Fault Detection for ZPW-2000A Jointless Track Circuit Based on Deep Belief Network Optimized by Grey Wolf Optimizer Algorithm

Abstract

At present, the fault detection for jointless track circuit still depends on the experience of electrical personnel, which leads to low maintenance efficiency and the phenomenon of misjudgment and missed judgment. A fault detection method for track circuit based on deep belief network (DBN) is proposed in this paper. According to the working principle and transmission characteristics of track circuit, 12 voltage and current monitoring parameters are selected to detect 15 types of track circuit fault. However, the selection of structural parameters for deep belief network is time-consuming, so grey wolf optimizer algorithm (GWO) is proposed to optimize DBN, which can adaptively select the number of neurons in each hidden layer. The simulation results show that by introducing the GWO algorithm to optimize the DBN detection model, the fault classification accuracy of the ZPW-2000A jointless track circuit can reach 96.96%, which significantly improves the fault detection level of the track circuit.