GA-GRGAT: A novel deep learning model for high-speed train axle temperature long term forecasting

Long-term axle temperature prediction plays a significant role in train condition assessment and daily maintenance. However, most of methods make predictions for short-term conditions. In this paper, a method named GA-GRGAT is introduced that uses GAT and GAN to forecast the long-term axle temperature. In the proposed method, a GRGAT framework is used as a spatiotemporal fusion in temperature prediction. The GAN network with the GRGAT framework is used to construct a temporal conditional sequence after analyzing the periodic variation of the axle temperature, which can fuse the historical axle temperature information to improve the long-term prediction accuracy of the GA-GRGAT model. Our method in Python software and using the actual high-speed trains datasets in spring and summer. We using MAE, RMSE, MAPE, PCC to evaluate the accuracy of prediction. The accuracy of the GA-GRGAT is more than 90% on long-term predictions (1 day), and more than 80% on super long-term predictions (2 week). The GA-GRGAT method outperforms and is more accurate than the classical forecasting methods such as GRU, GOAMLP, DCNN, SVR and HA. In addition, the cost time of the proposed method is less than 5 min, which meets the requirements of high accuracy and long time.

Automated summary

This paper introduces a method named GA-GRGAT that uses GAT and GAN for long-term axle temperature prediction. By fusing historical axle temperature information, the proposed method improves prediction accuracy. Evaluation on actual high-speed trains datasets demonstrates that the method achieves high accuracy with short computation time.