Abstract
Aiming at the significant deficiencies of traditional traffic signal control algorithms in multi-intersection collaboration, unexpected event response and system generalization ability, this paper proposes an intelligent traffic signal control method that integrates a bidirectional gated recurrent unit BGRU with deep reinforcement learning DRL. The method adopts BGRU to model the historical traffic flow data in time sequence and accurately predict the traffic state; and based on this, it constructs deep Q-network intelligences to dynamically optimize the multi-intersection signal timing strategy. The experimental validation on SUMO simulation platform shows that the proposed method effectively improves the control performance. Compared with the traditional fixed-cycle control and adaptive control methods, the proposed method reduces the average vehicle waiting time by 58.8% and 42.2%, and improves the intersection access efficiency by 83.8% and 52.4%, respectively. The study provides new ideas for building an efficient and intelligent urban traffic management system.