Arti cial Intelligence Techniques for Dynamic

Security Assessments – A Survey

Miguel Cuevas
University of Chile
Ricardo Alvarez-Malebrán

Universidad Técnica Federico Santa María

Claudia Rahmann
University of Chile
Diego Ortiz
Universidad de las Fuerzas Armadas ESPE
José Peña
Universidad Técnica Federico Santa María
Rodigo Rozas-Valderrama
Universidad Técnica Federico Santa María

Research Article

Keywords: Dynamic security assessments, Power system stability, Arti cial intelligence, Machine
learning, Deep learning

Posted Date: January 31st, 2024

DOI: https://doi.org/10.21203/rs.3.rs-3903334/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.

Version of Record: A version of this preprint was published at Arti cial Intelligence Review on October
21st, 2024. See the published version at https://doi.org/10.1007/s10462-024-10993-y.
 Artificial Intelligence Techniques for Dynamic
Security Assessments – A Survey

Miguel Cuevas1 , Ricardo Álvarez-Malebrán2*, Claudia Rahmann1 ,

Diego Ortiz3 , José Peña2 , Rodigo Rozas-Valderrama2
1 Department of Electrical Engineering, University of Chile, Santiago,
8320000, Chile.
2* Department of Electrical Engineering, Universidad Técnica Federico

Santa Marı́a, Santiago, 8940000, Chile.

3 Department of Electrical Engineering, Universidad de las Fuerzas

Armadas ESPE, Sangolqui, 171103, Ecuador.

*Corresponding author(s). E-mail(s): [email protected];

Contributing authors: [email protected];
[email protected];

Acknowledgments
This work was supported in part by the Chilean Council of Scientific and Tech-
nological Research, ANID/FONDECYT/1201676, ANID/FONDECYT 1231739,
ANID/FONDEF IT19I0129 and ANID/FONDAP/1522A0006.

Abstract
The rise of converter interface generation (CIGs) technologies with fast response
times has introduced new challenges for the stability of electrical power sys-
tems. Traditional dynamic security assessment (DSA) methods of power systems
are based on dynamic simulations within a time domain. Because of the high
computational efforts and human resources required to carry out these kinds
of simulations, they are performed for a limited number of adverse (critical)
operational conditions in terms of their stability. However, further incorporation
of CIGs in power systems not only results in a shift of the critical operating
conditions that may threaten the system’s stability, but also the number of crit-
ical conditions increases. Consequently, there is an urgent need to develop new
methods for evaluating system stability in a wide range of operating conditions
with a reasonable amount of human and computational effort. A complementary

1
 strategy to the traditional methods for DSA is to use artificial intelligence (AI)
techniques. Unlike traditional methods, AI techniques use a data set that cap-
tures the non-linear relationships between the system’s operational conditions
and their stability, without needing to solve the algebraic-differential equations
modeling the power system. Once these relationships are established, the system’s
stability can be evaluated for other operational conditions much faster (within
the order of the hundreds of milliseconds) and accurately. This, in turn, allows for
the consideration of an extended range of operational conditions for DSA stud-
ies. This article surveys the state of the art regarding the use of AI techniques
for DSA in electrical power systems. Focus is made on algorithms, the kind of
stability being addressed, data processing and applications. Finally, limitations,
challenges, and future trends are discussed.

Keywords: Dynamic security assessments, Power system stability, Artificial

intelligence, Machine learning, Deep learning

1 Introduction
1.1 Motivation
Electric power systems worldwide are going through a significant energy transition in
order to combat climate change. This energy transition is characterized by a massive
integration of non-conventional renewable energies and, in some cases, the decom-
missioning of coal-fired power plants. Under this global drive, converter-interfaced
generation (CIGs) technologies, such as photovoltaic and wind power, appear as key
pillars in future power systems. In fact, from 2012 to 2021, the total installed capacity
of wind power increased from 270 GW to 825 GW, while the total installed capacity
of photovoltaic power increased from 102 GW to 843 GW (IRENA, 2022). Despite
its benefits, increasing the share of CIG brings several challenges to power systems,
especially from a stability viewpoint. Most of these challenges arise due to some fun-
damental differences between conventional synchronous generators (SG) and CIGs,
including limited short-circuit current capability (Jia, Yang, & Nielsen, 2018), inertia-
less characteristics and fast responding times of CIGs (Fang, Li, Tang, & Blaabjerg,
2019). Accordingly, future power systems with high shares of CIG will face more crit-
ical situations than those dominated by SG. Consequently, it will be crucial to apply
dynamic security assessment (DSA) to future electrical systems to ensure a secure
energy transition.
Historically, DSAs of conventional power systems has been performed using
detailed time domain simulations (TDS) (F. Luo et al., 2015). For realistic-size
power systems, TDSs require solving hundreds of thousands of non-linear differential-
algebraic equations, involving high human and computational efforts (Dong, Xu,
Zhang, & Wong, 2013). Therefore, traditional DSAs are only performed on some of the
most critical scenarios, which are selected based on the system’s historical performance
and operator experience (R. Liu, Verbič, Ma, & Hill, 2018). However, the integration of
CIG increases the uncertainty and variability of the generation feed-in, which results

2
in an increase and shift of the critical operating conditions. Hence, the traditional
worst-case approach for DSA, based on a limited number of critical operating condi-
tions may fail to identify the most critical situations in which the system’s stability
may be threatened (Rahmann, Ortiz-Villalba, Álvarez, & Salles, 2017).Furthermore,
the focus of traditional DSA has been on the time scale of the electromechanical tran-
sients, neglecting fast phenomena in the network and other fast-response devices are
neglected (Hatziargyriou et al., 2021). However, as the number of converter-based
devices increases, fast transients start to dominate the system’s dynamic response,
and neglecting them may lead to erroneous conclusions. In summary, the current DSA
practices may not be valid much longer as the share of CIGs continues to increase.
One strategy to evaluate power system stability that has received increasing atten-
tion in the last years is the use of AI techniques. Unlike traditional analytical methods,
AI techniques can capture non-linear relationships between power system operating
states and stability utilizing predefined datasets. Once the AI establishes said rela-
tionships, it can evaluate different operating conditions and contingencies without
having to solve complex sets of non-linear differential-algebraic equations and, thus,
significantly lowering the computational efforts. These characteristics enable intelligent
systems (IS) to produce results with high degrees of accuracy within a few hundred
milliseconds to a few seconds (Dong et al., 2013), noticeably reducing the computa-
tional time required in comparison to analytical methods. DSAs using AI have been
developed in both offline and real-time applications. While offline applications focus
on maximizing accuracy (Lin, 2013), online applications aim to perform quickly and
make accurate assessments so that the operator can take preventive/remedial actions
(Sun, Likhate, Vittal, Kolluri, & Mandal, 2008). DSAs using AI for online applications
have recently seen a significant increase due to the massification of synchrophasor mea-
surement units (PMUS), which allow fast, accurate and synchronized data collection
(H. Yang, Zhang, Chen, & Wang, 2018).

1.2 Objetive and main contributions

This work presents a review of the use of AI-based techniques for DSA in power
systems. The main contributions are the following:
• A comprehensive review of works that have use AI, considering the revisited and
extended classification of stability of the IEEE PSDPC Task Force (Hatziargyriou
et al., 2021).
• A review of the AI techniques that have been used for DSA, including works that
have compared different techniques.
• The identification of the main features that have been used as input data for the
AI models for each type of stability being addressed.
• The identification of gaps in the research and
• Future research opportunities regarding AI techniques, type of stability and
challenges for practical applications.
Even though a review of works that use AI techniques for power system security
and stability assessments was already presented in (Alimi, Ouahada, & Abu-Mahfouz,

3
2020), this review focuses more specifically on power quality disturbance, SCADA net-
work vulnerabilities and threats, transient stability, and voltage stability. However, this
work does not review studies focusing on frequency stability, resonance stability and
converter-driven stability. The remainder of this paper includes the following sections:
Section 2 presents an overview of intelligent systems for power system stability assess-
ments. Section 3 reviews technical issues related to the AI-based models used for DSA,
such as the learning algorithm being explored, the generation and treatment of input
data and other advanced topics. Section 4 reviews the practical applications for which
the IS was developed. Finally, Section 5 presents the conclusions and summarizes the
research gaps.

2 Intelligent-based stability assessment systems -

overview
For this review, we characterized an IS for evaluating power system stability according
to key definitions that must be made in two mains aspects: the technical implementa-
tion and its practical application in actual power systems (see Figure 1). On the one
hand, technical implementations refer to learning algorithms, input data generation
and processing, and other advanced issues, such as possible strategies for updating the
models when new input data becomes available and dealing with noise, incorrect data,
or missing variables. On the other hand, practical implementations refer to the pur-
pose of the IS and why it was developed. In this regard, we include topics as the kind
of stability that is being addressed, the type of power system model used for validating
the proposal, and the kind of applications (online or offline stability assessment).

Intelligent-based stability
assessment systems

Technical Practical
implementations applications

Learning algorithm Kind of stability

Generation and
Type of system
treatment of input data

Advanced topics Type of application

Fig. 1 Key issues for IS’s technical implementations and practical applications in stability assess-
ment.

The database of this review comes from journal and conference papers mainly from
IEEE Xplore Digital Library and Elseviers’ ScienceDirect. The search criteria consisted
of works that, in their abstract or at in the title, include one of the following keywords:

4
 Table 1 Number of
publications found

Period Publications
≤ 2005 11
2006 – 2011 15
2012 – 2017 64
>2017 75

“intelligent systems,”, “artificial intelligence”, “machine learning”, or “deep learning”,

and which bear the name of the specific AI techniques, together with stability-related
terms such as “voltage stability”, “resonance stability”, “converter-driven stability”,
and “dynamic security assessment”. Table 1 summarizes the total number of works
included in the database. In total, 165 papers were using the search criteria above
were found (after filtering those unrelated to this review).

3 Technical implementation
3.1 Learning algorithm
3.1.1 Types of technique
The learning algorithm is the core of an IS (Dong et al., 2013). Machine Learning
(ML) techniques are generally divided into supervised, unsupervised and reinforcement
learning algorithms (Taeho, 2021). There is also a subcategory called semi-supervised
learning that combines supervised and unsupervised learning (Goodfellow, Bengio, &
Courville, 2016). In total, 145 papers that proposed supervised techniques were found:
eight papers proposed both, supervised and unsupervised techniques, five papers pro-
posed unsupervised learning, three papers with both, reinforcement and supervised
learning, two papers with reinforcement learning and two papers with semi-supervised
learning techniques. Next, we briefly describe these techniques, providing examples in
each case to facilitate understanding. Then, we present the review including the ML
techniques used in each case.
Supervised learning algorithms use labeled data to generate a model that predicts
a targeted attribute. For example, in (Y. Zhang, Xu, Dong, & Zhang, 2019), opera-
tional variables such as wind power and load demand are used as input data to predict
voltage stability. The target attribute is classified as ”stable” or ”unstable” in this
case. Detailed TDS are needed to classify each operating condition as stable/unsta-
ble. Thus, the target attribute is known for each input data and used to train the
models. Then, the trained IS can predict whether the system is stable for new opera-
tional conditions. In contrast, unsupervised learning algorithms seek hidden structures
in the data without using any labels. Clustering is a common unsupervised learning
algorithm used in DSA, which consists of breaking down a data set into groups so
that the elements within one group are similar and different from the elements of the
other groups. For example, in (Rahmann et al., 2017), a clustering technique is used
for grouping operational conditions (8760 operating points) based on operational and
system-wide characteristics such as generator dispatch, ROCOF, and systemic inertia.
The objective is to arrange groups so that the operational conditions within a group

5
have similar dynamic behavior in case of disturbances. Finally, reinforcement learning
is a training method aiming to maximize a reward signal. Unlike supervised learning,
it does not need labeled data or direct instructions. An example of this technique is
used in (Hadidi & Jeyasurya, 2013), where a trained algorithm can determine the opti-
mal limits of power system stabilizers (PSS) to improve the transient stability margin
during three-phase faults. The impact of the angular deviation in the corresponding
busbar is evaluated for each change in the PSS limits. If the action has a positive/neg-
ative impact, it receives a reward/penalty, which allows it to adjust the new control
actions in the future in an iterative manner.
Regarding the ML-based techniques applied to power system stability assessments,
most of the proposals found in the technical literature use supervised learning. One
of the most common techniques is Decision Tree (DT) because of its simplicity, accu-
racy, and interpretability. DTs work by sequentially splitting the data into smaller
groups using an if-then logic. The following works have used DTs as supervised learn-
ing for stability assessments: (Adewole & Tzoneva, 2017; N.G. Baltas, Mazidi, Ma, de
Asis Fernandez, & Rodriguez, 2018; Chang, Lu, & Hsiao, 2005; Diao et al., 2009; Diao,
Vittal, & Logic, 2010; Genc, Diao, Vittal, Kolluri, & Mandal, 2010; Guo & Milanović,
2013, 2014; M. He, Vittal, & Zhang, 2013; M. He, Zhang, & Vittal, 2013; Hosseini,
Olamaei, Gholami, & Jahangir, 2012; Karapidakis, 2007; Krishnan & Thampatty,
2020; Lin, 2013; Y. Luo, Lu, Zhu, & Song, 2023; H. Mohammadi & Dehghani, 2015;
H. Mohammadi, Khademi, Simon, & Dehghani, 2016; M. Mohammadi, Gharehpetian,
& Niknam, 2010; Mukherjee & De, 2020; Naderi, Javadi, Mazhari, & Chung, 2023;
Nandanwar, Kolhe, Warkad, Patidar, & Singh, 2018; Nandanwar & Warkad, 2016;
Nie, Yang, Centeno, & Jones, 2017; Nuqui, Phadke, Schulz, & Bhatt, 2001; Pannell,
Ramachandran, & Snider, 2018; Rahmatian, Chen, Palizban, Moshref, & Dunford,
2017; Ren, Wang, Yu, Xu, & Dong, 2023; Ren, Yuan, Li, Zhang, & Xu, 2023; Rovnyak,
Kretsinger, Thorp, & Brown, 1994; Rovnyak, Taylor, & Sheng, 2000; Senroy, Heydt, &
Vittal, 2006; Sun et al., 2008; Teeuwsen, Erlich, & El-Sharkawi, 2005; Y. Yang et al.,
2017; R. Zhang, Xu, Dong, & Hill, 2012; R. Zhang et al., 2023a, 2023b; Y. Zhang, Xu, &
Dong, 2018; Y. Zhang, Xu, Zhang, & Dong, 2019; Zheng, Malbasa, & Kezunovic, 2013;
Zhu, Lu, Dong, & Hong, 2017). Another commonly used technique is Random Forest
(RF) (Kaci, Kamwa, Dessaint, & Guillon, 2014; Kamwa, Samantaray, & Joos, 2010;
Krishnan & Thampatty, 2020; S. Liu et al., 2020; Malbasa, Zheng, Chen, Popovic,
& Kezunovic, 2017; Ortiz-Villalba, Rahmann, Alvarez, Canizares, & Strunck, 2020;
Suprême, Dessaint, Kamwa, & Heniche-Oussédik, 2018; R. Zhang et al., 2012, 2023a;
Y. Zhang, Xu, Zhang, & Dong, 2019), an algorithm that implements an ensemble of
DTs obtained from random subsamples of the training data. Like RF, Isolation Forest
(iF) (Gao et al., 2023) creates an ensemble of DTs suitable for managing outliers. Sup-
port Vector Machine (SVM) is another popular method used for binary classification
in a supervised fashion: (N.G. Baltas et al., 2018; Bo, Wang, & Liu, 2014; Echeverrı́a,
Cepeda, & Colomé, 2017; Geeganage, Annakkage, Weekes, & Archer, 2015; Gomez,
Rajapakse, Annakkage, & Fernando, 2011; Krishnan & Thampatty, 2020; Malbasa et
al., 2017; H. Mohammadi, Khademi, Dehghani, & Simon, 2018; M.V & C.K., 2014;
Pannell et al., 2018; Pérez-Londoño, Olivar-Tost, & Mora-Florez, 2017; Sajan, Kumar,
& Tyagi, 2015a; Tang, Cui, & Wang, 2017; Tian et al., 2019; Yuanhang, Lei, Weiling,

6
& Yong, 2015; R. Zhang et al., 2012, 2023a, 2023b; Y. Zhang, Dong, Zhang, & Xu,
2018; Z. Zhang, Sun, Deng, Kang, & Chow, 2023; Zhou, Wu, Yu, Ji, & Hao, 2016;
Zhou & Zhang, 2023). SVM finds the hyperplane that separates the data while max-
imizing its distance to any data point. Support vector regression (SVR) (Alizadeh &
Amraee, 2014; Cepeda, Rueda, Colomé, & Echeverrı́a, 2014; Suganyadevi, Babulal,
& Kalyani, 2016), and Least-squares support vector machines (LSSVM) (Maihemuti,
Wang, Wang, Wu, & Zhang, 2021; H. Yang et al., 2018) use the same principle as
SVM, but instead of finding a hyperplane to classify the data, they aim at finding
a hyperplane that fits the data. Hence, both techniques are used to solve regression
problems. Other variations of the SVM are the Aggressive and Conservative support
vector machines (ASVM and CSVM) (Hu et al., 2019), Core vector machines (CVM)
(B. Wang, Fang, Wang, Liu, & Liu, 2016), and Ball Vector Machines (M. Mohammadi
et al., 2010). Other algorithms are K-Nearest Neighbor (KNN) (C. He, Guan, & Mo,
2016; Krishnan & Thampatty, 2020; R. Liu, Verbič, & Ma, 2019; Y. Xu, Guan, Dong,
& Wong, 2010; R. Zhang et al., 2023a), Gaussian Process Regression (Tan & Zhao,
2023; Zhai, Nguyen, & Zong, 2023), Times Series Shapelet Classification (Zhu & Hill,
2022; Zhu et al., 2017), Quantile Regression (Naderi et al., 2023), Multivariate adap-
tive regression splines (Rahmatian et al., 2017), Multi-Label Learning (Ren, Yuan, et
al., 2023), Mahalanobis kernel regression (X. Liu, Zhang, Chen, Xu, & Feng, 2020),
Bayesian multiple kernel learning (Gu & Li, 2013), Naı̈ve Bayes (Pannell et al., 2018),
Logistic regression (Krishnan & Thampatty, 2020; Z. Zhang et al., 2023), Adapta-
tive AR (C. Xu, Liang, Yun, & Zhang, 2005), Artificial Immune Systems (Suliman &
Rahman, 2010), eXtreme Gradient Boosting classifier (XGBoost) (Chen et al., 2019;
Krishnan & Thampatty, 2020; R. Zhang et al., 2023a). Other widely used techniques
for stability studies are those based on Artificial Neural Networks (ANN). This tech-
nique implements layers of artificial neurons (also called nodes). An ANN comprises at
least one input layer and one output layer, with any number (if any) of hidden layers
in between the input and the output. An ANN with no hidden layers (or perceptron)
(Z. Zhang et al., 2023) resembles a linear regression, while adding hidden layers allows
the ANN to learn more complex non-linear behavior from the data. The most common
ANN is the Multiple Layer Perceptron (MPL), also referred to as Feed Forward Neural
Network (FFNN) since its neurons receive information only from the previous layers.
Because of the significant computation burden required for training an ANN, most
MPLs use a single hidden layer ((Alizadeh & Amraee, 2014; Bahbah & Girgis, 2004;
Frimpong, Okyere, & Asumadu, 2017; Jayasankar, Kamaraj, & Vanaja, 2010; Jensen,
El-Sharkawi, & Marks, 2001; Karapidakis, 2007; Lin, 2013; Mahdi & Genc, 2017; Nakas
et al., 2023; Sajan, Kumar, & Tyagi, 2015b; Sajan, Tyagi, & Kumar, 2014; Sharifian &
Sharifian, 2015; Y. Zhang, Li, Na, Li, & Li, 2015)). Extreme Learning Machine (ELM)
is one of the most widely used types of single hidden layer MPL, found in (G.N. Bal-
tas, Perales-González, Mazidi, Fernandez, & Rodrı́guez, 2018; F. Li, Wang, Tang, Xu,
& Dang, 2021; Y. Li & Gu, 2013; Y. Li, Li, & Wang, 2015; Y. Li & Yang, 2017; R. Liu
et al., 2019; Ren, Wang, et al., 2023; Ren & Xu, 2019, 2020; Ren, Xu, Zhang, & Zhang,
2020; Ren, Yuan, et al., 2023; Sulistiawati, Priyadi, Qudsi, Soeprijanto, & Yorino, 2016;
Tang et al., 2017; Q. Wang, Li, Tang, & Xu, 2019; Y. Xu, Dai, Dong, Zhang, & Meng,
2013; Y. Xu, Dong, Meng, Zhang, & Wong, 2011; Y. Xu, Dong, Zhao, Zhang, & Wong,

