The GRA-two algorithm for massive-scale feature selection problem in power system scenario classification and prediction

Feature selection is a powerful tool for choosing a feature subset of relevant attributes and has been widely used in many research fields, including power system. In this paper, we have introduced a two-step feature selection algorithm that combines the advantages of Grey Relation Analysis (GRA) and Binary Particle Swarm Optimization (BPSO) search method. The proposed method aims to solve the problem of massive-scale feature selection in power system and find these attributes which are highly related to the target power system scenario. This algorithm would eliminate some features based on GRA correlation coefficient in step 1, and the remaining features would accept further selection in step 2, in the meanwhile, the modified initialization rule based on GRA coefficient would be used to enhance the optimization speed and improve the performance of the final feature subset. The effectiveness of the selected feature subset is evaluated using the classification and prediction accuracy. After some experiments based on actual power system scenario data, our method has shown strong ability to find a subset with high classification accuracy and low dimension, while the predictor also has better forecasting performance when using the selected feature subset, which would help operators to judge the state of the power system, so that they could make some more accurate decisions to improve the safety and stability of the grid. (C) 2021 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the International Conference on Power Engineering, ICPE, 2020.