A PINN-Based Controller for Dual-Plane Voltage Predictive Control of Five-Phase PMSMs

This article proposes a novel physics-informed neural network (PINN) controller for five-phase permanent magnet synchronous motors (PMSMs) that addresses the limitations of existing dual-plane current control methods. The notable constraints of existing machine learning controllers for PMSMs, particularly with respect to interpretability and stability, are examined to enable offline interpretability and online deployability. A neural network (NN) control model is derived for dual-plane operation of five-phase PMSMs by extending the NN control frameworks of three-phase PMSMs. The motor’s prior physical information is deeply integrated into the training process to enhance generalization. A hybrid optimization strategy is proposed to improve the fit between the predicted and actual operating voltages and to suppress prediction fluctuations, thereby better aligning the predicted values with the exact operational requirements. Experimental results validate the excellent steady-state performance and robust antiinterference performance of the proposed controller.