Neural Network Based On-Time Control of Boundary Conduction Mode Boost Power Factor Corrector

Abstract

Additional on-time control in boundary conduction mode (BCM) boost power factor correction (PFC) circuits has been a practical solution to improve power quality. However, optimal on-time profiles are difficult to obtain using conventional model-based approaches due to nonlinearities in both hardware and control. These nonlinearities require extensive analytical modeling, parameter extraction, and tuning, which leads to a complex development process. Data-driven on-time control can estimate accurate on-time without analytical modeling because it inherently includes hardware nonlinearities. However, conventional approximation methods, such as lookup tables and polynomial fitting have limitations in terms of memory requirement and computation time. Although artificial intelligence (AI)-based methods have recently been used in power-electronics applications to reduce the burden of model analysis and optimize, the application of AI to additional on-time control in BCM boost PFC converters remain limited. This article proposes a lightweight neural network (NN)-based on-time control method using pretrained NN, without requiring analytical modeling. A complete design flow—data collection, training, and real-time deployment—is presented. Since the proposed NN can be rapidly retrained with newly collected data, it can be easily adapted to different device selections or circuit parameter variations without extensive redesign effort. Experimental results from a prototype with 110–220 VRMS input and 400 V/150 W output validate the effectiveness of the proposed NN controller in achieving competitive performance in terms of power factor and total harmonic distortion of the input current with the real time application on a low-cost microcontroller unit.