摘要详情

ID / 提交时间

80 / 2023-10-12 15:52:43

标题

Research on vehicle rectifier control strategy based on reinforcement learning

关键字

vehicle rectifier； reinforcement learning； dq current decoupling control ；optimal control .

主题及专题

2.Power electronic technology and application > (1)、Power converter topologies and design

状态

终稿

作者

Mingwei Tang / Southwest Jiaotong University

Zhigang Liu / School of Electrical Engineering； Southwest Jiaotong University

摘要

The vehicle rectifier includes various linear, nonlinear and intelligent control strategies. Reinforcement learning compensates traditional control strategies, but these control strategies have various shortcomings. This paper proposes a replacement control strategy based on reinforcement learning, which can effectively solve the shortcomings of previous control strategies. Based on the traditional dq current decoupling control, the voltage loop is removed and all PI controllers are replaced. The reward function, state observation and action output of the dq axis are designed according to the performance index and effect. The double rectifier control system is designed, trained and verified. Finally, in order to increase the explainability of the control based on reinforcement learning, the optimal control theory is used to explain.