关键词: S Zorb工艺, 辛烷值损失, BP神经网络, 模糊C均值聚类算法, 遗传算法

Abstract: There is usually a certain degree of octane number loss in FCC gasoline upgrading process. Based on the data over the years from a domestic S Zorb unit, 21 modeling variables from a total of 368 variables including the properties of feedstock,adsorbent,product and operating variables were screened out by grey correlation analysis and SPSS method. On the basis of clustering the feedstocks into three categories by fuzzy C-means clustering algorithm, the 21-20-1, 21-18-1, and 21-17-1 types of BP neural network models were established respectively. The verification results indicated that the accuracies of the three models established were high. The BP neural network models and genetic algorithm were combined to optimize the operating conditions for reducing the octane number loss of product on the premise of ensuring the desulfurization effect of gasoline. The predicted results can give an important reference for industrial production.

Key words: S Zorb process, octane loss, BP neural network, fuzzy C-means clustering algorithm, genetic algorithm