Fire Evacuation Path Planning Based on Improved MADDPG (Multi-Agent Deep Deterministic Policy Gradient) Algorithm

The lack of a scientific and reasonable optimal evacuation path planning scheme is one of the main causes of casualties in fire accidents. In addition to the high temperature and harmful smoke in the fire environment, the crowding problem caused by the change of the position of the crowd in the evacuation process will also affect the evacuation effect. Therefore, by improving the multi-agent depth deterministic strategy gradient algorithm, an AMADDPG (Adjacency Multi-agent Deep Deterministic Policy Gradient) model suitable for fire evacuation is proposed. First, the dangerous grid area is defined, and the influence of congestion degree and nearest exit is considered at the same time. The learning framework of "distributed execution and centralized local learning" is adopted to realize experience sharing among neighboring agents. Improve the learning efficiency and evacuation effect of the model. The experimental results show that the model can basically adapt to the complex and dynamic fire environment well, achieve the optimal path planning within 30, and ensure that the degree of congestion on the evacuation path is maintained within 0.5, which can achieve the safe evacuation goal. Meanwhile, compared with the MADDPG algorithm, the model has obvious advantages in terms of training efficiency and stability. It has good application value.