Research on text classification method for metro train control onboard equipment logs based on Capsule network

Log analysis of metro onboard equipment in train control system, which heavily relies on manual retrieval, leads to inefficient fault localization and negatively impacts the safe-ty and efficiency of train transportation. A classification model for onboard equipment logs based on multi-head attention and Capsule Network is proposed. Considering the characteristics of onboard log texts, the preprocessed and augmented text is transformed into vector representations by bidirectional encoder representations from transformers, which are then fed into the multi-head attention layer to focus on strongly correlated features. Finally, these features are input into the Capsule Network for text classification. To address the severe class imbalance in the dataset, a novel class-balanced focal loss function is proposed, which dynamically adjusts sample weights during the training phase to mitigate the imbalance. Experimental results demonstrate that the model effectively mitigates the impact of data im-balance on classification performance and achieves accurate classification of onboard equipment log texts. Compared with other models, the proposed model outperforms in terms of precision, recall, and F1-score, validating its effectiveness in providing decision support for fault localization in metro onboard equipment.