Power Supply Vehicle Scheduling Method Based on Multi-Agent Reinforcement Learning with Shared Attention

With the frequent occurrence of extreme weather events caused by climate change, cities are facing the risk of increasingly frequent large-scale power outages. Due to the lack of intelligence and collaborative dispatching capabilities, the traditional power supply vehicle dispatching method is difficult to quickly restore urban power supply after a disaster, resulting in serious economic losses and social impact. Therefore, improving the efficiency and response speed of power supply vehicle dispatching has become an urgent problem to be solved. This paper proposes an innovative method named multi-agent reinforcement learning method based on a shared attention mechanism (SARL) for efficient dispatching of urban power supply vehicles after disasters. This method comprehensively considers user priorities and actual needs by integrating multiple information sources, including meteorological conditions, geographic information system (GIS) data, user power demand, traffic conditions, etc. During the scheduling process, each power supply vehicle agent shares information and collaborates through the shared attention module to optimize the scheduling path and power supply strategy to maximize power supply efficiency and minimize power outage time. Experimental results show that the proposed method exhibits superior prediction accuracy and decision-making capabilities in complex and changeable environments, and significantly improves the recovery speed and efficiency of urban lifelines after disasters. This not only ensures the life safety of residents and the normal operation of the city, but also provides new ideas and technical support for intelligent emergency resource dispatching.