Multi-Task Learning and Attention Mechanism Based Long Short-Term Memory for Temperature Prediction of EMU Bearing

The traction motor is one of the key components that plays an important role in ensuring the safety and stability of the running EMU (Electric Multiple Units). The running state of the traction motor can be determined through monitoring and predicting the change of EMU bearing temperature. In this paper, we propose a Long Short-Term Memory Neural Network based on Multi-task Learning and Attention Mechanism for the bearing temperature prediction in view of the complex influencing factors of bearing temperature in train operation. The model learns the characteristics of temperature sensors in different positions jointly through multi-task learning. And the Long Short-Term Memory Neural Network based on Attention Mechanism is used to consider the influence of current operating conditions and previous train records on bearing temperature in different degrees. So the model takes various influencing factors and spatial-temporal correlation into consideration. The experimental results with actual EMU datasets show that our method outperforms the baseline approaches.