Reinforcement learning-based efficiency optimization for dual active bridge converters in charging stations

This doctoral thesis focuses on the development and efficiency optimization of dual active bridge (DAB) converter topologies used in fast charging stations for electric vehicles. In this study, the mathematical model of the DAB converter was presented, and single-phase shift (SPS) and triple-phase shift (TPS) modulation techniques were comparatively analyzed. A detailed mathematical analysis was conducted for the six different operating modes of the TPS modulation, deriving equations for current stress, reactive power, and active power flow. The developed MARL controller was compared with traditional methods (PID, Fuzzy Logic, Optimal Control, Adaptive PID). According to the results obtained, the MARL controller achieved an average efficiency of %96.3, showing improvements of %52.4 over PID, %0.21 over Fuzzy Logic, %1.9 over Optimal Control, and %3 over Adaptive PID. The ZVS success rate was %85, and the voltage regulation error was achieved as 20V with a %5 deviation. When a special reward function was used for minimizing current stress, the current stress was reduced by %86.2 to 1.74A. Additionally it was demonstrated that the MARL approach can adapt to variable load profiles in electric vehicle charging stations. In conclusion, this thesis makes significant contributions to the advancement of electric vehicle charging technologies and provides practical solutions for the design of high efficiency charging stations.