Hybrid Deep Learning Optimization for Dual-Stage Interleaved EV Onboard Charger Systems

Abstract

The rapid adoption of electric vehicles has intensified the demand for efficient, adaptive, and intelligent onboard charging systems capable of meeting stringent performance and grid compliance requirements. Dual-stage interleaved onboard chargers have emerged as a preferred architecture due to their ability to achieve high power factor, reduced current ripple, galvanic isolation, and improved efficiency across varying battery conditions. This paper presents a comprehensive review of advanced control and optimization techniques applied to dual-stage interleaved onboard chargers. It highlights the integration of deep learning approaches, including convolutional neural networks, recurrent models, and reinforcement learning, for tasks such as real-time control, fault detection, and battery state estimation. Additionally, the study focuses on the Hybrid Adaptive Genghis Khan Shark Gold Rush (HAGKSGR) algorithm for optimizing PIDD2-PD controllers, enabling improved transient response, reduced overshoot, and enhanced robustness in nonlinear operating environments. Applications include high-performance electric vehicle charging systems, vehicle-to-grid integration, and intelligent energy management. Comparative analyses demonstrate that deep learning-assisted and metaheuristic-optimized controllers outperform conventional methods in efficiency, adaptability, and accuracy. However, challenges such as computational complexity, real-time deployment, and hardware constraints remain. This review emphasizes the potential of combining deep learning and hybrid optimization techniques to develop next-generation intelligent onboard charging systems.