A heterogeneous transfer learning method for fault prediction of railway track circuit

Abstract

Prediction and identification of faults in track circuits are crucial for improving the safety and efficiency of railway transportation. However, due to the absence of real data, the task of track circuit fault prediction through deep learning methods facing significant challenges. This paper proposed a novel heterogeneous transfer learning network structure for track circuit deep learning fault prediction. The proposed transfer learning network can reduce the reliance on track circuit data in the process of deep learning models training by utilizing public datasets from other similar tasks. In this paper, an index describing the data distribution is used to demonstrate the transferability between heterogeneous data firstly. Then a heterogeneous transfer learning network structure is proposed to help the deep learning model training on the track circuit fault prediction task. Finally, the effect of transfer learning is comprehensively examined. The simulation experimental results show that the proposed heterogeneous transfer learning network structure can transfer useful knowledge in other similar fields for tasks in track circuit fault prediction, and the resulting model can distinguish between nine different classes with a high accuracy level over 99% on the test dataset while reducing the amount of required training data to 10% of the traditional training methods.