Stabilizing DC bus voltage using CEEMDAN-XGBoost based adaptive filtering technique in EV chargers

Abstract

The need for clean AC current, transient-free DC bus voltage, and reactive power support (RPS) in EV chargers has increased due to the widespread use of power electronics for current loads. Addressing these challenges, this manuscript proposes an adaptive filtering technique (AFT) that synergizes Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and XGBoost, an AI-driven framework, to stabilize DC-link voltage under nonlinear and dynamic load conditions. The CEEMDAN-XGBoost approach leverages CEEMDAN’s robust signal decomposition capability to isolate transient disturbances, while XGBoost’s machine learning prowess adaptively optimizes voltage regulation and transient response. This integration achieves enhanced grid stability and power quality, reducing total harmonic distortion (THD) to 2.44% in grid-to-vehicle (G2V) and 3.83% in vehicle-to-grid (V2G) modes. Further, a fourth-order Quadrature Signal Generator (QSG) filter is embedded within EV chargers to augment harmonic attenuation, suppress DC offsets, and accelerate settling times during abrupt load transitions. The efficacy of the proposed control strategy is rigorously validated through MATLAB/Simulink simulations and experimental testing on a laboratory-scale prototype. Results demonstrate superior DC bus voltage stabilization, improved dynamic performance, and compliance with power quality standards, underscoring the viability of CEEMDAN-XGBoost as a transformative solution for next-generation EV charging systems.