Reinforcement Learning and Singular Perturbation-Based Optimal Speed Synchronous Control of a Flexible Coupling Dual-PMSM System

The dual-motor system is leveraged as a feasible option to improve both fault tolerance and power output in practical applications. The conventional control methods either rely on model parameters or lack optimization capabilities. In this article, a model-free reinforcement learning (RL)-based optimal speed synchronous control methodology is imposed for the flexible coupling dual permanent magnet synchronous motor (PMSM) system with time-varying reference signals. Singular perturbation theory is used to capture the reduced-order slow time-scale dynamics of the dual-PMSM system. To track the time-varying reference signal, we formulate an optimal output regulation problem for speed regulator design regarding such a reduced-order model and then an optimal dynamic regulator is obtained. In addition, an advanced model-free RL algorithm, which does not require the knowledge of model parameters, is developed to learn the optimal speed regulator iteratively. Comparative experimental results indicate that our control strategy can achieve better speed synchronous tracking and ameliorate transient response performance than existing schemes.