| Title |
Optimal Design of PCB Planar Transformers for DAB Converters Based on Reinforcement Learning |
| Authors |
Min-Seung Kim ; Eun-Soo Lee |
| DOI |
https://doi.org/10.6113/TKPE.2025.30.4.340 |
| Keywords |
Machine Learning; Reinforcement Learning; Dual Active Bridge (DAB) Converter; Planar PCB Transformer; Integrated Transformer |
| Abstract |
The conventional high-frequency transformer used in dual active bridge (DAB) converters typically employs Litz wire, which presents manufacturability challenges and necessitates the use of additional inductors. This paper proposes a reinforcement learning-based design optimization methodology that provides optimal design results for planar transformers with PCB windings, eliminating the need for external inductors and overcoming their inherent limitations. The proposed optimization framework incorporates a deep Q-network algorithm to determine the optimal coil widths on the primary and secondary sides, thereby minimizing copper losses while achieving the desired leakage inductance. The effectiveness of the reinforcement learning approach was validated through the design of integrated PCB-based transformers for DAB converters, providing optimal solutions that minimize copper loss. A prototype transformer designed using the proposed methodology was experimentally verified in a DAB converter operating at 50 kHz, with 150 V input and output voltages, and a 1 kW load. |