Dissertation > Industrial Technology > Oil and gas industry > Oil, natural gas processing industry > Petroleum refining > The composition, properties and analysis of oil > Chemical Properties and Analytical Methods

Soft Sensor of Naphtha Dry Point on Support Vector Machines Regression

Author LiAng
Tutor WangQingChao
School Harbin Institute of Technology
Course Chemical Process Equipment
Keywords Dry point of Naphtha Soft sensor modeling SVM LS_SVM
CLC TE622.1
Type Master's thesis
Year 2008
Downloads 66
Quotes 1
Download Dissertation

Atmospheric and vacuum distillation unit is the first procedure of petroleum process, the stable of production affects the quality and draw ratio of petroleum products heavily. By using kinds of facilities, oil is fractionated into many products like gasoline, kerosene, diesel oil, etc. Naphtha is one of the main products of distillation process; dry point is an index reflecting its quality. It cannot be measured by conventional instruments. The purpose of this paper is to estimate the dry point value by applying soft sensor. Soft sensing technique consists of 4 parts: variables selection, data processing, modeling, model rectification, among them soft sensing model is the key point. This paper emphasis on the modeling: Support Vector Machine (SVM).Support Vector Machine is a kind of powerful learning method which put Statistical Learning Theory to use. In the paper support vector machine actuality and development is discussed particularly from such four sides: theory study of support vector machine, algorithm improvement, choosing kernel function and parameters, support vector machine expander. And then, standard support vector machine is used to predict the dry point of Naphtha in an oil refinery and analysis the related results. Testing results with data collected from field show SVM is able to meet certain technical requirements in limited accuracy of the online naphtha dry point demanded estimate.Then use LS_SVM machine to predict the dry point of Naphtha in an oil refinery, and compare the test results with standard SVM. Testing results shows LS_SVM are also able to meet certain technical requirements in limited accuracy of the online naphtha dry point, and the LS-SVM has a better ability in non-linear modeling with high learning speed and accuracy. Under the conditions in the same samples, the LS-SVM has a better approximation of the model and the generalization performance than standard SVM, and can save a amount of computing time. Compare with standard SVM, sparseness is lost in the LS_SVM. In this issue, improve the LS_SVM to overcome the drawback, and get the relevant simulation test results. Then use the PCA to improve the input data,make the input data processed easier, and the model compute faster, and use the training data after the application of PCA to be the soft sensor modeling of Naphtha dry point. The application results show that although the training results get bad, but the test accuracy has improved, so the PCA features with good approximation and well generalization ability. Finally, correct Soft-sensor model and use the corrected model in the simulation. Compare the results of corrected model with the standard results. It shows that most points after the correction showed a better approximation, and the whole model has improved.

Related Dissertations
More Dissertations