【作者】 李艳；

【导师】 朱学峰；

【作者基本信息】 华南理工大学 ， 控制理论与控制工程， 2003， 博士

【摘要】 蒸煮是制浆过程中的一个重要环节,是复杂的化学工业过程,在蒸煮过程中稳定纸浆的Kappa 值是稳定纸浆质量的关键,而且有助于减少蒸汽和化学品的消耗,减少环境污染,提高生产效益。要控制纸浆的Kappa 值,需要对其进行在线测量或者估计,但是至今国内外尚未开发出准确、可靠、价廉的蒸煮过程在线、商用纸浆Kappa 值测量仪表,因此研究纸浆Kappa 值的软测量技术具有很大的理论意义和实用价值。软测量技术是一门新兴的工业技术,发展前景广阔。它利用易测过程变量(辅助变量)以及这些变量与难以直接测量的待测变量(主导变量)之间的关系(软测量模型),通过各种计算和估计方法实现对主导变量的测量。从广义的信息获取角度来看,软测量技术也是一种信息利用和发现规律的方法,在软测量建模过程中要综合利用各种理论、方法,充分挖掘数据中的有用信息,以达到软测量的目的并为进一步设计基于软测量的先进控制打下基础。本论文就以下主要内容进行了深入的研究并取得了以下结果:1) 在分析硫酸盐间歇蒸煮过程机理和生产实际情况的基础上,强调了脱 木素过程是一个分段线性化的过程,指出整个蒸煮过程的单一模型具 有很多局限性,在Hatton 简化模型的基础上,提出了分段机理回归模 型的预测方法。2) 在分析了软测量模型预测误差构成的基础上,结合制浆蒸煮数据样本 的特点,给出了一种综合判别异常样本数据的方法。该方法基于聚类 分析和工艺机理发掘矛盾数据组,并结合回归分析和统计分析,定位 异常样本数据并解释这些异常样本对建模的影响大小。把该方法用于 来自于工厂的实际数据分析,收到了良好效果。来自于实际生产的测 量数据存在着复杂性与准确性等问题,特别是一定数量的矛盾数据和 异常数据的存在,有必要考察数据产生的背景,通过对生产过程的其 他信息的分析,获得校正模型或补偿模型预测残差的方法。3) 通过对蒸煮过程的工艺流程分析,将小波变换引入升温过程分析中, 采用Daubechies 小波变换来获得样本数据所对应的升温过程的特征向 量,构造了一个简化的五维特征向量。根据五维特征向量可以对样本 的大量脱木素的主要阶段的升温过程进行特征描述,并可以进一步用 来进行工况的分类研究。4) 在分析了实际工业过程中经验模型预测精度下降的可能原因的基础更多还原

【Abstract】 Cooking is an important stage in Kraft pulping process. It is also a verycomplicated chemical industrial process. During the cooking process to stabilize theKappa number is the key to stabilize the quality of paper pulp. The steady Kappanumber is also helpful to decrease the consumption of stream and chemical products,to decrease the environment pollution and enhance production efficiency. In order tocontrol the Kappa number of pulp, it must be measured or estimated online.Regrettably, until now the Kappa number online measurement instrument, which isprecision, dependable, low-cost and commercial, has not been developed throughoutinland and overseas. Therefore, it is significant in theory and application to developsoft sensing technology of Kappa number in cooking process.As a rising industrial technology, soft sensing has great developing space. Itemploys easily measured variables (auxiliary variables) and their relationship (softsensing model) with process variables to be measured (primary variables), which ishard to measure directly, by computation and estimation models.From the point of Data-mining, soft sensing technology is an importantmethodology to analyze available information and find out rules. To act as an exactand sensitive soft sensor for the design of the advanced control system based on softsensing technology, during the soft sensing modeling process, it is necessary tointegrate different theories and methods to dig out useful information from theoriginal data.This dissertation concentrated on the research work listed below and achievedsome creative results.1) Based on the technical analysis and the condition of actual product process of thebatch pulp cooking, the dissertation points out the limits of single model for the wholecooking process, since the process of delignification is linearization for differentphase. A new subsection model is presented based on the simplified Hatton model.2) After analyzing the composing of prediction error of soft sensing model, a methodof abnormal data discovery for data processing of Kappa number soft sensing ispresented. The new data processing method digs out incompatible data based on dataclustering and mechanism analysis, as well as finds out the outlier data by regressionanalyzing and statistical analysis. It also can explain the impact of abnormal data onsoft sensing. The method is validated by data analyzing from actual factory cookingprocess. Since there are many problems hiding in the measured data from the actual更多还原