Fault Prediction Method for on-board Equipment of CTCS based on Seq2Seq Model using LSTM

Efficient fault prediction can improve system reliability and reduce maintenance costs. With the rapid development of deep learning, more and more people apply related methods in the field of fault prediction. However, there are few effective methods to predict the fault for on-board equipment of Chinese Train Control System (CTCS). In this paper, we proposed a fault prediction method for on-board equipment of CTCS based on the seq2seq model using Long Short-Term Memory (LSTM). Firstly, we use a time series generator to capture the possible feature of the time to the subsequent fault from historical fault data. Then, these features are fed into the seq2seq model, which has an encoder-decoder structure. The encoder and decoder each use a layer of LSTM network, and a linear layer is used to realize the fault prediction of the on-board equipment of CTCS. Finally, we verified the method on the maintenance data from 2015 to 2019. The result shows that compared with other methods, this method has better effect on three performance metrics.