I. Introduction
Substantial research has been performed for dual active bridge (DAB) converters since it was proposed in [1]. Through decades of development, the DAB converter is widely employed in various industrial applications such as electric vehicles, fuel cell power conversion, renewable energy storage system, etc. [2], [3]. With the desire for higher efficiency and higher power density power converters, the operating frequency is usually increased to megahertz (MHz) range [4], [5]. Consequently, soft-switching has become more important to high frequency DAB converters due to the increase of switching loss. There are various DAB modulation schemes, e.g., single-phase-shift (SPS), dual-phase-shift (DPS), triple-phase-shift (TPS), and multiphase-shift (MPS) are different modulation strategies. Comparing DPS, TPS, and MPS with SPS, the duty ratio as an additional degree of freedom is adopted mainly for shaping the inductor current to fulfil zero-voltage switching (ZVS) condition. Most analysis of ZVS typically depends on the resonance between the inductance along the ac link and switch output capacitance Coss while the resonant loop also varies under different modulation strategies [6]. However, under high operating frequency, the transformer's parasitic capacitances become non-negligible, which can cause the current resonance and the narrowing of the ZVS region [7]. To mitigate the current resonance, the approach of distributing the external inductance on both sides of the transformer is introduced and analyzed in [2]. With the consideration of more parasitics, the converter model becomes more complicated and makes it harder to derive a precise analytical model. Thus, the whole system can be regarded as a gray-box. To obtain the targeted design inside the gray box more efficiently, the use of artificial intelligence (AI) becomes a promising choice.