【作者】 罗贺；

【导师】 杨善林；

【作者基本信息】 合肥工业大学 ， 计算机应用技术， 2009， 博士

【摘要】 在企业的生产和管理过程中，不断地涌现出大量的信息，这些多源异构的信息增加了进行信息融合与协商处理的难度。目前，Agent理论与方法已经在科学研究、互联网、生产控制、金融服务、企业管理等诸多领域中有了广泛的应用。在Agent理论与方法的研究过程中，逐渐形成了以个体Agent为基础，以多Agent系统为核心，以面向Agent的程序设计为应用工具的研究体系。如何有效地对多源异构信息融合，并在此基础上与系统中的其他Agent进行协商成为目前亟待解决的问题之一。因此，研究多Agent的信息融合与协商问题对提高Agent在复杂环境下的适应能力具有重要的理论意义与实际价值。本文针对多Agent信息的冲突性以及协商过程的不确定性，研究了多Agent信息融合与协商问题及其在故障诊断中的应用。不仅扩展了经典信息融合与协商的理论和方法，也进一步增强了其在实际中的应用能力，具体研究成果有以下几个方面：(1)针对Agent所处的动态、不确定性环境，从焦元的角度分析了多源证据合成过程中存在的合成悖论、“一票否决”以及鲁棒性差等问题，在经典的证据合成规则的基础上，提出了焦元距离和焦元相似度的概念，并构建了基于焦元相似度的证据理论合成规则。该规则将证据合成分解为相关性证据合成与冲突性证据合成两部分，根据焦元相似度的大小对冲突证据进行分配，并通过实验进行比较分析。(2)研究了多Agent协商过程的特点，提出了基于联合意图的多Agent协商协议。分析了多Agent在协商过程中的不同角色，针对顺序协商与并行协商的优劣，利用CAM对多个协商过程进行管理，在此基础上，构建了协商状态和协商动作集合，提出了基于CAM的多Agent并行协商模型。在仿真环境下对多Agent并行协商模型进行实验，分析了不同终止方案对多Agent并行协商的影响，以及单步动作对协商过程中决策次数和协商成功率的影响。(3)利用面向主体的建模方法分析多Agent系统的建模需求，对系统中的Agent角色进行了划分，并分析了多种角色的功能及形式化模型，构建了面向主体的多Agent系统模型框架。提出了基于Q学习的自主Agent模型，将Q学习方法引入Agent的建模，通过对环境的感知获取包括其它Agent在内的环境状态，作为动作选择的依据之一。这种Agent具有很强的独立性以及较高的学习效率，能够适应于动态环境。(4)探讨了多Agent信息融合与协商在故障诊断领域中的应用。针对诊断对象系统设备的复杂性和分布性等特点，将面向焦元的证据合成规则作为智能故障诊断系统的信息融合算法，较好地解决了多Agent信息融合过程中的冲突证据对合成结果的不利影响，建立了相应的智能故障诊断决策规则。利用面向主体的建模方法分析了智能在线故障诊断系统中的角色及功能，构建了智能在线故障诊断系统的体系结构。在此基础上，对系统的功能、结构等进行了分析和设计，并在Visual C++环境下实现了原型系统。更多还原

【Abstract】 Huge Information is generated continuously during the process of production and management among the enterprises. Usually, this kind of information is heterogeneous and coming from multi sources, so, it is more complex for fusing the information and handling negotiation between different agents. At the present time, the theories and methods of agent have been widely developed during the scientific researches, and have also been applied into several areas, such as internet, production control, financial services, business management, and so on. Now, the theories and methods of agent have gradually formed their own research system, which is built on the basis of the single agent, multi agent systems, and agent oriented programming. However, how to analyze and fuse multi sources information effectively and negotiate with other agents in the system reasonably is still problem to sovle. Therefore, the research on the information fusion and negotiation of multi agents are more useful to strengthen the agent ability of adapting to the complex environment, which are not only significant for the theory, but also for the application.This dissertation mainly focuses on the research of multi agent information fusion and negotiation, and its application on fault diagnosis, according to the conflicts of the information and the uncertainty of negotiation process. The dissertation has extended the classical theory and method of multi agent information fusion and negotiation; it has also impoved the ability for application. The details are as follows:(1) According to the dynamic and uncertainty environment, the dissertation analyzes the problems of combination paredox, "veto by one vote", and robustness during the process of information fusion from the focal element view. The concepts of focal element distance and similarity of focal element are proposed on the basis of the classical evidence combination rule. The dissertation suggests a new evidence combination rule base on the similarity of focal element. The new rule divides the combination process into two parts, the process of relevance evidence combination and the process of conflict evidence combination, during which the evidences are allocated according to the different similarity. Four contrast experiments are done to test the performance among different rules under different situations.(2) The dissertation proposes the negotiation protocol for multi agent based on the joint intentions according to the characteristic of multi agent negotiation process, and it also analyzes the different roles during the negotiation process. According to the comparing of sequential and concurrent negotiation, the dissertation also builds the set of negotiation state and negotiation action, and suggests using CAM to manage multiple negotiation processes. The framework for multi agent concurrent negotiation based on CAM is proposed and analyzed by the given example. The different termination schemes are discussed in the negotiation experiments to analyze the influences on the multi agent negotiation process, number of decisions during the negotiation and successful rates of the negotiation.(3) The modeling requirements of multi agent system are discussed by the agent oriented modeling methods, and the agent roles in the system are also divided. The dissertation builds the framework of multi agent system based on the function models and formulation models of the agent roles. Q-Learning method is also introduced into the single agent modle to enhance its ability of learning. Agent can not only obtain the information of the envinroment but also regonize its inside state. All the information is used for choosing action under current circumstances. This kind of agent has a good flexibility and learning efficiency, and can adapt to the dynamic environment.(4) The dissertation investigates the application of multi agent information fusion and negotiation in the fields of fault diagnosis. On the basis of the complexity and distributivity of the object systems, and the focal element oriented combination rule is set as the information algrithrim in the intelligent fault diagnosis systems, which solved the conflicts among the process of fusing multi agent information. The intelligent fault diagnosis decision rules are also suggested correspondingly. Then, the architecture of the intelligent fault diagnosis system is built based on the analysis of the roles and functions in the system using agent oriented modeling methods. The dissertation also implements a prototype system after designing the structures and interfaces of different agents using Visual C++.更多还原