【作者】 李丹；

【导师】 陈利；

【作者基本信息】 华中师范大学 ， 计算机应用技术， 2008， 硕士

【摘要】 协作是多Agent系统(Multi-Agent System,简称MAS)研究的核心问题之一,而交互是协作的基础。由Randall Davis和Reid G.Smith借鉴商业活动中合同招标的运作模式,对任务和资源分配提出的经典交互协议——合同网协议已成为多Agent协作的关键技术。由于合同网协议任务的产生、分配以及管理者与承包商角色都是动态产生的,灵活性好,已广泛应用于MAS。但经典的合同网协议存在通信量大等不足,众多的研究者对其进行了扩展研究,但大多是用符号逻辑从形式理论上针对诚实友好合作型的协作问题进行的研究,形式理论虽然严谨,但晦涩难懂,不易实现。同时,在非面对面的、动态的网络环境中,Agent的诚实度不一样,这必然导致交互的过程中存在不安全隐患。因此,有必要对合同网协议的安全性进行建模研究,使其简便易懂,能适应MAS中复杂多变的协作问题。本文将面向对象技术和有色Petri网相结合,采用面向对象有色Petri网(Object-Oriented Colored Petri Net,简称OOCPN),从合同网协议的宏观和微观两个方面进行了建模,得到安全的动态合同网协议(Secure-Dynamic Contract Net Protocol,简称SDCNP)模型。首先,微观上,结合Anand S.Rao和Michael P.Georgeff提出的Agent的认知型结构——BDI(Belief-Desire-Intention)的思想,在目前研究的基础上,引入信任度和阈值,对交互个体——BDI Agent的结构进行建模,通过信任度和阈值来限制发放标书的范围和控制评价标书的数量,达到减少通信量的目的。并采用Petri网的分析方法对模型进行了分析验证,得出模型具有有界性、安全性、活性和可达性;其次,宏观上,借鉴人类社会系统,对合同网协议的交互过程进行了改进,从“法律”和“道德”两个方面引入安全机制——仲裁惩罚机制和行为评估学习机制,一方面对自私Agent的不诚实行为采用“法律”手段进行仲裁惩罚,另一方面通过对自身行为的评估和学习,充分利用历史行为信息,尽量做出有益的行为决策。通过这两方面的安全保证,减少Agent对承诺的否认,增强了合同网协议的安全性。通过对其进行面向对象有色Petri网的建模,验证了模型的正确性;最后,采用REPAST仿真工具对不同模型进行了仿真模拟,对结果进行比较分析,表明SDCNP具有一定的高效性和安全性。更多还原

【Abstract】 Collaboration is one of the most important issues in the Multi-Agent System (MAS). But the interaction is the basis for collaboration. The Contract Net Protocol (CPN) is a classical interactive protocol, which has been proposed by Randall Davis and Reid G Smith. It borrows ideas from invitation for bid in commercial activity, and it has been applied to tasks and resources distribution. The CPN has become the key technology in collaboration of MAS. In the CPN, the formation of task, distribution and Manager Agents and Contractor Agents are dynamically generated. It is flexibility, and has been widely used in MAS. But the classical CPN, there are many deficiencies, such as the excess communication and so on. Although now many researchers have proposed many methods to expanding and improving it, most of them studied by the logic symbols and the friendly cooperation. The logic symbols are rather obscure and also uneasy to program. At the same time, in the non-face-to-face, dynamic network environment, each Agent’s honesty is not same. This will lead to insecurity in the interactive among the agents. Therefore, in order to adapt the dynamic, complex and volatile environment better, it is nessary to research a safe, simple and understandable model to the CPN.So in this paper, we proposed SDCNP (Secure-Dynamic Contract Net Protocol) model using OOCPN (Object-Oriented Colored Petri Net), which is combined by Object-Oriented and Colored Petri Net to meet this needs. First, from the micro, with belief, desire and intention of BDI Agent which is proposed by Anand S.Rao and Michael P.Georgeff, also used other researchersbresearch results" Trust and Threshold, we model the internal structure of the interaction of the individual. And then use the analytical method of the Petri Net to analyze the model and make sure the model has the boundedness, security, reachability and liveness. Sencond, from the macroscopy, we improve the interaction process based on CPN. From two aspects, which are ! law5 and ! morality5 to guarantee the secrity of CPN. On the one hand, to punish the dishonest actions; on the other hand, according to learning and evaluating its own action and taking full advantage of its historical information. And then it can choose a beneficial action to execute. Through the two aspects of security assurances, reduced the disavowal to its own action, enhanced the security of the CPN. At last, through the simulation to different models using REPAST, the analysis and comparison indicate that the improved CPN, which is SDCNP, has a certain high efficiency and security.更多还原