7
2012; Y. Xu et al., 2016; R. Zhang, Xu, Dong, & Wong, 2015; R. Zhang, Xu, Dong,
Zhang, & Wong, 2013; Y. Zhang, Dong, et al., 2018; Y. Zhang et al., 2015; Y. Zhang,
Xu, & Dong, 2018; Y. Zhang, Xu, Dong, Xu, & Wong, 2017; Y. Zhang, Xu, Dong,
& Zhang, 2019). ELM uses fixed random weights for the input layer, which allow to
obtain the output weights analytically, substantially reducing the computational bur-
den. However, random weights can pose a problem for the robustness of the results in
prediction (R. Zhang et al., 2013). Thus, many works implement ELMs in ensembles.
Closely related to ELMs are Random Vector Functional Link (RVFL) models (Y. Luo
et al., 2023; Ren, Wang, et al., 2023; Ren & Xu, 2019, 2020; Ren et al., 2020; Ren,
Yuan, et al., 2023; Y. Zhang, Xu, Zhang, & Dong, 2019). The only difference with an
ELM is that an RVFL has its output layer connected to the input layer. Another spe-
cific type of ANN is the Radial Basis Function neural network (RBFNN), in which the
response of neurons is given by a radial basis function (Bahbah & Girgis, 2004; Innah
& Hiyama, 2011; Jain, Srivastava, & Singh, 2003; Siddiqui, Verma, Niazi, & Fozdar,
2018; Velayati, Amjady, & Khajevandi, 2015; Y. Xu et al., 2011, 2016). There is a par-
ticular type of ML that many authors refer to as Deep Learning (DL). DL models are
ANNs of multiple layers, able to capture more complex characteristics from the data
than traditional ML algorithms. DL can incorporate any of the three types of learn-
ing previously mentioned. While an ANN with no hidden layers cannot be considered
a DL system, multiple connected layers of artificial neurons are the algorithms’ basic
architecture. An example of DL can be found in (Xie & Sun, 2021), which proposes
a technique to predict the system’s frequency evolution in real-time after the occur-
rence of a disturbance. Using DL enabled it to extract spatial and temporal features
from the input data and combine local features with global ones, which ML algorithms
such as SVM or simpler ANN structures could not. MPL applications with more than
one hidden layer only started to become prevalent in recent years, as seen in (Abbass,
Lis, & Mushtaq, 2023; Ali, El-Amary, Ibrahim, & Mekhamer, 2015; Anderson et al.,
2023; Bahmanyar & Karami, 2014; N.G. Baltas et al., 2018; Gurung, Naetiladdanon,
& Sangswang, 2021; Hashiesh, Mostafa, Khatib, Helal, & Mansour, 2012; Hossain &
Kumar, 2023; Krishnan & Thampatty, 2020; Naderi et al., 2023; Ren & Xu, 2019;
Sawhney & Jeyasurya, 2006; Sepehr, Gomis-Bellmunt, & Pouresmaeil, 2022; Shah &
Verma, 2016; Tan & Zhao, 2023; Tang et al., 2017; Y. Wang & Pal, 2023; R. Zhang,
Wu, Shao, Li, & Lu, 2018; Zhao, Yue, & Wang, 2023a; Zhou & Zhang, 2023). Some
papers refer to ANNs generically without specifying the model’s architecture, proba-
bly referring to single hidden layer MLP (Boudour & Hellal, 2005; Goh et al., 2015;
Malbasa et al., 2017; M.V & C.K., 2014; Velayati et al., 2015; Yousefian, Bhattarai, &
Kamalasadan, 2017; R. Zhang et al., 2012, 2013; Y. Zhang, Dong, et al., 2018; Zhao,
Yue, & Wang, 2023b). Other variations of the classic MLP are Entropy Network (Kara-
pidakis, 2007) (which emulates a DT), Probabilistic Neural Network ((Teeuwsen et
al., 2005; Velayati et al., 2015)) and Fuzzy Neural Network ((C.-W. Liu, Su, Tsay, &
Wang, 1999; M.V & C.K., 2014; Sharifian & Sharifian, 2015; Teeuwsen et al., 2005)).
Finally, the most famous among the advanced ANNs are Convolutional neural net-
work (CNN) (Arteaga, Hancharou, Thams, & Chatzivasileiadis, 2019; Azman, Isbeih,
Moursi, & Elbassioni, 2020; Bashiri Mosavi, Amiri, & Hosseini, 2018; Y. Luo et al.,
2023; Ren, Wang, et al., 2023; Ren & Xu, 2019, 2023; Ren, Yuan, et al., 2023; Tan

8
et al., 2017; C. Wang, Li, Liu, & Li, 2021; Xie & Sun, 2021; R. Zhang et al., 2023a,
2023b; Zhao et al., 2023b), Graph Neural Network (Y. Luo et al., 2023; R. Zhang et
al., 2023b), and Recurrent Neural Network (RNN) (Bahbah & Girgis, 2004; Nakas et
al., 2023; Song et al., 2020). In the latter, unlike FFNN, each neuron is fed by the
input and by its own response in a recursive way. The most famous form of an RNN
is the Long short-term memory (LSTM) model (Azman et al., 2020; Y. Luo et al.,
2023; Nakas et al., 2023; Ren, Wang, et al., 2023; Ren & Xu, 2023; Ren, Yuan, et al.,
2023; Xie & Sun, 2021; Yu, Hill, Lam, Gu, & Li, 2018; Yu, Lam, Hill, & Li, 2017;
R. Zhang et al., 2023a, 2023b; Zhao et al., 2023a, 2023b; Zhu, Hill, & Lu, 2021; Zhu &
Luo, 2021). The last AI technique found in this review is Quantum Machine Learning
(QML) (Zhou & Zhang, 2023) which implements a Variational Quantum Circuit (an
ordered sequence of quantum gates characterized by variable parameters that can be
tuned during the training procedure) that takes advantage of the ability of quantum
machines to perform complex tensor operations and utilize quantum operators (such
us superposition and entanglement) to capture complex non-linear behavior (Zhou &
Zhang, 2023) is the first (and only work so far) that implements QML for DSA. In
particular, they use QML to classify different system states into stable or unstable,
obtaining degrees of accuracy higher than 96%, which is very similar to the accu-
racies found using MLP and SVM. As for unsupervised, reinforcement learning and
semi-supervised techniques, only a few works have used them in the context of power
system stability assessments. In total, we found 13 publications that use unsuper-
vised algorithms. The techniques used are K-means (R. Liu et al., 2018; Naderi et al.,
2023; Pérez-Londoño et al., 2017; Y. Wang, Silva-Saravia, & Pulgar-Painemal, 2019),
Growing hierarchical self-organizing map (GHSOM) (Boudour & Hellal, 2005), Auto
Encoders ((Mahdi & Genc, 2018; Sepehr et al., 2022; Tan et al., 2017)), Generative
adversarial network (GAN) (Ren & Xu, 2019), Complete Linkage Clustering (Ortiz-
Villalba et al., 2020; Rahmann et al., 2017) and Bayesian Neural Network (T. Liu
et al., 2020; Yousefian & Kamalasadan, 2016). Five works that use Reinforcement
Learning (RL) were found (Y. Wang & Pal, 2023; Yousefian et al., 2017; Yousefian &
Kamalasadan, 2016, 2018; Zhao et al., 2023a). Finally, only two works based on semi-
supervised learning were found. The techniques used were Deep Belief Network (DBN)
(R. Zhang et al., 2018) and semi-supervised SVM (S3VM) (Zhu, Hill, & Lu, 2022). It
is worth mentioning that several papers use more than one algorithm for the training
structure or to compare the performance of the algorithms against another one.
Figure 2 summarizes the type of techniques implemented throughout the years.
This figure shows that stability assessments using AI-based techniques have increased
significantly over the last years, mainly through supervised learning techniques. Unsu-
pervised, semi-supervised, and reinforcement learning techniques are scarce and have
been proposed only in recent years.
Figure 3 presents the number of works that have used DL and ML techniques for
DSA over the years. Note that single hidden layer ANN is considered an ML technique,
not a DL one. This figure shows that the number of works that use an ML technique
increased by 13% from 2012-2017 to 2018-2023, while the number of works using a
DL technique for DSA increased by 146% in the same period. These results show
growth in both techniques but highlight a greater interest in DL. The reason might

9
 140 300
120
120

Accumulated number of works

250

100

Number of works
200
83
80
150
60
100
40

20 16 17 50
9
1 23 23
0 0
≤2005 2006-2011 2012-2017 2018-2023
Period
Semi-supervised learning Reinforcement learning
Unsupervised learning Supervised learning
Total accumulated
Fig. 2 Distribution of categories of Machine Learning algorithms in works over periods of years.

be the ability of DL to employ multiple learning categories and capture more complex
features compared to traditional algorithms, together with the rapid development of
computational capacity.

3.1.2 Ensemble algorithms

One strategy to improve prediction capacity is to use ensemble models. An ensemble
combines individual predictors (i.e., single models) to form a single high-quality pre-
dictor (Adewole & Tzoneva, 2017). Individual predictors can use the same technique
and be trained using different training data and parameters, or the ensemble can con-
sist of a mix of different techniques. Ensemble learning algorithms have benefited from
several advantages, such as a decreased risk of over-fitting, avoidance of falling into a
local optimum, and enlargement of the space for possible fitting hypotheses (Tang et
al., 2017). The works that employ ensemble techniques for DSA are: (Abbass et al.,
2023; Adewole & Tzoneva, 2017; An, Yu, Li, Zhou, & Mu, 2020; Cremer & Strbac,
2021; Gu & Li, 2013; Guo & Milanović, 2013; M. He, Vittal, & Zhang, 2013; M. He,
Zhang, & Vittal, 2013; Y. Li & Yang, 2017; X. Liu et al., 2020; Mukherjee & De,
2020; Nie et al., 2017; Niu et al., 2021; Ren et al., 2020; Tang et al., 2017; Y. Xu et
al., 2012, 2016; Y. Yang et al., 2017; Yu et al., 2017; R. Zhang et al., 2013; Y. Zhang,
Xu, Dong, & Zhang, 2019; Y. Zhang, Xu, Zhang, & Dong, 2019; Zhou et al., 2016;
Zhu et al., 2022). It is worth mentioning that in this review, RF and iF techniques
were not considered ensembles. Within an ensemble algorithm, the traditional aggre-
gation strategy for classification problems is majority voting, whereas, for regression
problems, it is direct averaging. As the names suggest, with majority voting, each
classifier predicts a class, and the class with the most votes is selected. For regres-
sion problems, direct averaging consists of selecting the average values delivered by

10
 80
70
70
62 64
60

Number of works
50

40

30 26

20
11
10 8 9 6

0
≤2005 2006-2011 2012-2017 2018-2023
Period

Deep learning Machine Learning

Fig. 3 Comparison of the growth of Deep Learning and traditional Machine Learning in different
periods.

each predictor. Next, we present some example proposals using ensemble models. The
work in (Mukherjee & De, 2020) proposes an ensemble of decision tree predictor mod-
els for real-time DSA of power systems. The proposed method reached a prediction
accuracy of 91.44%, which is much higher than the accuracy obtained by individual
classifiers such as the Method of Least Square (77.3%), NN (78.4%), vector quanti-
zation (80.54%), and SVM (81.7%). Work in (Tang et al., 2017) proposes an online
frequency prediction model using ensemble learning consisting of DT, multivariable
linear regression (MLR), ANN, LSSVM, and ELM as base learners. The strategy
combines individual learners by weighted averaging, using modified cross-entropy to
quantify the contribution of an individual sample in the dataset. The model is tested
on the IEEE 39-bus network using a limited data set (200 training and 108 test data).
As a result, the relative error of the predictions ranged between 17.1% and 17.8% for
individual algorithms, while the error for ensemble learning was 13.6%. These results
show the potential of ensemble learning to obtain better prediction results than indi-
vidual algorithms. In (Adewole & Tzoneva, 2017), the authors develop two ensemble
methods for evaluating voltage stability for classifier and predictor problems: a boost-
ing ensemble, and a bagging ensemble. On the one hand, the boosting ensemble uses a
sequential procedure to train individual predictors. At every boosting step, the weights
of the observations misclassified by a DT increase, while the weights of the correctly
classified observations decrease. Then, the updated weights help train the next learner
to improve the prediction accuracy of misclassified observations. On the other hand,
the bagging ensemble generates bootstrap replicas of the original dataset (i.e., a set
with random samples of the observations) and trains individual DTs on each of these
replicas. To evaluate the performance of the ensemble methods, the authors developed

11
a classifier and a predictor for each operating condition. The classifier’s output was
the system state (stable, alert, or unstable) from a voltage stability viewpoint, and
the output of the predictor was the margin to the voltage collapse point. The results
obtained in a case study based on the New England IEEE 39-bus showed that while
the bagging ensemble method had the best performance for the classifier problem
(above 99.8% accuracy), the boosting ensemble method showed the best performance
for the regression problem (less than 2 MSE in the predicted value of the margin to
the voltage collapse point). In (Y. Zhang, Xu, Dong, & Zhang, 2019), the authors
propose an ELM ensemble model with a hierarchical structure for short-term voltage
stability (STVS) prediction. In the first stage, the trained model classifies the opera-
tional conditions as stable or unstable. Control actions are triggered to stabilize the
system if the prediction model detects an unstable condition. A second-stage model
predicts the severity of fault-induced delayed voltage recovery (FIDVR) if the condi-
tion is stable. Using a self-adaptive aggregation strategy, if the prediction’s accuracy is
unacceptable (“Not sure”), the STVS assessment will be postponed to the next time
point, where more information (measurements) is available. The results obtained using
the New England IEEE 39-bus test system showed that the self-adaptive aggregation
structure optimally balances prediction accuracy with evaluation speed. Specifically,
for 1987 test samples, including 1003 stable and 984 unstable samples, in the first hier-
archy (STVS), 82.39% of the samples were successfully classified within 0.2 s, while to
achieve a 99.09% accuracy, the proposed method takes 0.8 s. Figure 4 shows the num-
ber of works that have proposed single and ensemble models over the years. Sixty-six
papers published from 2018 to 2023 used individual models, while only nine papers
proposed ensemble models during the same period. Note that the use of ensemble
models began in 2012. Even though most works still propose single models, ensem-
ble models are gaining interest. It is important to note that most works that propose
ensemble models also develop single models for comparison purposes.
Choosing the proper algorithm for DSA is a challenging task. In the last few years,
several authors have compared the performance of different techniques. For example,
in (Y. Xu et al., 2011) the authors compare the performance of ELM, DT, SVM,
and RBFNN for transient stability. In (Y. Zhang, Dong, et al., 2018), the authors
compare the performance of ANN, Regression Tree (RT), SVM, and ELM to predict
short-term voltage stability (STVS); in (R. Zhang et al., 2018), the authors compared
the performance of DBN, two ANNs, SVM, DT, and RF, to predict the STVS. In
(Y. Xu et al., 2016), the authors compared the performance of ELM, SVM, DT, and
RBFNN to assess short term voltage stability. These studies show that, while some
techniques outperform others for a specific type of problem, e.g., type of stability, test
system, amount, and type of input data, the same technique may perform worse for
other problems. In this regard, a research gap is to identify which type of problems
and input data techniques perform better than others and why.

12
 80

70 66

60

Number of works
50
50

40

30

20 15 14
11 9
10

0
≤2005 2006-2011 2012-2017 2018-2023
Period

Ensemble model Single model

Fig. 4 Number of publications that have proposed ensemble models and single models

3.2 Generation and treatment of input data

3.2.1 Introduction
One of the basic requirements to implement an IS is to have good-quality data with
a sufficiently large dataset to train and validate the models. If the database is biased,
contains too few operating conditions, or lacks diversity, the extracted knowledge can
be insufficient for an adequate stability classification or prediction, and the model will
perform poorly (Dong et al., 2013). Hence, a key challenge is to generate enough high-
quality input data for training and testing the models. Once the system’s operating
conditions are determined, the usual approach is to evaluate the system’s dynamic
performance using TDS. Then, each operating condition is either labeled (e.g., stable,
unstable) for classification problems or valued by a stability index (e.g., VSM, TSI,
FSM), for regression problems.
Given the high human and computational resources involved, one of the main
barriers to developing IS-based models for stability assessments in realistic-sized power
systems is having to perform numerous TDS. Hence, most proposals are developed
and tested in small test systems (e.g., (Adewole & Tzoneva, 2017; Hashiesh et al.,
2012; R. Liu et al., 2018; M. Mohammadi et al., 2010; Y. Zhang, Xu, Dong, & Zhang,
2019)) or use only a few snapshots of the system operation. In this latter case, a
usual strategy to generate a wide variety of input data is to perform minor changes
in the operating conditions, for example choosing contingencies and clearing times
randomly (see (Adewole & Tzoneva, 2017; Guo & Milanović, 2014; Hashiesh et al.,
2012; Y. Zhang, Xu, Dong, & Zhang, 2019)). Only a few works use realistic-size power
system models (e.g., (C. He et al., 2016; M. He, Vittal, & Zhang, 2013; C. Liu et al.,
2014; Xie & Sun, 2021)). Next, we will present and discuss the literature reviewed

13
involving strategies to generate good-quality input data for training and testing AI-
based models.

3.2.2 Class imbalance issue

The challenge of class imbalance arises when system states are unevenly distributed
within the dataset. For example, if the objective is to predict voltage stability, input
data should have enough cases where the system is both stable and unstable. For real-
world power systems, this is challenging because the number of operating conditions
where the system remains stable after a contingency is significantly larger than the
cases where the power system becomes unstable. Such an imbalance can considerably
deteriorate the prediction’s performance if not appropriately treated, since the models
could be biased toward the majority class, thus overlooking the minority class (Zhu
et al., 2017). Several authors have proposed methods for dealing with the class imbal-
ance problem. One option is to artificially generate operating conditions where the
system is unstable after a fault, thus achieving balanced classes (Nakas et al., 2023;
Sepehr et al., 2022; R. Zhang et al., 2012; Y. Zhang, Xu, Dong, & Zhang, 2019; Zhao
et al., 2023b). HHowever, this strategy may not be suitable for real-world power sys-
tems applications since the input data may contain unrealistic operating conditions.
Another option to deal with the class imbalance problem is to introduce synthetic
samples of the minority class (e.g., unstable cases) from existing data. This approach
involves randomly selecting a sample from the minority class and creating new data
from the same class, for example by interpolation with its close neighbors. For example,
in (Zhu et al., 2017), the authors mitigate the class skewness with a forecasting-based
non-linear synthetic minority oversampling technique (FN-SMOTE). In addition, they
employ cost-sensitive learning to impose more bias on scarce unstable instances. In
this work, they also implemented other methods like ROS, SMOTE, B-SMOTE, and
ADASYN. As a result, they showed that the prediction accuracy improved from 80.6%
(without any data treatment) to 97.8% by using FN-SMOTE and ADASYN to bal-
ance the classes. When using only the SMOTE technique, the accuracy improved to
97.2%. Finally, ROS and B-SMOTE improved the accuracy to 94.4% and 95%, respec-
tively. The authors in (Kamwa, Samantaray, & Joos, 2012) propose another strategy:
replicating as many non-secure cases as needed to balance the unsafe/safe ratio. The
original data set of 60836 operational conditions only included 22.5% unstable cases.
To balance the unsafe/safe ratio, the authors increased the number of unstable oper-
ating conditions by replicating the unstable cases three times, bringing the dataset to
98800 cases. However, this strategy introduces too many conditions overlapping each
other, thus leading to overfitting problems (Zhu et al., 2017).

