Soft Sensor Study Of Gasoline Dry Point On Atmospheric Tower Top

Atmospheric and vacuum distillation is a complex process of physical and chemical change.Due to the large number of products and the serious coupling between many variables,the correlation between variables increases,and it is often necessary to change the crude oil production plan according to the current market conditions.The dry point of the atmospheric tower directly affects the quality,output and energy consumption of the products.On this premise,the soft sensor method can be used to estimate the dry point of the top of the screened atmospheric tower on line,so as to realize the inferential control of the dry point of the screened atmospheric tower.Its specific contents include:1.Selection of auxiliary variablesAccording to the mechanism law of atmospheric and vacuum distillation process,based on the understanding of the operation rules of a refinery,combined with the field production process data,this paper preliminarily selects 14 variables that affect the dry point of atmospheric tower top.2.Data collection and processingDue to the complexity of field data collection variables,there are more or less gross errors and random errors caused by human or process complexity.Therefore,box chart is used to check whether the data distribution conforms to the operating standard of working conditions,and then Laida criterion is used to process gross errors and correct the data.Then the digital filtering method is analyzed and compared,and finally the weighted recursive average filtering method is used for data smoothing.3.Dimension reduction processing of original dataDue to the complexity of atmospheric and vacuum distillation process,there are serious coupling and nonlinearity among the variables.Sparse Principal Component Analysis（SPCA）is introduced into KPCA algorithm（KPCA）during data preprocessing in this paper,and the input variables of the model are selected by KSPCA algorithm,which realizes the nonlinear dimensionality reduction between data,simplifies the structure of KPCA,and increases the sparsity among KPCA variables.4.Establishment of soft sensor modelBased on support vector machine（SVM）and least square method,this paper analyzes and studies the soft sensing model of gasoline dry point on the top of atmospheric tower with the least square SVM.The data preprocessed by PCA,KPCA and SKPCA in the previous chapter were input into the least-squares support vector machine（LSSVM）model of atmospheric tower top gasoline dry point prediction.The feasibility of SKPCA algorithm in the previous chapter was verified through MATLAB simulation analysis,and the superiority of SKPCA-LSSVM model was also proved.