Machine Learning-Driven Optimization of MTPA in IPMSM Drives for Electric Vehicles

This research presented in this paper outlines a pioneering optimization framework concentrated on perfecting the Maximum Torque per Ampere (MTPA) control mechanism in Interior Permanent Magnet Synchronous Motor (IPMSM) drives for the realm of electric vehicles (EVs). Through the integration of supervised learning methodologies for parameter forecasting alongside reinforcement learning (RL) techniques for real-time adaptive regulation, the proposed methodology dynamically enhances torque output and energy efficiency across fluctuating driving conditions. The framework employs a Deep Q-Network (DQN) agent to perpetually recalibrate the direct-axis and quadrature-axis currents, establishing an equilibrium between torque production and energy expenditure in a real-time context. The simulation data reveals a potential rise of up to 15% in torque efficiency, a 10% lessening in energy expended, and a 15% prolongation in battery life when juxtaposed with traditional fixed-parameter approaches. This technique surpasses traditional motor control methodologies by prioritizing real-time adaptability and resilience, thus presenting itself as a viable solution for the forthcoming generation of EVs. Subsequent research will investigate the practical application of the proposed framework in real-world scenarios and its adaptability to more intricate operating conditions.