Improvement of PMSM Control Using Reinforcement Learning Deep Deterministic Policy Gradient Agent
Marcel Nicola, Claudiu-Ionel Nicola
Abstract
Based on the advantage of using the reinforcement learning on process control, provided by the fact that it is not necessary to know the exact mathematical model and the structure of its uncertainties, this article approaches the possibility of improving the performances of the PMSM (Permanent Magnet Synchronous Motor) control system based on the FOC (Field Oriented Control) control strategy, by using the correction signals provided by a trained reinforcement learning agent, which will be added to the control signals u <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d</inf> , u <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">q</inf> , and i <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">qref</inf> . The type of reinforcement learning used is the Deep Deterministic Policy Gradient (DDPG). The combination possibilities of these control structures are presented, and their superiority over the FOC-type control strategy is validated by numerical simulations.