Neural Network-Based Design of a Buck Zero-Voltage-Switching Quasi-Resonant DC–DC Converter

Abstract

In this paper, a design method using a neural network of a zero-voltage-switching buck quasi-resonant DC–DC converter is presented. The use of this innovative approach is justified because the design of quasi-resonant DC–DC converters is more complex compared to that of classical DC–DC converters. The converter is a piecewise linear system mathematically described by Kirchhoff’s laws and represented through switching functions. In this way, a mathematical model is used to generate data on the behavior of the state variables obtained under various design parameters. The obtained data are appropriately normalized, and a neural network is trained with them, which in practice serves as the inverse model of the device. An example is considered to demonstrate how this network can be used to design the converter. The key advantages of the proposed methodology include reducing the development time, improving energy efficiency, and the ability to automatically adapt to different loads and input conditions. This approach offers new opportunities for the design of advanced DC–DC converters in industries with high efficiency and performance requirements, such as the automotive industry and renewable energy sources.