Resilient Supervisory Multiagent Systems
Resilient Supervisory Multiagent Systems
复制标题
弹性监管多代理系统
DOI：
10.1109/tro.2021.3108074
复制
发表时间：
2022
影响因子：
7.8
通讯作者：
Tanner, Herbert G.
中科院分区：
文献类型：
--
作者：
Baxevani, Kleio;Zehfroosh, Ashkan;Tanner, Herbert G.
多智能体系统中协调功能的意外或故意中断已经在社会科学文献中被讨论并称为领导者斩首;本文概述了一种使多智能体网络对这种类型的故障具有弹性的方法，通过利用机器学习技术及时恢复正常运行。该方法涉及赋予代理人级联的独立学习模块，使他们能够随着时间的推移发现自己在整个系统协调策略中的作用，使他们能够自主地执行它时，中央协调抓住功能。通过这些机器学习算法，智能体逐步识别整个系统的任务规范，同时优化其策略以服务于共同目标。
Accidental or deliberate disruption of the coordination function in a multiagent system has been discussed and referred to in the social sciences literature asleader decapitation; this article outlines a methodology for making multiagent networks resilient to this type of failure, enabling a timely restoration of operation normalcy by leveraging machine learning techniques. The approach involves endowing the agents with a cascade of independent learning modules that enable them to discover over time their role in the overall system coordinating strategy, so that they are able to autonomously implement it when central coordination seizes to function. Through these machine learning algorithms, the agents incrementally identify the overall system’s task specification and simultaneously optimize their strategy to serve the common goal.