3.2.3 Amount of input data

To address the challenge of acquiring many TDS to train an IS, a sound strategy is to
develop models that require fewer input data. This strategy can be crucial in reduc-
ing the barriers when developing AI-based stability assessments for real-world power
system applications. Unfortunately, this issue remains unresolved. In this regard, ref-
erence (Malbasa et al., 2017) is the only work that attempts to reduce the number of

14
TDS for training an IS. The authors propose an active learning approach for a multi-
class problem. The training dataset is built iteratively, starting from a single operating
condition for each class. Using TDS, new operating conditions are selected and labelled
in each iteration. Then, the IS is re-calibrated with this newly acquired information.
Three approaches were proposed for choosing the new operating conditions for each
iteration: 1) random selection, 2) uncertainty sampling, and 3) margin sampling. As a
result, they show that the IS can accurately classify operating conditions in terms of
voltage stability with only a few operating conditions. For example, when using only
100 operating conditions, the prediction accuracy reached 85%, whereas —using 1000
operating conditions resulted in a maximum accuracy of around 90%. While the mar-
gin sampling strategy exhibited the highest performance among the three, the random
sampling strategy achieved the worst results. The main drawbacks of this work are:
1) it does not consider contingencies; and 2) in the case study, the classes were well-
balanced, which is unrealistic. However, the good results suggest that choosing the
operating conditions for performing TDS strategically based on a stability criterion
rather than randomly may render benefits for improving the IS accuracy when there
is limited availability of TDS.

3.2.4 Dimensionality reduction

Dimensionality reduction is a data pre-processing step aiming at transforming the
input data into an effective domain, eliminating data redundancy while maintaining
most of the information. Feature extraction and feature selection are two methods that
can achieve dimensionality reductions. While feature extraction transforms the input
data into a reduced set of new features, feature selection finds a subset of the original
variables. Applying a dimensionality reduction technique reduces the training data
size, which results in faster learning and, in many cases, may even lead to better predic-
tion accuracy (Geeganage et al., 2015; Jensen et al., 2001; Y. Li & Yang, 2017; Sawhney
& Jeyasurya, 2006; Y. Xu et al., 2011; C. Zhang, Li, Yu, & Tian, 2016; R. Zhang et
al., 2012). EExamples of techniques used for dimensionality reduction are RELIEF
(R. Liu et al., 2019, 2018; S. Liu et al., 2020; Y. Xu et al., 2012, 2010, 2016; Y. Zhang,
Xu, Dong, & Zhang, 2019), Principal Component Analysis (PCA) (H. Mohammadi et
al., 2016; Sawhney & Jeyasurya, 2006), divergence analysis (Voumvoulakis, Gavoyian-
nis, & Hatziargyriou, 2006), Fischer discrimination (Jain et al., 2003; Jensen et al.,
2001), Gram–Schmidt orthogonalization (Bahmanyar & Karami, 2014), and Wapper
models (T.w. Wang, Guan, & Zhang, 2008). In addition, in (M. Mohammadi et al.,
2010)DTs are also used to select relevant features. In (R. Zhang et al., 2012), the
performance of eight feature selection approaches, which are various combinations of
evaluation criteria and search methods, were tested for a transient stability assessment
problem. Concretely, the techniques explored were RELIEF, Chi-square, k-NN, Fisher,
and Divergence. The search methods were Genetic algorithm, Back-track search, For-
ward, and Select-by-rank. For a case study using the IEEE 145-bus 50-generator test
system, they obtained that, although applying a feature selection reduced the train-
ing time significantly, it only achieved marginal accuracy improvements. Without any
treatment of the input data, the prediction accuracy using ANN, SVM, and DT was
93.0%, 97.1%, and 95.9%, respectively, with training times of 586.6 s, 1369.4 s, and

15
58.7 s. When applying the best feature selection techniques (RELIEF, Select-by-rank,
and Genetic algorithm), the prediction accuracy increased marginally to 94.7%, 97.4%,
and 96.4%. However, the training time decreased to 61.2 s, 75.5 s, and 9.4 s, respec-
tively. In contrast, when applying the worst techniques (Divergence and back-tracking
search), a slight increase in accuracy was observed for ANN (93.7%) with a training
time of 55.8 s. However, for SVM and DT, the accuracy decreased to 95.3% and 95.9%,
respectively, with 72.1 s and 9.8 s training times. The results of this work reinforce that
implementing a dimensionality reduction technique allows for reducing the training
times, but it does not necessarily lead to better performance in prediction accuracy.

3.3 Advanced topics

3.3.1 Model update
Power systems are constantly changing. New components such as lines, substations
and generating units are added to the system every year. In addition, renewable ener-
gies may exhibit different power feed-in patterns throughout the year. This results
in changes in the operating conditions and the system’s dynamics after a contin-
gency. Because of the changing conditions in real-world applications, prediction models
must be updated periodically, which usually requires generating new training samples
(R. Liu et al., 2019). However, incorporating them efficiently to update the models
is an active research area. Among the techniques used for model updating are active
learning (Malbasa et al., 2017), transfer learning (Xie & Sun, 2021; R. Zhang et al.,
2023a, 2023b), and confidence-based models (T. Zhang et al., 2021). AActive learning
consists of creating new training samples or enhancing the existing ones iteratively,
upon demand, instead of collecting and labelling new data in chunks, which is com-
putationally costly. An example of this technique is presented in (Malbasa et al.,
2017) for a voltage stability prediction problem. In this work, the proposed approach
actively searches for operating conditions, in which inaccurate predictions occur. It
then performs dynamic simulations around these identified conditions and adds them
to the existing pool of training datasets. On the other hand, transfer learning consists
of utilizing the characteristics of an already developed model as a starting point for
developing another one. In neural network-based models, transfer learning is usually
implemented by freezing some layers of neurons from the source model (i.e., the model
trained initially) and then re-training a subset of its layers (typically the last ones)
using new data, which is also known as fine-tuning. This way, models can be updated
with new data without having to train the model from scratch, which helps reduce the
amount of data sampling and training time. The same strategy was used in (Xie &
Sun, 2021) for an online frequency stability assessment problem. First, a CNN LSTM
model was trained on an IEEE 118-bus test system using around 2000 samples. Then,
the model was successfully transferred and applied to the New England 39-bus system
and the South Carolina 500-bus system using only 300 samples. In (Zhu & Luo, 2021),
this strategy was implemented for a voltage stability prediction problem to update the
model to adjust it to unexpected topological changes and changes in the load and gen-
eration dispatch. In a case study on the 77-bus Nordic test system, the original model
was trained with 54000 samples and then adjusted to new operating conditions using

16
1500 samples. Lastly, confidence awareness refers to the ability of a machine learning
model to quantify how confident the model is in its prediction upon a given dataset
(T. Zhang et al., 2021). If the likelihood that the output is correct does not satisfy
minimum requirements, then the model is updated. In (T. Zhang et al., 2021), the
authors propose a Conditional Bayesian Deep Auto-Encoder (CBDAC) based secu-
rity assessment framework to compute a confidence metric of the prediction, which
allows the operator to judge whether the model needs updating. Another strategy for
developing an IS that can adjust to changing operating conditions is to identify and
group similar operating conditions, for example, using a clustering technique, and then
train specific models for each group. Once the models are built, it is possible to assess
new operating conditions with the model trained with the data that best matches the
operating condition under evaluation. Such a strategy was implemented in (Y. Xu et
al., 2012) for a real-time dynamic security assessment problem and in (X. Liu et al.,
2020) for a transient stability assessment problem, where the authors propose a simi-
lar strategy that addresses changes in the network topology. Here, the proposed model
is a weighted sum of several models, each trained using data collected from a specific
topology. The topological similarity and the numerical similarity between the samples
determined the weights. Other works that re-train a model for adjusting to new data
are (Diao et al., 2010; M. He, Vittal, & Zhang, 2013; Y. Li & Gu, 2013; Naderi et al.,
2023; Nakas et al., 2023; Y. Xu et al., 2012; Zhai et al., 2023; R. Zhang et al., 2015;
Zhao et al., 2023b; Zhu et al., 2017; Zhu & Luo, 2021).

3.3.2 Dealing with noise and bad data

Many IA-based proposals for DSA are developed to be implemented in real-world
power systems. However, in real-world power systems, the quality of the data collected
by measurement units may not always meet the requirements for developing or using
IA-based models. Bad data can come in many forms and refer to noise, bias, or incon-
sistencies, such as a measurement unit failure. In particular, noise from local, fast, or
random load changes can cause measurements peaks (S. Liu et al., 2020). Hence, an
important research line is how to deal with noise and bad data. Works that have dealt
with these issues are (Gao et al., 2023; R. Liu et al., 2019; Naderi et al., 2023; Nakas
et al., 2023; Ren, Wang, et al., 2023; Ren & Xu, 2023; Sharifian & Sharifian, 2015; Shi
et al., 2020; Y. Wang & Pal, 2023; Yu et al., 2018, 2017; Zhai et al., 2023; R. Zhang
et al., 2023a; Y. Zhang, Dong, et al., 2018; Zhao et al., 2023b; Zhou & Zhang, 2023).
For example, in (Sharifian & Sharifian, 2015), for estimating the critical clearing time
(CCT) of a fault, the authors propose a Type-2 fuzzy neural network-based method
to address the uncertainty and noisy nature of the measured data. The Type-2 fuzzy
layer converts uncertain and noisy inputs into more reliable linguistic variables that
become useful inputs in an MLP ANN layer. As a result, the proposed strategy could
estimate the CCT with higher precision and at a lower computational cost than the
MLP ANN method without treatment of the input data. In (Yu et al., 2018), to study
the influence of noise on PMU samples in a transient stability assessment problem, a
random noisy signal was added to the measured voltages in the data set for training
and testing. The results obtained from a case study on the IEEE 39-bus New Eng-
land system with 5000 samples showed that adding noisy data to the measurements

17
decreased the average response time (ART) from 1,466 cycles to 1,448 cycles and the
accuracy decreased from 100% to 99.92%, in comparison to the test data without
noise. These results indicate that, for real-world applications, noise in the input data
can reduce the model’s prediction accuracy..

3.3.3 Dealing with missing variables

Another important issue for implementing AI-based models for DSA in real-world
power systems is to deal with missing variables. Most AI-based data-driven stability
assessments proposals assume that input data is fully available. However, in practice,
some PMUs may be unavailable due to failure, which may either deteriorate the stabil-
ity assessment’s performance or invalidate it. Studies that consider the unavailability
of some input data are (Adewole & Tzoneva, 2017; Guo & Milanović, 2013; M. He,
Vittal, & Zhang, 2013; Mahdi & Genc, 2018; Ren & Xu, 2019; Y. Xu et al., 2011;
R. Zhang et al., 2018; Y. Zhang, Xu, & Dong, 2018; Y. Zhang, Xu, Zhang, & Dong,
2019). For example, (Ren & Xu, 2019) develops a GAN model that can generate sam-
ples of the missing data by using the Adam algorithm. On the other hand, for real-time
STVS assessments the authors in (Y. Zhang, Xu, Zhang, & Dong, 2019) designed an
ensemble learning method to adjust its structure so it uses only the available inputs.
Under scenarios of a PMU loss or topology changes, the proposal strategically selects
the observable buses in the system to maintain a high level of network observability.
As seen, even though there are methods proposed to deal with missing variables, they
are still scarce, which is a gap that needs solutions for real-world applications.

4 Practical applications
This section reviews the works in terms of the practical application for which the
AI-based models for DSA were conceived. This review focuses on three main topics:
the kind of stability it addresses, the type of system where proposals were developed
and validated, and the type of targeted applications. The latter differentiates between
offline and online applications.

4.1 Kind of stability

Power system stability is defined in (Kundur et al., 2004) as “the ability of an electric
power system, for a given initial operating condition, to regain a state of operating
equilibrium after being subjected to a physical disturbance, with most system variables
bounded so that practically the entire system remains intact”. Recently, (Hatziargyriou
et al., 2021) proposed an extended stability classification, covering the effects of fast-
response power electronic devices up to electromagnetic transients. The new proposal
classifies the stability phenomena as 1) rotor angle stability, 2) voltage stability, 3)
frequency stability, 4) resonance stability, and 5) converter-driven stability. However,
power system stability is essentially a single problem (Kundur et al., 2004). Indeed, in
many situations, one form of instability will probably not occur in its pure form, par-
ticularly in stressed systems with high levels of CIG, in which one form of instability
may ultimately lead to another form. Accordingly, while the classification of system

18
stability is an effective and convenient means to deal with the high dimensionality
and complexity of stability problems, overall system stability should always be kept
in mind. Regarding the kind of stability addressed, most works in the technical liter-
ature focus on short-term transient stability and short-term large-disturbance voltage
stability. Only a few works focus on short-term small-disturbance rotor-angle stability,
long-term small-disturbance voltage stability, and frequency stability (both short and
long-term). Only one work focused on resonance stability and two on converter-driven
stability (fast and slow interaction).
It is worth mentioning that some works claim that their proposal is general enough
to be used for any type of stability (see, for example, (R. Liu et al., 2019; Y. Xu et
al., 2011; R. Zhang et al., 2023a; T. Zhang et al., 2021; Zhao et al., 2023b; Zhu & Hill,
2022)). However, the following review only considers the kind of stability addressed
within the case studies. The main reason is that even though a model for stability
assessment can be developed to consider a wide range of stability issues, different
kinds of stability issues may have different challenges for their prediction. Therefore,
the only way to demonstrate the prediction capability is through specific case studies.

4.1.1 Rotor angle stability

Rotor angle stability refers to the ability of synchronous machines of an interconnected
power system to remain in synchronism after being subjected to a disturbance (Kun-
dur et al., 2004). Rotor angle stability can be subcategorized into small-disturbance
angle stability and transient stability. These kinds of stabilities are both considered
short-term phenomena. IS-based proposals for evaluating transient stability are (An et
al., 2020; Anderson et al., 2023; Azman et al., 2020; Bahbah & Girgis, 2004; G.N. Bal-
tas et al., 2018; N.G. Baltas et al., 2018; Bashiri Mosavi et al., 2018; Boudour & Hellal,
2005; Cepeda et al., 2014; Chen et al., 2019; Dharmapala, Rajapakse, Narendra, &
Zhang, 2020; Echeverrı́a et al., 2017; Frimpong et al., 2017; Geeganage et al., 2015;
Genc et al., 2010; Glavic, 2019; Gomez et al., 2011; Gu & Li, 2013; Guo & Milanović,
2013, 2014; Hashiesh et al., 2012; C. He et al., 2016; M. He, Vittal, & Zhang, 2013;
M. He, Zhang, & Vittal, 2013; Hu et al., 2019; Huang et al., 2020; Jensen et al., 2001;
Kaci et al., 2014; F. Li et al., 2021; Y. Li & Gu, 2013; Y. Li et al., 2015; Y. Li &
Yang, 2017; Lin, 2013; C.-W. Liu et al., 1999; J. Liu et al., 2021; R. Liu et al., 2019;
S. Liu et al., 2020; T. Liu et al., 2020; X. Liu et al., 2020; F. Luo et al., 2015; Mahdi &
Genc, 2017, 2018; Maihemuti et al., 2021; Mukherjee & De, 2020; Naderi et al., 2023;
Nakas et al., 2023; Niu et al., 2021; Pannell et al., 2018; Qiao et al., 2021; Rahmatian
et al., 2017; Ren, Wang, et al., 2023; Ren & Xu, 2019, 2020; Ren, Yuan, et al., 2023;
Rovnyak et al., 1994, 2000; Sawhney & Jeyasurya, 2006; Senroy et al., 2006; Shahzad,
2021, 2022a, 2022b, 2022c; Sharifian & Sharifian, 2015; Shi et al., 2020; Siddiqui et al.,
2018; Sulistiawati et al., 2016; Sun et al., 2008; Suprême et al., 2018; Tan et al., 2017;
Tan & Zhao, 2023; Tian et al., 2019; B. Wang et al., 2016; T.w. Wang et al., 2008;
Y. Xu et al., 2011, 2012, 2010; Y. Yang et al., 2017; Yousefian et al., 2017; Yousefian
& Kamalasadan, 2016, 2018; Yu et al., 2018, 2017; Yuanhang et al., 2015; C. Zhang
et al., 2016; R. Zhang et al., 2018, 2012, 2015, 2023a, 2023b; T. Zhang et al., 2021;
Y. Zhang et al., 2015; Y. Zhang, Xu, & Dong, 2018; Y. Zhang et al., 2017; Zhao et al.,
2023a, 2023b; Zhou et al., 2016; Zhou & Zhang, 2023; Zhu & Hill, 2022; Zhu et al.,

19
2022). Most of these works were developed for online-transient stability assessments,
except a few developed for offline applications (e.g., (Arteaga et al., 2019; G.N. Baltas
et al., 2018; C. He et al., 2016; Jensen et al., 2001; Y. Li et al., 2015; Lin, 2013; X. Liu
et al., 2020; Yuanhang et al., 2015; C. Zhang et al., 2016; R. Zhang et al., 2012)).
The main purpose of the works focusing on online transient stability assessment is to
reach a prediction as fast as possible to trigger corrective measures. The capabilities
of IS in this regard have been widely demonstrated. For example, the average compu-
tation time for the transient stability assessments in (Yu et al., 2018) was around two
cycles with a 99.98% accuracy. In (R. Zhang et al., 2015), , also for transient stability
assessment, their proposal took 2.5 cycles to reach a prediction, with an accuracy of
above 99.4%. References (Gao et al., 2023; R. Liu et al., 2018; M. Mohammadi et al.,
2010; Teeuwsen et al., 2005; Zhai et al., 2023) propose small-disturbance rotor-angle
stability assessment using IS techniques. In (R. Liu et al., 2018), the authors propose
a self-adaptive approach combining K-means (a traditional clustering technique) with
Particle Swarm Optimization (PSO) to identify representative operating conditions
for a fast small-signal stability exploration and voltage evaluation. In a case study
performed on the 59-bus IEEE test system with data from the Australian National
Electricity Market, the authors state that the proposed model overcomes two criti-
cal drawbacks of K-means, which are that the results depend on the results on the
initial centroids and the risk of getting trapped in local optima due to the use of gra-
dient descent. The proposed method reduced the number of operating conditions from
8760 to 555 representative ones, reaching a dimensionality reduction of 95.2%. Com-
pared to K-Means, the average error of the proposed clustering decreased from 4,5%
to 3.2%, meaning that the computational burden of performing dynamic simulations
can significantly decrease for practical applications since dynamic simulations must
only take place in representative operating conditions. Work in (M. Mohammadi et
al., 2010) presents another online small-disturbance rotor-angle stability assessment
model. The proposal uses a Ball Vector Machine to classify operating conditions into
four classes according to their small-signal stability margin: normal, alert, correctable
emergency, and uncorrectable emergency. As a result, operating conditions were clas-
sified with an accuracy of over 97%. One option to evaluate small-signal rotor-angle
stability without performing TDS is using modal analysis. In this case, the damping
ratio is the most widely used variable as a stability index, for example in (Teeuwsen et
al., 2005) which evaluated small signal rotor angle stability. Here, the IS: i) classifies
the states of the system according to its damping; ii) estimates the minimum damp-
ing; iii) predicts the positions of the dominant eigenvalues; iv) classifies the eigenvalue
regions where the dominant eigenvalues typically occur; and v) predicts the eigenvalue
regions, which is a combination of eigenvalue prediction and eigenvalue region classi-
fication. Regarding the features utilized for characterizing operating conditions, rotor
angle (Guo & Milanović, 2014; Senroy et al., 2006), rotor speed and acceleration of
the rotor (C.-W. Liu et al., 1999) are the most widely used for transient stability. In
(Rovnyak et al., 2000), the authors the rotor speed and acceleration, apparent resis-
tance, and its change (Rdot) (Rovnyak et al., 2000). As for the stability index, the
most common index utilized is the maximum angle deviation between two generators
in the transient period (target output for prediction) (Rahmatian et al., 2017). This

20
index is also utilized for calculating the Transient Stability Index (TSI) (G.N. Baltas
et al., 2018; Cepeda et al., 2014; Genc et al., 2010; Gomez et al., 2011; M. He, Zhang,
& Vittal, 2013; F. Li et al., 2021; X. Liu et al., 2020; Senroy et al., 2006; R. Zhang et
al., 2015; Y. Zhang et al., 2017). If the TSI is positive, the system is stable; otherwise,
it is unstable. Another index for assessing small-signal angular stability is the damping
ratio (R. Liu et al., 2018; M. Mohammadi et al., 2010; Teeuwsen et al., 2005).

