用倾向最大回报的协同进化优化多Agent合作

Using Co-evolutionary of Biasing Maximal Rewards Search for Optimal Multi-agent Behaviors

进化计算是多Agent系统学习的一个有用技术。在多Agent系统研究中的某些领域,一种常用的方法是协同进化多Agent合作。研究已经指出:在某些领域,协同进化系统更倾向于稳定而不是成效(即收敛到局部优化解)。这与多Agent系统研究的目的(追求利益最大化)是不相符的。为此,文章提出了一种基于混沌机制的倾向于最大回报的协同进化算法,改进了Wiegand等人的工作,。理论分析和仿真实验表明,这种基于混沌机制的倾向能促使协同进化向更优化的全局稳定点收敛,从而帮助协同进化算法在某些合作的多Agent领域发现更好的解(甚至是最优解)。

Evolutionary computation is a useful technique for learning behaviors in multi-agent systems.In some research of multi-agent systems,one natural and popular method is to co-evolve multi-agent behaviors,Recent research has suggested that co-evolutionary systems may favor stability rather than performance in some domains （convergence to local optimal point）.This is not suitable with the aim of multi-agent systems research （the aim is reaching for optimal benefits ） .This paper presents the search method of biasing maximal rewards based on chaos system,improves the work of Wiegand et al.Theory analysis and experiment indicates that this bias can make co-evolutionary to convergence upon the whole optimal stability point.