Compensation Capacitor Status Monitoring Research Based on Feature Fusion and SVM

In order to meet the needs of railway electrical departments for “state repair” of track circuit compensation capacitors and timely and effective monitoring of compensation capacitor status, this paper proposes a new method that combines the feature quantities decomposed from CEEMD and LMD algorithms and utilizes support vector machines for compensation capacitor status monitoring. Firstly, a ZPW-2000A track circuit model is established using KCL, KVL, and transmission line theory. By changing the capacitance value of the compensation capacitor, the shunt current curves of the compensation capacitor in each state are simulated. Then, the shunt current curves are decomposed into each order component using CEEMD and LMD, and fuzzy entropy is calculated and combined into a new feature vector. Finally, it is input into a trained multi-class SVM model for state monitoring. The experimental results show that the accuracy of compensating capacitor state monitoring is improved to 91% for a single decomposed feature after fusion.