As the fleet of high-speed rail vehicles expands, ensuring train safety is of the utmost importance, emphasizing the critical need to enhance the precision of axel temperature warning systems. Yet, the limited availability of data on the unique features of high thermal axis temperature conditions in railway systems hinders the optimal performance of intelligent algorithms in alarm detection models. To address these challenges, this study introduces a novel dynamic principal component analysis preprocessing technique for tolerance temperature data to effectively manage missing data and outliers. Furthermore, a customized generative adversarial network is devised to generate distinct data related to high thermal axis temperature, focusing on optimizing the network’s objective functions and distinctions to bolster the efficiency and diversity of the generated data. Finally, an integrated model with an optimized transformer module is established to accurately classify alarm levels, provide a comprehensive solution to pressing train safety issues, and, in a timely manner, notify drivers and maintenance departments (DEPOs) of high-temperature warnings.
The accuracy of axle-box bearing temperature prediction is important for monitoring the operating condition of high-speed electric multiple units. However, temperature data collected in service often contain missing values, abnormal fluctuations, and noise, which can affect the reliability of the prediction results. To address this problem, this paper proposes an Attention-BP-LSTM framework that combines data preprocessing with temperature prediction. First, missing data are completed by linear interpolation. The dynamically augmented temperature sequence is then processed using time-delay low-rank sparse decomposition to separate the normal temperature variation trend, sparse anomalies, and measurement noise, based on which the abnormal data are reconstructed. After Z-score normalization, the processed time series is converted into supervised learning samples using a sliding window. The prediction model consists of LSTM, an attention mechanism, Dropout, and a BP network. LSTM is used to extract the temporal dependencies in the temperature sequence, the attention mechanism assigns corresponding weights to different historical features, Dropout alleviates model overfitting, and the BP network completes the nonlinear mapping from temporal features to the predicted temperature. Experimental results based on axle-box bearing temperature data show that, compared with BP, LSTM, and BP-LSTM, the proposed model achieves lower MAE and RMSE and a higher R2. This indicates that improving data quality before prediction, together with attention-based temporal feature extraction, helps improve the reliability of short-term axle-box bearing temperature prediction.