Multi-Objective Optimization Design of Dual-Bridge Series-Resonant Converter Based on Deep Reinforcement Learning

Inhalt:
To address the challenges of parameter optimization in Dual-Bridge Series-Resonant Converters (DBSRC), which traditionally rely heavily on expert experience and suffer from low design efficiency under wide input/output voltage ranges, this paper proposes an automated design methodology based on Deep Reinforcement Learning (DRL). By constructing a DRL-based multi-objective optimization framework that integrates system modeling with a co-optimized reward function considering efficiency, volume, and cost, rapid parameter tuning is achieved. Experimental results on a 6.6 kW prototype demonstrate that the optimized parameters yield an average efficiency of 96.7% over a wide output voltage range (200–500 V).