This paper introduces an innovative control strategy for Permanent Magnet Synchronous Motors (PMSMs) by combining the established Direct Torque Control (DTC) technique with Deep Reinforcement Learning (DRL), utilizing the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. The adaptation of the TD3 algorithm is particularly effective in managing the complex state and action spaces encountered in PMSM drive systems. By substituting the traditional Proportional-Integral (PI) controller in DTC's outer loop with a DRL-based TD3 agent, this approach addresses the inherent limitations of PI controllers, such as reliance on accurate models and the difficulty of tuning parameters. The DRL framework enables the TD3 agent to develop optimal control strategies through learned experiences, catering to the intricate and non-linear dynamics of control tasks. Moreover, this work incorporates Model Reference Adaptive System (MRAS) for refined speed estimation, enriching the control accuracy and reducing the costs associated with speed sensors. Comprehensive simulation results validate the TD3-based control method's superior performance in precise speed control across various conditions, contributing significant advancements to the field of PMSM drive systems and setting a foundation for further research in electric drive control technologies.
Enhanced Direct Torque Control of Sensorless PMSM Drive with TD3 Agent-Based Speed Controller
2024-07-31
1790337 byte
Conference paper
Electronic Resource
English