Abstract: Aiming at the non-linearity and complexity of electromagnetic coil temperature rise during differential operation and the problems existing in the prediction of the traditional BP neural network, the improved grey wolf optimization(GWO) was adopted to optimize the BP neural network. According to the data of the field tracking experiment in a room temperature environment, a network prediction model is established, which takes running time, DC current and running power as input and the temperature rise difference between the real-time temperature and the initial ambient temperature after eight hours continuous operation of the electromagnetic coil of an electronic lock differential as output. MAE, MAPE and MSE were selected as system evaluation indexes, and compared with the traditional BP network model and the network model optimized by particle swarm optimization algorithm (PSO-BP), the results showed that GWO-BP neural network model had better prediction ability and smaller error accuracy. It provides a method and idea for realizing the accurate prediction of the temperature rise of the electromagnetic coil on the automobile axle differential.