Comparative Evaluation of RL-TD3 and PI Control for Voltage Regulation in a DAB DC-DC Converter

Lopez D.G.; Apaza-Cruz J.; Belizario A.; Azcue J.L. · 2025 16th IEEE International Conference on Industry Applications, INDUSCON 2025 - Proceedings (2025) · DOI: 10.1109/induscon66435.2025.11241663

Resumen

This article presents an analysis of a Dual Active Bridge (DAB) converter operating with Single Phase Shift (SPS) modulation, in which a Reinforcement Learning (RL)based controller is proposed using the advanced Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm for output voltage regulation. This controller is compared to a traditional PI controller. The study is conducted through simulations in the Matlab®/Simulink environment, utilizing the PLECS Blockset along with thermal and semiconductor models provided by Onsemi. The results obtained allow for a comparison of both controllers in terms of reference tracking and disturbance rejection. The evaluated parameters include overshoot, settling time, rise time, and the ITAE (Integral of Time-weighted Absolute Error) index. The results show that the RL-TD3-based controller improves the dynamic performance of the system compared to the PI control, particularly in terms of response speed and overshoot reduction.