Construction of a Data-Driven Model for Boost Converter Loss Based on Convolutional Neural Networks and Current Spectrograms

Power electronic DC-DC converters have extensive applications in numerous fields. As nonlinear systems, the inductor current, input current and output current of these converters exhibit strong periodicity, containing rich information. Traditional linearized modeling inevitably introduces deviations in loss estimation compared to actual values. This paper proposes a loss modeling and analysis method for synchronous rectifier (SR) digitally controlled boost converters based on convolutional neural network (CNN) and current spectrograms. The operational mechanism and loss sources of the synchronous rectifier boost converter are analyzed to determine the optimal data acquisition points for the data-driven model. Subsequently, a data-driven modeling approach based on CNN and current spectrograms is introduced. And a 300 W SR Boost converter prototype is constructed for experimental validation. Unlike traditional mechanism-based modeling, this method utilizes periodic current information to estimate losses, enabling users to achieve more precise parameter design through loss estimation.