Stator winding short-term temperature prediction of EMU traction motors based on DTCN-CBAM

Accurately predicting the temperature of the stator winding of the traction motor and preventing the accelerated deterioration of the insulation system caused by high temperature plays an important role in prolonging the service time of the motor and reducing the operating cost of the EMU. The condition monitoring data of the traction motor has both temporal and spatial characteristics. However, most of the existing temperature prediction methods do not carefully consider the specific characteristics of time and space in the condition monitoring data of traction motors, resulting in the prediction effect not meeting expectations. Therefore, this paper proposes a prediction model based on the Deep Temporal Convolution Network (DTCN) and Convolution Block Attention Mechanism (CBAM), considering long-term time dependence and global spatial characteristics. First, DTCN is designed to extract multi-dimensional long-term time features in motor parameters. Secondly, CBAM is used to extract spatial features, and the attention weight distribution of multiple channels and spaces is realized at the same time. Finally, the temperature of the motor is predicted by fusing all the features of the nonlinear fully connected layer. Experiments on the motor temperature public dataset on Kaggle verify the superiority of the algorithm.