AI-assisted power loss modelling and efficiency optimization for power converters

This dissertation proposes an AI-assisted framework for power loss modeling and efficiency optimization of Boost converters. Leveraging LTspice simulation data for pre-training and real-world experimental data for calibration, a multi-layer perceptron (MLP) model is developed to accurately predict power losses, achieving and a 70.79% accuracy improvement over simulation-only models. The framework integrates analog switch-mediated parameter regulation with a synergistic DSP-ADC data acquisition system, validated for non-resonant topologies.This integrated approach significantly improves loss prediction accuracy and enables efficient, scenario-adaptive optimization of power converter performance.