4.1.2 Voltage stability

Voltage stability refers to the ability of a power system to maintain steady voltages at
all buses in the system after being subjected to a disturbance (Kundur et al., 2004).
Voltage stability issues can be short-term or long-term depending on how long it takes
for the disturbance to cause a system collapse. As for short-term large-disturbance
voltage stability assessments, several IS have been proposed in recent years (see, for
example (Abbass et al., 2023; Ali et al., 2015; Dharmapala et al., 2020; Hossain &
Kumar, 2023; Hosseini et al., 2012; Y. Luo et al., 2023; Nakas et al., 2023; Nuqui et
al., 2001; Ren & Xu, 2023; Sajan et al., 2015a; Y. Xu et al., 2016; H. Yang et al.,
2018; R. Zhang et al., 2023a, 2023b; Y. Zhang, Dong, et al., 2018; Y. Zhang, Xu,
Dong, & Zhang, 2019; Y. Zhang, Xu, Zhang, & Dong, 2019; Zhao et al., 2023b; Zhu
et al., 2017)). All these methods were proposed to detect fast (online) instabilities
after a contingency, where prediction speed is crucial. In (Y. Zhang, Xu, Dong, &
Zhang, 2019), the authors proposed a hierarchical self-adaptive method using ELM for
assessing voltage stability. They used the Root-Mean-Squared Voltage-Dip Severity
Index (RVSI) as a stability index. With the proposal, instabilities were identified
within 0.17 s after the fault with an accuracy of 99.1%. The model can also determine
the Fault-induced delayed voltage recovery index (FIDVR) when the system remains
stable. The model in these cases predicted the FIDVR index within 0.39 s with an
accuracy of 97.6%. Transient Voltage Collapse Index (TVCI) is another commonly used
index for assessing short-term voltage stability (Adewole & Tzoneva, 2017; Y. Xu et
al., 2016; Zhu et al., 2017)and wide area voltage stability assessment index (vcaRVSA)
(Adewole & Tzoneva, 2017) aand the Transient Voltage Severity Index (TVSI) (Y. Xu
et al., 2016) are other indices. While the vcaRVSA index predicts the margin to collapse
from the maximum available reactive power of the generators, the TVSI quantifies the
magnitude of the voltage violation and the associated duration time. IS for evaluating
long-term (small-disturbance) voltage stability are proposed in (Adewole & Tzoneva,
2017; Diao et al., 2009; Innah & Hiyama, 2011; Jain et al., 2003; Malbasa et al., 2017;
H. Mohammadi et al., 2016; Nandanwar et al., 2018; Nandanwar & Warkad, 2016;
Nie et al., 2017; Ren et al., 2020; Shah & Verma, 2016; Suganyadevi et al., 2016;
C. Xu et al., 2005; R. Zhang et al., 2013; Zheng et al., 2013; Zhu et al., 2021; Zhu
& Luo, 2021). In (Malbasa et al., 2017), the proposed model classifies the system
operating conditions as stable, alert or critical, depending on the Voltage Stability
Margin (VSM) index. Their results showed a prediction accuracy between 86% and
90% out of 1000 operating conditions of each class. In (Zheng et al., 2013), the authors
proposed a Regression Tree model to estimate the Oscillatory Stability Margin (OSM)
and the VSM index. The OSM index was estimated with 12572 records, and the VSM
index with 15303. Their results showed that the model could predict the OSM index

21
with an accuracy of 48.6%, 79.6%, and 92.3% for 2%, 10%, and 100% of the training
database, respectively. For predicting the VSM index, the model reached an accuracy
of 75.9%, 86.9%, and 93.2% for the 2%, 10%, and 100% of the training database,
respectively. Real-time measurements from PMU have also been used as input data
for online stability assessments in real-world applications, as in (Malbasa et al., 2017;
H. Yang et al., 2018; Zheng et al., 2013).

4.1.3 Frequency stability

Frequency stability refers to the ability of a power system to maintain a steady fre-
quency following a disturbance that results in imbalances between generation and
load (Kundur et al., 2004). Models for evaluating frequency stability were proposed
in (Alizadeh & Amraee, 2014; Bo et al., 2014; Chang et al., 2005; Karapidakis, 2007;
Nakas et al., 2023; Ortiz-Villalba et al., 2020; Song et al., 2020; Tang et al., 2017;
C. Wang et al., 2021; Q. Wang et al., 2019; Xie & Sun, 2021; Z. Zhang et al., 2023).
For example, in (Y. Xu et al., 2013), the authors proposed an ELM-based model that
uses the generators feed-in, the load demand, and the total power generation and
demand as input data for predicting the system’s frequency stability margin (FSM)
in a given set of contingencies. Their results showed that for 360 operating conditions,
the training time is 0.156s, and the accuracy calculated through the mean average
percentage error (MAPE) is 0.63%. In (Tang et al., 2017), the authors proposed a
cross-entropy ensemble algorithm that uses PMU measurements as input data for pre-
dicting the system’s frequency nadir. The prediction error of the ensemble algorithm
is 13.6%, which is better than the accuracy reached by individual base learners (ANN,
LSSVM, and ELM), which reached prediction errors between 17.1% and 17.8%. In
(Q. Wang et al., 2019), the proposal combines a model for system frequency response
(SFR) with an ELM-based data-driven one for fast online frequency stability assess-
ment. The ELM-based model is used to fit and correct the error of the SFR model.
Their results showed that integrating both models improves the accuracy of the pre-
dictions. For example, the MAPE obtained by SFR, ELM, and the integrated model
for the maximum frequency deviation is 43.2%, 9.6%, and 5.8%, respectively.

4.1.4 Resonance stability

Resonance generally occurs when energy exchanges between devices and components
of a power system oscillate. Particularly for sub-synchronous resonance (SSR), it can
manifest in two possible forms: due to a resonance occurring between the series com-
pensation and the mechanical torsional frequencies of the turbine-generator shaft, or
due to a resonance between the series compensation and the electrical characteristics
of the generator. When these oscillations grow because of insufficient energy dissipa-
tion in the flow path and manifest as amplified voltage, current, or torque magnitudes
exceeding the specified thresholds, a resonance instability is said to have occurred.
Hence, resonance stability can be torsional resonance (effect of HVDC links and
FACTs) or electrical resonance (effect of DFIG) (Hatziargyriou et al., 2021). Because
resonance stability is a relatively new phenomenon, only one work addresses this type
of stability using IS (Krishnan & Thampatty, 2020), in which the authors trained dif-
ferent techniques to predict unstable sub-synchronous resonance (SSR) oscillations.

22
The techniques explored were Logistic Regression, ANN, KNN, DT, SVM, XGBoost,
and RF. As a result, RF achieved the highest accuracy (96.18%) in a case study based
on the IEEE First Benchmark Model.

4.1.5 Converter-driven stability

Converter-driven stability refers to stability issues associated with the CIG’s cross-
couplings with the electromechanical dynamics of SGs and the network’s electromag-
netic transients. This interaction may lead to unstable power system oscillations over
a wide frequency range (Hatziargyriou et al., 2021). This phenomenon can be a fast
or slow interaction (Hatziargyriou et al., 2021). Since this phenomenon has gained
importance recently, only two works address this issue with AI techniques (Sepehr et
al., 2022) and (Y. Wang & Pal, 2023). In (Sepehr et al., 2022),the authors proposed an
ANN model composed of stacked autoencoders (SAE) for the early detection of Power
Synchronization Control (PSC) instability in a system with a weak grid. In an unstable
condition, determined by the predicted voltage limits at the point of common coupling
(PCC), a phase freeze mode is activated in the converter synchronization loop, allow-
ing other grid-following converters to stay connected to the PCC and back up the grid
during contingencies. In a case study based on a power system with renewable-based
CIGs connected to a weak AC grid through two transmission lines, the SAE algorithm
obtained an accuracy of 99.5%, slightly higher than the 98.9% accuracy obtained with
a simpler ANN.

4.1.6 Summary
Figure 5 presents the number of papers by kind of stability using AI. This figure shows
that most papers focus on angle rotor stability and, secondly, on voltage and frequency
stability. In comparison, only one paper focused on converter and one on resonance
stability. Figure 5 also shows that only a limited number of papers focus on more than
one type of stability, which are (Cremer & Strbac, 2021; Diao et al., 2010; Kamwa et
al., 2010; R. Liu et al., 2018; Nakas et al., 2023; Y. Wang et al., 2019; R. Zhang et al.,
2023b; Zhao et al., 2023b).

4.2 Type of system

The database generation that the models use can come from synthetic or real test
systems. A common research practice is to test the model’s performance on small test
systems and then evaluate its performance on more realistic ones. However, many
works are only validated using small test systems, in which the proposed models may
exhibit difference performances than if tested in realistic-sized power systems. This
section reviews test systems used for validating the proposals found in the technical
literature, which readers can use to select test systems for developing and testing their
proposals.

4.2.1 Test systems

Synthetic systems are characteristically test networks developed by IEEE, WECC,
and other institutions. Based on the number of buses considered, these systems can

23
 60

50
45

Number of works
40 37

30
22
20
14

10 7 8 7 6
4 5
3 2 1 2 1
1
0
≤2005 2006-2011 2012-2017 2018-2023
Period
Converter-driven stability Resonance stability
More than one kind of stability Frequency stability
Voltage stability Rotor angle stability
Fig. 5 Number of works by kind of stability

be classified as small (≤50 busbars),medium (between 50 and 500 busbars), or large

(¿500 busbars). Small test systems used in the literature are the IEEE 4-bus test
system (Abbass et al., 2023), the IEEE 9-bus test system (Bahbah & Girgis, 2004;
G.N. Baltas et al., 2018; N.G. Baltas et al., 2018; Boudour & Hellal, 2005; Goh et al.,
2015; Hossain & Kumar, 2023; C.-W. Liu et al., 1999; C. Xu et al., 2005), the IEEE 14-
bus power system (Adewuyi, Shigenobu, Ooya, Senjyu, & Howlader, 2019; Ali et al.,
2015; Dharmapala et al., 2020; Goh et al., 2015; Innah & Hiyama, 2011; Jayasankar et
al., 2010; Ortiz-Villalba et al., 2020; Pannell et al., 2018; Pérez-Londoño et al., 2017;
Y. Zhang, Xu, Zhang, & Dong, 2019; Z. Zhang et al., 2023), the IEEE 24-bus power
system (Y. Zhang, Xu, Zhang, & Dong, 2019), the IEEE 30-bus power system (Ali
et al., 2015; Jain et al., 2003; Mukherjee & De, 2020; Nandanwar et al., 2018; Pérez-
Londoño et al., 2017; Suganyadevi et al., 2016; Suliman & Rahman, 2010; Y. Zhang,
Xu, Zhang, & Dong, 2019), the IEEE 39-bus New England (Adewole & Tzoneva,
2017; Azman et al., 2020; Bahmanyar & Karami, 2014; Bo et al., 2014; Cepeda et al.,
2014; Chen et al., 2019; Echeverrı́a et al., 2017; Frimpong et al., 2017; Geeganage et
al., 2015; Gomez et al., 2011; Gu & Li, 2013; Hashiesh et al., 2012; M. He, Vittal, &
Zhang, 2013; M. He, Zhang, & Vittal, 2013; Hossain & Kumar, 2023; Hu et al., 2019;
F. Li et al., 2021; Y. Li & Gu, 2013; Y. Li et al., 2015; Y. Li & Yang, 2017; R. Liu
et al., 2019; S. Liu et al., 2020; T. Liu et al., 2020; X. Liu et al., 2020; Y. Luo et al.,
2023; H. Mohammadi et al., 2018; M. Mohammadi et al., 2010; M.V & C.K., 2014;
Naderi et al., 2023; Nakas et al., 2023; Ren, Wang, et al., 2023; Ren & Xu, 2019, 2020,
2023; Ren et al., 2020; Ren, Yuan, et al., 2023; Rovnyak et al., 1994; Sajan et al.,
2015b, 2014; Sawhney & Jeyasurya, 2006; Shah & Verma, 2016; Sharifian & Sharifian,
2015; Siddiqui et al., 2018; Sulistiawati et al., 2016; Suprême et al., 2018; Tan et al.,
2017; Tan & Zhao, 2023; Tang et al., 2017; Velayati et al., 2015; B. Wang et al., 2016;

24
Q. Wang et al., 2019; T.w. Wang et al., 2008; Y. Wang et al., 2019; Xie & Sun, 2021;
Y. Xu et al., 2013, 2011, 2016; H. Yang et al., 2018; Y. Yang et al., 2017; Yu et al.,
2018, 2017; Yuanhang et al., 2015; Zhai et al., 2023; C. Zhang et al., 2016; R. Zhang et
al., 2018, 2015; Y. Zhang, Dong, et al., 2018; Y. Zhang, Xu, & Dong, 2018; Y. Zhang
et al., 2017; Y. Zhang, Xu, Dong, & Zhang, 2019; Y. Zhang, Xu, Zhang, & Dong,
2019; Zhu & Hill, 2022), the 9-bus WSCC dynamic test system (Alizadeh & Amraee,
2014; Tang et al., 2017; Q. Wang et al., 2019), the 9-bus WECC system (Tian et al.,
2019), the 11-bus Two-area system (Zhou & Zhang, 2023), the 42-bus SE Australian
Power System (Bahmanyar & Karami, 2014), the 32-bus Nordic system (Zhu et al.,
2017; Zhu & Luo, 2021), the 36-bus 8 machine system (R. Zhang et al., 2023a), the 25-
bus Java-Bali system (Sulistiawati et al., 2016), the 36-bus system CEPRI (R. Zhang
et al., 2023b), and the 41-bus Nordic system (Ren & Xu, 2023; Ren et al., 2020).
Medium-sized systems used are the IEEE of 68 buses and 16 machines (Azman et
al., 2020; Gao et al., 2023; Guo & Milanović, 2013, 2014; Gurung et al., 2021; F. Li
et al., 2021; C. Wang et al., 2021; Yousefian et al., 2017; Yousefian & Kamalasadan,
2016, 2018; Z. Zhang et al., 2023; Zhao et al., 2023a; Zhou et al., 2016), the 74-bus
Nordic System (Naderi et al., 2023), the 77-bus Nordic System (Zhu et al., 2021), the
simplified Australian Electric System of 59 buses and 14 generators (R. Liu et al.,
2018; Suprême et al., 2018), the Indian 75-bus system (Chang et al., 2005; Jain et al.,
2003), the Iranian 66-bus power grid (H. Mohammadi et al., 2018, 2016), the IEEE
standard 118-bus system (Anderson et al., 2023; Dharmapala et al., 2020; Maihemuti
et al., 2021; Mukherjee & De, 2020; Nie et al., 2017; Y. Wang et al., 2019; Xie & Sun,
2021; R. Zhang et al., 2013; Zhao et al., 2023b), the 123-bus IEEE system (Y. Wang
& Pal, 2023), the 145-bus IEEE 50-generator test system (Azman et al., 2020; F. Luo
et al., 2015; Velayati et al., 2015; Y. Xu et al., 2012; Yu et al., 2018; R. Zhang et al.,
2012, 2015), the 162-bus IEEE 17 machine (Yu et al., 2018), the 300-bus IEEE system
(Zhou & Zhang, 2023), the 179-bus WECC system (Mahdi & Genc, 2018; Malbasa et
al., 2017; Senroy et al., 2006; Zheng et al., 2013), the 127-bus WSCC (Mahdi & Genc,
2017), the 181-bus Indian system (Suganyadevi et al., 2016), the 120-bus dynamic
equivalent system of Chinas power grid load (Y. Xu et al., 2012), the 176-bus Pacific
AC Intertie (PACI) (Rovnyak et al., 2000), the 140-bus NPCC (Q. Wang et al., 2019;
Zhou & Zhang, 2023), the 200-bus Illinois system (Ren, Wang, et al., 2023; Ren, Yuan,
et al., 2023; Tan & Zhao, 2023; Zhao et al., 2023b), the 360-bus subsystem within
AEP (Nuqui et al., 2001), and the 153-bus North GZ Power Grid in China (Zhu &
Luo, 2021). Among the large systems used, there is only the 1648-bus PSS/E (S. Liu
et al., 2020). Of the 35 test systems, the most widely used is the 39-bus IEEE New
England, with 72 publications.

4.2.2 Real systems

Several works have also used real-world system models to develop and validate their
proposals. In these cases, the operational conditions are usually extracted from actual
data during real events. Examples of real-world medium-size system models are the
600-bus WECC System (M. He, Vittal, & Zhang, 2013), the 400-bus Western Danish
power system (C. Liu et al., 2014), the 380-bus China Southern Power grid in 2014

25
(C. He et al., 2016), the 500-bus South Carolina System (Xie & Sun, 2021), the 162-
bus NESTA (Arteaga et al., 2019), and the 246-bus Indian Northern Region Power
Grid (NRPG) (Sajan et al., 2015a). Large system models used are the 876-bus Greek
mainland system projected to 2005 (Voumvoulakis et al., 2006), the 783-bus Hydro-
Québec network (Kamwa et al., 2010, 2012), the 750-bus Power system of Liaoning
province (Y. Li & Yang, 2017), the 17724-bus WECC bulk transmission (Rahmatian et
al., 2017), the 16100-bus Eastern Interconnection in North America (Genc et al., 2010),
the 2941-bus Turkish power system (Mahdi & Genc, 2018), the 2100-bus Entergy
system 240-generator (Sun et al., 2008), the 2400-bus subsystem inside AEP (Diao
et al., 2009), the 1140-bus Venezuela Power Grid (Gomez et al., 2011), the 1800-bus
Chilean Northern Interconnected System (Ortiz-Villalba et al., 2020), the 2000-bus
Northeast China Power Grid (R. Zhang et al., 2023a), the 2000-bus Guangdong Power
Grid in South China (Zhu et al., 2021), and the 1700-bus Realistic provincial power
grid in China (Hu et al., 2019).
Figure 6 summarizes the number of systems used to validate proposed models,
depending on the size of the system and whether the system is real or synthetic. The
figure shows that most proposals were validated in synthetic and small systems. Only
a few works used large-scale or real-world power system models. The problem with
using small systems to validate the developed models is that the accuracy obtained
may not be sustained when implemented in real systems.

140

120

100
Numbers of works

80

60

40

20

0
Small Medium Large
Test systems

Real system Test system

Fig. 6 Test systems applied

26
4.3 Type of application
This subsection reviews the type of application for the IS, differentiating between off-
line and on-line applications. In addition, there are preventive and corrective actions
that the results of the IS may trigger. It is worth noting that many works focus on
only assessing stability, without triggering any control actions.

