Adaptive Fractional-Order Model-Free Control for Permanent-Magnet Linear Synchronous Motor Drive System via Reinforcement Learning

In this paper a novel adaptive fractional-order model-free integral sliding-mode control (AFO-MFISMC) scheme for permanent magnet linear synchronous motor (PMLSM) drive is proposed. First, A fractional-order ultra-local model (FOULM) is developed to formulate PMLSM dynamics for simplifying the controller design. Next, a FO-ISMC is designed to stabilize the PMLSM. To enhance the transient behavior, the total disturbance is estimated using high-accuracy fractional-order extended state observer (FO-ESO). For improving the robustness of the proposed control scheme, a new adaptive updating algorithm based on reinforcement learning (RL) approach is developed for online estimation of the FO-ESO parameters and control input gains. An actor/critic learning structure is utilized for the adaptive autotuning mechanism. Besides, a single radial-basis function network (RBFN) is established to approximate the policy function of actor and the value function of critic concurrently for increasing the learning efficiency and decreasing the required storage space. The temporal difference (TD) error performance index and gradient-descent algorithm are utilized for updating rules of RBFN kernel function and network weights in an adaptive manner. The stability and convergence conditions of the proposed AFO-MFISMC is proved by Lyapunov function. The validation results ensure the effectiveness and superiorities of the AFO-MFISMC compared with sliding-mode control.