Optimizing Product Distribution Of FCC MIP Process By Data Mining Technology

Based onNo.2 catalytic cracking MIP unit from China Sinopec Jiujiang Branch, product distribution optimization models were established using data mining technology. The variables including feedstock properties, regenerated catalyst properties and operating conditions were selected according to reaction mechanism and characteristics of MIP process. Industrial real-time data of these variables were collected and the raw data were pretreated. The clustering models of feedstock properties were established using K-means clustering method and fuzzy c-means clustering method. An optimization model for product distribution was thus established according to the clustering models, which was aimed at finding the historical optimum operating conditions for one special feedstock oil. The research results show that the two clustering methods have good applicability to feedstock properties. By using Pearson correlation coefficient method, the collected variables were reduced, and then the product yield modes were established by BP, RBF and GRNN neural network on the basis of the reduced variables. The results show that BP and GRNN were better than RBF neural network. On the basis of the established BP neural network for MIP product yield, the influences of the feedstock preheating temperature, reaction temperature, reaction pressure, recycle oil flow and pre-upgrade steam flow on product distribution were investigated. PSO and GA algorithm were used to optimize the operating conditions to get the maximum gasoline yield and the optimization results for the operating conditions were consistent with FCC process principle. GA method was used to optimize the operating conditions to get the maxmum gasline yield while to get the minimum coke yield. Compared with the optimization results for only getting the maxmum gasline yield, although the gasline yield drops, the coke yield decreases substantially, and the overall optimization results were more ideal. Therefore, these results could be used to guide the optimization of practical operation of FCC MIP unit.