【作者】 胡文斌；

【导师】 王少梅；

【作者基本信息】 武汉理工大学 ， 机械设计及理论， 2004， 博士

【摘要】 随着经济的发展及科学技术的进步，各行各业面临的是一个日益复杂和不断变化的环境，决策问题变得越来越复杂，特别是对于具有多层次、多属性的多目标评价指标体系的复杂系统决策问题，其涉及到多种专业知识，存在大量定性的、病态结构的信息，需要依靠群体的力量才能很好的解决。因此，分布式智能群决策便成为当前决策支持系统研究的重要课题。 本文针对基于多Agent的分布式智能群决策支持系统(DIGDSS)的关键技术进行研究，主要内容和成果体现在以下几个方面： (1) 提出了多层次、多属性的多目标群决策模型，采用“三维结构体系”对群决策系统的构成进行描述，提出了系统的决策过程，并建立了系统的功能结构模型。 (2) 提出一种基于贝叶斯网络的权重自学习方法，解决了多层次、多属性的多目标评价指标体系中权重难以确定和自学习的问题。并提出指数函数变换式的归一化处理方法，该方法使归一化处理后的规范化值更能反映该属性对方案决策的影响程度。 (3) 对数据包络法的C~2GS~2模型进行了改进，在模型中加入松弛向量来分析输入因素对方案的评价结果的影响，首先判断方案是否处在生产可能集中(即满足C~2GS~2模型的应用前提)，然后判断是哪些属性影响了方案的评价结果，并依此来调整方案，使方案达到更满意的效果，论文给出了评价及调整方案的算法步骤，最后通过实例验证了此模型的实用性。 (4) 针对多Agent系统建模的复杂性，本文提出采用UML语言对系统建模，并从不同的视角来描述多Agent系统各种静态和动态行为方面的特征，然后对多Agent系统组织的角色定义、静态结构和交互行为进行分析和建模。设计了基于多Agent的DIGDSS的总体结构框架，给出了基于B／S和J2EE框架的系统软件结构和系统软件的实现方法。 (5) 设计和开发了结构化多Agent通信体，并定义了它的接口，在KQML规范下，结合物流决策支持系统开发的特点，设计了基于Java的Agent通信原语集。提出了基于J2EE技术的系统通信模式，并采用NetDDE、多线程技术方法实现了系统中多Agent之间的通信。 (6) 提出了基于对策论的多Agent协商机制，在引入多边特征值的基础上，给出了针对多Agent协商联盟的构成算法，在Nash平衡概念下对协武汉理工大学博士学位论文商联盟稳定性进行了检验，并在基于JZEE框架下，实现了多个Agent之间的协商。 (7)采用Java作为开发语言，Jbuilder7.O作为开发平台，初步研制了基于多Agent的物流D工GDSS系统，实现了系统的协商、通讯等基本功能，并应用于物流系统决策中，为实现物流分布式智能群决策打下了基础。更多还原

【Abstract】 With the development of social economy and advancement of technology, every walk of life faces an increasingly complicated and variation environment. Decision-problem becomes much more complicated, especially for the decision-problem of intricate system with judge index system of multi-hierarchy, multi-attribute and multi-object, it involves multifold professional acknowledge and has a mass of qualitative, morbid structure information which depends on mass wisdom to deal with. So the research on distributed intelligent group decision becomes to be an important task in DSS.The key technologies of distributed intelligent group decision supported system (DIGDSS) based on multi-agent are research in this paper; the main contents and results are as fellow:(1) Advancing group decision model with multi-hierarchy, multi-attribute and multi-object. Uses "Three Dimension Structure System" to describe the structure of GDSS. The decision process of the system is advanced, and establishes the function structural model of the system.(2) A new ascertaining weight method of weight self-learning based on Bayes Net are put forward, and the problem of difficulty in ascertaining weight and self-learning of judge index system of multi-hierarchy, multi-attribute and multi-object is solved. A grouping together method with exponential function transforms is advanced, the standardized value by using this method can better express the degree that attribute influences the project decision-making.(3) Ameliorating the C2GS2 model of DEA, add relax vector to the model to analysis the influence of input factor to the project. Firstly judges whether the project is in the producing possible collect, then estimates that attribute influence the evaluating result of the project, and according to this to adjust the project to make it better. This paper presents the algorithmic process of evaluating and adjusting project, in the end by applying an example validates the practicability of the model.(4) Aim at the complication of modeling multi-agent system, this paper use UML language to model the system. The frame structure of DIGDSS based onmulti-agent is designed, moreover the software structure based on B/S and J2EE, realizing process are put forward.(5) The structural multi-agent communicating body is designed and developed in this paper, and defines it’s interface. Under the KQML criterion, combined the developing characteristics of logistics decision supported system, the agent primitive set of the system based on Java is designed. The communicating mode of the system base on J2EE technology is advanced, and uses NetDDE, multi-thread technology to realize the communication among multi-agent.(6) Negotiating mechanism of multi-agent based on game theory is brought forward. Based on import the Bilateral Self-Value, the composing arithmetic of negotiating alliance of multi-agent system is presented, and import Nash concept to verify stability of negotiating alliance. It realizes the negotiation of system under J2EE frame.(7) Adopts Java as developing language, Jbuilder7.0 as developing platform to establish the framework of logistics system intelligent group decision-making supported system based on Multi-agent, realizes the communicating, negotiating and so on basal function, and applies it in logistics decision problem, it settles the basic of realizing the logistics distributed intelligent group decision.更多还原