Deep Reinforcement Learning-based Intelligent Sliding-mode Control for Permanent Magnet Synchronous Motor Speed Regulation

This paper proposes a novel intelligent controlframework for precise speed regulation of permanent magnetsynchronous motors (PMSMs). To achieve fast convergence ofthe tracking error, a novel sliding-mode manifold is designedby employing the inverse tangent function, resulting in thecorresponding baseline controller and disturbance observer. Thisdesign ensures that the controlled speed tracking system theoretically converges rapidly after precise compensation of thelumped disturbance. To further reduce the conservatism of theproposed baseline controller, the Deep Q-Network (DQN) isemployed to autonomously optimize the speed tracking dynamics,simultaneously enhancing both transient response and steadystate accuracy. After a detailed stability analysis of the closedloop system, physical experiments demonstrate the superior speedtracking performance and practical applicability of the proposedmethod with a reasonable setting of the reward function.