Adaptive traffic signal control for road networks based on dueling double deep q-network

The Adaptive Traffic Signal Priority (ATSP) model is an intelligent traffic signal control technology designed to dynamically adjust traffic signals to prioritize bus transit, thereby enhancing public transportation efficiency and reducing travel delays. Addressing the challenge of implementing ATSP in complex multi-intersection scenarios, this study proposes a Multi-Agent Adaptive Traffic Signal Priority (MA-ATSP) model based on the Dueling Double Deep Q Network (D3QN) algorithm. This model comprehensively accounts for traffic flow interactions between intersections, aiming to minimize the total passenger waiting time across the network for both buses and general vehicles. Using Discrete Traffic State Encoding (DTSE), intersections are divided into grids to capture spatial vehicle information. The SUMO simulation software is then used to construct a simulation scenario based on four actual intersections in Jinan. Experimental results indicate that the MA-ATSP model reduces cumulative passenger waiting time by 24.94% and cumulative vehicle queue length by 10.98% compared to traditional multi-agent deep reinforcement learning models. The findings demonstrate that the extended MA-ATSP model effectively improves bus transit efficiency and significantly reduces network-wide vehicle waiting times, offering a novel approach to urban traffic management.