石油学报(石油加工) ›› 2019, Vol. 35 ›› Issue (4): 807-817.doi: 10.3969/j.issn.1001-8719.2019.04.024

• 研究报道 • 上一篇    下一篇

基于人工智能算法的催化裂化装置汽油收率预测模型的构建与分析

杨帆1,周敏2,戴超男1,曹军3   

  1. 1. 联想数据智能应用实验室,四川  成都, 610011;
    2. 四川理工学院 过程装备与控制工程四川省高校重点实验室, 四川 自贡 643000;
    3. 华东理工大学 机械与动力工程学院,上海 200237
  • 收稿日期:2018-06-11 修回日期:2019-03-11 出版日期:2019-07-25 发布日期:2019-09-24
  • 通讯作者: 曹军, 男,助理研究员,博士,从事石油化工及工业大数据相关研究工作,Tel: 021-64253810, E-mail:caojun@ecust.edu.cn E-mail:caojun@ecust.edu.cn
  • 作者简介:第一作者:杨帆,男,硕士,从事大数据及工业智能相关研究工作
  • 基金资助:
    上海市自然科学基金项目(18ZR1409000)和过程装备与控制工程四川省高校重点实验室开放基金资助项目(GK201818)资助

Construction and Analysis of Gasoline Prediction Model for FCC Unit Based on Artificial Intelligence Algorithm

YANG Fan1, ZHOU Min2, DAI Chaonan1, CAO Jun3   

  1. 1. Data Intelligence Application Lab, Lenovo Group, Chengdu 610041, China;
    2. Sichuan Provincial Key Lab of Process Equipment and Control, Sichuan University of Science and Engineering, Zigong 643000, China;
    3. School of Mechanical and Power Engineering, East China University of Science and Technology, Shanghai 200237, China
  • Received:2018-06-11 Revised:2019-03-11 Online:2019-07-25 Published:2019-09-24

摘要: 基于某石化企业的LIMS(Laboratory information management system)及DCS(Distributed control system)系统中的工业生产数据,结合工业经验中已知的影响催化裂化产品收率的重要因素,通过分析监控指标与实际汽油收率的相关性,筛选出与汽油收率的正负相关性较高的分析指标。在此基础上,基于梯度提升决策树GBDT算法构建了催化裂化汽油收率的预测模型,并预测了相应的汽油产率。结果表明:由GBDT算法构建的汽油收率预测模型预测结果的准确率为98.9%,R2系数为0.236,平均绝对误差为0.531%;模型预测结果与实际汽油产率相比,误差率小于1%,表明构建的模型能精确预测催化裂化装置中汽油等产品收率,有助于在实际生产中优化催化裂化装置的操作条件,从而进一步提升催化裂化装置的经济性能。

关键词: 人工智能, 催化裂化, 预测模型, GBDT算法

Abstract: Industrial data were collected from a petrochemical company’s DCS (Distributed control system) and LIMS (Laboratory information management system) systems. Together with the essential factors affecting the product yield of catalytic cracking from industrial experience, indicators with high correlations were filtered out by analyzing their correlations with actual gasoline yield. Subsequently, a prediction model based on gradient-growth decision tree (GBDT algorithm) was constructed to predict the gasoline yield on the catalytic cracking unit. The results show that the accuracy of the GBDT gasoline yield prediction model is 98.9%, the R2 value is 0.236, and the mean absolute error is 0.531%. Compared to actual gasoline yield, the error of prediction result is less than 1%, indicating that the presented model could predict the product yield such as gasoline on the catalytic cracking unit accurately. This will optimize the operation conditions of FCC, and further enhance the economic performance of FCC unit.