【作者】 王树锋；

【导师】 吴耿锋；

【作者基本信息】 上海大学 ， 计算机应用技术， 2008， 博士

【摘要】 目前,大部分的知识系统都只能处理那些具有良好的结构、封闭的范围、协调的内涵、明确的外延、完整的内容等特征的规范知识,一旦出现超出系统边界的问题或稍稍违背这些特征的知识,系统就可能崩溃,表现出其极端的脆弱性。由于实际应用的迫切需要,计算机科学的研究发生了许多重大的变化。人们从注重研究对象的形式转向研究对象的内容;从注重研究良性结构问题转向研究病构问题;从注重研究封闭性世界转向研究开放性世界;从研究内涵完整协调和精确的问题转向研究内涵不完整、不协调和不精确的问题。这些趋势在知识处理的研究中表现为一个过去研究得较少的十分困难的课题,即非规范知识处理。所谓知识的非规范性,是指知识内涵的难处理性,包括知识的不确定性(模糊、不确定、随机和不精确的知识),或知识的不完整性(内容不完整的知识和结构不完整的知识),或知识的不协调性(含矛盾的知识、带噪音的知识和含冗余的知识),或知识的非恒常性(时变知识和启发式知识)。非规范知识处理的最典型应用领域是因特网上知识的处理。因特网上的知识大部分是非结构或半结构的,它们以各种媒体形式存在,以自然语言为载体,分布在几亿个网页上,每天以百万网页的数量级在增长、消失或改变内容,它充满了各种矛盾的事实、数据和观点。因特网的快速发展与广泛应用要求在开放、动态环境下实现灵活的、可信的、协同的、深层次的知识共享和利用,这个目标的实现在很大程度上依赖于非规范知识处理技术的进步。盾构隧道工程等大型工程是一个具有明显多变量、非线性、不确定性和时滞特征的复杂系统,涉及到海量的实时数据。从这些海量数据中发现(非规范)知识是另一个典型的应用领域。这些海量的实时数据需要用多种方法和数学模型去刻画,任意单一模型的使用效果均有局限,各种模型所得结果的综合是一大问题。这又提出了各种非规范知识的融合问题。本文以因特网和盾构隧道施工风险知识管理中的知识处理为主要背景,围绕非规范知识获取和融合问题,探索和研究了若干关键技术,并把它们应用于实际问题的解决中。本文首先介绍了非规范知识的定义,综述了非规范知识处理在基础研究和应用研究方面的研究现状。第二章基于粗糙集理论,研究了内容不完整的非规范知识的获取。提出了基于属性扩展的规则抽取算法(AEA)和基于粗糙集的近似序列决策规则挖掘算法(ASDREA)。AEA算法选择原始决策表中的最重要属性形成优化属性表,从该表中抽取的规则一般是不一致的,通过添加新的属性就可消除这种不一致。ASDREA算法处理粗糙边界数据,按用户需求挖掘出边界区域中的全部知识或满足特定要求的知识。第三章基于神经网络及神经网络集成,研究了带噪音的非规范知识的获取。提出了从以阶跃函数作为激活函数的多层感知机神经网络模型(SAMLP)中获取规则的方法和基于SAMLP模型的网络集成及集成网络规则获取算法。提出了一种混合粗糙集和神经网络方法的决策规则挖掘模型,直接从内容不完整的和带噪音的信息表中获取规则。第四章基于信念修正理论,研究了矛盾知识的融合。提出了一种基于群体信念协商的矛盾知识融合模型,讨论了该模型的公理系统和模型的实现。研究了可能逻辑语义下的知识融合方法。根据被融合资源是否一致、矛盾、冗余或独立,研究了多种融合操作符,设计了一个融合知识的框架。第五章研究了非恒常知识的融合。基于信息过滤技术,将非恒常知识转换为用概率论、证据理论和可能性理论表示的不确定结构化知识。提出了融合同种类型的不确定结构化知识和不同类型的不确定结构化知识的方法。第六章开展了非规范知识获取和融合技术在隧道施工风险控制知识中的应用研究。将本文的研究成果应用到《盾构施工风险控制知识管理系统》项目中,建立了一个盾构施工风险知识管理的示范性平台。该平台实现了从风险辨识、风险评估、风险预警、风险控制以及风险后处理的风险管理流程,从全新的视角,重新思考和探索盾构隧道工程风险控制的技术和策略,促进风险控制技术的创新。更多还原

