Abstract:

Fluid catalytic cracking (FCC) is a highly non-linear and strong coupled operation system and is too hard to be described by traditional mechanism model. The artificial neural network provides a promising way to solve the problem because of its strong nonlinear prediction and self-learning ability. In a practical application of this method for a 1Mt/a MIP unit, a 19-24-4 type of BP neural network to predict the yields of liquid petroleum gas (LPG), gasoline, diesel and coke was established using nineteen input variables including properties of feedstock, regenerated catalyst and operating variables and so on. Based on the BP neural network, the influences of the feedstock preheating temperature, outlet temperatures of two reaction zones and reaction pressure on product distribution were investigated and the operating variables are optimized using genetic algorithm(GA) with a view to maximize gasoline yield. The industrial data agree well with the predicted results and a significant improvement in the gasoline yield was gained under the optimized conditions.

Key words: fluid catalytic cracking, MIP process, BP neural network, genetic algorithm