A Computation Method for Voltage Conversion Gains of an LLC Converter Using Physics Model and Machine Learning

This paper describes a modeling method for rapid computation of input-output voltage conversion gains of an LLC converter with high accuracy. This method quickly predicts the gain close to that of an actual circuit, even when circuit component values, switching frequency, and load conditions vary. Conventionally, the Fundamental Harmonic Approximation (FHA) method calculates the gain rapidly thanks to a simple equation. However, since FHA relies on an approximation, its accuracy decreases as an operating frequency deviates from the series resonant frequency. To address this issue, this study employs an Artificial Neural Network (ANN), a type of machine learning (ML), to estimate prediction errors and combines them with the gain obtained from FHA. This paper presents a procedure for dataset creation, an ANN training process, and a method for synthesizing the gain. In this paper, circuit simulation results of an LLC converter with a half-bridge circuit are considered as reference values (= ground truth). Test results demonstrate that selecting appropriate features for the input and output layers of the ANN decreases the prediction error.