TD3 Algorithm Based Reinforcement Learning Control for Multiple-Input Multiple-Output DC-DC Converters

• Published in: IEEE transactions on power electronics 2024-10, Vol.39 (10), p.12729-12742
• Main Authors: Ye, Jian, Guo, Huanyu, Zhao, Di, Wang, Benfei, Zhang, Xinan
• Format: Article
• Language: English
• Subjects: Cross-regulation; DC-DC power converters; Modulation; Multiplexing; multiport dc–dc converters; reinforcement learning (RL); Resource management; Steady-state; Switches; Vectors
• ISSN: 0885-8993; 1941-0107
• DOI: 10.1109/TPEL.2024.3416911

This article presents a reinforcement learning (RL) controller based on the twin delayed deep deterministic (TD3) policy gradient algorithm for single-inductor multiple-input multiple-output (SI-MIMO) dc-dc converters. The controller aims to address the power allocation challenges arising from parallel input sources and mitigate cross-regulation among multiple output channels. The objective is to enable the converter to exhibit outstanding performance in both steady-state and dynamic regulation during operation. The proposed RL controller is trained using the TD3 algorithm. It directly generates duty cycle control signals based on the observed states from the controlled converter. After applying input-side power allocation modulation and output-side time-multiplexing modulation, the controller generates the switching signals for each switch, completing the closed-loop control of the SI-MIMO converter. To validate the effectiveness of the proposed controller, stability analysis of the RL controller is conducted, and an experimental platform is established. Experimental results demonstrate that the proposed controller exhibits superior control performance. It effectively addresses the challenges associated with input-side power allocation and output voltage cross-regulation in SI-MIMO dc-dc converters.