关键词: OCC工艺, BP神经网络, XGBoost, 遗传算法

Abstract: In order to improve the yield of propylene and ethylene in the olefin catalytic cracking (OCC) unit, based on the collection of industrial data from a 100 kt/a OCC unit of a domestic enterprise, 835 sets of data samples were obtained through data preprocessing. 13 input feature variables were selected using the maximum mutual information coefficient and Pearson correlation coefficient methods. With the yield of diene (propylene and ethylene) as the target variable, BPNN and XGBoost methods were used to establish XGBoost and BPNN classification prediction models with a structure of 13-64-13-1, respectively. Comparing the classification prediction results of the two models, it was found that the BPNN model had a better overall prediction performance; Furthermore, by combining genetic algorithm to optimize the operating variables of the BPNN model, it was found that after optimizing the operating variables, the model's predicted diene yield increased by an average of 2.8 percentage points. This indicates that the established BPNN model can provide guidance for optimizing the operation of OCC industrial facilities.

Key words: OCC process, BP neural network, XGBoost, genetic algorithm