Temperature Prediction of a Driving Motor based on Particle Swarm Optimization and Long Short-Term Memory Network

Temperature prediction of the driving motor during various operating conditions is crucial to its control and diagnosis. This study proposes a Long Short-Term Memory (LSTM) method for forecasting the temperature of a high-power motor rated at 350 kW. By analyzing the characteristics of motor temperature variation, speed, torque, voltage, and current are selected as input data for temperature prediction. The Particle Swarm Optimization (PSO) algorithm is incorporated into the neural network to analyze the best number of hidden layer nodes and initial learning rate, thereby improving the accuracy of the network. Lastly, results show that the PSO-LSTM method reduces the error (RMSE) by 16.5% on the training dataset and 2.4% on the testing dataset compared to the LSTM network. This approach not only effectively captures the complex dynamics of temperature changes but also provides important references for the safe operation and maintenance of driving motors.