多Agent系统问题求解的广义粒子模型方法

提出一种多Agent系统分布式问题求解的新的广义粒子模型,将复杂环境下多Agent系统资源分配和任务规划的优化问题转变为广义粒子模型中的粒子运动学和动力学问题.广义粒子模型可以描述和处理的复杂环境包括多Agent系统中的Agent之间存在的随机、并发、多类型的交互行为.各Agent有不同的个性、自治性、生命周期、拥塞程度和故障几率等.本文讨论了广义粒子模型和多Agent系统分布式问题求解的关系,提出了广义粒子模型的数学物理模型和多Agent系统分布式问题求解算法,并且证明了它们的正确性、收敛性、稳定平衡性等基本性质.通过复杂环境下多Agent系统资源分配和任务规划问题的实验和比较,证实了广义粒子模型方法的有效性及其特点.

A Generalized Particle Model For Problem-Solving in Multi-Agent Systems

This paper is devoted to a novel generalized particle model (GPM) approach to distributed problem-solving in MAS, which transforms the optimization problem of resource assignments and task allocations of MAS in complex environment into the kinematics and dynamics in GPM. The complex environment in MAS that the proposed GPM approach may deal with includes: A variety of interactions randomly and concurrently occurring among agents; different personality and autonomy of distinct agents; different life-cycle period, congestion degree and failure rate for distinct entities in MAS. At first, the relation between the GPM and MAS in the context of distributed problem-solving is expatiated. Then the mathematical-physical formalization for GPM and the parallel algorithm GPMA are presented. The basic properties of the GPMA algorithm, including the feasibility, convergency and stability, are discussed. Through a number of simulation experiments and comparisons related to resource assignments and task allocations in MAS in complex environment, the authors demonstrate many advantages of the proposed GPM approach over other coalition methods for MAS problem-solving in terms of the parallelism and the suitability for complex environment.