Prediction-Guided Distributed Signal–Trajectory Coordination for Heterogeneous Cooperative Traffic at Signalized Intersections

Abstract

Cooperative traffic control at signalized intersections must accommodate human-driven vehicles (HDVs) and connected and automated vehicles (CAVs) with heterogeneous cooperation capabilities while meeting roadside real-time constraints. This study develops a prediction-guided, distributed signal–trajectory coordination framework for such mixed traffic. A two-layer long short-term memory model forecasts five-minute traffic demand and adaptively selects the rolling number of green phases. A cooperation-class-aware entering-time algorithm uses the information-sharing and decision rights defined by SAE J3216, while trajectory construction is executed on on-board units and intersection-level signal and scheduling tasks remain at roadside equipment (RSE). Experiments in Simulation of Urban Mobility (SUMO) across ten traffic-composition scenarios show that, as HDV penetration decreases from 100% to 20%, average fuel consumption, stopped time, and delay decrease by 25%, 99%, and 51%, respectively. Under full CAV penetration, raising the cooperation class primarily improves stability, reducing conflict participation from 1.57% to 0.01% and trajectory reconstructions from 5.21 to 3.65 per vehicle. The proposed architecture keeps maximum RSE computation below 0.15 s per 1 s simulation step. A component-wise ablation further indicates that traffic-flow prediction and adaptive phase-horizon adjustment each contribute to delay reduction, and their integration yields a 5–10% reduction in average delay relative to the double-ablation baseline. These results support scalable cooperative control for heterogeneous connected traffic.