基于多agent系统的分布式问题求解中的重尾现象  

Heavy-Tailed Phenomena in Multi-agent-based Distributed Problem Solving

作　　者：唐屹[1] 

机构地区：[1]广州大学数学与信息科学学院,广州510405

出　　处：《计算机工程与应用》2005年第30期81-83,共3页Computer Engineering and Applications

摘　　要：通过随机给变量赋初值,统计采用基于多agent系统的分布式逃逸算法求解约束满足问题所需的求解步数,不同类型问题的求解性能具有很大的差别。已发现一类着色问题的求解,显现出重尾分布的现象。该文提出了一个新的全局调度策略,通过权重的重置,保持权重的相对平衡性,有效地消除了重尾现象。By randomly initializing the variables,we apply the Distributed Breakout Algorithm to constraint satisfaction problems based on a multi-agent system.This method exhibits a large variability in performance.We observe the heavy-tailed distribution of cycles in solving some graph coloring instances.We present a new global scheduling strategy, weights resetting,to balance the weights of constraints and eliminate heavy-tailed behavior in distributed problem solving.

关 键 词：分布式逃逸 重尾分布 权重重置 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]