【作者】 宋博；

【导师】 刘建昌；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 伴随着人工智能技术,模型构建语言,合同网,通讯机制以及规则库的发展,多智能体系统作为复杂控制系统的解决方案越来越发挥出自己的优势。在多智能体系统的实现过程中,大多数设计和开发者只在模型搭建完成以后添加安全问题的考虑。这在很大程度上导致了多智能体系统的脆弱性。基于此,本文引入了基于UML元模型的2×2四视窗多智能体模型搭建语言,对设计过程,运行过程,智能体内部和智能体外部分别进行考虑,实现了一般性的多智能体系统建模。更加有利于系统开发者对于智能体整体功能的把握,加强系统的容错能力。本文针对多智能体系统的协商特点,引入合同网,但是合同网的节点智能体缺乏处理模糊信息的能力,然而心智状态则需要模糊化描述,才能更好地使单个智能体实现智能决策,所以本文在引入合同网作为各个智能体之间协调的基础后,在合同网中引入竞标模糊化,使得各个单智能体能够实现更加智能,更加合理地决策和协调。不仅如此,在分析了传统合同网的不足后,针对通讯资源浪费和计算资源浪费,在管理智能体中添加了信任度挥发机制,在智能体中添加了缓冲区功能,初步实现工作智能体能够选择任务的优化策略,管理智能体可以选择等待回应的工作智能体优化策略。板形控制系统是典型的多输入多输出,强耦合的复杂控制系统,多智能体控制系统就是把复杂的系统(软硬件系统)建造成小的彼此相互通讯及协调的,易于管理的系统,并且通过自己的协调机制和控制思想使整个系统达到最优化。在本文中针对板形控制系统中存在的板形缺陷,设计不同的控制策略,然后将这些控制策略封装为单个的工作智能体,赋予智能体相应的动作和智能性,它们之间的协调机制,采用本文设计的改进的合同网机制,完成智能体的封装之后,利用基于UML的2x2四视窗多智能体模型搭建语言在软件staruml中实现建模,并利用visual studio使模型实现代码化,结合MATLAB进行多智能体板形控制系统的仿真,基于安全建模的控制系统较没有基于安全建模的控制系统,容错能力更强,稳态误差更小。基于改进合同网的控制系统较基于传统合同网的控制系统,响应速度更快,超调更合理。更多还原

【Abstract】 Along with the development of artificial intelligence technology, modeling language, contract net, communication mechanism and knowledge base. As a solution of complex systems, Multi-Agent system is becoming more and more popular. In the implementation of Multi-Agent system, almost all of the MAS designers consider security problems when the implementation process of MAS is over. This may lead to vulnerability of MAS.Based on this, this paper recommend a2x2element model with four views based on UML element model to consider modeling process in designing process, running process, agent internal and agent external, achieve a general MAS model. It is good to designer to handle the whole system. It is good to the fault tolerant ability of the system.Considering coordination of MAS, contract net is a good solution, but agents in contract net is not good at handling of fussy information, but mental statement needs fuzzy description, then single agent can achieve intelligent decision, so, in this thesis, contract net is a good base, but not enough, using fuzzy theory in the competitivebidding to make agent more intelligent, not only that, realizing the weakness of contract net including corresponding waste and computational waste, this paper add trust level volatilize in master agent and buffer in all agent. Based on this, a worker agent can choose the task, a master agent can choose a worker agent who is waiting for a response.Flatness control system is a classical control system that with many input, many output, strong coupling and complicated system. Multi-Agent system can make a big and complicated system into small systems, these small systems have a good relationship with each other, they are easy to manage, using coordination mechanism and control theory, it is more likely to get a good result. In this thesis, in connection with different flatness defect, design different control tactics and then package these control tactics into worker agent, give these agents corresponding action and intelligence. Meanwhile, use the modified contract net as the coordination mechanism. When packaging is over, use2×2four view element model language in staruml software. Further, code specific in visual studio software. Finally, make a simulation about Multi-Agent flatness control system combining with MATLAB. Comparing with control system without security modeling, the control system with security modeling has a better fault-tolerant ability, and the steady-state error is smaller. Comparing with control system with traditional CNP, the control system with modified CNP has a fast corresponding speed and the overshoot is smaller.更多还原