文献类型：期刊文献

中文题名：基于工业数据的溶剂脱沥青装置多工况建模

英文题名：Multi-conditional Modeling of Solvent Deasphalting Device Based on Industrial Data

作者：陈鹏宇[1];隆建[1];杨明磊[1];钱锋[1]

机构：[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

年份：2020

卷号：27

期号：11

起止页码：2002

中文期刊名：控制工程

外文期刊名：Control Engineering of China

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：国家自然科学基金重点项目(61333010);国家自然科学基金青年项目(21506050);中央高校基本科研业务费。

语种：中文

中文关键词：炼油过程;溶剂脱沥青;栈式降噪自编码器;模糊C均值聚类;集成学习

外文关键词：Refining process;solvent deasphalting;SDAE;FCM;ensemble learning

摘要：作为重油处理的重要单元之一,溶剂脱沥青装置由于进料成分复杂,萃取过程中两相平衡数据过于庞大且难以获取,传统通过机理的建模方法较难适用。提出一种SDAE-FCM工况分类法,借助于深度神经网络的自学习功能提取高维输入特征的同时降低输入维度,减少噪声对后续模型的影响;结合通过隶属度函数定义的模糊C均值聚类(FCM)算法对工况进行划分,缓解了由于进料性质波动和操作条件改变带来的工况漂移问题,较全局分析更具优势;采用基于树模型的集成学习方法针对不同工况分别建立产品收率和性质的模型。现场工业数据验证结果表明,该方法建立的模型,在预测脱沥青油(DAO)收率、残炭、硫含量、四组分等方面有较好的性能,可为实际装置的优化提供指导。
Solvent deasphalting unit is one of the most important heavy oil processing units in refining process.As the two-phase equilibrium data in extraction process is too large and too difficult to obtain,traditional mechanism model can hardly apply.In order to solve this problem,SDAE-FCM is proposed.Firstly,with the help of the self-learning ability of deep neural network,stacked denoising auto-encoder(SDAE)is introduced to reduce the dimensionality of input space.The extracted features can reduce the influence of noise on the subsequent model.Then,the fuzzy C-means clustering(FCM)algorithm is used to identify different working conditions,which is more advantageous than global analysis.Finally,ensemble learning method based on the tree model is used to establish models of product yield and properties for different working conditions.Field industrial data verification results show that the model established by this method has good performance in predicting the yield of deasphalted oil(DAO),residual carbon,sulfur content,four components and so on,and it can provide guidance for the optimization of practical devices.