Application Of Data Mining In Gasoline Yield Optimization For MIP Process

Based on LIMS data of feedstock oils and regenerated catalysts,and DCS real-time operating data from No.1 and No.2 catalytic cracking MIP units of Jiujiang Petrochemical Company,gasoline yield was optimized with MIP process operation experiences and Data Mining.Firstly,according to the actual production conditions and pretreatment principles,the data from No.1 and No.2 MIP units was pretreated.Secondly,the cluster analyses for feedstock oils data from No.1 and No.2 MIP units were performed with K-MEANS clustering,two-step clustering,and system clustering respectively.The results show that the best clustering method is system clustering,and feedstock oils data were both clustered into two types.Thirdly,with joint method of the Pearson correlation coefficient analyses and operation experiences,correlation analyses for all basic data of feedstock oils,regenerated catalyst and real-time operating variables were done,and 13 variables from No.1 and No.2 MIP units respectively were selected to be the input variables of BP neural network models.Finally,different BP neural network models were respectively built on all the data and the clustering data.The results show that the models based on the clustering data were more reliable.Besides,genetic algorithm was used to optimize the gasoline yield by the models on the clustering data,and the optimization results could be the guidance for actual production.