4.3.1 Online and off-line applications

A typical off-line application is to perform a stability screening as part of operational
planning and expansion studies (see, for example, (R. Liu et al., 2018; Sun et al.,
2008)). The main idea is to support energy regulators and system operators in validat-
ing the results of these studies from a stability perspective, which confirms the validity
of different operation schedules or expansion plans from a stability viewpoint. Another
example of an off-line application is to design a specific control strategy (see, for exam-
ple, (Adewuyi et al., 2019; Arteaga et al., 2019; G.N. Baltas et al., 2018; Chang et
al., 2005; Gurung et al., 2021; C. He et al., 2016; Krishnan & Thampatty, 2020; Y. Li
et al., 2015; Lin, 2013; X. Liu et al., 2020; Ortiz-Villalba et al., 2020; Y. Wang et al.,
2019; Yuanhang et al., 2015; C. Zhang et al., 2016; R. Zhang et al., 2012)). The main
focus of off-line applications lies on the model’s accuracy since the response time of
the IS is non-critical. For online applications, the goal is to perform fast, real-time
stability assessments to identify risky conditions so the system operator can take pre-
ventive/remedial actions. Works focused on online stability assessments are (Abbass
et al., 2023; Ali et al., 2015; Alizadeh & Amraee, 2014; Anderson et al., 2023; Azman
et al., 2020; Bahbah & Girgis, 2004; Bahmanyar & Karami, 2014; N.G. Baltas et al.,
2018; Bashiri Mosavi et al., 2018; Bo et al., 2014; Boudour & Hellal, 2005; Cepeda et
al., 2014; Chen et al., 2019; Dharmapala et al., 2020; Diao et al., 2009; Echeverrı́a et
al., 2017; Frimpong et al., 2017; Gao et al., 2023; Geeganage et al., 2015; Genc et al.,
2010; Goh et al., 2015; Gomez et al., 2011; Gu & Li, 2013; Guo & Milanović, 2013,
2014; Hashiesh et al., 2012; M. He, Zhang, & Vittal, 2013; Hossain & Kumar, 2023; Hu
et al., 2019; Innah & Hiyama, 2011; Jain et al., 2003; Jayasankar et al., 2010; Kaci et
al., 2014; Kamwa et al., 2010; Karapidakis, 2007; F. Li et al., 2021; Y. Li & Gu, 2013;
Y. Li & Yang, 2017; C.-W. Liu et al., 1999; R. Liu et al., 2019, 2018; T. Liu et al., 2020;
F. Luo et al., 2015; Y. Luo et al., 2023; Mahdi & Genc, 2017, 2018; Maihemuti et al.,
2021; Malbasa et al., 2017; H. Mohammadi & Dehghani, 2015; H. Mohammadi et al.,
2018, 2016; M. Mohammadi et al., 2010; Mukherjee & De, 2020; M.V & C.K., 2014;
Naderi et al., 2023; Nakas et al., 2023; Nandanwar et al., 2018; Nie et al., 2017; Nuqui
et al., 2001; Pannell et al., 2018; Pérez-Londoño et al., 2017; Rahmatian et al., 2017;
Ren, Wang, et al., 2023; Ren & Xu, 2019, 2020, 2023; Ren et al., 2020; Ren, Yuan,
et al., 2023; Rovnyak et al., 1994, 2000; Sajan et al., 2015a, 2015b, 2014; Sawhney &
Jeyasurya, 2006; Senroy et al., 2006; Shah & Verma, 2016; Sharifian & Sharifian, 2015;
Siddiqui et al., 2018; Song et al., 2020; Suganyadevi et al., 2016; Suliman & Rahman,
2010; Sulistiawati et al., 2016; Sun et al., 2008; Suprême et al., 2018; Tan et al., 2017;
Tan & Zhao, 2023; Tang et al., 2017; Teeuwsen et al., 2005; Tian et al., 2019; Velayati
et al., 2015; B. Wang et al., 2016; C. Wang et al., 2021; Q. Wang et al., 2019; Y. Wang
& Pal, 2023; Xie & Sun, 2021; C. Xu et al., 2005; Y. Xu et al., 2013, 2011, 2012,

27
2010, 2016; H. Yang et al., 2018; Y. Yang et al., 2017; Yousefian et al., 2017; Youse-
fian & Kamalasadan, 2016, 2018; Yu et al., 2018, 2017; Zhai et al., 2023; R. Zhang
et al., 2018, 2015, 2013, 2023a, 2023b; Y. Zhang, Dong, et al., 2018; Y. Zhang et al.,
2015, 2017; Y. Zhang, Xu, Dong, & Zhang, 2019; Y. Zhang, Xu, Zhang, & Dong, 2019;
Z. Zhang et al., 2023; Zhao et al., 2023b; Zheng et al., 2013; Zhou et al., 2016; Zhou
& Zhang, 2023; Zhu & Hill, 2022; Zhu et al., 2021, 2017; Zhu & Luo, 2021). In the
case of online applications, the response time of the IS is critical, as it must be able
to identify unstable conditions as quickly as possible.

4.3.2 Type of action

Stability assessments can be carried out preventively, before a contingency occurs
(pre-disturbance), or done after the occurrence of a disturbance (post-disturbance).
Pre-disturbance assessment focuses on evaluating the stability of the power system
for a predefine set of disturbances that have not occurred, and the results may trig-
ger preventive control actions (R. Zhang et al., 2015). Preventive controls involve
changing the operating condition if a contingency could lead to instability. The most
typical preventive action is a generation rescheduling (Bo et al., 2014). For example,
a generation rescheduling is proposed in (Genc et al., 2010) for a transient stability
assessment, using DTs to evaluate the system stability, using SVM in (Tian et al.,
2019) and a Bayesian Neural Network in (T. Liu et al., 2020). Work in (Lin, 2013)
proposes two preemptive methods for offline transient stability margins based on DT
and MLP. Both methods use variations of active power and terminal voltages at each
generator and variations of active power and reactive power demand to increase the
CCT rate and improve stability margins. Although both preventive controls perform
well, DT-based preventive control requires triggering simultaneous control actions to
work. In contrast, MLP-based control only determines one control action, thus being
more flexible. Furthermore, (Gurung et al., 2021) proposes an MLP architecture and a
metaheuristic optimization to maximize the probabilistic small-signal stability (PSSS)
margin. Control actions are changes in the parameters of the power oscillation damp-
ing controllers (PODC). Work in (Jayasankar et al., 2010) presents another example of
preventive control within power system planning, where an FFNN designed to monitor
the voltage stability margin of lines. If it identifies a weak line, it will run a genetic opti-
mization algorithm to locate a thyristor-controlled series capacitor (TCSC) optimally
and improve the line’s stability margin. In contrast, post-disturbance assessment aims
to predict a system’s stability after the occurrence of a contingency. The objective is to
identify such instability as soon as possible to trigger corrective control actions. Typical
corrective control actions are disconnections of load and generation (Bo et al., 2014),
the activation of under-frequency load disconnection (UFLS) (Alizadeh & Amraee,
2014; Q. Wang et al., 2019; Xie & Sun, 2021) and the activation of under-voltage load
disconnection (UVLS) (Ren et al., 2020; Zhu & Luo, 2021). Other corrective control
actions are the modification of control system parameters (Hossain & Kumar, 2023;
Yousefian et al., 2017; Yousefian & Kamalasadan, 2016, 2018; Zhao et al., 2023a) and
load shedding (Genc et al., 2010; Naderi et al., 2023; Tian et al., 2019). As indicated
earlier, many works only focus on evaluating system stability, without triggering any
actions. We identified following works that falls into this case: (Abbass et al., 2023; Ali

28
et al., 2015; Anderson et al., 2023; Azman et al., 2020; Bahbah & Girgis, 2004; Bah-
manyar & Karami, 2014; N.G. Baltas et al., 2018; Bashiri Mosavi et al., 2018; Bo et
al., 2014; Boudour & Hellal, 2005; Cepeda et al., 2014; Chen et al., 2019; Dharmapala
et al., 2020; Diao et al., 2009; Echeverrı́a et al., 2017; Frimpong et al., 2017; Gao et
al., 2023; Geeganage et al., 2015; Goh et al., 2015; Gomez et al., 2011; Gu & Li, 2013;
Guo & Milanović, 2013, 2014; Hashiesh et al., 2012; M. He, Zhang, & Vittal, 2013; Hu
et al., 2019; Innah & Hiyama, 2011; Jain et al., 2003; Jayasankar et al., 2010; Kaci et
al., 2014; Kamwa et al., 2010; Karapidakis, 2007; F. Li et al., 2021; Y. Li & Gu, 2013;
Y. Li & Yang, 2017; C.-W. Liu et al., 1999; R. Liu et al., 2019, 2018; T. Liu et al.,
2020; F. Luo et al., 2015; Y. Luo et al., 2023; Mahdi & Genc, 2017, 2018; Maihemuti
et al., 2021; Malbasa et al., 2017; H. Mohammadi & Dehghani, 2015; H. Mohammadi
et al., 2018, 2016; M. Mohammadi et al., 2010; Mukherjee & De, 2020; M.V & C.K.,
2014; Nakas et al., 2023; Nandanwar et al., 2018; Nie et al., 2017; Nuqui et al., 2001;
Pannell et al., 2018; Pérez-Londoño et al., 2017; Rahmatian et al., 2017; Ren, Wang,
et al., 2023; Ren & Xu, 2019, 2020, 2023; Ren, Yuan, et al., 2023; Rovnyak et al., 1994,
2000; Sajan et al., 2015a, 2015b, 2014; Sawhney & Jeyasurya, 2006; Senroy et al.,
2006; Shah & Verma, 2016; Sharifian & Sharifian, 2015; Siddiqui et al., 2018; Song et
al., 2020; Suganyadevi et al., 2016; Suliman & Rahman, 2010; Sulistiawati et al., 2016;
Sun et al., 2008; Suprême et al., 2018; Tan et al., 2017; Tang et al., 2017; Teeuwsen et
al., 2005; Velayati et al., 2015; B. Wang et al., 2016; C. Wang et al., 2021; Y. Wang
& Pal, 2023; C. Xu et al., 2005; Y. Xu et al., 2013, 2011, 2012, 2010, 2016; H. Yang
et al., 2018; Y. Yang et al., 2017; Yu et al., 2018, 2017; Zhai et al., 2023; R. Zhang
et al., 2018, 2015, 2013, 2023a, 2023b; Y. Zhang, Dong, et al., 2018; Y. Zhang et al.,
2015, 2017; Y. Zhang, Xu, Dong, & Zhang, 2019; Y. Zhang, Xu, Zhang, & Dong, 2019;
Z. Zhang et al., 2023; Zhao et al., 2023b; Zheng et al., 2013; Zhou et al., 2016; Zhou
& Zhang, 2023; Zhu & Hill, 2022; Zhu et al., 2021, 2017). Figure 7 shows the different
actions articles studied during this investigation have applied. This figure shows that
most articles focus on evaluating system stability or estimating stability margin and
less on performing corrective or preventive actions. Finally, we found only one paper
that determines both preventive actions (generation rescheduling) and corrective ones
(Load shedding schemes) (Genc et al., 2010).

5 Conclusion
The use of AI-based techniques for DSA has gained increasing attention in the past
years. The main reason is that predictive models based on AI have been shown to pro-
vide high-accuracy results in very short times, thus providing a reliable and helpful
tool that complements traditional DSA based on TDS. For offline DSA, AI-based tech-
niques allow evaluation of a wide range of operating conditions with very low human
and computational efforts (once the models are built), which can be useful as a screen-
ing tool for power system operational and expansion planning and control design. On
the other hand, for online DSA, rapid DSA can be extremely helpful in determining
corrective control actions once a contingency occurs, which otherwise cannot be done
using traditional TDS. The latter has also been possible due to the wide incorporation
of PMUs in power systems. Despite of its benefits, there are still significant research

29
 180

160
145
140

Number of works
120

100

80

60

40

20 14
5 1
0
Measure margin Corrective actions Preventive Preventive and
of stability actions corrective actions
Application

Fig. 7 Type of applications

gaps. As seen throughout this review, most proposals use small test systems and syn-
thetic (unrealistic) data to develop and validate their proposals, which questions the
validity of the models for their application in real-world power systems since they may
not exhibit the same performance. Another important characteristic of AI-based tech-
niques is that training AI-based models requires many operating conditions labeled
using TDS, which poses a significant barrier to the development of such models, espe-
cially for realistic-size power system models. In this regard, significant research is
needed to develop strategies and models that require few operating conditions labeled
with TDS. One possibility for filling this gap is to develop models using steady-state
or quasi steady-state stability indices, which are cheap to compute, and then correct
the models using a few operating conditions labeled with TDS, for example, through a
transfer learning strategy. An interesting feature identified in this review is the use of
more advanced DL models. Here, we refer to DL models as multiple-layer ANNs that
can capture more complex characteristics from the data than traditional ML algo-
rithms. Even though most works still propose traditional ML-based models, the use of
DL models has significantly increased in the past few years. Similarly, a growing trend
exists to develop ensemble models instead of single ML models. The good results in
terms of prediction accuracy obtained by ensemble models, compared to their single-
model counterparts, suggest that this trend towards ensemble models should continue
in the future. Regarding the type of stability being addressed, most works focus on
transient stability and large disturbance voltage stability. Only a few works have
focused on small disturbance angle stability, small disturbance voltage stability, and
frequency stability. Furthermore, we only found one work focusing on resonance and
converter-driven stability, respectively. Given the importance of these latter stability
issues in future power systems dominated by CIG, there is an urgent need to integrate

30
such stability issues within AI-based models for DSA. From a practical perspective,
i.e., for the development of AI-based models for real-world power system applications,
there is a significant research gap in addressing challenges associated with such appli-
cations: most works assume full availability of PMU measurement, neglecting relevant
issues such as measurement failures, noise, delays, and bad data. In addition, only a
few works have dealt with adapting the prediction models to changes in power sys-
tems. The key is to develop advanced strategies to identify the need for adaptation
and to adjust the models with minimum human and computational resources, espe-
cially in labeling new operating conditions using TDS. For applications in real-time,
training time for adapting the models is also crucial. In summary, using AI for DSA
is a promising and growing research area that still requires further efforts to fill the
research gap, especially toward using the proposed models in real-world power system
applications.

References
Abbass, M.J., Lis, R., Mushtaq, Z. (2023). Artificial Neural Network (ANN)-
Based Voltage Stability Prediction of Test Microgrid Grid. IEEE Access,
11 , 58994–59001, https://doi.org/10.1109/ACCESS.2023.3284545 Retrieved
from https://ieeexplore.ieee.org/document/10147228 (Conference Name: IEEE
Access)

Adewole, A.C., & Tzoneva, R. (2017, March). Surrogate-splits ensembles for real-
time voltage stability assessment in the presence of missing synchrophasor
measurements. IET Science, Measurement & Technology, 11 (5), 545–552,
https://doi.org/10.1049/iet-smt.2016.0431 Retrieved from https://digital-
library.theiet.org/content/journals/10.1049/iet-smt.2016.0431 (Publisher: IET
Digital Library)

Adewuyi, O.B., Shigenobu, R., Ooya, K., Senjyu, T., Howlader, A.M. (2019, July).
Static voltage stability improvement with battery energy storage considering
optimal control of active and reactive power injection. Electric Power Systems
Research, 172 , 303–312, https://doi.org/10.1016/j.epsr.2019.04.004

Ali, A.K., El-Amary, N.H., Ibrahim, A., Mekhamer, S.F. (2015, October). Volt-
age instability detector based on Phasor Measurement Units using artifi-
cial neural network. 2015 Workshop on Engineering Applications - Inter-
national Congress on Engineering (WEA) (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/7370151

Alimi, O.A., Ouahada, K., Abu-Mahfouz, A.M. (2020). A Review of Machine

Learning Approaches to Power System Security and Stability. IEEE Access,
8 , 113512–113531, https://doi.org/10.1109/ACCESS.2020.3003568 Retrieved

31
 from https://ieeexplore.ieee.org/document/9121208 (Conference Name: IEEE
Access)

Alizadeh, M., & Amraee, T. (2014, February). Adaptive scheme for local prediction
of post-contingency power system frequency. Electric Power Systems Research,
107 , 240–249, https://doi.org/10.1016/j.epsr.2013.10.014 Retrieved from
https://www.sciencedirect.com/science/article/pii/S037877961300285X

An, J., Yu, J., Li, Z., Zhou, Y., Mu, G. (2020, November). A Data-driven Method
for Transient Stability Margin Prediction Based on Security Region. Journal of
Modern Power Systems and Clean Energy, 8 (6), 1060–1069, https://doi.org/
10.35833/MPCE.2020.000457 (Conference Name: Journal of Modern Power
Systems and Clean Energy)

Anderson, A.A., Jefferson, B.A., Kincic, S., Wenskovitch, J.E., Fallon, C.K., Baweja,
J.A., Chen, Y. (2023). Human-Centric Contingency Analysis Metrics for
Evaluating Operator Performance and Trust. IEEE Access, 11 , 109689–
109707, https://doi.org/10.1109/ACCESS.2023.3322133 Retrieved from
https://ieeexplore.ieee.org/document/10273136 (Conference Name: IEEE
Access)

Arteaga, J.-M.H., Hancharou, F., Thams, F., Chatzivasileiadis, S. (2019, June). Deep
Learning for Power System Security Assessment. 2019 IEEE Milan PowerTech
(pp. 1–6). Retrieved from https://ieeexplore.ieee.org/document/8810906

Azman, S.K., Isbeih, Y.J., Moursi, M.S.E., Elbassioni, K. (2020, November).

A Unified Online Deep Learning Prediction Model for Small Signal and
Transient Stability. IEEE Transactions on Power Systems, 35 (6), 4585–
4598, https://doi.org/10.1109/TPWRS.2020.2999102 Retrieved from
https://ieeexplore.ieee.org/document/9105102 (Conference Name: IEEE Trans-
actions on Power Systems)

Bahbah, A., & Girgis, A. (2004, May). New method for generators’ angles and angu-
lar velocities prediction for transient stability assessment of multimachine power
systems using recurrent artificial neural network. IEEE Transactions on Power
Systems, 19 (2), 1015–1022, https://doi.org/10.1109/TPWRS.2004.826765
Retrieved from https://ieeexplore.ieee.org/document/1295012 (Conference
Name: IEEE Transactions on Power Systems)

Bahmanyar, A.R., & Karami, A. (2014, June). Power system voltage sta-
bility monitoring using artificial neural networks with a reduced set of

32
 inputs. International Journal of Electrical Power & Energy Systems, 58 ,
246–256, https://doi.org/10.1016/j.ijepes.2014.01.019 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061514000325

Baltas, G.N., Perales-González, C., Mazidi, P., Fernandez, F., Rodrı́guez, P. (2018,
October). A Novel Ensemble Approach for Solving the Transient Stabil-
ity Classification Problem. 2018 7th International Conference on Renewable
Energy Research and Applications (ICRERA) (pp. 1282–1286). Retrieved from
https://ieeexplore.ieee.org/document/8566815 (ISSN: 2572-6013)

Baltas, N.G., Mazidi, P., Ma, J., de Asis Fernandez, F., Rodriguez, P. (2018, Septem-
ber). A Comparative Analysis of Decision Trees, Support Vector Machines and
Artificial Neural Networks for On-line Transient Stability Assessment. 2018
International Conference on Smart Energy Systems and Technologies (SEST)
(pp. 1–6). Retrieved from https://ieeexplore.ieee.org/document/8495872

Bashiri Mosavi, A., Amiri, A., Hosseini, H. (2018). A Learning Framework

for Size and Type Independent Transient Stability Prediction of Power Sys-
tem Using Twin Convolutional Support Vector Machine. IEEE Access, 6 ,
69937–69947, https://doi.org/10.1109/ACCESS.2018.2880273 Retrieved
from https://ieeexplore.ieee.org/document/8540347 (Conference Name: IEEE
Access)

Bo, Q., Wang, X., Liu, K. (2014, October). Minimum frequency prediction of power
system after disturbance based on the v-support vector regression. 2014 Interna-
tional Conference on Power System Technology (pp. 614–619). Retrieved from
https://ieeexplore.ieee.org/document/6993789

Boudour, M., & Hellal, A. (2005, September). Combined use of super-

vised and unsupervised learning for power system dynamic security map-
ping. Engineering Applications of Artificial Intelligence, 18 (6), 673–
683, https://doi.org/10.1016/j.engappai.2005.01.005 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0952197605000096

Cepeda, J.C., Rueda, J.L., Colomé, D.G., Echeverrı́a, D.E. (2014, February).
Real-time transient stability assessment based on centre-of-inertia estimation
from phasor measurement unit records. IET Generation, Transmission &
Distribution, 8 (8), 1363–1376, https://doi.org/10.1049/iet-gtd.2013.0616
Retrieved from https://digital-library.theiet.org/content/journals/10.1049/iet-
gtd.2013.0616 (Publisher: IET Digital Library)

Chang, R.-F., Lu, C.-N., Hsiao, T.-Y. (2005, November). Prediction of frequency
response after generator outage using regression tree. IEEE Transactions

33
 on Power Systems, 20 (4), 2146–2147, https://doi.org/10.1109/TPWRS.2005
.857259 Retrieved from https://ieeexplore.ieee.org/document/1525147?denied=
(Conference Name: IEEE Transactions on Power Systems)

Chen, M., Liu, Q., Chen, S., Liu, Y., Zhang, C.-H., Liu, R. (2019). XGBoost-
Based Algorithm Interpretation and Application on Post-Fault Transient
Stability Status Prediction of Power System. IEEE Access, 7 , 13149–
13158, https://doi.org/10.1109/ACCESS.2019.2893448 Retrieved from
https://ieeexplore.ieee.org/document/8620201 (Conference Name: IEEE
Access)

Cremer, J.L., & Strbac, G. (2021, June). A machine-learning based probabilistic

perspective on dynamic security assessment. International Journal of Electrical
Power & Energy Systems, 128 , 106571, https://doi.org/10.1016/j.ijepes.2020
.106571

Dharmapala, K.D., Rajapakse, A., Narendra, K., Zhang, Y. (2020). Machine Learn-
ing Based Real-Time Monitoring of Long-Term Voltage Stability Using Voltage
Stability Indices. IEEE Access, 8 , 222544–222555, https://doi.org/10.1109/
ACCESS.2020.3043935 (Conference Name: IEEE Access)

