Feature Selection Method for Power System Transient Stability Status Prediction Considering Class Imbalanced Characteristic

Power system transient stability status prediction (TSSP) has class imbalanced characteristic, including the imbalance of sample size and class importance. Therefore, a new feature selection method is proposed for TSSP. On the basis of analyzing the imbalanced characteristic of TSSP, the basic criteria of feature selection are put forward, and then the index of class weighted accuracy is proposed as the evaluation index. Under the guidance of the new evaluation index, the wrapper method combing sequential forward search and extreme learning machine is utilized for TSSP feature selection. The effectiveness of the proposed method is verified on Northeast Power Coordinated Council (NPCC) 48-machine 140-bus system.