Deep Reinforcement Learning-Driven Traffic Signal Control Strategy for Emergency Vehicle Scenarios in Fog Computing Framework

With the increasingly complex urban transportation system and the continuous growth of vehicle ownership, it is impossible to effectively guarantee the priority of police cars, ambulances, and other emergency vehicles, threatening the timeliness of tasks and public safety. Given the limitations of traditional signal control strategies in ensuring the priority of emergency vehicles, this paper proposes a traffic signal control strategy based on the three-layer architecture of fog computing (DNLight), which utilizes deep reinforcement learning methods to enable continuous strategy adjustments. In this architecture, the bottom fog node is responsible for collecting real-time traffic status and vehicle interaction information; the middle layer fog node agent uses the deep reinforcement learning algorithm, introduces dynamic noise for hybrid exploration, combines Dueling network diversion architecture to improve stability, and introduces the penalty coefficient C to ensure fairness between emergency vehicles and social vehicles. In terms of the reward function, this paper modifies the calculation of reward value to be dynamic, allowing it to adjust the weight of emergency vehicles and social vehicles according to traffic conditions, thereby improving flexibility. The top layer cloud service layer is mainly responsible for receiving the data backed up at the bottom layer and conducting training. The simulation experiment demonstrates that the DNLight strategy enhances the traffic efficiency of emergency vehicles by more than 40% under various traffic scenarios and traffic flow intensities, and reduces the negative impact on social vehicle traffic by up to 50%, providing strong support for urban traffic optimization and emergency management.