A dual-time-scale eco-driving strategy for connected vehicles: Integrating traffic light information and precursor prediction to achieve long-horizon planning

Eco-driving strategies that co-optimize speed planning and energy management have emerged as a key research direction in plug-in hybrid electric vehicles (PHEVs). The energy optimization performance of eco-driving strategies is determined by the planning horizon. To explore the feasibility and optimization performance of a long planning horizon in car-following scenarios, this study proposed a dual-time-scale eco-driving control strategy. First, two deep neural networks are employed to predict the long-horizon position and short-horizon speed of the preceding vehicle (PV). Integrating these predictions with Intelligent Traffic System (ITS) data, a novel Coupled Spatiotemporal Dynamic Programming (CST-DP) algorithm generates long-horizon speed profiles and the corresponding battery SOC reference trajectories (RTs), enabling near-simultaneous arrival with the PV without close-following. Subsequently, a nonlinear Model Predictive Control (NMPC) strategy safely tracks the speed profile, while an Adaptive Equivalent Consumption Minimum Strategy (A-ECMS) allocates power to follow the planned SOC reference. Comparative simulation results show an average terminal time difference of 5.3 s relative to the PV and only 0.8 s against the prevailing car-following algorithms across four traffic scenarios, while achieving an average 10.7% reduction in cumulative tractive energy consumption, validating its energy-saving performance. Additionally, ride comfort is substantially enhanced through reduced acceleration and jerk. The system exhibits strong robustness even under extreme prediction errors, demonstrating its adaptability to highly stochastic traffic environments. Hardware-in-the-loop (HIL) validation ultimately confirms its real-time feasibility, supporting practical online deployment.