A control approach for traffic congestion based on multipath propagation model

The spatiotemporal propagation of urban traffic congestion constitutes a complex system phenomenon involving multi-factor coupling mechanisms, particularly the dynamic interplay between road network structure and navigation information. In this study, we address these complexities by integrating traffic road and social networks to establish a multipath propagation model that describes the coupled propagation process of traffic congestion and information-guided control. In addition, We propose two innovative control frameworks: (1) A model-based architecture employing a stacked ensemble learning algorithm (RLLS: RF-LSBoost-LR Stacked) for data-driven control in known system environments, and (2) A model-free framework utilizing Proximal Policy Optimization (PPO) algorithm capable of handling stochastic dynamic propagation control in unknown environments. Finally, Comprehensive simulation experiments based on the real peak-hour traffic flow patterns validate the model’s correctness and the effectiveness of the control scheme. Comparative analysis reveals that the proposed PPO algorithm achieves near-optimal performance (1.69% cost deviation) while significantly enhancing the adaptability of traditional optimal control theory in dynamic scenarios.