Diao, R., Sun, K., Vittal, V., O’Keefe, R.J., Richardson, M.R., Bhatt, N., . . .
Sarawgi, S.K. (2009, May). Decision Tree-Based Online Voltage Security
Assessment Using PMU Measurements. IEEE Transactions on Power Systems,
24 (2), 832–839, https://doi.org/10.1109/TPWRS.2009.2016528 Retrieved
from https://ieeexplore.ieee.org/document/4813188 (Conference Name: IEEE
Transactions on Power Systems)

Diao, R., Vittal, V., Logic, N. (2010, May). Design of a Real-Time Secu-
rity Assessment Tool for Situational Awareness Enhancement in Mod-
ern Power Systems. IEEE Transactions on Power Systems, 25 (2), 957–
965, https://doi.org/10.1109/TPWRS.2009.2035507 Retrieved from
https://ieeexplore.ieee.org/document/5342461 (Conference Name: IEEE Trans-
actions on Power Systems)

Dong, Z.Y., Xu, Y., Zhang, P., Wong, K.P. (2013, July). Using IS to Assess
an Electric Power System’s Real-Time Stability. IEEE Intelligent Sys-
tems, 28 (4), 60–66, https://doi.org/10.1109/MIS.2011.41 Retrieved from
https://ieeexplore.ieee.org/document/6682954 (Conference Name: IEEE Intel-
ligent Systems)

34
Echeverrı́a, J.D.E., Cepeda, C.J.C., Colomé, D.G. (2017, September). Real-time
transient stability assessment of electric power systems using predictive-SIME
based on machine learning. 2017 IEEE PES Innovative Smart Grid Technologies
Conference - Latin America (ISGT Latin America) (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/8126760

Fang, J., Li, H., Tang, Y., Blaabjerg, F. (2019, December). On the Inertia of Future
More-Electronics Power Systems. IEEE Journal of Emerging and Selected Top-
ics in Power Electronics, 7 (4), 2130–2146, https://doi.org/10.1109/JESTPE
.2018.2877766 Retrieved from https://ieeexplore.ieee.org/document/8506338
(Conference Name: IEEE Journal of Emerging and Selected Topics in Power
Electronics)

Frimpong, E.A., Okyere, P.Y., Asumadu, J. (2017, June). On-line determination

of transient stability status using MLPNN. 2017 IEEE PES PowerAfrica (pp.
23–27). Retrieved from https://ieeexplore.ieee.org/document/7991194

Gao, H., Yang, D., Cai, G., Chen, Z., Ma, J., Wang, L., Duan, F. (2023, September).
Machine Learning-Based Reliability Improvement of Ambient Mode Extraction
for Smart Grid Utilizing Isolation Forest. IEEE Transactions on Power Systems,
38 (5), 4752–4760, https://doi.org/10.1109/TPWRS.2022.3226263 Retrieved
from https://ieeexplore.ieee.org/document/9970387 (Conference Name: IEEE
Transactions on Power Systems)

Geeganage, J., Annakkage, U.D., Weekes, T., Archer, B.A. (2015, July).
Application of Energy-Based Power System Features for Dynamic Secu-
rity Assessment. IEEE Transactions on Power Systems, 30 (4), 1957–
1965, https://doi.org/10.1109/TPWRS.2014.2353048 Retrieved from
https://ieeexplore.ieee.org/document/6902826?denied= (Conference Name:
IEEE Transactions on Power Systems)

Genc, I., Diao, R., Vittal, V., Kolluri, S., Mandal, S. (2010, August). Decision Tree-
Based Preventive and Corrective Control Applications for Dynamic Security
Enhancement in Power Systems. IEEE Transactions on Power Systems, 25 (3),
1611–1619, https://doi.org/10.1109/TPWRS.2009.2037006 Retrieved from
https://ieeexplore.ieee.org/abstract/document/5374083 (Conference Name:
IEEE Transactions on Power Systems)

Glavic, M. (2019, January). (Deep) Reinforcement learning for electric power system
control and related problems: A short review and perspectives. Annual Reviews
in Control , 48 , 22–35, https://doi.org/10.1016/j.arcontrol.2019.09.008

35
Goh, H.H., Chua, Q.S., Lee, S.W., Kok, B.C., Goh, K.C., Teo, K.T.K.
(2015, January). Evaluation for Voltage Stability Indices in Power Sys-
tem Using Artificial Neural Network. Procedia Engineering, 118 , 1127–
1136, https://doi.org/10.1016/j.proeng.2015.08.454 Retrieved from
https://www.sciencedirect.com/science/article/pii/S1877705815021098

Gomez, F.R., Rajapakse, A.D., Annakkage, U.D., Fernando, I.T. (2011, August). Sup-
port Vector Machine-Based Algorithm for Post-Fault Transient Stability Status
Prediction Using Synchronized Measurements. IEEE Transactions on Power
Systems, 26 (3), 1474–1483, https://doi.org/10.1109/TPWRS.2010.2082575
Retrieved from https://ieeexplore.ieee.org/document/5617329 (Conference
Name: IEEE Transactions on Power Systems)

Goodfellow, I.J., Bengio, Y., Courville, A. (2016). Deep Learning. Cambridge, MA,
USA: MIT Press.

Gu, X., & Li, Y. (2013, July). Bayesian multiple kernels learning-based transient
stability assessment of power systems using synchronized measurements. 2013
IEEE Power & Energy Society General Meeting (pp. 1–5). Retrieved from
https://ieeexplore.ieee.org/document/6672810 (ISSN: 1932-5517)

Guo, T., & Milanović, J.V. (2013, October). The effect of quality and availability
of measurement signals on accuracy of on-line prediction of transient stability
using decision tree method. IEEE PES ISGT Europe 2013 (pp. 1–5). Retrieved
from https://ieeexplore.ieee.org/document/6695256 (ISSN: 2165-4824)

Guo, T., & Milanović, J.V. (2014, January). Probabilistic Framework for
Assessing the Accuracy of Data Mining Tool for Online Prediction of
Transient Stability. IEEE Transactions on Power Systems, 29 (1), 377–
385, https://doi.org/10.1109/TPWRS.2013.2281118 Retrieved from
https://ieeexplore.ieee.org/document/6606932 (Conference Name: IEEE Trans-
actions on Power Systems)

Gurung, S., Naetiladdanon, S., Sangswang, A. (2021). A Surrogate Based

Computationally Efficient Method to Coordinate Damping Controllers for
Enhancement of Probabilistic Small-Signal Stability. IEEE Access, 9 , 32882–
32896, https://doi.org/10.1109/ACCESS.2021.3060502 Retrieved from
https://ieeexplore.ieee.org/document/9358149 (Conference Name: IEEE
Access)

Hadidi, R., & Jeyasurya, B. (2013, March). Reinforcement Learn-

ing Based Real-Time Wide-Area Stabilizing Control Agents to Enhance
Power System Stability. IEEE Transactions on Smart Grid , 4 (1),

36
 489–497, https://doi.org/10.1109/TSG.2012.2235864 Retrieved from
https://ieeexplore.ieee.org/abstract/document/6410470 (Conference Name:
IEEE Transactions on Smart Grid)

Hashiesh, F., Mostafa, H.E., Khatib, A.-R., Helal, I., Mansour, M.M. (2012,
June). An Intelligent Wide Area Synchrophasor Based System for Predict-
ing and Mitigating Transient Instabilities. IEEE Transactions on Smart
Grid , 3 (2), 645–652, https://doi.org/10.1109/TSG.2012.2187220 Retrieved
from https://ieeexplore.ieee.org/document/6184356 (Conference Name: IEEE
Transactions on Smart Grid)

Hatziargyriou, N., Milanovic, J., Rahmann, C., Ajjarapu, V., Canizares, C., Erlich,
I., . . . Vournas, C. (2021, July). Definition and Classification of Power Sys-
tem Stability – Revisited & Extended. IEEE Transactions on Power Systems,
36 (4), 3271–3281, https://doi.org/10.1109/TPWRS.2020.3041774 Retrieved
from https://ieeexplore.ieee.org/document/9286772 (Conference Name: IEEE
Transactions on Power Systems)

He, C., Guan, L., Mo, W. (2016, October). A method for transient stability assess-
ment based on pattern recognition. 2016 International Conference on Smart
Grid and Clean Energy Technologies (ICSGCE) (pp. 343–347). Retrieved from
https://ieeexplore.ieee.org/document/7876081

He, M., Vittal, V., Zhang, J. (2013, May). Online dynamic security assessment
with missing pmu measurements: A data mining approach. IEEE Transac-
tions on Power Systems, 28 (2), 1969–1977, https://doi.org/10.1109/TPWRS
.2013.2246822 Retrieved from https://ieeexplore.ieee.org/document/6475215
(Conference Name: IEEE Transactions on Power Systems)

He, M., Zhang, J., Vittal, V. (2013, November). Robust Online Dynamic Security
Assessment Using Adaptive Ensemble Decision-Tree Learning. IEEE Transac-
tions on Power Systems, 28 (4), 4089–4098, https://doi.org/10.1109/TPWRS
.2013.2266617 Retrieved from https://ieeexplore.ieee.org/document/6547746
(Conference Name: IEEE Transactions on Power Systems)

Hossain, R.R., & Kumar, R. (2023, April). Machine Learning Accelerated Real-
Time Model Predictive Control for Power Systems. IEEE/CAA Journal of
Automatica Sinica, 10 (4), 916–930, https://doi.org/10.1109/JAS.2023.123135
Retrieved from https://ieeexplore.ieee.org/document/10085977 (Conference
Name: IEEE/CAA Journal of Automatica Sinica)

37
Hosseini, S.M.H., Olamaei, J., Gholami, V., Jahangir, M. (2012, March). Decision
Tree Based Online Identification of Critical Voltage Control Area. 2012 Asia-
Pacific Power and Energy Engineering Conference (pp. 1–4). Retrieved from
https://ieeexplore.ieee.org/document/6307562 (ISSN: 2157-4847)

Hu, W., Lu, Z., Wu, S., Zhang, W., Dong, Y., Yu, R., Liu, B. (2019, Jan-
uary). Real-time transient stability assessment in power system based on
improved SVM. Journal of Modern Power Systems and Clean Energy,
7 (1), 26–37, https://doi.org/10.1007/s40565-018-0453-x Retrieved from
https://ieeexplore.ieee.org/document/9028808 (Conference Name: Journal of
Modern Power Systems and Clean Energy)

Huang, J., Guan, L., Su, Y., Yao, H., Guo, M., Zhong, Z. (2020). Recurrent
Graph Convolutional Network-Based Multi-Task Transient Stability Assessment
Framework in Power System. IEEE Access, 8 , 93283–93296, https://doi.org/
10.1109/ACCESS.2020.2991263 (Conference Name: IEEE Access)

Innah, H., & Hiyama, T. (2011, October). A real time PMU data and neural net-
work approach to analyze voltage stability. 2011 International Conference on
Advanced Power System Automation and Protection (Vol. 2, pp. 1263–1267).
Retrieved from https://ieeexplore.ieee.org/document/6180572

IRENA (2022). Renewable Capacity Statistics 2022 (Tech. Rep.). Abu Dhabi: The
International Renewable Energy Agency.

Jain, T., Srivastava, L., Singh, S. (2003, November). Fast voltage contingency
screening using radial basis function neural network. IEEE Transactions
on Power Systems, 18 (4), 1359–1366, https://doi.org/10.1109/TPWRS.2003
.818607 Retrieved from https://ieeexplore.ieee.org/abstract/document/1245558
(Conference Name: IEEE Transactions on Power Systems)

Jayasankar, V., Kamaraj, N., Vanaja, N. (2010, October). Estimation of

voltage stability index for power system employing artificial neural net-
work technique and TCSC placement. Neurocomputing, 73 (16), 3005–
3011, https://doi.org/10.1016/j.neucom.2010.07.006 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0925231210003309

Jensen, C., El-Sharkawi, M., Marks, R. (2001, November). Power system security
assessment using neural networks: feature selection using Fisher discrimination.
IEEE Transactions on Power Systems, 16 (4), 757–763, https://doi.org/10
.1109/59.962423 Retrieved from https://ieeexplore.ieee.org/document/962423
(Conference Name: IEEE Transactions on Power Systems)

38
Jia, J., Yang, G., Nielsen, A.H. (2018, April). A Review on Grid-
Connected Converter Control for Short-Circuit Power Provision Under Grid
Unbalanced Faults. IEEE Transactions on Power Delivery, 33 (2), 649–
661, https://doi.org/10.1109/TPWRD.2017.2682164 Retrieved from
https://ieeexplore.ieee.org/abstract/document/7878663 (Conference Name:
IEEE Transactions on Power Delivery)

Kaci, A., Kamwa, I., Dessaint, L.-A., Guillon, S. (2014, July). Phase angles as pre-
dictors of network dynamic security limits and further implications. 2014 IEEE
PES General Meeting | Conference & Exposition (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/6939281 (ISSN: 1932-5517)

Kamwa, I., Samantaray, S.R., Joos, G. (2010, September). Catastrophe Predictors

From Ensemble Decision-Tree Learning of Wide-Area Severity Indices. IEEE
Transactions on Smart Grid , 1 (2), 144–158, https://doi.org/10.1109/TSG
.2010.2052935 Retrieved from https://ieeexplore.ieee.org/document/5545486
(Conference Name: IEEE Transactions on Smart Grid)

Kamwa, I., Samantaray, S.R., Joos, G. (2012, March). On the Accuracy Versus
Transparency Trade-Off of Data-Mining Models for Fast-Response PMU-Based
Catastrophe Predictors. IEEE Transactions on Smart Grid , 3 (1), 152–161,
https://doi.org/10.1109/TSG.2011.2164948 (Conference Name: IEEE Transac-
tions on Smart Grid)

Karapidakis, E.S. (2007, January). Machine learning for frequency

estimation of power systems. Applied Soft Computing, 7 (1), 105–
114, https://doi.org/10.1016/j.asoc.2005.04.002 Retrieved from
https://www.sciencedirect.com/science/article/pii/S1568494605000281

Krishnan, K.C., & Thampatty, K.S. (2020, July). Prediction of Sub-synchronous

Resonance Oscillations - A Machine Learning Approach. 2020 IEEE Stu-
dents Conference on Engineering & Systems (SCES) (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/9236775

Kundur, P., Paserba, J., Ajjarapu, V., Andersson, G., Bose, A., Canizares, C., . . . Vit-
tal, V. (2004, August). Definition and classification of power system stability
IEEE/CIGRE joint task force on stability terms and definitions. IEEE Transac-
tions on Power Systems, 19 (3), 1387–1401, https://doi.org/10.1109/TPWRS
.2004.825981 (Conference Name: IEEE Transactions on Power Systems)

Li, F., Wang, Q., Tang, Y., Xu, Y., Dang, J. (2021). Hybrid analytical and
data-driven modeling based instance-transfer method for power system online

39
 transient stability assessment. CSEE Journal of Power and Energy Sys-
tems, 1–10, https://doi.org/10.17775/CSEEJPES.2020.03880 Retrieved
from https://ieeexplore.ieee.org/document/9420351 (Conference Name: CSEE
Journal of Power and Energy Systems)

Li, Y., & Gu, X. (2013, December). Power system transient stability assessment based
on online sequential extreme learning machine. 2013 IEEE PES Asia-Pacific
Power and Energy Engineering Conference (APPEEC) (pp. 1–4). Retrieved
from https://ieeexplore.ieee.org/document/6837163 (ISSN: 2157-4847)

Li, Y., Li, G., Wang, Z. (2015, June). Rule extraction based on extreme learning
machine and an improved ant-miner algorithm for transient stability assessment.
PLOS ONE , 10 (6), e0130814, https://doi.org/10.1371/journal.pone.0130814
Retrieved from http://arxiv.org/abs/1810.00658 (arXiv:1810.00658 [eess])

Li, Y., & Yang, Z. (2017). Application of EOS-ELM With Binary Jaya-Based Fea-
ture Selection to Real-Time Transient Stability Assessment Using PMU Data.
IEEE Access, 5 , 23092–23101, https://doi.org/10.1109/ACCESS.2017.2765626
Retrieved from https://ieeexplore.ieee.org/document/8081764 (Conference
Name: IEEE Access)

Lin, Y.-J. (2013, February). Comparison of CART- and MLP-based

power system transient stability preventive control. International
Journal of Electrical Power & Energy Systems, 45 (1), 129–
136, https://doi.org/10.1016/j.ijepes.2012.08.066 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061512005169

Liu, C., Sun, K., Rather, Z.H., Chen, Z., Bak, C.L., Thøgersen, P., Lund, P. (2014,
March). A Systematic Approach for Dynamic Security Assessment and the
Corresponding Preventive Control Scheme Based on Decision Trees. IEEE
Transactions on Power Systems, 29 (2), 717–730, https://doi.org/10.1109/
TPWRS.2013.2283064 (Conference Name: IEEE Transactions on Power
Systems)

Liu, C.-W., Su, M.-c., Tsay, S.-S., Wang, Y.-J. (1999, May). Application of a
novel fuzzy neural network to real-time transient stability swings prediction
based on synchronized phasor measurements. IEEE Transactions on Power
Systems, 14 (2), 685–692, https://doi.org/10.1109/59.761898 Retrieved
from https://ieeexplore.ieee.org/document/761898 (Conference Name: IEEE
Transactions on Power Systems)

40
Liu, J., Peng, X., Wu, L., Li, Y., Liu, J., Niu, S., . . . Ren, C. (2021, December).
Incremental learning method for online transient stability assessment of power
system based on KKT condition. 2021 IEEE Sustainable Power and Energy
Conference (iSPEC) (pp. 3968–3974).

Liu, R., Verbič, G., Ma, J. (2019, July). A new dynamic security assessment framework
based on semi-supervised learning and data editing. Electric Power Systems
Research, 172 , 221–229, https://doi.org/10.1016/j.epsr.2019.03.009 Retrieved
from https://www.sciencedirect.com/science/article/pii/S0378779619301014

Liu, R., Verbič, G., Ma, J., Hill, D.J. (2018, January). Fast Stability Scanning
for Future Grid Scenario Analysis. IEEE Transactions on Power Systems,
33 (1), 514–524, https://doi.org/10.1109/TPWRS.2017.2694048 Retrieved
from https://ieeexplore.ieee.org/document/7898836 (Conference Name: IEEE
Transactions on Power Systems)

Liu, S., Liu, L., Fan, Y., Zhang, L., Huang, Y., Zhang, T., . . . Mao, D. (2020, July).
An Integrated Scheme for Online Dynamic Security Assessment Based on Partial
Mutual Information and Iterated Random Forest. IEEE Transactions on Smart
Grid , 11 (4), 3606–3619, https://doi.org/10.1109/TSG.2020.2991335 Retrieved
from https://ieeexplore.ieee.org/document/9082004 (Conference Name: IEEE
Transactions on Smart Grid)

Liu, T., Liu, Y., Liu, J., Wang, L., Xu, L., Qiu, G., Gao, H. (2020, September).
A Bayesian Learning Based Scheme for Online Dynamic Security Assessment
and Preventive Control. IEEE Transactions on Power Systems, 35 (5), 4088–
4099, https://doi.org/10.1109/TPWRS.2020.2983477 (Conference Name: IEEE
Transactions on Power Systems)

Liu, X., Zhang, X., Chen, L., Xu, F., Feng, C. (2020, November). Data-
driven Transient Stability Assessment Model Considering Network Topol-
ogy Changes via Mahalanobis Kernel Regression and Ensemble Learn-
ing. Journal of Modern Power Systems and Clean Energy, 8 (6), 1080–
1091, https://doi.org/10.35833/MPCE.2020.000341 Retrieved from
https://ieeexplore.ieee.org/document/9275596 (Conference Name: Journal of
Modern Power Systems and Clean Energy)

Luo, F., Dong, Z., Chen, G., Xu, Y., Meng, K., Chen, Y., Wong, K. (2015, April).
Advanced Pattern Discovery-based Fuzzy Classification Method for Power
System Dynamic Security Assessment. IEEE Transactions on Industrial Infor-
matics, 11 (2), 416–426, https://doi.org/10.1109/TII.2015.2399698 Retrieved

41
 from https://ieeexplore.ieee.org/document/7029678 (Conference Name: IEEE
Transactions on Industrial Informatics)

Luo, Y., Lu, C., Zhu, L., Song, J. (2023, January). Graph Convolu-
tional Network-Based Interpretable Machine Learning Scheme in Smart
Grids. IEEE Transactions on Automation Science and Engineering, 20 (1),
47–58, https://doi.org/10.1109/TASE.2021.3090671 Retrieved from
https://ieeexplore.ieee.org/document/9502909 (Conference Name: IEEE Trans-
actions on Automation Science and Engineering)

Mahdi, M., & Genc, V.M.I. (2017, April). Artificial neural network based algorithm
for early prediction of transient stability using wide area measurements. 2017
5th International Istanbul Smart Grid and Cities Congress and Fair (ICSG) (pp.
17–21). Retrieved from https://ieeexplore.ieee.org/document/7947611

Mahdi, M., & Genc, V.M.I. (2018, November). Post-fault prediction of transient
instabilities using stacked sparse autoencoder. Electric Power Systems Research,
164 , 243–252, https://doi.org/10.1016/j.epsr.2018.08.009

Maihemuti, S., Wang, W., Wang, H., Wu, J., Zhang, X. (2021). Dynamic Security
and Stability Region Under Different Renewable Energy Permeability in IENGS
System. IEEE Access, 9 , 19800–19817, https://doi.org/10.1109/ACCESS
.2021.3049236 Retrieved from https://ieeexplore.ieee.org/document/9314149
(Conference Name: IEEE Access)

Malbasa, V., Zheng, C., Chen, P.-C., Popovic, T., Kezunovic, M. (2017, November).
Voltage Stability Prediction Using Active Machine Learning. IEEE Transactions
on Smart Grid , 8 (6), 3117–3124, https://doi.org/10.1109/TSG.2017.2693394
Retrieved from https://ieeexplore.ieee.org/document/7898513 (Conference
Name: IEEE Transactions on Smart Grid)

Mohammadi, H., & Dehghani, M. (2015, January). PMU based voltage secu-
rity assessment of power systems exploiting principal component analysis and
decision trees. International Journal of Electrical Power & Energy Systems,
64 , 655–663, https://doi.org/10.1016/j.ijepes.2014.07.077 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061514005213

Mohammadi, H., Khademi, G., Dehghani, M., Simon, D. (2018, December).

