Abstract: Based on the commercial data from a FCC MIP unit, two variables, aromatics content and outlet temperature, are deleted from the 18 variables including properties of feedstocks, properties of regenerated catalysts and operating conditions by using Pearson correlation coefficient method. Using 16 variables as input variable and 4 main product yields as output variables, a 16-20-4 type of BP neural network model was established. The BP neural network model and genetic algorithm model were combined to optimize the operating conditions for the maximum gasoline yield and the maximum gasoline yield plus the lowest coke yield. The verification results indicated that the calculated values are good consistent with FCC actual values. The multi-objective optimizations can also be obtained by the combined model. Compared with the optimization results for only getting the maximum gasoline yield，though the gasoline yield drops slightly，the coke yield decreases substantially.

Key words: FCC, MIP process, BP neural network, genetic algorithm