Font Size: a A A

Soft-sensing Modeling Based On SVM And Its Application In Polypropylene Melt Index

Posted on:2009-12-15Degree:MasterType:Thesis
Country:ChinaCandidate:Z S ChiFull Text:PDF
GTID:2178360245499628Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
As a rising industrial technique, soft-sensing technology has great developing space. In this paper, soft-sensing research of melt index (MI) was investigated for the polypropylene unit of Petrochemical Branch in North China of China Petroleum.Firstly, according to the principle of propylene and the experience of industrial field operator, auxiliary variables were chose. Measurements of auxiliary variables were acquired and preprocessed to remove the effect of outliers and stochastic error. In industrial operation, the production grades changes frequently and the range of variables for two production grades is large. For the reasons above, the data collected from production grade transitions process were used to determine the interval between the process measurements time and its corresponding test results time.There are some special points in the modeling data which only accounted for a small part. They can't reflect the majority cases of the production process and can't be removed simply because much important information may be contained. According to this, the clustering weighted SVM (CWSVM) was proposed, The CWSVM could establish more reasonable model than SVM when the special points was considered. Compared to SVM, the proposed approach makes the model contain more information of the production process. The simulation results on non-linear function fitting and soft-sensing of crude petrol of FCCU demonstrate that the proposed methods are more effective than SVM. When applied to on-line identification and other field, SVM often has long training time. The feasible solution construction (FSC) based on the positive definition quadratic programming tight constraints algorithm (PDQPTCA) was proposed to solve the problem by taking the features of the construction and application fields. The solving speed of SVM can be accelerated by constructing the initial feasible solution. For online study of the CWSVM, two clustering methods were proposed, compared and analyzed. The soft-sensing results on a simulated continuous stirred tank reactor (CSTR) show that the FSC can accelerate the solving speed of SVM.Finally, the CWSVM was applied to soft-sensing of MI of polypropylene for Spheripol technics. The simulation results on polypropylene MI demonstrated that the CWSVM is more accurate than SVM in soft-sensing modeling.
Keywords/Search Tags:soft-sensing, weighted SVM, cluster analysis, polypropylene melt index
PDF Full Text Request
Related items