A soft sensor of stator winding temperature prediction for PMSMs based on extreme learning machine

Permanent magnet synchronous motor (PMSM) plays an effective role in electric vehicle applications. Monitoring PMSM's temperature in real time is critical to its safety and reliability. The traditional method of PMSM temperature monitoring is to install temperature sensors into the motor, and it is a very expensive method. Currently, the lumped-parameter thermal networks (LPTNs) are the appropriate alternative for determining PMSM components' temperatures. However, they lack physical interpretability once the degrees of freedom are reduced in order to meet real-time requirements. The approach based on soft sensors is an efficient and economical way to solve such problems. In this paper, a soft sensor is developed to predict the stator winding temperature using an extreme learning machine (ELM). Furthermore, the Principal Component Analysis (PCA) technique is used to select effective and relevant variables. The performance of the model is evaluated based on five statistical indicators: the correlation coefficient (R
²
), the root relative squared error (RRSE), the mean square error (MSE), the mean absolute error (MAE), and the root-mean-squared error (RMSE). The results showed that the PCA-ELM model has a high efficiency to predict the stator winding temperature with RMSE = 0.0622, MAE = 0.0480, MSE = 0.0039, RRSE = 6.73% and R2 = 0.9955. Moreover, it has low computational complexity. Due to its low computational complexity and high performance, this application could have a direct influence and economic savings on the development and design of PMSMs temperature monitoring systems.