Adaptive Energy Management Strategy for Connected Plug-In Hybrid Electric Vehicles Based on Graph Neural Network Traffic Prediction

As intelligent transportation systems mature and connected plug-in hybrid electric vehicles (C-PHEVs) proliferate, energy management strategies can leverage vehicle-to-everything data to remain effective. In this context, accurate short-term traffic forecasting becomes critical for proactive energy management. This paper proposes a traffic-grade-aware adaptive equivalent consumption minimization strategy (TGAA-ECMS) to optimize the power split of PHEVs in real-time. First, regional traffic grades are defined into five discrete grades, capturing conditions from free-flow to heavy congestion via K-means clustering on mean speed and flow, and a two-layer Graph Sample and Aggregate method is trained to predict short-term congestion grades for each area with high accuracy, augmented by a constant-time boundary-robust decoder that resolves adjacent-grade uncertainty. Next, offline particle swarm optimization refines the equivalent factor across discretized State of Charge (SOC) and traffic-grade spaces, yielding a compact energy-distribution lookup table keyed to both traffic conditions and battery SOC. High-fidelity simulations of real-world urban drive cycles demonstrate that TGAA-ECMS reduces fuel consumption by 1 1. 9 % compared to a conventional rule-based baseline, surpassing adaptive equivalent consumption minimization strategy and model-predictive control benchmarks while preserving battery charge. The results confirm that TGAA-ECMS seamlessly couples real-time traffic-grade foresight with adaptive power-split optimization, delivering a more accurate, deployable energymanagement solution for C-PHEVs navigating dynamic urban congestion.