A Neural Network Based Model-Free Online-Training Controller for Single Switch DC-DC Converter

Numerous control strategies have been proposed for DC-DC converters. Among them, Neural network controller has gained significant attention for its ability to approximate functions without the need for precise mathematical model, making it advantageous for dealing with nonlinear and uncertain control systems. However, existing research still necessitates a detailed modeling process for the converter. This paper introduces a model-free online training control scheme for a single switch DC-DC converter by combining the concept of a model-free system with neural network control. The fundamental concept involves leveraging real-time operational data to facilitate the online training of the neural network. This empowers the converter with effective control capabilities while bypassing the need for conventional modeling approaches. A series of simulations and experiments are performed on a Buck converter, demonstrating improved dynamic performance compared to the PI controller.