CIESC Journal ›› 2018, Vol. 69 ›› Issue (1): 464-471.DOI: 10.11949/j.issn.0438-1157.20170841

Previous Articles     Next Articles

Select Filter-Wrapper characteristic variables for yield prediction of fluid catalytic cracking unit

WANG Jie, CAO Daofan, LAN Xingying, GAO Jinsen   

  1. State Key Laboratory of Heavy Oil Processing, China University of Petroleum-Beijing, Beijing 102249, China
  • Received:2017-06-29 Revised:2017-08-12 Online:2018-01-05 Published:2018-01-05
  • Contact: 10.11949/j.issn.0438-1157.20170841
  • Supported by:

    supported by Science Foundation of China University of Petroleum, Beijing (C201606).

用于催化裂化装置产率预测的Filter-Wrapper特征变量选择方法

王杰, 曹道帆, 蓝兴英, 高金森   

  1. 中国石油大学(北京)重质油国家重点实验室, 北京 102249
  • 通讯作者: 蓝兴英
  • 基金资助:

    中国石油大学(北京)青年创新团队C计划项目(C201606)。

Abstract:

Fluid catalytic cracking is an important means to improve quality of heavy oil. Proper mathematical models are required to investigate influence of operating conditions and raw material properties on product distribution. Selection of suitable input variables largely affects model performances. The input variable selection is currently dependent on understanding mechanisms of FCC process. From the viewpoint of data-driven modeling, a method of selecting Filter-Wrapper characteristic subset was proposed by combination of Filter method using classical RReliefF algorithm and Wrapper method using GA-SVR algorithm. With no requirement of prior FCC knowledge, this method chose input variables by spontaneous data-driven selection of characteristic variables. A model with good prediction accuracy and proper number of input variables was established by taking advantage of a FCCU operating data and selecting input variables for prediction model of the unit dry gas and coke yield. Present work can not only provide a method to variable selection in FCCU modeling and process analysis, but also extend to other industrial process analysis.

Key words: model reduction, algorithm, genetic algorithm, feature selection, yield prediction

摘要:

催化裂化过程是重质油轻质化的重要手段,为了研究操作条件、原料性质等因素对产品分布的影响,通常需要对催化裂化过程建立准确可靠的数学模型。选择合适的输入变量对模型预测效果有着较大的影响,而在现有的催化裂化装置模型中,输入变量的选取主要依赖于对催化裂化机理的理解。本文从数据驱动建模的角度出发,提出一种Filter法与Wrapper法联合使用的特征子集选择方法。该方法在输入变量选取的过程中不依赖于催化裂化的先验知识,是一种数据驱动的自发的特征变量选择过程。以某炼油厂催化裂化装置为研究对象,利用该装置的生产数据分别选择用于干气和焦炭产率预测模型的输入变量,建立了预测精度高、输入变量数目适中的模型。此外,该方法为催化裂化装置建模的变量选取提供了新角度。

关键词: 模型简化, 算法, 遗传算法, 特征选择, 产率预测

CLC Number: