Comparative Analysis of Learning-Based Methods for Transient Stability Assessment

TL;DR

This paper compares learning-based methods for predicting critical clearing time in power systems, introducing new definitions and a hybrid feature selection strategy to improve efficiency and accuracy.

Contribution

It introduces new engineering-based definitions of transient stability and CCT, and employs a hybrid feature selection method to enhance model performance.

Findings

Hybrid feature selection improves model efficiency.

Learning-based methods effectively predict CCT under uncertainties.

New definitions provide clearer engineering insights.

Abstract

Transient stability and critical clearing time (CCT) are important concepts in power system protection and control. This paper explores and compares various learning-based methods for predicting CCT under uncertainties arising from renewable generation, loads, and contingencies. Specially, we introduce new definitions of transient stability (B-stablilty) and CCT from an engineering perspective. For training the models, only the initial values of system variables and contingency cases are used as features, enabling the provision of protection information based on these initial values. To enhance efficiency, a hybrid feature selection strategy combining the maximal information coefficient (MIC) and Spearman's Correlation Coefficient (SCC) is employed to reduce the feature dimension. The performance of different learning-based models is evaluated on a WSCC 9-bus system.