An Improved TimesNet-Based Temperature Estimation Framework for PMSMs Using Feature Selection and Transfer Learning Methods

High-power-density design is the developing trend of permanent magnet synchronous motors (PMSMs), and the accurate prediction of internal temperature is a practical and necessary scientific research. In this article, the TimesNet model is utilized for the high-precision temperature estimation of PMSMs to capture multiscale thermal dynamics. It is observed that the generalization capability of the TimesNet model could be significantly improved by incorporating a transfer learning (TL) strategy with efficient fine-tuning, which guides the model adaptation across cross-condition and cross-motor scenarios. In addition, the maximal information coefficient (MIC) method is chosen to select dominant thermal features from nonlinear coupled variables, which reduces the computational redundancy and improves the physical interpretability of the selected features. Experimental results demonstrate that the proposed method achieves superior estimation accuracy and robust generalization with reduced training samples.