Data-driven design of MHz resonant converters for full-freedom soft-switching optimization

Abstract

Achieving soft switching is critical for MHz resonant circuits, including Class E, EF, and Φ-based resonant topologies. Traditional parameter design methods heavily rely on equation-derived analytical modeling, which is highly complex and infeasible to get direct analytical solutions. Existing approaches typically simplify the problem by constraining the model’s degrees of freedom, leading to locally optimal solutions and limiting design space exploration. To address these challenges, we propose a data-driven surrogate modeling approach that eliminates the need for analytical modeling and enables direct parameter optimization for soft switching without imposing constraints on the design space. To the best of our knowledge, this is the first approach that allows global exploration of the full-freedom soft-switching space for resonant converters, unlocking more optimal design solutions. Inspired by the understanding of resonant converters (similarity of inverter and rectifier and resonant capacitor as the crucial soft-switching component), we proposed mirrored circuit simulation and resonant-capacitor-based parameter decouple strategies to improve data density and diversity, achieved high quality data preparation. Experimental validations across three case studies demonstrate its effectiveness. Beyond the circuits studied in this paper, the proposed methodology is broadly applicable to other resonant topologies, offering a scalable framework for automated design optimization in high-frequency power electronics.