Reinforcement learning-based H-infinity current control method and system for permanent magnet synchronous motor

Abstract

The invention discloses a reinforcement learning-based H-infinity current control method and system for a permanent magnet synchronous motor. The method comprises the steps: converting an H-infinity optimal control problem into a double-player zero-sum game problem based on a saddle point theory in combination with a linear discretization mathematical model of the permanent magnet synchronous motor, constructing a GARE equation, and solving the GARE equation; and updating a Q function and a strategy by adopting a reinforcement learning algorithm based on an Actor-Critic framework, and training to obtain an optimal H-infinity controller meeting Nash equilibrium. According to the method, reinforcement learning and H-infinity control are combined, a GARE solution is learned on line in a data driving mode, a motor mathematical model is completely not needed, H-infinity optimal control of a time-varying system is successfully achieved, the method is applied to current control of an IPMSM, the robustness is high, and the performance is far better than that of PI control.

Description

technical field [0001] The present application relates to the technical field of motor control, in particular to a reinforcement learning-based H∞ current control method and system for permanent magnet synchronous motors. Background technique [0002] The interior permanent magnet synchronous motor (Interior Permanent Magnet Synchronous Motor, IPMSM) has the advantages of high power density and large speed range, and is widely used in the field of motion control. The control strategies of the IPMSM drive system mainly include vector control and direct torque control, but they cannot meet the application scenarios with high robustness requirements, such as aircraft control. Although the sliding mode control has strong robustness, the chattering problem caused by the switching of the sliding mode surface cannot be completely eliminated. Active Disturbance Rejection Control (ADRC) can observe and compensate the system disturbance in real time, and the anti-disturbance ability ...

AI Technical Summary

Problems solved by technology

However, the pure model-free reinforcement learning method has a large amount of data and low efficiency, and cannot guarantee control performance and reliability.

Images