一种新的基于复合弹簧网络的多Agent系统分布式问题求解方法

当多Agent系统（MAS）中Agent之间存在多种复杂的随机的社会交互行为时，当各Agent表现出不同程度的自治性和理性时，难以用现有的方法描述和求解MAS问题，即使对仅仅存在竞争和合作这两种社会交互行为，并且不考虑Agent之间自治程度的本质性差异时，现有的基于结盟的MAS问题求解算法也具有极高的计算复杂性，该文提出一种新的复合弹簧网络模型和方法，利用分布式弹性动力学方程，将MAS分布式问题求解过程转变对应的复合弹簧网络形变过程，这种模型和方法能够处理各种社会交互行为以及Agent不同程度的自治性，分析和仿真实验表明，在计算复杂性和适用性等许多方面，该文的分布并行算法优于文献7,8]的Shehory-Kraus算法。

A New Composite-Spring-Net-Based Approach to Distributed Problem Solving in Multi-Agent System

So far there are various methods for coalition formation and problem solving in multi-gent system (MAS), but most of them have very high computational complexity and many limitations. This paper presents a new approach to distributed problem solving for MAS, which makes use of the distributed elastic dynamics of a composite spring network model. By presented approach, the process of problem solving for MAS is transformed into a deformation process of the corresponding composite spring network. The approach can deal with various complicated social interactions occurred among agents in MAS and can describe the behavior dynamics involved in quite different autonomy between individual agents, which may be formidably difficult problems to solve by using other traditional methods. The simulation and comparison with Shehory-Kraus's algorithm demonstrate many advantages of the proposed approach.