Adaptive Reinforcement Learning-Based Speed and Torque Control for Intelligent Electrical Drive Systems
Abstract
Intelligent electrical drive systems play a pivotal role in modern industrial automation, electric vehicles, and robotics. Traditional linear controllers, such as Proportional-Integral (PI) and Proportional-Integral-Derivative (PID), have been the industry standard for decades. However, they inherently struggle with non-linearities, unknown external load disturbances, and internal parametric variations (such as thermal stator resistance drift and magnetic saturation). To overcome these limitations, this comprehensive research paper proposes a novel Adaptive Reinforcement Learning (RL) framework for continuous, real-time speed and torque control in Permanent Magnet Synchronous Motor (PMSM) drive systems. Specifically, a Deep Deterministic Policy Gradient (DDPG) algorithm is synthesized and trained within a Markov Decision Process (MDP) environment to formulate optimal voltage command signals, entirely replacing the traditional cascaded speed current loop architecture. Through extensive hardware-calibrated numerical simulations, the proposed RL agent demonstrates superior dynamic performance compared to finely tuned Field Oriented Control (FOC) PI strategies. The RL controller reduces speed overshoot by 100%, decreases steady-state settling time by 68.4%, and suppresses torque ripple by 55.2% during sudden load torque injections. Furthermore, the adaptive nature of the neural network continuous actor-critic architecture ensures robust operational stability even under 40% variations in stator resistance and rotor inertia. By bridging advanced machine learning techniques with complex electromechanical physics, this research establishes a highly resilient, autonomous control paradigm for next-generation intelligent electrical drives.
KEYWORDS: Adaptive Control, Deep Deterministic Policy Gradient, Intelligent Electrical Drives, Reinforcement Learning, Field Oriented Control, Torque Ripple Minimization.
Full Text:
PDF 67-80Refbacks
- There are currently no refbacks.