Designing a Double Filter-Encircled Wrappers Feature Selection Algorithm for Transient Stability Prediction Based on Transient Time Series Data

Abstract

Integrating the feature selection process into machine learning-based transient analysis (MLTA) is essential for ensuring precise and timely transient stability prediction (TSP). Therefore, it’s important to prioritize the development of a hybrid feature selection scheme (FSS) to find optimal transient features (OTFs) from high-dimensional transient data within MLTA projects. This study introduces the double filter-encircled wrappers FSS (DFEWFSS) to survive OTFs from transient 28-trajectory data. Each fold of DFEWFSS consists of interconnected encircled wrappers by filters. The filter phase is exerted by relevancy score (RS) and conditional RS (CRS) which stem from mutual information (MI) and conditional MI (CMI). The wrapper phase is conducted by kernelized hyperplane-based classifiers, which are mounted on incremental wrapper subset selection (IWSS) and IWSS with replacement (IWSSr). Following the completion of OTFs, the evaluation of OTFs in TSP is carried out through a cross-validation scenario. The results show that DFEWFSS-based OTFs have a prediction accuracy of 99 % and a processing time of 102.155 milliseconds for TSP.

Keywords

• Optimal transient features (OTFs)
• hybrid feature selection
• transient stability prediction (TSP)