基于分合闸线圈电流和触头行程融合的断路器异常辨识方法研究OA

This paper addresses the problem of multi-source signal acquisition and analysis in circuit breaker mechanical abnormal diagnosis.A data acquisition system for closing and opening coil current and contact travel signals is designed,along with corresponding abnormal simulation schemes.Feature extraction is performed using segmented filtering and cyclic difference discrimination,and the CatBoost machine learning algorithm is employed for abnormal diagnosis based on single-source signals.Parameter optimization is achieved using a genetic algorithm(GA),enabling highly accurate diagnosis based on coil current and contact travel signals.Furthermore,linear discriminant analysis(LDA)is used for feature-level fusion to enhance diagnostic performance.The diagnostic results of various fusion methods are analyzed,revealing that both the LDA-GA-CatBoost feature-level fusion method and the improved dempster-shafer(D-S)evidence theory-based decision-level fusion method achieve the highest abnormal diagnosis accuracy of 95.82%.However,the model training time for LDA-GA-CatBoost is only half that of the improved D-S evidence theory method,indicating a clear advantage in practical applications.