基于粗糙集和神经网络的属性约简方法研究-控制理论与控制工程专业论文.docxVIP

信息T程人学碗I学位论文摘 信息T程人学碗I学位论文 摘 要 粗糙集理论1982年由z．Pawlak提出，经历了近20年的发展，已经在理论和应用上 取得了丰硕的成果。它不依赖于数据集之外的附加信息，是处理含有噪声、不精确、不完 整数掘的有力工具，是一种新的数据挖掘技术。通过对约简算法的研究，提出了一种基于 属性重要度和属性依赖度的简单有效的约简算法，并从理论与实验两方面验证了它的作确 性及有效性。 人工神经网络是对人脑生物神经网络的一种模拟，它通过将若干具有处理功能的神经 元节点按一定结构连接起来而构成，适用于处理模糊、不精确数据和复杂的非线性映射问 题。 敏感性分析是精简网络结构的一种常见的简单易行的方法。论文在分析各类神经网络 的基础上，着重研究了RBFNN在删除冗余属性方面的研究。依据概率测度收敛的定义给 出了随机变量函数的导数的求法，并给出了径向基函数神经网络在逼近函数的过程中，对 各个输入变量的敏感性的计算式。经过与文献[121中所提出的敏感性定义作比较，说明了 这种敏感性的定义方式是文献[12]dP所提出的定义的推广。 关键词：粗糙集属性约简径向基函数神经网络敏感性分析 第1贞 竹息T程人学颀I：学位论文Abstract 竹息T程人学颀I：学位论文 Abstract Rough Set(RS)theory Wilts put forward by Z．Pawlak is law in 1982．After about 20 years of developing，it has received fruitful achievements in both of theory and applications．RS doem’t depend 0n additional information beyond the data set，which is a potent tool for dealing、^，ith vague，imprecise，incomplete and uncertain data,and is a new technology in Data Mimng．After the research in the reduction algorithm，a new reduction algorithm based on Attribute importance and Attribute dependencies is proposed．This algorithm proves its validity in theory and in experiment． Artificial neural network(ANN)is a kind of simulation of the biological neural network of human brain．It is constituted by several neurons，which are connected according to sortie rules and have processing capacities．ANN is suitable for dewing with the蜮problem that inaccuracy data and complicated nonlinearity shine upon． Sensitivity analysis is a common easy method to simplify network structure．The paper have studied RBFNN research in deleting redundant attribute emphatically on the basis of analyzing all kinds of neural networks in thesis．Provide random variable law of asking of derivative of function according to probability definition disappeared to estimate，provide the base function nerve network of the radial during the process of approaching the function，calculation type to each sensitivity of inputting the variable．Through define，make comparisons with sensitivity put