A IPMSM Current Control Method Based on Reinforcement Learning

The control problem of Interior Permanent Magnet Synchronous Motor (IPMSM) under external disturbance has always been a difficult problem in the industry, although H∞ control can effectively improve the robustness, it is difficult to solve the Game Algebra Riccati Equation (GARE) analytically due to the time-varying uncertainty of motor parameters. To solve this problem, this paper proposes an off-policy reinforcement learning method, which uses the data-driven way to learn the solution of GARE online, completely without the mathematical model of the motor, successfully realizes the H∞ optimal control of the time-varying system, and applies it to the current control of IPMSM. Firstly, using the saddle point theory of game theory and the linear discrete mathematical model of the motor, the H∞ optimal control problem was transformed into a two-player zero-sum game problem, and the GARE equation was constructed. Then, the reinforcement learning algorithm based on Actor-Critic framework is used to update the Q function and the strategy by using the input and output data of the system, and the optimal H∞ controller satisfying Nash equilibrium is learned. The test results of Processor In Loop (PIL) prove the feasibility of the proposed scheme, and its performance is far superior to PI control.