文献类型：期刊文献

中文题名：集总动力学模型结合神经网络预测催化裂化产物收率

英文题名：Prediction of the product yield from catalytic cracking process by lumped kinetic model combined with neural network

作者：欧阳福生[1];刘永吉[1]

机构：[1]华东理工大学石油加工研究所,上海200237

年份：2017

卷号：46

期号：1

起止页码：9

中文期刊名：石油化工

外文期刊名：Petrochemical Technology

收录：CSTPCD;;Scopus;北大核心:【北大核心2014】；CSCD:【CSCD2017_2018】；

语种：中文

中文关键词：催化裂化;MIP工艺;集总模型;神经网络

外文关键词：catalytic cracking;MIP process;lumped model;neural network

摘要：根据催化裂化反应机理和多产异构烷烃的重油催化裂化(MIP)工艺的特点,结合大量的工业数据,开展了MIP工艺过程集总动力学模型与BP神经网络模型相结合提高目标产物预测精度的研究,建立了饱和分、芳香分、胶质+沥青质、柴油、汽油、液化气、干气和焦炭8个集总反应网络,结合龙格库塔法与遗传算法求得该集总模型的47个动力学参数。实验结果表明,所求得的动力学参数能较好地体现催化裂化反应规律;模型对产物分布的模拟计算相对偏差均小于5%,采用14-7-5结构的BP神经网络与集总模型相结合,可进一步提高模型对产物分布的预测精度,为重油催化裂化的模拟优化提供了一个新的方向。
Based on the reaction mechanism of heavy oil catalytic cracking and the characteristics of maximizing iso-paraffins（MIP） process and combined with a large amount of industrial data,the reaction network of an 8-lump kinetic model including saturates,aromatics,asphaltenes＋resins,diesel,gasoline liquefied gas,dry gas and coke for the process was developed. And then the 47 kinetic parameters for the model were calculated by the combination of Runge-Kutta method and genetic algorithm. The results showed that good consistence with the reaction mechanism of heavy oil catalytic cracking,the average relative errors between calculated values and actual values of products are all less than 5%. Combining the lumped model with the 14-7-5 type of BP neural network can further improve the prediction accuracy of the product distribution,which provides a new direction for simulation and optimization for heavy oil catalytic cracking.