【作者】 唐李真；

【导师】 杨君锐；

【作者基本信息】 西安科技大学 ， 计算机应用技术， 2004， 硕士

【摘要】 本文研究了模糊智能系统的建模过程，并结合模糊逻辑、神经网络和知识工程等理论和方法，提出了一种利用模糊知识建立智能系统的可行途径，并依照此思想设计和实现了基于模糊知识的智能系统开发环境。本文从知识表示和推理机制两个方面阐述了智能系统开发环境的设计和实现。知识表示方面。实现了知识的异构表示。在知识获取模块，知识表示采用了基于XML的模糊知识框架表示方法；在推理模块，把XML框架解析为JAVA知识对象。这两种知识形态的存在使得系统在存储时具有可读性、用户友好性、平台无关性和网络传输性；在推理时具有封装性、智能性和处理速度快等特点。推理方面。扩展了传统的Mamdani推理模型，可以同时对多种模糊知识（精确值、隶属函数、语言值等）进行匹配推理；在推理控制策略上选择了反向推理和回溯算法；采用了模糊变量的语义距离来计算模糊变量之间的匹配程度；同时设计了语气算子处理模块，使得系统推理更加符合人类的思维习惯。隶属函数生成器。采用模糊C－均值聚类算法对模糊变量的样本值进行分类，利用基于Levenberg—Marquardt优化算法的神经网络实现隶属函数的生成，并把学习结果导入XML知识文件。传统的智能系统开发工具大都采用命令行的方式，设计了窗口命令的方式实现与用户之间的交互，这降低了系统的使用难度，增强了系统的透明度。更多还原

【Abstract】 Based on researching the modeling of fuzzy intelligent system, combined with such theories and methods as fuzzy logic, neutral network, knowledge engineering and so on, a practicable approach of establishing an intelligent system making use of fuzzy knowledge is presented in this paper and according to this idea, a fuzzy knowledge based environment of intelligent system is designed and implemented. It is from two aspects of both knowledge representation and reasoning mechanism that the design and implementation of the environment are detailed.On the knowledge representation, isomerous representation of knowledge is realized. In knowledge-acquiring module, the notation of fuzzy knowledge frame based on XML is adopted; in reasoning module, the XML frame is parsed into JAVA knowledge objects. The existence of these two forms of knowledge makes system knowledge readable, friendly, platform-independent and network-transmittable when stored, and meanwhile makes it encapsulated, intelligent, quick-processing and etc. when reasoned.On the reasoning, the traditional reasoning model of Mamdani is expanded and various kinds of fuzzy knowledge can be matched and reasoned at one time; on the control strategy of reasoning, reverse reasoning and trace algorithm are selected; semantic distance of fuzzy variable is used to calculate the matching extent between fuzzy variables; at the same time, a processing module of mood operator is designed, making the reasoning system more accordant to human thinking custom.On the member function generator, the sampling values of fuzzy variable are classified by the algorithm of fuzzy C-mean clustering, the member functions are generated through BP network based on the L-M optimized algorithm and the learning results are imported into files of XML language.In stead of using the conventional method of command line adopted by most intelligent developing tools, the method of window command is designed to achieve the intercourse with <WP=4>users, which reduces using difficulty and enhances the transparency of the system.更多还原