Deep Reinforcement Learning-Aided Modulation Optimization for Single-Stage Interleaved Totem-Pole Bidirectional AC-DC DAB Converter

The growing adoption of electric vehicles (EVs) has increased the demand for efficient and reliable charging infra-structure. Conventional two-stage AC-DC converters, widely used to transfer power from the AC grid to EV batteries, often require bulky electrolytic capacitors at the DC bus, which reduce power density, efficiency, lifespan, and overall system robustness. As an alternative, the electrolytic-capacitorless single-stage totem-pole bidirectional AC-DC dual active bridge (DAB) converter offers notable advantages, including high power density, improved effi-ciency, enhanced robustness, and lower cost. However, its simpli-fied architecture imposes stricter requirements on the modulation strategy, which must simultaneously achieve power factor correction (PFC), output voltage regulation, and efficiency optimization. To address this challenge, this paper proposes a three-degree-of-freedom modulation method based on frequency-domain modeling, enabling fine-grained control of key operating parameters. In addition, a soft actor-critic (SAC) deep reinforcement learning-aided strategy is employed to adaptively optimize modulation per-formance in real time under varying conditions. Simulation results verify the effectiveness of the proposed approach, achieving a power factor of 0.99 and a peak efficiency of 89.7% at full load, thus demonstrating its feasibility and superior performance.