Fault Diagnosis of Compensation Capacitor in Joint-less Track Circuit via Long Short-Term Memory Network

The fault detection and diagnosis of compensation capacitors in the ZPW-2000 Joint-less Track Circuit (JTC) are quite crucial and challenging for Chinese railway safety. In this work, we present a novel method based on Long Short-Term Memory (LSTM) network for the compensation capacitor disconnection fault detection. We tackle the problem based on the compensation capacitors relationship in spatial dimension and the temporal correlation between the normal and fault state of the capacitors. Specifically, to acquire more realistic sequence data, the JTC is firstly modeled and verified with measured data, and then extract the feature data from shunt current based on the JTC model. The obtained data are fed into the LSTM network to detect the compensation capacitor disconnection fault. The experiments show that the proposed method achieves competitive and robust results.