【作者】 王澜；

【导师】 何华灿；

【作者基本信息】 西北工业大学 ， 计算机软件与理论， 2006， 博士

【摘要】 本课题的研究来源：国家自然科学基金项目“经验知识推理理论研究”(60273087)、北京市自然科学基金项目“不确定推理理论研究”(4032009)。多Agent系统(Multi-Agent System，MAS)研究的是由多个智能Agent通过相互交互作用而组成的系统。在对MAS的研究中，最重要的是理解发生在这些Agent之间的交互的类型。现有的成果仅仅对交互作用中的合作、协商、协调等行为进行了孤立地研究，而没有从系统的观点对这些行为之间的内在联系进行深入探讨。此外，对于同样存在于MAS交互作用中的敌对关系的研究还是一片空白。何华灿教授提出的泛逻辑学思想，充分考虑到存在于事物之间的关系柔性——广义相关性对推理运算具有的影响，并以广义相关系数来刻画广义相关性的连续变化。本文将广义相关性概念引入到MAS交互作用中，定量研究了合作、自利、竞争、敌对等关系，取得了以下创新性成果：1)提出目标相容性概念，并应用于NAS交互作用分类现有的MAS合作、协调、协商等概念不能明确地描述Agent间的相互关系，本文提出了目标相容性的概念，并以此为基础，重新界定合作、协调、非协调关系，从而明确了各种交互作用，为后续的量化研究打下了基础。2)提出基于广义相关性的多Agent交互作用量化表示方法将泛逻辑学中广义相关性的思想引入到MAS的交互作用中，建立了基于广义相关性的MAS交互作用量化表示方法，可以根据广义相关系数的大小来定量地研究各种交互作用。该表示方法不仅包括了现有研究中的协作、自利、竞争等关系，而且还能涵盖敌对关系。3)提出基于广义相关性的Agent决策模型London大学Jennings教授等提出了社会责任Agent的决策函数，本文针对其中加权系数存在的难以取值、求解算法复杂、未考虑敌对关系等问题，提出了基于广义相关性的Agent决策模型，该模型具有加权系数取值方便、通过线性运算即可求解的优点。运用该模型对博弈论“囚徒两难”等问题进行仿真，验证了Agent可以在合作和敌对之间进行灵活的行为选择。4)设计并实现了广义相关性MAS系统——GCC-MAS对基于广义相关性的Agent的思维模型进行形式化，将决策函数引入到形式化过程中，并建立了广义相关性MAS系统。在Swarm仿真环境下，验证了该系统中Agent在不同目标导向下的决策行为效果。本文对基于广义相关性的Agent交互作用及决策进行了初步研究，今后还需要进一步深入探讨在实时系统中，广义相关系数根据历史经验进行学习，合理调节等问题。更多还原

【Abstract】 This thesis is supported by the National Natural Science Foundation of China(grant 60273087) and Natural Science Foundation of Beijing(grant 4032009).Multi-Agent System(MAS) focuses on the system composed of several intelligent agents which interact with each other. The most important thing of research on MAS is to understand the type of interaction among them. The current researches only discuss the cooperation, negotiation and coordination separately, and have not realized the inherent relationship among them. Besides, there is no study on antagonistic relationship.Universal Logic, proposed by Prof. He hua_can, considers the relational flexibility between propositions in full, and calls it general correlativity. To depict the continuous changeability, a coefficient, named general correlation coefficient, is introduced.In this dissertation, the general correlativity is introduced into the study on interaction in MAS. The behaviors which are cooperative, self-interested, competitive and antagonistic are investigated quantitatively. The main innovative achievements are as follows:1) A new taxonomy of interaction among MAS is proposedCurrent concepts of cooperation, coordination and negotiation in MAS can not describe the relation between agents. A concept of Goal Compatibility is proposed to define several different relations: cooperation, coordination and non-coordination. This endeavor builds the foundation for the following work of quantization.2) A method of quantifying agent interaction based on general correlativity is proposedThe concept of general correlativity in universal logic is introduced in the research on interaction of agents. A method of quantization is proposed. It is possible to quantitatively study on various relations by the values of general correlation coefficients. By this method, relation of cooperation, self-interest, competition are covered, further more, the relation of antagonism is included too.3) A model of decision making based on general correlativity is proposedSocially responsible agents was put forward by Prof. Jennings of London University. There are several defects in the decision making function of the agent, such as difficulty in coefficient selection, complexity in calculation and neglect of antagonism. By the model of decision making based on general correlativity proposed in this dissertation, the value of coefficients are related to the change of general correlativity and are easy to compute via liner operation. What is more, through empirical evaluation, it is shown that the agent makes more flexible decision using this model.4) The agent based on general correlativity is designedIn the formalization of mind model, decision making function is introduced. Architecture of agent based on general correlativity is designed. The behaviors of agent are simulate based on swarm and the results verify the effects of the decision making directed by different goals.更多还原