative conditional attributes variables, in order to
provide as much information as possible by using conditional power system transient stability, which is based on the
attributes. Some features related to transient stability can be measured data rather than the power system model and
extracted. Machine learning algorithm is applied to assessment parameters. However, this method does not take into account
of transient stability, the key problems including initial feature the physical mechanism of the dynamic response of the power
input, feature selection, model learning and training. The system, and requires precise learning samples when
difficulty lies in obtaining the characteristic sample set which calculating. However, it is difficult to provide a large number
can represent the physical essential characteristics and of valid samples matching the actual operation through off-line
designing the classifier with the lowest error rate for transient simulation. Therefore, artificial intelligence method applied to
stability assessment. The transient stability assessment process power system transient stability has the difficulties including
of power electronics dominated power system is shown in deep mining of massive WAMS data, off-line simulation and
Figure 9. errors in the use of measured data.

2019 Prognostics & System Health Management Conference—Qingdao

(PHM-2019 Qingdao)
 In addition to the traditional power system analysis method, corrected transient energy function,” IET Generation, Transmission &
a new analysis method of power system transient stability Distribution, vol. 10, pp. 470-476, February 2016.
should be proposed according to the characteristics of power [8] S. M. Ashraf, A. Gupta, D. K. Choudhary, et al. “Voltage stability
monitoring of power systems using reduced network and artificial neural
electronics dominated power system. The analysis method network,” International Journal of Electrical Power & Energy Systems,
combined with bifurcation theory can be studied. vol. 87, pp. 43-51, 2017.
In the aspect of data acquisition, simulation method is used [9] K. W. Kow, Y. W. Wong, R. K. Rajkumar, et al. “A review on
performance of artificial intelligence and conventional method in
to obtain transient stability analysis data instead of relying on mitigating PV grid-tied related power quality events,” Renewable and
actual fault data. The main reason is that the probability of Sustainable Energy Reviews, vol. 56, pp. 334-346, 2016.
actual power system failure, especially transient instability, is [10] Y. Xu, J. V. Milanoviü, “Artificial-intelligence-based methodology for
low. Because the power system vary with time, the load disaggregation at bulk supply point,” IEEE Trans. on Power
applicability of historical data decreases and it is difficult to Systems, vol. 30, pp. 795-803, February 2015.
provide high-quality training data for artificial intelligence [11] S. M. A. Elazim, E. S. Ali, “Optimal power system stabilizers design via
algorithm. From this point of view, how to improve the cuckoo search algorithm,” International Journal of Electrical Power &
Energy Systems, vol. 75, pp. 99-107, 2016.
consistency between simulation data and actual fault data is an
urgent challenge to be solved. [12] S. M. Zahraee, M. K. Assadi, R. Saidur, “Application of artificial
intelligence methods for hybrid energy system optimization,” Renewable
and Sustainable Energy Reviews, vol. 66, pp. 617-630, 2016.
V. CONCLUSION [13] S. C. Savulescu, Real-time stability in power systems: techniques for
early detection of the risk of blackout, New York: Springer, 2006. pp.
This paper presents an overview of the power system 147-165.
transient stability assessment methods. A comprehensive [14] B. Wang, B. Fang, Y. Wang, et al. “Power system transient stability
analysis and comparison of the deterministic assessment and assessment based on big data and the core vector machine,” IEEE Trans.
probabilistic assessment is presented. Compared with on Smart Grid, vol. 7, pp. 2561-2570, May 2016.
traditional AC power system, the characteristics of power [15] C. J. Ye, M. X. Huang, “Multi-objective optimal power flow considering
electronics dominated power system have changed transient stability based on parallel NSGA-II,” IEEE Trans. on Power
dramatically, but it is essentially still a time-varying and non- Systems, vol. 30, pp. 857-866, February 2015.
linear complex system. The characteristics of power- [16] Y. Zhou, H. Huang, Z. Xu, et al. “Wide-area measurement system-based
transient excitation boosting control to improve power system transient
electronized power system are analyzed, and the artificial stability,” IET Generation, Transmission & Distribution, vol. 9, pp. 845-
intelligence methods of transient stability for power- 854, September 2015.
electronized power system have been presented. In addition, [17] M. Panteli, D. N. Trakas, P. Mancarella, et al. “Boosting the power grid
the AI methods which have been used to analyze power system resilience to extreme weather events using defensive islanding,” IEEE
transient stability are reviewed, and data acquisition, feature Trans on Smart Grid, vol. 7, pp. 2913-2922, June 2016.
extraction and algorithm application are discussed. [18] C. Peng, J. Zhang, “Delay-distribution-dependent load frequency control
of power systems with probabilistic interval delays,” IEEE Trans. on
Power Systems, vol. 31, pp. 3309-3317, April 2016.
REFERENCES [19] M. Vatani, T. Amraee, A. M. Ranjbar, et al. “Relay logic for islanding
[1] P. Kundur, N. J. Balu, and M. G. Lauby, Power system stability and detection in active distribution systems,” IET Generation, Transmission
control, New York: McGraw-hill, 1994. pp.30-46. & Distribution, vol. 9, pp. 1254-1263, December 2015.
[2] F. Li, W. Qiao, H. Sun, et al. “Smart transmission grid: Vision and [20] T. Gonen, Electrical power transmission system engineering: analysis
framework,” IEEE Trans. on Smart Grid, vol. 1, pp. 168-177, February and design, CRC press, 2015, pp. 300-328.
2010. [21] S. Liu, P. X. Liu, X. Wang, “Stochastic small-signal stability analysis of
[3] M. Pavella, “Power system transient stability assessment-traditional vs grid-connected photovoltaic systems,” IEEE Trans. on Industrial
modern methods,” Control Engineering Practice, pp. 1233-1246, Electronics, vol. 63, pp. 1027-1038, February 2016.
October 1998. [22] K. P. Wong, "Artificial intelligence and neural network applications in
[4] M. L. Crow, Computational methods for electric power systems, Crc power systems," International Conference on Advances in Power System
Press, 2015. Control, pp. 37-46, 1993.
[5] J. Geeganage, U. D. Annakkage, T. Weekes, et al. “Application of [23] Q. C. Zhong, “Power-electronics-enabled autonomous power systems:
energy-based power system features for dynamic security assessment,” Architecture and technical routes,” IEEE Trans. on Industrial
IEEE Trans. on Power Systems, vol. 30, pp. 1957-1965, April 2015. Electronics, vol. 64, pp. 5907-5918, July 2017.
[6] T. L. Vu, K. Turitsyn, “Lyapunov functions family approach to transient [24] Y. Yang, H. Wang, A. Sangwongwanich, et al. Design for reliability of
stability assessment,” IEEE Trans. on Power Systems, vol. 31, pp. 1269- power electronic systems, Power Electronics Handbook. Butterworth-
1277, February 2016. Heinemann, 2018, pp. 1423-1440.
[7] J. Aghaei, M. Zarei, M. Asban, et al. “Determining potential stability
enhancements of flexible AC transmission system devices using

2019 Prognostics & System Health Management Conference—Qingdao

(PHM-2019 Qingdao)