[Paper Review] Artificial Neural Network For Transient Stability Assessment: A Review

Umair Shahzad|arXiv (Cornell University)|Jun 13, 2022

Power System Optimization and Stability4 citations

TL;DR

This review paper evaluates the application of artificial neural networks (ANNs) for transient stability assessment (TSA) in modern power systems, addressing challenges posed by high renewable integration and system uncertainty. It synthesizes existing research on ANN-based TSA methods, highlighting architectures, training strategies, and performance metrics, and provides a comprehensive foundation for advancing machine learning in power system security and stability.

ABSTRACT

Integration of large-scale renewable energy sources and increasing uncertainty has drastically changed the dynamics of power system and has consequently brought various challenges. Rapid transient stability assessment of modern power system is a vital requirement for accurate power system planning and operation. The conventional methods are unable to fulfil this requirement. Therefore, novel approaches are required in this regard. Machine leaning approaches such as artificial neural network can play a significant role in this regard. Therefore, this paper aims to review the application of artificial neural network for transient stability assessment of power systems. It is believed that this work will provide a solid foundation for researchers in the domain of machine learning applications to power system security and stability.

Motivation & Objective

• Address the growing challenge of transient stability assessment in modern power systems due to high penetration of renewable energy sources and system uncertainty.
• Identify limitations in conventional transient stability assessment methods that hinder real-time application.
• Review and synthesize recent advancements in artificial neural network (ANN) applications for transient stability assessment.
• Provide a structured foundation for researchers exploring machine learning in power system security and stability.
• Highlight key ANN architectures, data preprocessing techniques, and performance evaluation metrics used in TSA research.

Proposed method

• Systematically reviewed peer-reviewed literature on ANN-based transient stability assessment from 2000 to 2022.
• Categorized ANN models based on architecture (e.g., feedforward, recurrent, deep neural networks) and application context.
• Analyzed input feature selection methods, including power system state variables and post-fault trajectories.
• Evaluated training techniques such as backpropagation, optimization algorithms, and data augmentation strategies.
• Compared performance metrics including accuracy, computational speed, and generalization capability across studies.
• Discussed challenges such as interpretability, robustness to unseen contingencies, and real-time deployment constraints.

Experimental results

Research questions

• RQ1 What are the most effective artificial neural network architectures for transient stability assessment in modern power systems?
• RQ2 How do different input feature sets and data preprocessing techniques impact ANN performance in TSA?
• RQ3 What are the key performance metrics used to evaluate ANN-based TSA methods, and how do they compare to conventional approaches?
• RQ4 What are the main challenges in deploying ANNs for real-time transient stability assessment in practical power systems?
• RQ5 How do recent advancements in deep learning and transfer learning enhance the accuracy and robustness of ANN-based TSA?

Key findings

• Feedforward and deep feedforward neural networks are the most widely used architectures for transient stability assessment due to their simplicity and high accuracy.
• ANN-based methods achieve classification accuracies exceeding 95% in multiple benchmark systems, including the IEEE 39-bus and IEEE 118-bus systems.
• Recurrent neural networks (RNNs) and long short-term memory (LSTM) networks show improved performance in capturing temporal dynamics of fault-induced system trajectories.
• Data preprocessing techniques such as normalization and feature selection significantly enhance model generalization and reduce overfitting.
• Despite high accuracy, challenges remain in model interpretability, robustness to unseen contingencies, and real-time deployment on phasor measurement unit (PMU)-based data.
• The review identifies a research gap in standardized benchmarking and cross-system validation of ANN-based TSA models, urging for unified evaluation protocols.