Deep Reinforcement Learning-Aided Efficiency Optimized Dual Active Bridge Converter for the Distributed Generation System

• Published in: IEEE transactions on energy conversion Vol. 37; no. 2; pp. 1251 - 1262
• Main Authors: Tang, Yuanhong, Hu, Weihao, Zhang, Bin, Cao, Di, Hou, Nie, Li, Yunwei, Chen, Zhe, Blaabjerg, Frede
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.06.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Bridge circuits; Bridges; deep deterministic policy gradient (DDPG); Deep learning; deep reinforcement learning (DRL); Distributed generation; dual active bridge (DAB) converter; Efficiency; Efficiency optimization control; Electric bridges; Electric converters; Energy conversion; Inductors; Machine learning; Modulation; Optimization; Switches; Training; Zero voltage switching
• ISSN: 0885-8969; 1558-0059
• DOI: 10.1109/TEC.2021.3126754

With the development of power system electronization, distributed new energy generation-based DC microgrid system has become a hot spot of the current research. Aiming to implement the collection, transformation and transmission of the new energy sources, the dual active bridge (DAB) converter has become a key energy conversion device. Aiming to promote the transmitted efficiency for such DAB converter, a deep reinforcement learning (DRL) optimization scheme with the triple-phase-shift (TPS) is proposed. Specifically, the deep deterministic policy gradient (DDPG) algorithm is used for solving optimal solutions with the minimum power losses. Then, the trained DDPG agent is used as a fast surrogate predictor, which giving the suitable control decisions under whole operation conditions. The main control objective is to minimize the power losses for getting the target maximum efficiency. Besides, the soft switching constrains are added to the DDPG algorithm. Based on this, the proposed optimized method can be used for different operation conditions with excellent performance. The operation principles, the loss analyses, and the training consideration and evaluation of the DDPG algorithm are given. Finally, simulation and experimental results using a 200 W hardware prototype are displayed and compared to prove to effectiveness of theoretical analysis.