Rotor Temperature Prediction of PMSM Based on LSTM Neural Networks

Abstract

The rotor of the permanent magnet synchronous motor develops localized high temperatures at high-torque or high-speed operating conditions so that the demagnetization failure phenomenon may occur. To address this problem, a rotor temperature prediction model based on long-and-short-term memory (LSTM) neural networks is proposed. In addition, the effects of several hyperparameters on the network construction are investigated. To better improve the accuracy of prediction results, Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) are used to optimize the construction of the network parameters. The results of the study show that the LSTM model has a large error throughout the process, which ranges from − 2.66–6.64 °C. GA-LSTM has an error of − 1.71 ~ 3.91 ℃ throughout the process. The error of PSO-LSTM is − 1.78 ~ 0.96 ℃. Additionally, the proposed PSO-LSTM prediction model exhibits good accuracy and stability with RMSE of 0.7114 and MAPE of 1.22%.