Voltage stability assessment using multi-objective biogeography-based sub-
set selection. International Journal of Electrical Power & Energy Systems,
103 , 525–536, https://doi.org/10.1016/j.ijepes.2018.06.017 Retrieved from

42
 https://www.sciencedirect.com/science/article/pii/S0142061517319075

Mohammadi, H., Khademi, G., Simon, D., Dehghani, M. (2016, April). Multi-objective
optimization of decision trees for power system voltage security assessment.
2016 Annual IEEE Systems Conference (SysCon) (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/7490524?denied=

Mohammadi, M., Gharehpetian, G.B., Niknam, T. (2010, October). On-

line Small-signal Stability Assessment of Power Systems Using Ball Vec-
tor Machines. Electric Power Components and Systems, 38 (13), 1427–
1445, https://doi.org/10.1080/15325001003735150 Retrieved from
https://doi.org/10.1080/15325001003735150 (Publisher: Taylor & Francis
eprint: https://doi.org/10.1080/15325001003735150)

Mukherjee, R., & De, A. (2020, September). Development of an Ensemble Decision

Tree-Based Power System Dynamic Security State Predictor. IEEE Sys-
tems Journal , 14 (3), 3836–3843, https://doi.org/10.1109/JSYST.2020.2978504
Retrieved from https://ieeexplore.ieee.org/document/9044298 (Conference
Name: IEEE Systems Journal)

M.V, S., & C.K., B. (2014, 01). Fast assessment of voltage stability margin of a power
system. Journal of Electrical Systems, 10 , 305-316,

Naderi, S., Javadi, M., Mazhari, M., Chung, C.Y. (2023, January). A Machine
Learning-Based Framework for Fast Prediction of Wide-Area Remedial Con-
trol Actions in Interconnected Power Systems. IEEE Transactions on Power
Systems, 38 (1), 242–255, https://doi.org/10.1109/TPWRS.2022.3165210
Retrieved from https://ieeexplore.ieee.org/document/9750882 (Conference
Name: IEEE Transactions on Power Systems)

Nakas, G.A., Dirik, A., Papadopoulos, P.N., Matavalam, A.R.R., Paul, O., Tzelepis, D.
(2023). Online Identification of Cascading Events in Power Systems With Renew-
able Generation Using Measurement Data and Machine Learning. IEEE Access,
11 , 72343–72356, https://doi.org/10.1109/ACCESS.2023.3294472 Retrieved
from https://ieeexplore.ieee.org/document/10179902 (Conference Name: IEEE
Access)

Nandanwar, S.R., Kolhe, M.L., Warkad, S.B., Patidar, N.P., Singh, V.K.
(2018, July). Voltage Security Assessment by Using PFDT and CBR
Methods in Emerging Power System. Energy Procedia, 144 , 170–
181, https://doi.org/10.1016/j.egypro.2018.06.023 Retrieved from

43
 https://www.sciencedirect.com/science/article/pii/S1876610218301292

Nandanwar, S.R., & Warkad, S.B. (2016, December). Voltage Security Assessment
with Application of PMUs Using Decision Tree. 2016 8th International Confer-
ence on Computational Intelligence and Communication Networks (CICN) (pp.
365–369). Retrieved from https://ieeexplore.ieee.org/document/8082668 (ISSN:
2472-7555)

Nie, Z., Yang, D., Centeno, V., Jones, K.D. (2017, September). A
PMU-based voltage security assessment framework using hoeffding-tree-
based learning. 2017 19th International Conference on Intelligent Sys-
tem Application to Power Systems (ISAP) (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/8071402

Niu, S., Huo, C., Ke, X., Wei, P., Ren, C., Zhang, G., . . . Sun, H. (2021, December).
Research on Power System Transient Security Prediction Based on AdaBoost-
SVM. 2021 IEEE Sustainable Power and Energy Conference (iSPEC) (pp. 3975–
3981).

Nuqui, R., Phadke, A., Schulz, R., Bhatt, N. (2001, January). Fast on-line volt-
age security monitoring using synchronized phasor measurements and decision
trees. 2001 IEEE Power Engineering Society Winter Meeting. Conference Pro-
ceedings (Cat. No.01CH37194) (Vol. 3, pp. 1347–1352 vol.3). Retrieved from
https://ieeexplore.ieee.org/document/917282

Ortiz-Villalba, D., Rahmann, C., Alvarez, R., Canizares, C.A., Strunck, C.

(2020). Practical Framework for Frequency Stability Studies in Power
Systems With Renewable Energy Sources. IEEE Access, 8 , 202286–
202297, https://doi.org/10.1109/ACCESS.2020.3036162 Retrieved from
https://ieeexplore.ieee.org/document/9249237 (Conference Name: IEEE
Access)

Pannell, Z., Ramachandran, B., Snider, D. (2018, February). Machine learning

approach to solving the transient stability assessment problem. 2018 IEEE
Texas Power and Energy Conference (TPEC) (pp. 1–6). Retrieved from
https://ieeexplore.ieee.org/document/8312089

Pérez-Londoño, S.M., Olivar-Tost, G., Mora-Florez, J.J. (2017, April).

Online determination of voltage stability weak areas for situational
awareness improvement. Electric Power Systems Research, 145 , 112–
121, https://doi.org/10.1016/j.epsr.2016.12.026 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0378779616305399

44
Qiao, J., Wang, X., Ni, J., Shi, M., Ren, H., Chen, E. (2021, December). Graph Neural
Network Based Transient Stability Assessment Considering Topology Changes.
2021 International Conference on Power System Technology (POWERCON)
(pp. 1999–2003). (ISSN: 2642-6226)

Rahmann, C., Ortiz-Villalba, D., Álvarez, R., Salles, M. (2017, July). Methodology
for selecting operating points and contingencies for frequency stability studies.
2017 IEEE Power & Energy Society General Meeting (pp. 1–5). Retrieved from
https://ieeexplore.ieee.org/abstract/document/8274423 (ISSN: 1944-9933)

Rahmatian, M., Chen, Y.C., Palizban, A., Moshref, A., Dunford, W.G. (2017,
January). Transient stability assessment via decision trees and multivari-
ate adaptive regression splines. Electric Power Systems Research, 142 ,
320–328, https://doi.org/10.1016/j.epsr.2016.09.030 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0378779616303935

Ren, C., Wang, T., Yu, H., Xu, Y., Dong, Z.Y. (2023, September). EFed-
DSA: An Efficient Differential Privacy-Based Horizontal Federated Learning
Approach for Smart Grid Dynamic Security Assessment. IEEE Journal
on Emerging and Selected Topics in Circuits and Systems, 13 (3), 817–
828, https://doi.org/10.1109/JETCAS.2023.3293253 Retrieved from
https://ieeexplore.ieee.org/document/10183839 (Conference Name: IEEE Jour-
nal on Emerging and Selected Topics in Circuits and Systems)

Ren, C., & Xu, Y. (2019, November). A Fully Data-Driven Method Based on
Generative Adversarial Networks for Power System Dynamic Security Assess-
ment With Missing Data. IEEE Transactions on Power Systems, 34 (6),
5044–5052, https://doi.org/10.1109/TPWRS.2019.2922671 Retrieved from
https://ieeexplore.ieee.org/document/8736295 (Conference Name: IEEE Trans-
actions on Power Systems)

Ren, C., & Xu, Y. (2020, January). Transfer Learning-Based Power Sys-
tem Online Dynamic Security Assessment: Using One Model to Assess
Many Unlearned Faults. IEEE Transactions on Power Systems, 35 (1),
821–824, https://doi.org/10.1109/TPWRS.2019.2947781 Retrieved from
https://ieeexplore.ieee.org/document/8871201 (Conference Name: IEEE Trans-
actions on Power Systems)

Ren, C., & Xu, Y. (2023, May). A Universal Defense Strategy for Data-Driven
Power System Stability Assessment Models Under Adversarial Examples. IEEE
Internet of Things Journal , 10 (9), 7568–7576, https://doi.org/10.1109/JIOT
.2022.3202267 Retrieved from https://ieeexplore.ieee.org/document/9868826

45
 (Conference Name: IEEE Internet of Things Journal)

Ren, C., Xu, Y., Zhang, Y., Zhang, R. (2020, June). A Hybrid Random-
ized Learning System for Temporal-Adaptive Voltage Stability Assessment
of Power Systems. IEEE Transactions on Industrial Informatics, 16 (6),
3672–3684, https://doi.org/10.1109/TII.2019.2940098 Retrieved from
https://ieeexplore.ieee.org/document/8827296 (Conference Name: IEEE Trans-
actions on Industrial Informatics)

Ren, C., Yuan, H., Li, Q., Zhang, R., Xu, Y. (2023, November). Pre-Fault
Dynamic Security Assessment of Power Systems for Multiple Different Faults
via Multi-Label Learning. IEEE Transactions on Power Systems, 38 (6),
5501–5511, https://doi.org/10.1109/TPWRS.2022.3223166 Retrieved from
https://ieeexplore.ieee.org/document/9954911 (Conference Name: IEEE Trans-
actions on Power Systems)

Rovnyak, S., Kretsinger, S., Thorp, J., Brown, D. (1994, August). Decision trees
for real-time transient stability prediction. IEEE Transactions on Power
Systems, 9 (3), 1417–1426, https://doi.org/10.1109/59.336122 Retrieved
from https://ieeexplore.ieee.org/document/336122 (Conference Name: IEEE
Transactions on Power Systems)

Rovnyak, S., Taylor, C., Sheng, Y. (2000, October). Decision trees using apparent
resistance to detect impending loss of synchronism. IEEE Transactions on Power
Delivery, 15 (4), 1157–1162, https://doi.org/10.1109/61.891496 Retrieved
from https://ieeexplore.ieee.org/document/891496 (Conference Name: IEEE
Transactions on Power Delivery)

Sajan, K.S., Kumar, V., Tyagi, B. (2015a, December). Genetic algo-

rithm based support vector machine for on-line voltage stability monitor-
ing. International Journal of Electrical Power & Energy Systems, 73 ,
200–208, https://doi.org/10.1016/j.ijepes.2015.05.002 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061515002070

Sajan, K.S., Kumar, V., Tyagi, B. (2015b, November). ICA based Arti-
ficial Neural Network model for voltage stability monitoring. TENCON
2015 - 2015 IEEE Region 10 Conference (pp. 1–3). Retrieved from
https://ieeexplore.ieee.org/document/7372938 (ISSN: 2159-3450)

46
Sajan, K.S., Tyagi, B., Kumar, V. (2014, December). Genetic algorithm based
artificial neural network model for voltage stability monitoring. 2014 Eigh-
teenth National Power Systems Conference (NPSC) (pp. 1–5). Retrieved from
https://ieeexplore.ieee.org/document/7103798

Sawhney, H., & Jeyasurya, B. (2006, August). A feed-forward artificial

neural network with enhanced feature selection for power system tran-
sient stability assessment. Electric Power Systems Research, 76 (12),
1047–1054, https://doi.org/10.1016/j.epsr.2005.12.026 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0378779606000034

Senroy, N., Heydt, G., Vittal, V. (2006, November). Decision Tree Assisted
Controlled Islanding. IEEE Transactions on Power Systems, 21 (4), 1790–
1797, https://doi.org/10.1109/TPWRS.2006.882470 Retrieved from
https://ieeexplore.ieee.org/document/1717583 (Conference Name: IEEE Trans-
actions on Power Systems)

Sepehr, A., Gomis-Bellmunt, O., Pouresmaeil, E. (2022, May). Employing Machine

Learning for Enhancing Transient Stability of Power Synchronization Control
During Fault Conditions in Weak Grids. IEEE Transactions on Smart Grid ,
13 (3), 2121–2131, https://doi.org/10.1109/TSG.2022.3148590 Retrieved
from https://ieeexplore.ieee.org/document/9701593 (Conference Name: IEEE
Transactions on Smart Grid)

Shah, H., & Verma, K. (2016, March). PMU-ANN based approach

for real time voltage stability monitoring. 2016 IEEE 6th Interna-
tional Conference on Power Systems (ICPS) (pp. 1–5). Retrieved from
https://ieeexplore.ieee.org/document/7584137

Shahzad, U. (2021, September). Probabilistic Transient Stability Assessment of Power

Systems Using Artificial Neural Network. Journal of Electrical Engineering,
Electronics, Control and Computer Science, 8 (1), 35–42, (Number: 1)

Shahzad, U. (2022a, January). Application of supervised machine learn-

ing for prediction of probabilistic transient stability. Australian Jour-
nal of Electrical and Electronics Engineering, 19 (1), 65–78, https://doi
.org/10.1080/1448837X.2021.2013418 (Publisher: Taylor & Francis eprint:
https://doi.org/10.1080/1448837X.2021.2013418)

Shahzad, U. (2022b, June). Artificial Neural Network For Transient Stability

Assessment: A Review. arXiv. (arXiv:2206.06800 [cs, eess])

47
Shahzad, U. (2022c, April). A comparative analysis of artificial neural network
and support vector machine for online transient stability prediction consider-
ing uncertainties. Australian Journal of Electrical and Electronics Engineering,
19 (2), 101–116, https://doi.org/10.1080/1448837X.2021.2022999 (Publisher:
Taylor & Francis eprint: https://doi.org/10.1080/1448837X.2021.2022999)

Sharifian, A., & Sharifian, S. (2015, January). A new power system tran-
sient stability assessment method based on Type-2 fuzzy neural network
estimation. International Journal of Electrical Power & Energy Systems,
64 , 71–87, https://doi.org/10.1016/j.ijepes.2014.07.007 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061514004396

Shi, Z., Yao, W., Zeng, L., Wen, J., Fang, J., Ai, X., Wen, J. (2020, April). Convo-
lutional neural network-based power system transient stability assessment and
instability mode prediction. Applied Energy, 263 , 114586, https://doi.org/
10.1016/j.apenergy.2020.114586

Siddiqui, S.A., Verma, K., Niazi, K.R., Fozdar, M. (2018, January). Real-Time
Monitoring of Post-Fault Scenario for Determining Generator Coherency and
Transient Stability Through ANN. IEEE Transactions on Industry Applications,
54 (1), 685–692, https://doi.org/10.1109/TIA.2017.2753176 Retrieved from
https://ieeexplore.ieee.org/abstract/document/8038834 (Conference Name:
IEEE Transactions on Industry Applications)

Song, S., Jung, Y., Han, C., Jung, S., Yoon, M., Jang, G. (2020).
Recurrent Neural-Network-Based Maximum Frequency Deviation Prediction
Using Probability Power Flow Dynamic Tool. IEEE Access, 8 , 182054–
182064, https://doi.org/10.1109/ACCESS.2020.3028707 Retrieved from
https://ieeexplore.ieee.org/document/9212358 (Conference Name: IEEE
Access)

Suganyadevi, M.V., Babulal, C.K., Kalyani, S. (2016, February). Assessment of volt-

age stability margin by comparing various support vector regression models.
Soft Computing, 20 (2), 807–818, https://doi.org/10.1007/s00500-014-1544-x
Retrieved from https://doi.org/10.1007/s00500-014-1544-x

Suliman, S.I., & Rahman, T.K.A. (2010, June). Artificial immune system based
machine learning for voltage stability prediction in power system. 2010 4th
International Power Engineering and Optimization Conference (PEOCO) (pp.
53–58). Retrieved from https://ieeexplore.ieee.org/document/5559230

48
Sulistiawati, I.B., Priyadi, A., Qudsi, O.A., Soeprijanto, A., Yorino, N. (2016,
May). Critical Clearing Time prediction within various loads for tran-
sient stability assessment by means of the Extreme Learning Machine
method. International Journal of Electrical Power & Energy Systems, 77 ,
345–352, https://doi.org/10.1016/j.ijepes.2015.11.034 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061515004548

Sun, K., Likhate, S., Vittal, V., Kolluri, S., Mandal, S. (2008, July). An online
dynamic security assessment scheme using phasor measurements and decision
trees. 2008 IEEE Power and Energy Society General Meeting - Conversion and
Delivery of Electrical Energy in the 21st Century (pp. 1–1). Retrieved from
https://ieeexplore.ieee.org/document/4596398 (ISSN: 1932-5517)

Suprême, H., Dessaint, L.-A., Kamwa, I., Heniche-Oussédik, A. (2018, July). Eval-
uation for Voltage New Predictors Based on the Concept of Center of Power
for Transient and Dynamic Instability Detection. IEEE Transactions on Smart
Grid , 9 (4), 3605–3615, https://doi.org/10.1109/TSG.2016.2636816 Retrieved
from https://ieeexplore.ieee.org/document/7776913 (Conference Name: IEEE
Transactions on Smart Grid)

Taeho, J. (2021). Machine learning foundations: Supervised, unsupervised, and

advanced learning. Springer Cham.

