MODELOS DE PREDIÇÃO E ANÁLISE DE SENSIBILIDADE UTILIZANDO REDES NEURAIS ARTIFICIAIS E MODELOS ESTATÍSTICOS PARA TOPOLOGIAS HÍBRIDAS DE TRANSFERÊNCIA DE ENERGIA SEM FIO

Abstract

This work presents a parametric sensitivity comparative analysis between Double LCC and LCC-S topologies for Wireless Power Transfer (WPT) systems, while developing and evaluating secondary circuit parameter estimation methods using regression techniques and artificial neural networks. The study employs the Monte Carlo method to generate a sample space considering typical tolerances of commercial components, enabling the assessment of parametric variations' impact on system performance. The sensitivity analysis revealed that the coupling factor and main coils are the most influential elements in both topologies' performance, with LCC-S being more robust to parametric variations in terms of output current stability and more sensitive regarding other voltage and current stresses. For secondary parameter estimation, different techniques were implemented and compared, including Radial Basis Function Neural Network (RBF), Principal Component Regression (PCR), Ridge Regression, Kernel Regression, and Partial Least Squares Regression (PLS). Results demonstrate that all techniques can estimate secondary parameters using only primary circuit measurements, with PLS showing superior performance in dimensionality reduction and estimation accuracy. RBF demonstrated excellent capability in coupling factor estimation, while Ridge regression presented a good balance between complexity and precision. The developed techniques can be applied in adaptive control systems, contributing to the development of more efficient and reliable WPT systems. Keywords: Wireless Power Transfer, Sensitivity Analysis, Parameter Estimation, Machine Learning, Hybrid Topologies.