Short and long term memory method for predicting the temperature of motor stator based on Harris eagle algorithm optimization

Abstract

The temperature of a motor significantly affects its control and lifespan. However, due to the influence of motor structure and operating environment, precise temperature measurement of the motor is challenging with temperature sensors. Therefore, machine learning algorithms are often employed to predict the temperature more accurately. To enhance motor control, integrating machine learning algorithm models with the actual motor control terminal is highly beneficial. This paper proposes a Short and Long Term Memory (LSTM) algorithm model based on Harris's hawk optimization to predict the temperature of the motor stator, which is applied in actual motor control. Furthermore, it evaluates the tracking performance of motor control current. Firstly, an experimental platform for temperature measurement is established to acquire the temperature at different positions of the motor as raw data. Subsequently, the raw data is inputted into three algorithms: LSTM, PSO-LSTM, and HHO-LSTM, for comparison. By comparing evaluation metrics, it is demonstrated that HHO-LSTM exhibits excellent predictive performance. Furthermore, utilizing diverse segments of the motor as model input sets enhances the generalization capability and predictive accuracy of the model. Finally, applied in actual motor control, a comparative analysis of q-axis and d-axis currents reveals reduced fluctuation errors, showcasing improved motor control effectiveness.