Abstract:
Railway signal equipment has accumulated a large amount of text-recorded maintenance data during the operation and maintenance process. In order to realize efficient and precise classification, an automatic classification method of railway signal equipment fault text combining Word2vec, SMOTE algorithm and Convolutional Neural Network(CNN) was proposed in this paper. Firstly, the fault text was preprocessed by natural language methods, and Word2vec was used to train word vector, then text vector data of small category was generated automatically by SMOTE algorithm. Secondly, the generated word vectors were embedded in the input layer of CNN, then convolutional and pooling layer were used to extract high-level features of the local context of the fault text. Finally, softmax classifier was used to complete automatic classification of the fault text data. According to the test analysis of fault text of signal equipment recorded by a railway bureau and comparison with other methods, the test results indicate that this method can obviously upgrade the evaluation indexes, among which classification precision rate and recall rate can reach 95.26% and 94.32% respectively, and it can be used as an effective method for automatic classification of railway signal equipment faults.