Abstract: A deep catalytic cracking model of heavy oil based on BP neural network with structure 11-12-3 and Bayesian algorithm as learning algorithm was constructed and verified by using the experiment data of the heavy oil catalytic cracking reaction process and selecting 11 parameters such as raw material properties, catalyst activity, operation technology as input variables, and the yield of ethylene, propylene and light aromatics (BTX) as output variables. The results showed that the average relative errors of the model for the forecasts of the yields of ethylene, propylene and BTX were 4.59%, 3.92% and 2.28%, respectively. The established model has a good prediction effect on the yield of heavy oil catalytic cracking reaction products.

Key words: heavy oil, deep catalytic cracking, BP neural network, product yields