CBTC on-board signal fault diagnosis method based on LightGBM classification

Abstract

Aiming at the problem that the semantics of CBTC on-board equipment fault record text is not precise and the word redundancy, which makes it difficult to trace the cause of the fault, this paper proposes a CBTC on-board signal fault diagnosis method based on LightGBM classification. Firstly, the relationship between appearance and fault is analyzed by combining the knowledge graph search formed by manually combing the text; then, TF-IDF is used to extract the original text features, and Doc2vec is used to realize text vectorization. The actual fault text records are divided into training sets and testing sets. The LightGBM classifier is trained to obtain the classification and diagnosis model, and 1133 testing sets are tested and verified. The results show that the accuracy of classification diagnosis of the method proposed in this paper is 90.2%, which is 17.8% higher than that of SVM classification diagnosis and conforms to the manual graph fault analysis link.