基于DDQN的多智能体冲突消解方法  

A multi-agent conflict resolution method based on DDQN

作　　者：张翼 赵岭忠 翟仲毅 ZHANG Yi;ZHAO Lingzhong;ZHAI Zhongyi(School of Computer and Information Security,Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学计算机与信息安全学院,广西桂林541004

出　　处：《桂林电子科技大学学报》2022年第5期366-370,共5页Journal of Guilin University of Electronic Technology

基　　金：国家自然科学基金(61862014,61902086);广西高校中青年教师科研基础能力提升计划(2019KY0249)。

摘　　要：针对智能体在局部观测下无法有效决策的问题,提出了一种结合深度强化学习的冲突消解方法。该方法基于DDQN算法,利用强化学习的学习模式的特性,计算智能体的累计回报,通过回报值的大小确定智能体的优先级,从而达到冲突消解的目的。通过模拟现实生活中的堵车场景对该方法进行评估,实验结果表明,该方法能有效解决智能体的冲突。To solve the problem that agents cannot make effective decisions under local observation,a conflict resolution method combined with deep reinforcement learning is proposed.Based on DDQN algorithm,this method uses the characteristics of reinforcement learning mode to calculate the cumulative return of agent and determine the priority of agent through the return value,so as to achieve the purpose of conflict resolution.The method is evaluated by simulating the traffic jam in real life,and the experimental results show that the method can effectively solve the agent conflict.

关 键 词：多智能体系统 冲突消解 深度神经网络 深度学习 强化学习 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]