Optimizing Power Quality in EV Chargers Using Advanced Quadrature Signal Generators and AI-Driven Adaptive Filtering

Abstract

As electric vehicle (EV) adoption grows, vehicle-to-grid (V2G) technology enables bidirectional power flow in grid-interactive EV chargers. However, maintaining consistent power quality under nonideal grid conditions remains a challenge. Traditional PI controllers struggle to reduce total harmonic distortion (THD) and adjust to dynamic grid variations. This study explores machine learning techniques, including decision trees, artificial neural networks (ANN), and linear regression, as alternatives to conventional PI controllers. Decision trees emerge as the most advantageous due to their simplicity, interpretability, and ability to handle complex, nonlinear relationships with minimal data preprocessing. While ANN captures intricate patterns, it demands more computational resources and lacks transparency. Linear regression, though efficient, struggles with complex grid behaviors. The decision tree approach allows real-time adaptive control, improving THD reduction and grid stability. Additionally, a CNISOGI filter is implemented to enhance harmonic attenuation and DC-offset rejection. The system's effectiveness is validated through Matlab/Simulink simulations and a 1.1 kW hardware prototype. The results show that integrating decision tree-based controllers with advanced filtering techniques can significantly enhance power quality, grid stability, and operational efficiency in future smart grids.