Accurate and Computationally-Optimized Small-Signal Model Identification of LLC Resonant Converter Based on Machine Learning Techniques

LLC small-signal model is often identified via FHA (First Harmonic Approximation) and/or EDF (Extended Describing Function) in order to analyse the dynamical behaviour of the converter and hopefully fine tune the controller. These approaches fail easily and force to consider a resistive load, which is a restrictive method since most of the times resonant converters are used to feed current or stabilize a voltage (e.g., in battery charging applications). In this paper the small-signal output current response of the converter is approximated by a second order discrete-time transfer function, whose numerator and denominator coefficients change with the operating condition (i.e., output voltage and switching frequency). The coefficients are fitted using a sparse linear combination of functions in data-driven fashion (via simulation) adopting a well-known machine learning operator (Least Absolute Shrinkage and Selection Operator, LASSO). The aim of this paper is to report the first attempts made to obtain an accurate and computationally-optimized approximation of the output current response of a generic resonant converter based on machine learning techniques.