Proposed OLTEM model demonstrates reduced mean squared error in motor temperature prediction, suggesting enhanced accuracy over traditional models.
Key Points
OLTEM exhibits lower mean squared error than both the baseline thermal neural network and CNN–RNN models for PMSM temperature.
Results show that mean squared error stays below 4.0 °C², indicating reliable performance across thermal components of PMSM.
Ablation studies confirm that both the state-conditioned squeeze-and-excitation mechanism and enhanced power-loss module are essential for accuracy improvement.
The model's approach combining physics with learned mechanisms enhances temperature prediction while maintaining interpretability.