基于多智能体强化学习的边缘物联代理资源分配算法

Edge Io T Agent Resource Allocation Algorithm Based on Multi-Agent Reinforcement Learning

如何将电力物联网中多样化的任务动态、合理地部署在边缘节点,实现任务的高效执行是一个非常关键的问题。传统的方法往往采用静态、组合优化的方法进行服务部署,然而该方法不仅效率较低,同时也无法实现动态的资源调度。文章提出一种基于多智能体强化学习的动态资源调度方案,并提出值分解的方式来提升算法的效率。通过实验分析可以看出,提出的算法在不同场景下都可以取得最优的执行效率,同时也具有较好的可扩展性。

How to dynamically and reasonably deploy the diversified tasks in the power Internet of things at the edge nodes and realize the efficient execution of tasks is the key problem at present.Traditional methods often use static and combinatorial optimization methods for service deployment.This method is not only inefficient,but also can not achieve dynamic resource scheduling.In this paper,a dynamic resource scheduling scheme based on multi-agent reinforcement learning is proposed,and a value decomposition method is proposed to improve the efficiency of the algorithm.The experimental analysis shows that the algorithm can achieve the optimal execution efficiency in different scenarios,and has good scalability.