石油学报(石油加工) ›› 2021, Vol. 37 ›› Issue (1): 79-87.doi: 10.3969/j.issn.1001-8719.2021.01.008

• 研究报告 • 上一篇    下一篇

基于数据驱动的蜡油加氢装置产品预测与多目标操作优化

田水苗, 曹萃文   

  1. 华东理工大学 化工过程先进控制和优化技术教育部重点实验室,上海 200237
  • 收稿日期:2019-11-08 修回日期:2020-04-26 出版日期:2021-01-25 发布日期:2021-01-12
  • 通讯作者: 曹萃文,女,副教授,博士,从事复杂工业系统建模,分析与控制,流程工业生产计划与生产调度技术,供应链管理与优化,工业系统可靠性分析等研究,Tel:021-64251607; E-mail: caocuiwen@ecust.edu.cn E-mail:caocuiwen@ecust.edu.cn
  • 作者简介:第一作者:田水苗,女,硕士研究生,从事复杂工业系统建模,分析与控制,Tel:021-64251607,E-mail:303121868@qq.com
  • 基金资助:
    国家自然科学基金项目(61673175,61973120)资助

Data-Driven Product Prediction and Multi-objective Optimal Operations of Wax Oil Hydrotreating Unit

TIAN Shuimiao, CAO Cuiwen   

  1. Key Laboratory of Ministry of Education of Chemical Process Advanced Control and Optimization Technology, East China University of Science and Technology, Shanghai 200237, China
  • Received:2019-11-08 Revised:2020-04-26 Online:2021-01-25 Published:2021-01-12

摘要: 依据某炼油厂蜡油加氢装置生产数据,采用Aspen HYSYS对该装置进行机理建模,并用分层随机抽样法验证机理模型的有效性;然后以正常生产的28种减压蜡油和焦化蜡油进料量分区,运行机理模型,扩充了产品预测数据集。在此基础上,利用BP神经网络建立蜡油加氢装置的产品预测数据驱动模型,来预测精制蜡油流量,精制蜡油中硫、氮的质量分数,石脑油、液化气和燃料气的流量;最后以最小化精制蜡油中硫、氮的质量分数为目标进行在线操作优化。仿真结果表明:BP神经网络模型具有较高的产品预测精度,其平均绝对误差为6.286×10-3,均方误差为5.631×10-5;依据多目标优化结果调节操作参数,可降低精制蜡油中硫、氮的质量分数。

关键词: 蜡油加氢, Aspen HYSYS流程模拟 , 产品预测, BP神经网络, 多目标在线操作优化

Abstract: Based on the data of a wax oil hydrotreating unit in a refinery, a mechanism model was established based on Aspen HYSYS and the model was verified with the stratified random sampling method. Then, considering 28 various feedstock scenarios of vacuum gas oil (VGO) and coker gas oil (CGO) and running the mechanism model on Aspen HYSYS, the product prediction database were further expanded. Based on the above work, a product prediction data-driven model of the wax oil hydrotreating unit was established with utilizing a back propagation (BP) neural network. This model can predict mass flowrate of hydrotreated wax oil, naphtha, liquefied gas and gas products, as well as sulfur and nitrogen contents in the wax oil product. Finally, online optimal operations was carried out with the goal of minimizing sulfur and nitrogen contents in the wax oil product. The BP neural network model has significantly improved product prediction accuracy, with mean absolute error 6.286×10-3 and mean square error 5.631×10-5. With using multi-objective optimization, sulfur and nitrogen content in the wax oil product can be significantly reduced.

Key words: wax oil hydrotreating, Aspen HYSYS simulation, product prediction, BP neural network, multi-objective online optimal operation

中图分类号: