Abstract: Catalytic cracking unit has a significant impact on refinery production efficiency. Accurate prediction and optimization of its product yield and coke yield is important to improve the efficiency of the unit and the overall process flow of the refinery. In this paper, a yield prediction model was developed based on the production data from the FCC units in several refineries of SINOPEC by applying Gradient Boosting Decision Tree (GBDT) algorithm in deep learning algorithm and the neural network (ANN) algorithm to summarize the data processing experience for production data. The results show that the gradient tree algorithm based on deep learning performs better in prediction efficiency, accuracy and stability. Artificial intelligence methods can accurately predict product yield based on big data, help to carry out unit operation optimization and plant-wide overall process flow optimization based on data model, and improve plant-wide economic efficiency.

Key words: deep learning, catalytic cracking/FCC, artificial intelligence, operation optimization