Two-Layer Traffic Signal Optimization: A Edge-assisted Pressure Balance Approach Based on Cooperative Game

Traffic signal control is essential to efficient transportation networks since it can mitigate traffic congestion significantly. Trial-and-error approach in reinforcement learning will lead to traffic jams, even traffic accidents in the real scene, which is in violation of safety for traffic signal control. Besides, most signal control systems still rely on oversimplified information, which makes item challenging to adapt to dynamic traffic. In this paper, we focus on the edge coordinated optimization of large-scale traffic signal control, and propose a two-layeR edge-assisted pressUre balaNce (RUN) approach based on cooperative game. The external layer utilizes cooperative game to divide the traffic network into multiple coalitions. The internal layer uses pressure control and weighted queue to coordinate actions within each coalition and handle dynamic traffic situations over time. We derive a Pareto stable solution for the multi-intersection signal cooperative game with pressure control, and prove that it is non-superadditive. Moreover, we conduct extensive simulations to verify the significant performances of RUN based on both real data and synthetic data.