Soft Sensing Application To Dry Point Of Aviation Kerosene In Distillation Process

Crude oil distillation plant is at the beginning of the process in an oil refinery. By using kinds of facilities, oil is fractionated into many products like gasoline, kerosene, diesel oil, etc. Aviation kerosene is one of the main products of distillation process; dry point is an index reflecting its quality, which is obtained in laboratory every 8 hours. 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. The main method of modeling includes mechanism analysis, regression, artificial neural network, etc. This paper emphasis on the modeling method: Support Vector Machine (SVM). SVM is a newly developed method based on statistical learning theory, and is used in pattern recognition and non-linear function regression. By solving a set of linear equations instead of quadratic programming, Least Squares Support Vector Machine (LS-SVM) is an efficient way of nonlinear modeling. Soft sensor based on RBF neural network, conventional SVM and LS-SVM are discussed in this paper, and applied to predict the dry point of aviation kerosene in an oil refinery. Testing results with data collected from field show LS-SVM has a good ability in nonlinear modeling with high learning speed and accuracy. Methods discussed in this paper lay the foundation of dry point online estimation of aviation kerosene, and make it feasible to apply advanced control on quality parameters of aviation kerosene.