【作者】 谭天乐；

【导师】 李平； 宋执环；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2003， 博士

【摘要】 传统的工业过程建模、控制及故障诊断技术大多基于精确的数学模型，适用于已知运行机理和过程特性的工业系统。然而，工业系统的结构、参数和运行模式在不断的变化，对过程机理的透彻分析和精确数学模型的建立常常需要付出无法接受的代价，系统的控制和故障诊断也就因此面临一个精确建模问题。在实际工业生产中，工程技术人员在对系统机理和数学模型知之甚少的情况下通过观察和经验总结，仍然能够对系统进行良好的手动控制和及时的故障诊断，因此完全可以采用机器学习的智能方法模拟人的这种通过观察学习进行控制的能力。 粗糙集理论是一种通过等价关系和近似概念对数据进行约简以获取知识的方法。粗糙集信息系统是一个基于规则的知识系统，一个基于规则的系统不需要对过程进行数学描述，而是对过程进行经验总结，形式上是“如果…那么…”的规则的集合。规则系统提供了一种了解工业过程的更加直观、简单、易于理解和人性化、智能化的方法。粗糙集理论为基于知识的过程建模、控制与故障诊断提供了理论基础和研究思路。本文的主要目的是试图解决粗糙集在知识获取、机器学习以及工业生产实际运用中所遇到一些主要问题，利用粗糙集在分析处理不完全、不精确和不一致数据中所具有的优势，对复杂工业过程进行基于规则的非机理知识建模，在规则模型的基础上，结合现代控制理论中的有关概念和方法，构造粗糙控制的初步框架，并将粗糙集方法运用于故障诊断。 本文的主要贡献如下： 1、回顾和总结了粗糙集理论的发展和国内外的研究现状，讨论了其在复杂工业过程建模、控制与故障诊断中应用的一般方法。对于粗糙集的开发工具、研究及应用成果以及在工业控制中的发展过程作了系统深入的综述，重点分析和评述了粗糙集理论在工业过程建模、控制与故障诊断中所面临的问题。 2、介绍了粗糙集信息系统中数据预处理的几种方法，分析了连续属性离散化对知识获取的影响。基于布尔逻辑与粗糙集相结合的离散化方法，提出了连续／离散属性混合系统离散化的策略和启发式算法，在信息熵的基础上，提出了一种对离散化结果进行评判的标准，并对布尔逻辑离散化方法和混合离散化算法进行了评价。 3、在等价矩阵概念的基础上，分析了粗糙集知识系统中等价划分与等价摘要 矩阵的关系，采用等价矩阵来表示粗糙集的等价关系，提出了一种对 数据库知识系统进行数据清洗以及从中提取决策规则的矩阵算法，分 析了该算法的计算复杂性。基于已有的决策规则根据对象增减，提出 了在粗糙集信息系统中对决策规则进行动态调整的新策略，并进一步 给出了实现该策略的改进矩阵算法。4、从非线性系统分析方法出发，构造了基于粗糙集决策规则的非线性算 子方程和粗糙状态空间，分析了粗糙状态空间模型的一致性和完备性， 给出了对不一致和不完备的粗糙状态空间模型进行补充的方法。初步 讨论了粗糙状态空间模型的稳定性和可达性，提出了基于粗糙状态空 间模型的非线性过程建模和粗糙控制器设计方法。以一个具有非线性 特性的自来水加药混凝沉淀系统为背景进行了仿真研究，结果表明了 基于粗糙集决策规则的粗糙状态空间模型和粗糙控制器在非线性过程 建模与控制中的有效性。5、利用粗糙集理论自动获取过程工业生产系统中的故障知识，从信息嫡 的角度分析系统知识不确定性的变化，提出了一种基于粗糙集理论的 故障诊断新方法，在前向推理、反向推理以及混合推理的基础上，给 出了针对故障点建立决策表以及利用粗糙集约简所获得的诊断规则进 行正、反向故障诊断的步骤，讨论了这种故障诊断方法的诊断性能及 其在计算上的复杂度。利用粗糙集进行故障知识的发现，寻找系统各 个故障源信号之间合理的逻辑关系，提出了一种构造逻辑故障树的智 能方法，并给出了相应的故障树评价标准。以双效蒸发器系统为背景， 进行了基于粗糙集的故障诊断实例仿真。6、对下一步在基于粗糙集的过程建模、控制和故障诊断等方面将要进行 的工作进行了展望。更多还原

【Abstract】 The traditional approaches of industrial modeling, control and fault diagnosis are mainly based on analytical models. Those models are suitable for system that we known clearly how they work. When facing to complex, dynamic and nonlinear systems, it is usually hard to analysis the mechanism of processes and to build up mathematic models, or it will take us unbearable costs. In most situations a nonlinear system has special properties that are different from the others, so we can’t find out just one or few methods suitable to model and control all those nonlinear systems. An experienced human operator may have little knowledge about a complex system but can still doing good job in system control and fault diagnosis by observing signals of inputs and outputs. Therefore, the problem is, can we use some techniques of machine learning and artificial intelligent to mimic the human ability of "learn to control".Rough set theory is a new mathematical and AI technique in knowledge discovery. Rough set information system is a rule based knowledge system. A rule based system does not require a classical mathematical description of the process, but consist of sets of If....Then.... rules instead. By rule based system we can understand industrial processes in a visualized, simple, understandable, humanistic and intelligent way. The aims of this dissertation include: try to solve problems in rough set based knowledge discovery and machine learning; build up knowledge model for complex industrial processes; following the concepts and approaches of nonlinear system control, construct a control system framework based on rough state space; apply rough set theory to fault detection and diagnosis (FDD).The main contributions of the dissertation are as follows:1 Reviewed the developments and research situation of rough set theory. Discussed the common approaches in industrial modeling, control and FDD. Investigations about implemental system and achievements in industrial applications of rough set theory have been done. The problems of rough set theory in industrial modeling control and FDD are pointed .2, To be an improvement of discretization method based on rough set theory and Boolean reasoning, algorithms for discretization in mixed realand discrete attributes system are proposed, functions based on information entropy are presented to analysis the results of discretization quantitatively, and the effects of discretization for knowledge acquisition are discussed.3 The concept of equivalence matrix, which expresses equivalence relation in rough set information system, is introduced; the relations between equivalence matrix and equivalence classes are discussed. The algorithms for data cleaning and rules extraction in knowledge system based on matrix computation are proposed and their complexity of computation is analyzed. For online learning of a rule-based knowledge system, strategies of rule updating in rough set information system and matrix computation algorithms for dynamic modification of rules are given.4 Based on definitions of nonlinear system methodology, rough set based state space model is proposed. Definitions of rough set state space are given. Stability, reachability, consistency and completeness of this new state space model are discussed in detail, respectively. A rule-based structure of Internal Model Control (IMC) system is given, and a rule-based IMC controller is built up.5 Approach for rough set based FDD is proposed, and its applicability and computational complexity is discussed. Knowledge about the faults of a process control system is obtained automatically by rough set reduction. An entropy-based criterion is used to measure the uncertainty of it. Methods of forward and backward fault diagnosis and how to build up decision tables for each fault source are given. Further more, approach for FDD based on rough set theory and Logical Fault Tree ( LFT) is presented. The feasibility of forming a knowledge discovering and intelligent decision-making system for these a更多还原