基于偏差校正的建模法在BPA结晶过程中的应用

从推理控制的基本思路出发,采用推算出干扰量继而实现前馈控制的方法,本文提出了一种基于输出偏差校正(deviation correction)的建模方法.该方法与支持向量机(support vector machine,SVM)相结合,形成DC-SVM算法.将DC-SVM方法应用于双酚A结晶过程软测量模型中,所得到的回归模型输出与SVM方法相比.塔底双酚A组分的最大相对误差和平均相对误差分别从7.97%和2.4%降到5.51%和1.8%,塔底苯酚组分的最大相对误差和平均相对误差分别从4.93%和1.11%降到3.46%和0.96%.本文提出的DC-SVM建模方法,可以广泛应用在回归建模中,也适合应用于实践生产过程,但需要足够的数据作为建模的基础.

Modeling method based on deviation correction and its application to the crystallization procedure for BPA

According to the basic direction of inferential controlling,a new way of building model is referred in this thesis.It is based on the output of Deviation Correction,using the computative interference value to achieve the feed forward control.The new modeling method is linked with Support Victor Machine(SVM)to form an arithmetic called DC-SVM.DC-SVM can be applied to the test of the soft sensor of the crystallization procedure for Bisphenol-A.The result from the test is that,SVM compared with the output of regression model,the maximal and the average relative errors for the element of Bisphenol-A which on the bottom of the tower separated,reducing from 7.97% to 5.51% ,and 2.4% to 1.8%.The maximal and the average relative errors for the bottom of the tower of phenol are also falling from 4.93 % to 3.46%,and 1.11% to 0.96%.Based on enough data,DC-SVM modeling method will be popular on the appli- cation of regression model and practice of the manufacture.