Temperature-Aware Fractional-Order PID Scheduling for PMSM Drives Using Data-Driven Thermal Forecasts

Temperature limits critically affect the safety, efficiency, and lifetime of permanent-magnet synchronous motor (PMSM) drives in electric vehicles. While most prior work focuses on predicting motor temperatures, this paper goes further by embedding learned forecasts directly into the control loop. We propose a temperature-aware fractional-order PID (FO-PID) speed controller whose gains are scheduled online by a multi-target thermal predictor trained on the public Paderborn PMSM dataset. Sequence models (LSTM and Transformer) forecast rotor magnet and stator temperatures at short horizons, and the predictions drive a lightweight scheduling map that adapts Kp,Ki,Kd,λ,μ and applies soft torque limits near thermal boundaries. Step-response and drive-cycle evaluations demonstrate that, contrary to expectations, the scheduled FO-PID did not outperform classical PID. Both fixed FO-PID and FO-PID+Scheduler accumulated significantly higher tracking errors and led to unsafe winding temperature excursions, whereas PID achieved the lowest ITAE, IAE, and ISE, the fastest rise time, and maintained safe thermal margins. Scheduling reduced violations relative to fixed FO-PID but did not close the gap to PID performance. These findings provide two contributions: (i) a reproducible framework for integrating thermal forecasts into FO-PID scheduling, and (ii) a counter-intuitive result showing that classical PID may remain safer and more reliable than FO-PID under thermally constrained EV drive operation. This highlights that FO-PID’s benefits are not universal but depend strongly on tuning and operating conditions, and motivates future work on adaptive scheduling policies and hardware-in-the-loop validation.