【作者】 王昕；

【导师】 梁军；

【作者基本信息】 浙江大学 ， 系统工程， 2006， 硕士

【摘要】 在许多工业控制场合，存在着一大类变量，由于技术或经济的原因，目前尚难以或无法通过传感器进行检测，但同时又是需要加以严格控制的、与产品质量密切相关的重要过程参数。这将影响到产品质量及系统的稳定性，会给企业带来不可低估的经济损失。软测量技术应运而生。本文系统和深入的研究了软测量技术的若干重要方面，以偏最小二乘和神经网络的理论为基础，建立了多种软测量模型，并比较了各种模型的预测效果。以某石化企业的Spheripol工艺PP装置为例作为应用背景，通过从现场采集的数据建立聚丙烯熔融指数的软测量模型，并运用建立的模型对现场数据进行软测量预测，得到了较好的效果。本文主要研究内容如下：1．阐述了软测量技术的基本概念及其发展，分析了软测量的建模思想以及软测量建模的四个要素，软测量模型的工业实施。对工业聚丙烯的生产过程做了简要的介绍，特别介绍了Spheripol聚丙烯聚合工艺及其工艺流程。2．对主元分析(PCA)，偏最小二乘(PLS)，人工神经网络(ANN)等在软测量建模中要用到的一些方法做了理论介绍。为后面的应用打下理论基础。3．分别通过PLS算法和RBF神经网络算法建立了软测量模型，并以通过Matlab仿真双效蒸发工业过程得到的数据作为数据样本对模型进行检验。然后将PLS模型与RBF神经网络模型结合，以PLS算法作为外部模型，RBF神经网络算法作为内部模型得到PLS-RBFNN模型。最后对三种模型效果进行了比较。4．介绍了聚丙烯聚合反应机理，分析了影响聚丙烯熔融指数的多种因素，分别选择了建立均聚产品和共聚产品软测量模型的过程变量。通过从现场采集的数据建立PLS模型和PLS-RBFNN模型，并运用建立的模型对现场数据进行软测量预测，得到了较好的效果。最后，对全文做出总结，并对软测量的发展进行了展望。更多还原

【Abstract】 There exist a good many variables in the industrial process. At present, it is difficult, sometimes impossible, to check and measure the variables, which is closely related to the quality of production, through the sensor due to technological or economic restriction. It will affect the stability of the system, which will bring about large economic losses. Therefore, soft sensing comes into being.This paper goes deep into several important aspects of soft sensing technology, creating multiple models of soft sensing based on partial least square (PLS) and neural networks (NN). Then it compares the result of prospect. Combining the actual application of industrial polypropylene (PP) process, soft sensing models have been created with the data collected from the field and used to make prospects, which brings about good effects. Contents in this paper are as follows: 1 .It covers basic conceptions and their development of soft sensing technology andanalyzes the model-creation idea, the four factors and implement of soft sensing.The production process of industrial PP is briefly introduced. 2.The theories of PCA, PLS, and ANN are introduced to pave the way for laterapplication of soft sensing technology. 3.Soft sensing models are set up according to PLS and RBF and tested through thedata sample which is collected from the simulation of double evaporation. ThenPLS-RBFNN is gained by combining the model PLS with RBFNN. Finallycomparison is made among the three models. 4.Reaction mechanism of PP is introduced, multiple factors influencing melt index(MI) of PP are analyzed and process variables of equal polymerization andmultiple polymerization are set up. Model PLS and Model PLS-RBFNN are set upwith the data collected from the field and used to make prospects of other data.Good results have been received.Finally, conclusion is made, as well as the expectation for the future development of the soft sensing.更多还原