Abstract: S Zorb process is one of the main technologies to remove sulfur in fluid catalytic cracking gasoline. Based on the feedstock property data from a S Zorb gasoline adsorption desulfurization unit for 4 years, the characteristics and reaction mechanism of S Zorb process，the saturated hydrocarbon content, olefin content, aromatic content, sulfur content, density, and bromine number of the feedstocks were selected as the feedstock clustering variables. K-means and fuzzy C-means clustering algorithms were used to cluster the feedstocks by MATLAB programming platform. The results showed that both the K-means and fuzzy C-means clustering algorithms have a good classification effect on the feedstocks of the S Zorb unit and they both classify the raw oil into three categories. On this basis, the octane number prediction models can be established for different kinds of raw oils, which can provide guidance for finding the operating conditions to reduce the octane number loss of gasoline of S Zorb plant.

Key words: S Zorb process, FCC gasoline, K-means clustering algorithm, fuzzy C-means clustering algorithm, octane number