Comparative Analysis of AI Models for Capacitor Current Prediction in Power Factor Correction Topologies

Le résumé fourni par la source

This paper presents a comparative evaluation of AI-based prediction models for Root Mean Square (RMS) of DC link capacitors and harmonic currents in power factor correction (PFC) converters using only input voltage and current measurements. By eliminating the need for direct capacitor sensing, this approach enables scalable, non-invasive condition monitoring across large populations of power supplies, such as those in data centers or industrial plants. Experimental data were collected from a 90 W commercial single-phase boost PFC and a 3.6 kW totem pole PFC evaluation board over a wide input voltage range. Six regression models: XGBoost, artificial neural networks (ANN), support vector regression (SVR), random forest regression (RFR), Gaussian process regression (GPR), and a fine-tuned large language model (LLM), GPT-3.5-turbo, were trained to infer capacitor current RMS and low-order harmonics. Hyperparameters were tuned via exhaustive grid search, and model performance was assessed using mean absolute percentage error (MAPE). Results validate the feasibility of the proposed approach, paving the way for data driven PFC health monitoring without expensive instrumentation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.