Classification and Prediction of AC Contactor Degradation States Based on K-Means Clustering and Bayes-BiLSTM

Aiming at the problems of difficulty in characterizing the contactor degradation process by a single characteristic parameter, lack of basis for contactor degradation state delineation, and low state prediction accuracy, this paper proposes an AC contactor degradation state delineation and prediction method based on K-means clustering and Bayes optimized Bi-directional long-short term memory neural network (BiLSTM). Firstly, the signal data of the full-life test process of the contactor are collected to extract the electrical and mechanical parameters of the contactor. Then, the degradation indexes of mechanical and electrical parameters are constructed by principal component analysis (PCA), and the joint characteristic vectors are constructed. Secondly, the optimal contactor degradation state is classified by K-means clustering. Finally, the Bayes-BiLSTM method is built to predict the degradation state of the contactor. After comparative analysis, it can be seen that the method of this paper can better complete the division and prediction of the contactor degradation state, the degradation state of 4 presents the optimal clustering effect, and the average prediction accuracy of the degradation state is 97.3%.