Fault text classification of on-board equipment in high-speed railway based on labeled-Doc2vec and BiGRU

Abstract

Fault text classification is a prerequisite task for railway engineers based historical train operation data to diagnose vehicle on-board equipment (VOBE) faults and formulate maintenance strategies. Aiming at the low efficiency and accuracy of manual fault text classification, based on Bidirectional Gated Recurrent Unit (BiGRU) and improved attention mechanism (IAtt), an intelligent VOBE fault text classification method is proposed in this paper. Combining the characteristics of the VOBE faults text, also called application event log (AElog) files, the Labeled-Doc2vec is used to generate sentence embedding to realize the vectorized representation of the fault texts, then input sentence embedding into BiGRU to extract the fault text features as the improved attention mechanism layer. Finally, the high-dimensional fault text features outputted by hidden are input into Softmax to complete the fault text classification. The experimental results show that the proposed method can analyze the semantics of fault text according to the train running state before and after the fault time, that is, it can realize text classification by combining context. Compared with other methods, the method in this paper obtains the optimal accuracy, precision, recall and F1-score, which shows that the proposed method can be applied to fault text classification of VOBE, effectively reduces the labor cost of fault text classification in practice, and improves the efficiency of fault text classification of VOBE.