AC drive control design with reinforcement learning techniques

Abstract

Three-phase induction motors (IM) are widely known in industrial applications for their low cost and minimal maintenance. Control of AC induction motors based on Field-Oriented Control (FOC) and vector control with classic PID control laws are reliable techniques commonly used in industry. However, due to the nonlinear nature of electric machines and their susceptibility to external disturbances or parameter variations, conventional controllers often struggle to meet control requirements. Reinforcement Learning (RL) is an online learning technique, model-free, capable of handling parameter variations. These characteristics make reinforcement learning a potential candidate, acting as an adaptive controller that can replace conventional controllers.

This article proposes the design of a control system for three-phase induction motor drives based on reinforcement learning techniques. The proposed controller utilizes a reinforcement learning agent with the DDPG algorithm to replace the PI controllers of the current loop circuit in conventional FOC control. The performance of the controller has been validated under various operating conditions through computer simulations in MATLAB/SIMULINK.