Adaptive Control of EV Traction Inverter Using Deep Reinforcement Learning

Abstract

The traction inverter is a critical component in Electric Vehicle (EV) powertrains, responsible for efficient energy conversion and precise motor control. Traditional control strategies, such as Proportional-Integral (PI) controllers and Model Predictive Control (MPC), often struggle with parameter variations, non-linearities, and complex operating conditions. This paper proposes an adaptive control strategy for EV traction inverters utilizing Deep Reinforcement Learning (DRL). Specifically, w e e mploy a D eep D eterministic P olicy Gradient (DDPG) agent to dynamically optimize the switching signals of a three-phase Voltage Source Inverter (VSI) driving a Permanent Magnet Synchronous Motor (PMSM). The DRL agent learns to minimize current tracking errors and torque ripples directly from interaction with the environment, without relying on a linearized system model. Simulation results demonstrate that the proposed DRL-based controller achieves superior dynamic performance and robustness against parameter uncertainties compared to conventional Field Oriented Control (FOC) schemes.