Multi-Objective Traffic Dispatch Optimization with DQN-Based Deep Reinforcement Learning

With the acceleration of urbanization, traffic volumes have continuously increased, particularly during road construction and emergency incidents, leading to more severe traffic congestion. Bridge demolition operations generally require road closures, significantly impeding traffic flow. How to implement intelligent traffic scheduling during blasting operations to reduce urban traffic impacts has become a crucial issue demanding urgent solutions. Traditional traffic scheduling methods relying on rule-based strategies or historical data optimization struggle to adapt to complex and dynamic traffic environments.This paper constructs an intelligent traffic scheduling model based on deep reinforcement learning (DQN),enabling the system to dynamically learn optimal traffic control strategies. The DQN algorithm combines deep neural networks with reinforcement learning, obtaining optimal decisions through interactive training with traffic environments. Experiments were conducted using the VISSIM simulation tool, with comparisons against traditional fixed-signal control methods. Results indicate that the DQN scheduling scheme can significantly improve road traffic efficiency, alleviate blasting-induced congestion, and provide new insights for intelligent urban traffic management.