Tan, B., Yang, J., Pan, X., Li, J., Xie, P., Zeng, C. (2017, January). Representa-
tional learning approach for power system transient stability assessment based
on convolutional neural network. The Journal of Engineering, 2017 (13), 1847–
1850, https://doi.org/10.1049/joe.2017.0651 Retrieved from https://digital-
library.theiet.org/content/journals/10.1049/joe.2017.0651 (Publisher: IET
Digital Library)

Tan, B., & Zhao, J. (2023, September). Debiased Uncertainty Quantification Approach
for Probabilistic Transient Stability Assessment. IEEE Transactions on Power
Systems, 38 (5), 4954–4957, https://doi.org/10.1109/TPWRS.2023.3276207
Retrieved from https://ieeexplore.ieee.org/document/10124366 (Conference
Name: IEEE Transactions on Power Systems)

Tang, Y., Cui, H., Wang, Q. (2017, October). Prediction Model of

the Power System Frequency Using a Cross-Entropy Ensemble Algorithm.
Entropy, 19 (10), 552, https://doi.org/10.3390/e19100552 Retrieved from
https://www.mdpi.com/1099-4300/19/10/552 (Number: 10 Publisher: Multi-
disciplinary Digital Publishing Institute)

49
Teeuwsen, S., Erlich, I., El-Sharkawi, M. (2005, June). Small-signal stability assess-
ment for large power systems using computational intelligence. IEEE Power
Engineering Society General Meeting, 2005 (pp. 2661–2668 Vol. 3). Retrieved
from https://ieeexplore.ieee.org/document/1489375 (ISSN: 1932-5517)

Tian, F., Zhou, X., Yu, Z., Shi, D., Chen, Y., Huang, Y. (2019,
May). A preventive transient stability control method based on sup-
port vector machine. Electric Power Systems Research, 170 , 286–
293, https://doi.org/10.1016/j.epsr.2019.01.030 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0378779619300422

Velayati, M.H., Amjady, N., Khajevandi, I. (2015, July). Prediction of dynamic volt-
age stability status based on Hopf and limit induced bifurcations using extreme
learning machine. International Journal of Electrical Power & Energy Systems,
69 , 150–159, https://doi.org/10.1016/j.ijepes.2015.01.005 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0142061515000241

Voumvoulakis, E., Gavoyiannis, A., Hatziargyriou, N. (2006, June). Decision trees

for dynamic security assessment and load shedding scheme. 2006 IEEE Power
Engineering Society General Meeting (pp. 7 pp.–). (ISSN: 1932-5517)

Wang, B., Fang, B., Wang, Y., Liu, H., Liu, Y. (2016, September).
Power System Transient Stability Assessment Based on Big Data and
the Core Vector Machine. IEEE Transactions on Smart Grid , 7 (5),
2561–2570, https://doi.org/10.1109/TSG.2016.2549063 Retrieved from
https://ieeexplore.ieee.org/abstract/document/7445227 (Conference Name:
IEEE Transactions on Smart Grid)

Wang, C., Li, B., Liu, C., Li, P. (2021). Convolutional neural network-based
power system frequency security assessment. IET Energy Systems Inte-
gration, 3 (3), 250–262, https://doi.org/10.1049/esi2.12021 Retrieved
from https://onlinelibrary.wiley.com/doi/abs/10.1049/esi2.12021 ( eprint:
https://onlinelibrary.wiley.com/doi/pdf/10.1049/esi2.12021)

Wang, Q., Li, F., Tang, Y., Xu, Y. (2019, November). Integrating Model-
Driven and Data-Driven Methods for Power System Frequency Stability
Assessment and Control. IEEE Transactions on Power Systems, 34 (6),
4557–4568, https://doi.org/10.1109/TPWRS.2019.2919522 Retrieved from
https://ieeexplore.ieee.org/document/8723612 (Conference Name: IEEE Trans-
actions on Power Systems)

50
Wang, T.w., Guan, L., Zhang, Y. (2008, July). A modified pattern recognition algo-
rithm and its application in power system transient stability assessment. 2008
IEEE Power and Energy Society General Meeting - Conversion and Delivery of
Electrical Energy in the 21st Century (pp. 1–7). (ISSN: 1932-5517)

Wang, Y., & Pal, B.C. (2023, November). Destabilizing Attack and Robust Defense for
Inverter-Based Microgrids by Adversarial Deep Reinforcement Learning. IEEE
Transactions on Smart Grid , 14 (6), 4839–4850, https://doi.org/10.1109/TSG
.2023.3263243 Retrieved from https://ieeexplore.ieee.org/document/10089185
(Conference Name: IEEE Transactions on Smart Grid)

Wang, Y., Silva-Saravia, H., Pulgar-Painemal, H. (2019, July). Actuator

Placement for Enhanced Grid Dynamic Performance: A Machine Learn-
ing Approach. IEEE Transactions on Power Systems, 34 (4), 3119–
3128, https://doi.org/10.1109/TPWRS.2019.2895019 Retrieved from
https://ieeexplore.ieee.org/document/8625444 (Conference Name: IEEE Trans-
actions on Power Systems)

Xie, J., & Sun, W. (2021). A Transfer and Deep Learning-Based Method
for Online Frequency Stability Assessment and Control. IEEE Access, 9 ,
75712–75721, https://doi.org/10.1109/ACCESS.2021.3082001 Retrieved
from https://ieeexplore.ieee.org/document/9435326 (Conference Name: IEEE
Access)

Xu, C., Liang, J., Yun, Z., Zhang, L. (2005, August). The Small-disturbance Voltage
Stability Analysis through Adaptive AR Model Based on PMU. 2005 IEEE/PES
Transmission & Distribution Conference & Exposition: Asia and Pacific (pp.
1–5). Retrieved from https://ieeexplore.ieee.org/abstract/document/1547097
(ISSN: 2160-8644)

Xu, Y., Dai, Y., Dong, Z.Y., Zhang, R., Meng, K. (2013, March). Extreme
learning machine-based predictor for real-time frequency stability assess-
ment of electric power systems. Neural Computing and Applications, 22 (3),
501–508, https://doi.org/10.1007/s00521-011-0803-3 Retrieved from
https://doi.org/10.1007/s00521-011-0803-3

Xu, Y., Dong, Z.Y., Meng, K., Zhang, R., Wong, K.P. (2011, March).
Real-time transient stability assessment model using extreme learning
machine. IET Generation, Transmission & Distribution, 5 (3), 314–322,
https://doi.org/10.1049/iet-gtd.2010.0355 Retrieved from https://digital-
library.theiet.org/content/journals/10.1049/iet-gtd.2010.0355 (Publisher: IET
Digital Library)

51
Xu, Y., Dong, Z.Y., Zhao, J.H., Zhang, P., Wong, K.P. (2012, August). A
Reliable Intelligent System for Real-Time Dynamic Security Assessment
of Power Systems. IEEE Transactions on Power Systems, 27 (3), 1253–
1263, https://doi.org/10.1109/TPWRS.2012.2183899 Retrieved from
https://ieeexplore.ieee.org/document/6158623?denied= (Conference Name:
IEEE Transactions on Power Systems)

Xu, Y., Guan, L., Dong, Z.Y., Wong, K.P. (2010, October). Transient stability assess-
ment on China Southern Power Grid system with an improved pattern discovery-
based method. 2010 International Conference on Power System Technology (pp.
1–6). Retrieved from https://ieeexplore.ieee.org/document/5666045

Xu, Y., Zhang, R., Zhao, J., Dong, Z.Y., Wang, D., Yang, H., Wong, K.P. (2016,
August). Assessing Short-Term Voltage Stability of Electric Power Systems
by a Hierarchical Intelligent System. IEEE Transactions on Neural Networks
and Learning Systems, 27 (8), 1686–1696, https://doi.org/10.1109/TNNLS
.2015.2441706 Retrieved from https://ieeexplore.ieee.org/document/7284703
(Conference Name: IEEE Transactions on Neural Networks and Learning
Systems)

Yang, H., Zhang, W., Chen, J., Wang, L. (2018, July). PMU-based
voltage stability prediction using least square support vector machine
with online learning. Electric Power Systems Research, 160 , 234–
242, https://doi.org/10.1016/j.epsr.2018.02.018 Retrieved from
https://www.sciencedirect.com/science/article/pii/S0378779618300610

Yang, Y., Huang, Y., Liu, J., Liu, Y., Liu, T., Xiang, Y. (2017, Septem-
ber). Measurement-based cell-DT method for power system transient sta-
bility classification. CSEE Journal of Power and Energy Systems, 3 (3),
278–285, https://doi.org/10.17775/CSEEJPES.2015.01230 Retrieved from
https://ieeexplore.ieee.org/document/8057662 (Conference Name: CSEE Jour-
nal of Power and Energy Systems)

Yousefian, R., Bhattarai, R., Kamalasadan, S. (2017, November). Transient Stability

Enhancement of Power Grid With Integrated Wide Area Control of Wind Farms
and Synchronous Generators. IEEE Transactions on Power Systems, 32 (6),
4818–4831, https://doi.org/10.1109/TPWRS.2017.2676138 Retrieved from
https://ieeexplore.ieee.org/abstract/document/7867871 (Conference Name:
IEEE Transactions on Power Systems)

52
Yousefian, R., & Kamalasadan, S. (2016, August). A Lyapunov function based optimal
hybrid power system controller for improved transient stability. Electric Power
Systems Research, 137 , 6–15, https://doi.org/10.1016/j.epsr.2016.03.042

Yousefian, R., & Kamalasadan, S. (2018, March). Energy Function Inspired Value
Priority Based Global Wide-Area Control of Power Grid. IEEE Transactions
on Smart Grid , 9 (2), 552–563, https://doi.org/10.1109/TSG.2016.2555909
(Conference Name: IEEE Transactions on Smart Grid)

Yu, J.J.Q., Hill, D.J., Lam, A.Y.S., Gu, J., Li, V.O.K. (2018, January). Intelli-
gent Time-Adaptive Transient Stability Assessment System. IEEE Transactions
on Power Systems, 33 (1), 1049–1058, https://doi.org/10.1109/TPWRS.2017
.2707501 Retrieved from https://ieeexplore.ieee.org/document/7932885 (Con-
ference Name: IEEE Transactions on Power Systems)

Yu, J.J.Q., Lam, A.Y.S., Hill, D.J., Li, V.O.K. (2017). Delay Aware Intel-
ligent Transient Stability Assessment System. IEEE Access, 5 , 17230–
17239, https://doi.org/10.1109/ACCESS.2017.2746093 Retrieved from
https://ieeexplore.ieee.org/document/8017541 (Conference Name: IEEE
Access)

Yuanhang, D., Lei, C., Weiling, Z., Yong, M. (2015, November). Multi-
support vector machine power system transient stability assessment
based on relief algorithm. 2015 IEEE PES Asia-Pacific Power and
Energy Engineering Conference (APPEEC) (pp. 1–5). Retrieved from
https://ieeexplore.ieee.org/document/7381006?denied=

Zhai, C., Nguyen, H.D., Zong, X. (2023, April). Dynamic Security Assessment
of Small-Signal Stability for Power Grids Using Windowed Online Gaus-
sian Process. IEEE Transactions on Automation Science and Engineering,
20 (2), 1170–1179, https://doi.org/10.1109/TASE.2022.3173368 Retrieved
from https://ieeexplore.ieee.org/document/9773283 (Conference Name: IEEE
Transactions on Automation Science and Engineering)

Zhang, C., Li, Y., Yu, Z., Tian, F. (2016, October). Feature selection of power
system transient stability assessment based on random forest and recursive fea-
ture elimination. 2016 IEEE PES Asia-Pacific Power and Energy Engineering
Conference (APPEEC) (pp. 1264–1268).

Zhang, R., Wu, J., Shao, M., Li, B., Lu, Y. (2018, October). Transient Stability
Prediction of Power Systems Based on Deep Belief Networks. 2018 2nd IEEE
Conference on Energy Internet and Energy System Integration (EI2) (pp. 1–6).

53
 Retrieved from https://ieeexplore.ieee.org/document/8582609

Zhang, R., Xu, Y., Dong, Z.Y., Hill, D.J. (2012, July). Feature selec-
tion for intelligent stability assessment of power systems. 2012 IEEE
Power and Energy Society General Meeting (pp. 1–7). Retrieved from
https://ieeexplore.ieee.org/document/6344780 (ISSN: 1944-9925)

Zhang, R., Xu, Y., Dong, Z.Y., Wong, K.P. (2015, January). Post-disturbance
transient stability assessment of power systems by a self-adaptive intelligent
system. IET Generation, Transmission & Distribution, 9 (3), 296–305,
https://doi.org/10.1049/iet-gtd.2014.0264 Retrieved from https://digital-
library.theiet.org/content/journals/10.1049/iet-gtd.2014.0264 (Publisher: IET
Digital Library)

Zhang, R., Xu, Y., Dong, Z.Y., Zhang, P., Wong, K.P. (2013, July). Voltage sta-
bility margin prediction by ensemble based extreme learning machine. 2013
IEEE Power & Energy Society General Meeting (pp. 1–5). Retrieved from
https://ieeexplore.ieee.org/document/6672489 (ISSN: 1932-5517)

Zhang, R., Yao, W., Shi, Z., Ai, X., Tang, Y., Wen, J. (2023a, November). Encoding
Time Series as Images: A Robust and Transferable Framework for Power System
DIM Identification Combining Rules and VGGNet. IEEE Transactions on Power
Systems, 38 (6), 5781–5793, https://doi.org/10.1109/TPWRS.2023.3234287
Retrieved from https://ieeexplore.ieee.org/document/10006407 (Conference
Name: IEEE Transactions on Power Systems)

Zhang, R., Yao, W., Shi, Z., Ai, X., Tang, Y., Wen, J. (2023b, November). Towards
Multi-Scenario Power System Stability Analysis: An Unsupervised Transfer
Learning Method Combining DGAT and Data Augmentation. IEEE Transac-
tions on Power Systems, 38 (6), 5367–5380, https://doi.org/10.1109/TPWRS
.2022.3220569 Retrieved from https://ieeexplore.ieee.org/document/9942331
(Conference Name: IEEE Transactions on Power Systems)

Zhang, T., Sun, M., Cremer, J.L., Zhang, N., Strbac, G., Kang, C. (2021, Septem-
ber). A Confidence-Aware Machine Learning Framework for Dynamic Security
Assessment. IEEE Transactions on Power Systems, 36 (5), 3907–3920, https://
doi.org/10.1109/TPWRS.2021.3059197 (Conference Name: IEEE Transactions
on Power Systems)

Zhang, Y., Dong, Z.Y., Zhang, R., Xu, Y. (2018, May). A Noise-
Tolerant Temporal-Adaptive Approach for Short-Term Voltage Stabil-
ity Assessment of Power Systems. 2018 IEEE Innovative Smart
Grid Technologies - Asia (ISGT Asia) (pp. 62–67). Retrieved from

54
 https://ieeexplore.ieee.org/document/8467941?denied=

Zhang, Y., Li, T., Na, G., Li, G., Li, Y. (2015). Optimized Extreme Learning Machine
for Power System Transient Stability Prediction Using Synchrophasors. Math-
ematical Problems in Engineering, 2015 , 1–8, https://doi.org/10.1155/2015/
529724 Retrieved from http://arxiv.org/abs/1810.08652 (arXiv:1810.08652 [cs,
math])

Zhang, Y., Xu, Y., Dong, Z.Y. (2018, January). Robust Ensemble Data
Analytics for Incomplete PMU Measurements-Based Power System Sta-
bility Assessment. IEEE Transactions on Power Systems, 33 (1), 1124–
1126, https://doi.org/10.1109/TPWRS.2017.2698239 Retrieved from
https://ieeexplore.ieee.org/document/7913716 (Conference Name: IEEE Trans-
actions on Power Systems)

Zhang, Y., Xu, Y., Dong, Z.Y., Xu, Z., Wong, K.P. (2017, October). Intelligent
Early Warning of Power System Dynamic Insecurity Risk: Toward Optimal
Accuracy-Earliness Tradeoff. IEEE Transactions on Industrial Informatics,
13 (5), 2544–2554, https://doi.org/10.1109/TII.2017.2676879 Retrieved
from https://ieeexplore.ieee.org/document/7869388 (Conference Name: IEEE
Transactions on Industrial Informatics)

Zhang, Y., Xu, Y., Dong, Z.Y., Zhang, R. (2019, January). A Hierarchical Self-
Adaptive Data-Analytics Method for Real-Time Power System Short-Term
Voltage Stability Assessment. IEEE Transactions on Industrial Informat-
ics, 15 (1), 74–84, https://doi.org/10.1109/TII.2018.2829818 Retrieved
from https://ieeexplore.ieee.org/document/8345652 (Conference Name: IEEE
Transactions on Industrial Informatics)

Zhang, Y., Xu, Y., Zhang, R., Dong, Z.Y. (2019, September). A Missing-
Data Tolerant Method for Data-Driven Short-Term Voltage Stability Assess-
ment of Power Systems. IEEE Transactions on Smart Grid , 10 (5),
5663–5674, https://doi.org/10.1109/TSG.2018.2889788 Retrieved from
https://ieeexplore.ieee.org/document/8588381 (Conference Name: IEEE Trans-
actions on Smart Grid)

Zhang, Z., Sun, M., Deng, R., Kang, C., Chow, M.-Y. (2023, January).
Physics-Constrained Robustness Evaluation of Intelligent Security Assess-
ment for Power Systems. IEEE Transactions on Power Systems, 38 (1),
872–884, https://doi.org/10.1109/TPWRS.2022.3169139 Retrieved from

55
 https://ieeexplore.ieee.org/document/9761730 (Conference Name: IEEE Trans-
actions on Power Systems)

Zhao, T., Yue, M., Wang, J. (2023a, October). Deep-Learning-Based Koop-

man Modeling for Online Control Synthesis of Nonlinear Power System
Transient Dynamics. IEEE Transactions on Industrial Informatics, 19 (10),
10444–10453, https://doi.org/10.1109/TII.2023.3240939 Retrieved from
https://ieeexplore.ieee.org/document/10032610 (Conference Name: IEEE
Transactions on Industrial Informatics)

Zhao, T., Yue, M., Wang, J. (2023b, November). Robust Power System Sta-
bility Assessment Against Adversarial Machine Learning-Based Cyberattacks
via Online Purification. IEEE Transactions on Power Systems, 38 (6),
5613–5622, https://doi.org/10.1109/TPWRS.2022.3233735 Retrieved from
https://ieeexplore.ieee.org/document/10005029 (Conference Name: IEEE
Transactions on Power Systems)

Zheng, C., Malbasa, V., Kezunovic, M. (2013, May). Regression tree for stability mar-
gin prediction using synchrophasor measurements. IEEE Transactions on Power
Systems, 28 (2), 1978–1987, https://doi.org/10.1109/TPWRS.2012.2220988
Retrieved from https://ieeexplore.ieee.org/document/6331026?denied= (Con-
ference Name: IEEE Transactions on Power Systems)

Zhou, Y., Wu, J., Yu, Z., Ji, L., Hao, L. (2016, October). A Hierarchical Method
for Transient Stability Prediction of Power Systems Using the Confidence of
a SVM-Based Ensemble Classifier. Energies, 9 (10), 778, https://doi.org/10
.3390/en9100778 Retrieved from https://www.mdpi.com/1996-1073/9/10/778
(Number: 10 Publisher: Multidisciplinary Digital Publishing Institute)

Zhou, Y., & Zhang, P. (2023, January). Noise-Resilient Quantum Machine Learning for
Stability Assessment of Power Systems. IEEE Transactions on Power Systems,
38 (1), 475–487, https://doi.org/10.1109/TPWRS.2022.3160384 Retrieved
from https://ieeexplore.ieee.org/document/9737359 (Conference Name: IEEE
Transactions on Power Systems)

Zhu, L., & Hill, D.J. (2022, January). Networked Time Series Shapelet Learn-
ing for Power System Transient Stability Assessment. IEEE Transactions
on Power Systems, 37 (1), 416–428, https://doi.org/10.1109/TPWRS.2021
.3093423 Retrieved from https://ieeexplore.ieee.org/document/9468376 (Con-
ference Name: IEEE Transactions on Power Systems)

56
Zhu, L., Hill, D.J., Lu, C. (2021, October). Intelligent Short-Term Voltage Stability
Assessment via Spatial Attention Rectified RNN Learning. IEEE Transactions
on Industrial Informatics, 17 (10), 7005–7016, https://doi.org/10.1109/TII
.2020.3041300 Retrieved from https://ieeexplore.ieee.org/document/9273219
(Conference Name: IEEE Transactions on Industrial Informatics)

Zhu, L., Hill, D.J., Lu, C. (2022, May). Semi-Supervised Ensemble Learning
Framework for Accelerating Power System Transient Stability Knowledge Base
Generation. IEEE Transactions on Power Systems, 37 (3), 2441–2454, https://
doi.org/10.1109/TPWRS.2021.3117402 (Conference Name: IEEE Transactions
on Power Systems)

Zhu, L., Lu, C., Dong, Z.Y., Hong, C. (2017, October). Imbalance Learning Machine-
Based Power System Short-Term Voltage Stability Assessment. IEEE Transac-
tions on Industrial Informatics, 13 (5), 2533–2543, https://doi.org/10.1109/TII
.2017.2696534 Retrieved from https://ieeexplore.ieee.org/document/7906492
(Conference Name: IEEE Transactions on Industrial Informatics)

Zhu, L., & Luo, Y. (2021, July). Deep Feedback Learning Based Predictive Control
for Power System Undervoltage Load Shedding. IEEE Transactions on Power
Systems, 36 (4), 3349–3361, https://doi.org/10.1109/TPWRS.2020.3048681
Retrieved from https://ieeexplore.ieee.org/document/9312447 (Conference
Name: IEEE Transactions on Power Systems)

57