Deep Learning Based Online Diagnosis for Railway Interlocking System

Intelligent fault diagnosis based on operational data for railway interlocking systems is now an innovative approach to system maintenance. However, the main barriers to fault diagnosis are high-dimension and long-sequence features of the railway operational data, which makes general deep-learning models inefficient. In this paper, we present a novel deep-learning method of online fault diagnosis for interlocking systems. An online diagnosis server built on a self-attention recurrent neural network (SA-RNN) model was connected to the interlocking system. The problem of long dependency sequences of the data was mitigated by using route logic-based reprocessing. The high-dimensionality problem was resolved via a self-attention mechanism. Finally, experiments demonstrated that the proposed method is persuasive, and the results perform better than traditional models.