Multi-Objective Controller Design for Grid-Following Converters With Easy Transfer Reinforcement Learning
Yu Zeng, Shan Jiang, Georgios Konstantinou, Josep Pou, Guibin Zou, Xin Zhang
Abstract
This paper proposes an easy transfer reinforcement learning (ETRL) method that combines easy transfer learning with deep reinforcement learning to adapt a multi-objective controller tailor-made for one grid-following converter to other converters with different system parameters. The ETRL method contains five stages: system description, deep reinforcement learning, easy transfer learning, experimental data fine-tuning, and online implementation. The ETRL method can transfer knowledge effectively between controllers, offering a scalable solution for transferring knowledge between different converters without relying on extensive data or hyperparameter tuning. The ETRL method enhances controller adaptability, reduces training requirements by 96.4%, and ensures the stability of converter systems across diverse operating conditions. Experimental results validate the effectiveness of the proposed ETRL method, promising a new direction for power electronics controller design.