基于Crossformer高速铁路轨道质量指数预测方法OA

[Objective]The track quality index(TQI)directly reflects the service status of high-speed railway track structures,and accurately predicting its development trend can provide more precise guidance for the maintenance of high-speed railway lines.This study proposes a TQI prediction method based on the Crossformer artificial intelligence model.[Methods]First,the dimension hierarchical embedding method was used to embed temporal dependencies and dependencies among different features into the TQI sequence.Then,a hierarchical encoder centered on cross-time and cross-dimensional attention mechanisms was used to encode the dimension hierarchical embedded TQI sequence layer by layer at different scales,thereby extracting key information from the sequence.Finally,a hierarchical decoder was used to decode the encoded TQI sequence information at different scales layer by layer to obtain the final prediction results.The proposed Crossformer model was validated using TQI data from four different track lines,and its TQI prediction accuracy was compared with those of the SVM,CNN,and LSTM models.Additionally,the effects of different output lengths of the Crossformer model on TQI prediction accuracy were investigated.[Results]The results showed that,compared with the SVM,CNN,and LSTM models,the MAE of the Crossformer model decreased by 46.3％,34.7％,and 42.8％,respectively,and the RMSE decreased by 51.5％,43.3％,and 45.8％,respectively.This indicated that the Crossformer method achieved higher TQI prediction accuracy than the other three traditional models on all four track lines.The prediction performance was better when the input length of the Crossformer model was 4 or 6.[Conclusion]The proposed Crossformer method can effectively extract feature information from the TQI sequence at temporal and dimensional scales,thereby achieving accurate prediction of the TQI development trend and providing new technical support for the"condition-based maintenance"of high-speed railway lines.