基于强化学习的动态事故疏散

Dynamic Accident Evacuation Based on Reinforcement Learning

随着现代城市的发展,智能建筑系统对事故的决策尤为重要。事故发生时往往具有极高的随机性,这给疏散带来了巨大的挑战。本文提出一种网格化的马尔科夫决策过程的动态事故模型,其中对事故高发区域进行了合理的仿真。进一步地,本文应用Q学习方法,通过动态探索以及合理的更新率,实现了动态事故的自适应决策方法。

With the development of modern city,intelligent building system is very important for accident decision.Accidents tend to occur with a high degree of randomness,which makes evacuation a huge challenge.In this paper,a dynamic accident model of Markov decision process based on grid is proposed.Furthermore,this paper applies Q learning method to realize the adaptive decision-making method of dynamic accidents through dynamic exploration and reasonable update rate.