【作者】 周伟；

【导师】 高阳；

【作者基本信息】 中南大学 ， 管理科学与工程， 2004， 博士

【摘要】 随着信息技术的发展和全球经济一体化，现代企业面临新的挑战。虚拟企业能更好地适应市场、实现协作竞争，是网络化经济的理想组织模式。 多智能体系统(Multi-Agent Systems，MAS)是分布式人工智能(Distributed Artificial Intelligence，DAI)的研究热点，主要研究多个智能体(Agent)为了联合采取行动或求解问题，实现各自的目标和规划。MAS适合于解决具有分布、动态、开放特性环境中的协作问题，其问题域与虚拟企业成员协作进行调度和冲突消解完全吻合。 本文从虚拟企业的特性出发，提出虚拟企业体系结构，探索研究多智能体协作方法在虚拟企业任务调度和冲突消解中的应用，构建多智能体系统的信息框架，为虚拟企业运作提供新思路。 本文的研究始终沿着问题—理论—应用这一条主线展开并深入。本文的主要贡献在于： (1) 建立了由虚拟企业生命周期、多视图建模、系统平台组成的三维虚拟企业参考体系结构，使得虚拟企业在供应链管理模式下，采用业务过程建模，在协同商务的信息环境下运作，这为虚拟企业的组建和运作提供了整体思路。 (2) 提出了自学习贝叶斯协商模型，改进了合同网协议，将多智能体协作理论用于解决虚拟企业生命周期中各个环节的协作问题。通过算例可知，模型提高了协商效率，增加了协商双方的效用，实现虚拟企业的双赢。同时采用多属性效用函数，结合AHP分析方法和三角模糊数处理协商过程的决策支持，为多智能体协商提供了全面的解决方案。 (3)构建了分布、自治、并行的任务调度系统体系框架，采用项目分解结构方法对任务进行定义，通过能力匹配来实现任务的分解，结合自学习协商和带偏好的多属性决策，在系统原型中实现了虚拟企业调度中的任务分配，并结合多智能体调度方法和传统的智能算法，提出一种集成的调度算法，来解决调度优化和再调度问题。构造了混合遗传算法来进行实例分析，使得调度系统能够适应制造环境的不确定性、动态性和复杂性，快速地响应市场机遇。 (4)提出了在虚拟企业中，如何进行冲突消解的集成策略，从而保证虚拟企业协调运作;策略中采用模糊推理来进行消解方法的智能选择，同时采用聚类分析方法，将事例库中的事例分类，加速知识推理的过程;并实现了正向知识推理和协商的冲突消解算法，为虚拟企业快速响应市场、实现敏捷制造提供了有力支持。 (5)对虚拟企业领域知识进行了本体论描述，同时分析了智能体通信及互操作性的实现，构建了多智能体系统框架，为虚拟企业协同生产管理系统的构建提供了指导。开发出的系统原型运行结果表明，采用多智能体技术研究虚拟企业协作问题不仅可行，而且是有效的。更多还原

【Abstract】 With the developing of information technology and uniting of global economy, modern enterprises are facing with new challenges. The Virtual Enterprise (VE) becomes the idea organization modal adapting to the market and realizing the cooperation and competition at the same time.Multi-Agent Systems (MAS) is a hot topic of Distributed Artificial Intelligence (DAI). Researches on MAS focus on agents how to coordinate their knowledge, goals, strategies and plans to act or solve problems cooperatively. Theories of MAS are fit for solving interaction problems in distributed, dynamic and open systems. Its problem domain coincides with that of scheduling and conflict resolution in virtual enterprise.According to the properties of virtual enterprises, its system architecture is presented and ameliorated strategies of multi-agent cooperation and intelligent optimization algorithms are used to deal with task scheduling and conflict resolution in virtual enterprises. An information infrastructure of multi-agent system is built to support the operation of virtual enterprises.This paper is written along the line of problem-theory-application. The main contributions of the paper consist of:(1)A tri-dimension virtual enterprise system architecture is presented with life cycle of VE, views modeling and system plats. In the fashion ofsupply chain management and modeling with business process, the virtual enterprises can be organized and operated in collaborative commerce information environment.(2)A learning negotiation model with Bayesian Decision Making is set up which improves the contract net protocol (CNP) and makes it possible that the cooperative problems in virtual enterprises can be solved with cooperation theories of MAS. The multi-attribution utility theory (MAUT) is adopted to analyze the decision support during negotiation process with AHP method and trigonometric fuzzy function, which supplies the integrated resolution scheme to MAS negotiation.(3)A distributed, autonomic and collateral task scheduling model is built, in which the task decomposed with agent ability matching and scheduled with Partial-Global-Plan (PGPG) strategy. The Bayesian negotiation and MAUT are also used in task scheduling and re-scheduling with an integrated optimization algorithms base, which makes the scheduling system adapt to uncertain, dynamic, complex market environment.(4)Combining the conflict avoiding with conflict resolution, an integrated conflict strategy ensures that the virtual enterprises run coordinately. The fuzzy reasoning used to choose the strategy intelligently expedites the conflict resolution process and makes the VE response to the market faster.(5)The agent modeling technology is adopted to build VE model and domain knowledge is described by ontology. After analyzing agent communication, a mutually operation and plat independent soft resolution scheme is put forward for VE system. The information infrastructure of multi-agent system is built to realize the collaborative production management system of VE, which develops the Agent Oriented Programming (AOP) method.更多还原