【Abstract】 At present,most of the knowledge system can only deal with canonical knowledge which has the characteristics of well structure,enclosed area,coordinated content,clear extension and integrity content.If a system encounters the issues which are slightly beyond or slightly contrary to the knowledge of these features,it may face collapse,and show its extreme vulnerability.Due to the urgent need of practical applications,the research of computer science has had many tremendous changes.The people change the research from the problem of object-oriented to content-oriented;from the problem of well structure to ill structure;from the closed-oriented world to open world;from the problems of consistent and precise content to incomplete, inaccurate and inconsistent content.These trends in the knowledge processing research show an extremely difficult topics which was less studied in the past,namely non-canonical knowledge processing.The non-canonical knowledge is refers to the intractability of knowledge connotation, including the uncertainty knowledge(fuzzy and uncertain knowledge,random and imprecise knowledge),or the incomplete knowledge(contents incomplete knowledge and structure incomplete knowledge) or the non-consistent knowledge(contradiction knowledge,noise and redundant knowledge) or the non-permanent knowledge(time-varying knowledge and heuristic knowledge).The most typical application of the non.canonical knowledge is the knowledge processing on the Internet.However,the most knowledge on the internet is non-structural or semi-structural, and it exists in various forms.The knowledge takes natural language as the carrier and locates in the hundreds of millions of pages,which are increased,disappeared or changed by one million of magnitude per day,and are fraught with contradictions of the facts,data and viewpoints.The internet’s rapid development and wide application require a knowledge sharing and utilization with flexible,credible,concerted,in-depth way in an open,dynamic environment.Therefore,the realization of this goal to a great extent depends on the advancements in non-canonical knowledge processing technology.The tunnel project and other large projects are the significantly complex systems which have more variable,non-linear,uncertainty and delay characteristics.There is a huge of real-time data in these projects.Extracting non-canonical knowledge from these data is another typical application area.These data need to use various methods and the mathematical model to be described.The effects of using an arbitrary single model is limited,the results synthesis of all model is a big problem,so that the problem of canonical knowledge integration is proposed.The dissertation takes the knowledge processing of the Internet and the Shield Construction Risk Control Knowledge Management as the main background,circles the non-canonical knowledge extraction and fusing issues to explore and research a number of key technologies,and applies them in the solution of practical problems.The dissertation firstly introduces the definition of non-canonical knowledge,and then summarizes non-canonical knowledge processing research status in basic aspect and applied aspect.The second chapter of the dissertation studies the extraction of the content incomplete knowledge based on rough sets.We proposed an Attribute Extension Algorithm of irregular decision table(AEA) and an Extraction Algorithm of Approximate Sequence Decision Rules (EAASDR).AEA selects the important attributes from the original decision table and forms an optimal decision table.The rule extracts from this optimized table is possibly inconsistent and adds new attributes can eliminate such inconsistencies.EAASDR can extract complete knowledge or the knowledge met the specific request from the boundary region of decision information systems.The third chapter of the dissertation researches extraction of the noise knowledge based on neural networks and neural network ensemble.We propose the Staircase Activation function Multi-Layer Perceptron(SAMLP) using a step function as the activation function and extract rules from the SAMLP.We also propose an ensemble SAMLP and use it to extract rules.We present a Hybrid Model based on Rough sets and Neural networks(HMRN) extracting rules from the incomplete and sound information table.The forth chapter of the dissertation researches fusion the contradictions knowledge based on belief revision.We study a model of fusing inconsistency knowledge based on social belief negotiation and consider both a postulational and a procedural approach to social contraction. We also study the knowledge fusion method in the possible logical semantics.According to whether or not the information resources are consistent,contradictory,redundant or independent, we study each kind of fusion operator and design a fusion knowledge frame based on the standard possible logic and explain the frame implementation with an example.The fifth chapter of the dissertation researches fusion non-permanent knowledge.Based on the information filtering technology,non-permanent knowledge is converted into uncertainty structure knowledge expressed with three kinds of types(probability theory,evidence theory and possible theory).The dissertation introduces the method to fuse homogeneous type uncertain structure knowledge and the heterogeneous type uncertain structure knowledge.In the sixth chapter we have launched an application research of extraction and fusing non-canonical knowledge in the project of Shield Construction Risk Control Knowledge Management System.With the extracting and fusing technology of non-canonical knowledge, the project has established a model platform of Shield Construction risk knowledge management. The platform realizes the risk management process from the risk identification,risk assessment, risk warning,risk control and risk post-processing.The platform offers a new angle to rethink and explore shield tunnel project risk control techniques and strategies,and to promote risk-control technology innovation.更多还原