【作者】 杨丹；

【导师】 潘丰；

【作者基本信息】 江南大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 随着计算机技术与控制技术的发展与应用,系统设备的自动化程度大幅提高。故障原因、故障过程和故障现象错综复杂,计算智能领域的一些理论,如人工神经网络、粗糙集理论等已经在故障诊断中得到了广泛的应用。粗糙集理论是一种处理模糊和不确定知识的工具,它简化决策规则,提取有效的信息,能够消除知识冗余性的问题。但粗糙集的容错能力还不够理想,当核属性受噪声污染时有可能会出现误判的情况。因此为了提高粗糙集的容错能力,将粗糙集与神经网络相结合,构建粗糙集(RS)和神经网络(ANN)相结合的系统,充分利用粗糙集理论对知识的约简能力和神经网络的容错学习能力。本文利用二者的优势,将RS和ANN相结合,提出了一种基于粗糙集-神经网络的故障诊断模型。首先对粗糙集理论的数据约简问题进行了深入探讨。数据约简是粗糙集理论的核心内容之一。主要研究了运用粗糙集理论进行属性约简的方法,由于运用粗糙集理论进行属性约简是一个典型的NP(多项式复杂程度的非确定性)问题,提出了一种基于改进的二进制可辨识矩阵与布尔代数的属性约简方法,从而大大简化了可辨识矩阵,使属性约简计算量大幅度减小,可以快速得到给定要求下的属性约简。其次,对基于神经网络的连续数据离散化与故障诊断进行研究。粗糙集理论是一种基于离散数据进行处理的方法,连续数据的离散化直接影响到它的处理效果,为此提出了一种基于自组织映射神经网络的离散化算法,使得权值向量在输入向量空间中相互分离,形成各自代表的输入模式,实现特征自动识别的聚类分析功能。仿真结果验证了方法的有效性。最后,采用基于粗糙集-神经网络故障诊断方法,对田纳西-依斯曼过程(TEP)过程进行故障诊断,取得了良好的故障诊断效果。更多还原

【Abstract】 With the development and appliment of computer and control technical, the automation of system devices has being elevated sharply. The reason、process and phenomenon of fault is complicated. Some theory in computational intelligence fields, such as artificial neural network, Rough Set theory has been widely used in fault diagnosis. Rough Set theory is a tool for fuzzy and dubious knowledges. It can simplify decision rules and find useful information. It also can settle the problem caused by redundant data. But the fault-tolerance ability of rough sets is insufficiently ideal. When core’s attributes are polluted by noise, it may arouse error judgment. Therefore in order to enhance the fault-tolerance ability of Rough sets theory, unify Rough sets and neural network, construct an intelligent mixture system of Rough sets and neural networks, which fully develops the reduction ability of Rough sets and the classification ability of neural networks. In this paper, combining RS and ANN, using the both advantages, giving a fault diagnosis system model based on the rough set-neural network.Firstly, studied the attributes reduction methods used in rough set theory. The knowledge reduction is the core part of Rough sets, but it is a NP(Non-deterministic Polynomical) problem on theory. And on this basis giving a method based on the improved binary matrix and boolean algebra for attribute reduction, reduce the amount of computation, the request of attribute reduction can be quickly given.Secondly, researching the discrete method. Rough set theory is based on discrete data processing methods, discretization of continuous data directly impact on its effect. In this paper given a method based on SOM(Self-Organizing feature Map), transformated the input vector values which are detached from each other into a new input mode. The results of example demonstrated the effectiveness of this method.Finally, we used the fault diagnosis system by rough-neural network on Tennessee-Eastman process(TEP) fault diagnosis, it achieved good diagnosis result.更多还原