【作者】 朱振宇；

【导师】 苏宏业； 张泉灵；

【作者基本信息】 浙江大学 ， 控制理论与控制工程， 2006， 硕士

【摘要】 近年来数据挖掘技术在流程工业领域得到了大力的推广,一方面得益于数据挖掘技术的兴起和快速发展。随着数据库技术的迅速发展以及数据库管理系统的广泛应用,人们积累的数据越来越多。数据的丰富带来了对强有力的数据分析工具的需求,在这样的应用需求下,数据挖掘技术应运而生,并且逐步从实验室走向了现实应用,在商业、工程技术、科学研究等方面得到了广泛的应用。另一方面信息技术对流程工业的大力推动,过程检测和控制硬件设备的广泛应用,为数据挖掘技术在流程工业中的应用创造了十分有利的条件。但是由于流程工业数据的复杂性,这一应用又面临着许多的新难题。 论文以三唑磷合成工业为背景,以粗糙集理论为工具,对数据挖掘技术在流程工业中的应用进行了研究。 首先对三唑磷合成工业中的生产数据进行数据预处理。流程工业中数据存在着非常严重的空缺值、噪声和滞后问题。首先对生产数据进行压缩,将数据量控制在一定范围内,其次建立一个基于最小二乘法的二阶回归函数,以一天内的数据量为窗口,滑动窗口,对窗口内的数据进行拟合,去除空缺值和噪声。再次通过对生产数据时间序列的调整,消除滞后,计算得到比较准确的收率值。最后提出一种基于数据分布特征的全局有监督的离散化方法对连续的生产数据进行离散化处理,得到离散化数据。 运用基于粗糙集理论的数据挖掘算法对离散化数据进行数据挖掘。首先建立一个全局的决策系统,通过分析其不一致性,将其分解成一个完全一致决策系统和完全不一致决策系统。其次对完全一致决策系统,采用了一种面对决策属性的规则提取方法,简化了规则提取的复杂度。再次针对完全不一致决策系统则直接生成带粗糙算子的决策规则。最后得出了生产优化的方案。 三唑磷的生产实践证明了优化方案的有效性。更多还原

【Abstract】 In recent years, Data Mining becomes increasingly important and has widely applied in process industry, mainly for two reasons. On one hand, data mining technique is developing very quickly. Due to the development of data acquisition and database technique, a huge stock of data are accumulated in human’s activities, therefore, a powerful analysis tool is needed to deal with the data. Data mining is the solution for this problem and has found its applications in various areas such as business, engineering technology and scientific study. On the other hand, the phenomenon that the information system has greatly facilitated the process industry and the equipment of process check and process control widely applied in the process industry has created a favorable environment to apply data mining to the process industry. However there still exist many allication problems because of the complexity of process data.In this thesis, data mining technology based on rough sets for triazophos synthesis process is discussed as follows:Firstly, the production data of the triazophos synthesis process are preprocessed because there are many noisy, missing, and lagged data in the process industry. Condense the data and reduce them to a proper numbert. A nonlinear regression based on the method of least squares is estimated and polyfit the data in a window, thus eliminate the noisy and missing data, adjust the time series data, eliminate the time delay and get the accurate result. And finally a global supervisory discretization based on data distributed character is proposed to turn numeric attributes into discrete ones.The discrete data are mined by the arithmetic based on rough set theory. Firstly a global decision system is established and estimated and then divided into two parts: one is a consistency system and the other a non-consistency system. For the consistency system an algorithm for acquisition of decision rules to the decision attributes is proposed;for the non-consistency, the decision rules are got directly. Finally, the solution for the discrete data is obtained.This solution has been proved to be effective in a triazophos plant.更多还原