A Reinforcement Learning Method for Control Scheme of Permanent Magnet Synchronous Motor

This brief presents the application of Reinforcement Learning (RL) in control design of a Permanent Magnet Synchronous Motor (PMSM). Based on the conventional Field Oriented Control (FOC) structure, optimal control problem is integrated in the current sub-system with the advantage of satisfying not only the tracking effectiveness but also the optimality effectiveness. Due to the obstacle of directly solving partial Hamilton Jacobi Bellman (HJB) differential equation, the RL algorithm is mobilized to find the approximate optimal controller. In order to achieve the appropriate RL control scheme, the exact linearization is added to obtain the linear model of current sub-system. The actor/critic based RL algorithm is developed for linearized model with the satisfactory of not only the convergence but also the tracking problem. Theoretically, we illustrate that our method optimizes the given cost function. Empirically, the proposed RL based controller for PMSM is verified by simulation studies.