化工论文编号:HG219  论文字数:19945,页数:40

目  录
中文摘要 I
英文摘要 II
目录 III
1 绪论 1
1.1软测量模型及其方法............................................ 1
1.1.1 机理建模.................................................2
1.1.2 回归分析.................................................2
1.1.3 状态估计.................................................2
1.1.4 基于模式识别.............................................3
1.1.5 神经网络.................................................3
1.1.6 模糊系统.................................................4
1.2 存在的问题....................................................4
1.3 模型校正......................................................5
1.4 化工过程的操作优化............................................5
1.5 PTA装置溶剂脱水塔的现状......................................6
2. 软测量建模方法——回归分析 8
2.1引言..........................................................8
2.2多元线性回归(MLR).............................................9
2.2.1多元线性回归模型推导.....................................9
2.2.2实例验证................................................10
3. 软测量建模方法一人工神经网络.......................................13
3.1　引言.........................................................13
3.2　神经网络的特点和及应用.......................................14
3.3　多层前传网模型...............................................15
3.3.1　网络拓扑结构............................................15
3.3.2　目标函数和训练算法......................................16
3.4　实例验证.....................................................19
4.回归分析与ANN结合的混和建模........................................23
4.1　引言..........................................................23
4.2  PCA－BP神经网络的软测量建模.................................23
4.2.1　输入变量的初选...........................................23
4.2.2　隐层节点数的确定.........................................24
4.2.3  PCA－BP神经网络模型结构................................24
4.2.4  BP神经网络的LM优化算法................................24
4.3实例验证.......................................................25
5.实验结果与讨论.......................................................28
5.1 实验结果......................................................28
5.2 讨论..........................................................28
6. 总结与展望..........................................................31
6.1 本文工作小结..................................................31
 6.2 今后的研究发展方向............................................32
致谢 ...................................................................33
参考文献 ...............................................................34

摘 要：本文主要以PTA装置溶剂脱水塔为研究对象，运用统计回归方法和人工神经网络技术，建立溶剂脱水塔出口产品组成的PCA-BP神经网络软测量模型。本文完成的工作主要有：
1)本文在系统分析了基于PCA-BP的建模原理后，对人工神经网络模型的关键参数即隐层节点数的确定作了较为深入的探讨，总结出了比较有效的取值方法。并在PTA装置溶剂脱水塔产品组成的软测量建模中得到应用，取得了较为满意的结果。
2)研究了基于统计回归的溶剂脱水塔塔顶塔底产品组成的软测量建模方法。并选用多元线性回归(MLR)方法对溶剂脱水塔产品组成进行建模，对所建模型进行了验证，其结果表明模型的精度达到了预期的要求。
3)研究了基于BP算法的人工神经网络对溶剂脱水塔产品组成进行建模的方法，并用现场实测数据对所建模型进行验证测试。
本文最后，对统计回归和神经网络技术在过程软测量建模的应用进行了展望，并提出了今后建模过程中需要进一步深入研究的方向。

关键词：溶剂脱水塔；多元线性回归；软测量；主成分回归；人工神经元网络

Abstract: This paper is a study of PTA solvent dehydration tower. The author used a statistical regression method and the technology of artificial neural network, and built a soft-sensing model of the PCA-BP neural network which made up the solvent dehydration tower products for export. In this paper included the following aspects: firstly, the author analyzed the main parameter of the artificial neural network based on the principles of PCA-BP modeling, and summed up a more effective method. And this method has been applied to achieve a satisfactory result. Secondly, the author added up the soft-sensing approaches to both the top and bottom products of solvent dehydration tower. Based on these approaches, the author chose a multiple linear regression (MLR) method in modeling the solvent dehydration tower products, and also inspected those models. The result showed the accuracy of those models has achieved the expected standard. Thirdly, the author studied the method based on the BP artificial neural network in modeling the solvent dehydration tower products, and tested the models through the data on the scene. At last, the author took a long view on statistical regression method and the technology of artificial neural network used in soft-sensing modeling, and then pointed out some ideas for further work.

Keywords: Solvent Dehydration Tower; Multiple Linear Regression (MLR); Soft-sensor; Principal Component Regression(PCR); Artificial Neural Network.