基于深度学习的催化裂化过程建模方法  

Modelling Method Based on Deep Learning for Fluid Catalytic Cracking Processes

作　　者：陈琳 周利 吉旭[1] CHEN Lin;ZHOU Li;JI Xu(School of Chemical Engineering,Sichuan University,Chengdu,Sichuan 610065,China)

机构地区：[1]四川大学化学工程学院,四川成都610065

出　　处：《西安石油大学学报（自然科学版）》2023年第4期94-103,共10页Journal of Xi’an Shiyou University（Natural Science Edition）

基　　金：国家自然科学基金(22108178,21776183,21706220);中央高校基本科研业务费(YJ201838)。

摘　　要：针对催化裂化过程复杂的时空特性,提出了一种结合时间卷积网络(TCN)、注意力机制和反向传播网络(BPNN)的深度学习建模方法,用于预测催化裂化的产品收率。首先利用TCN对过程时序数据进行深度信息提取,其次利用注意力机制来自适应地捕捉时序过程中的时间权重,以增强模型的信息提取能力,最后利用BPNN将提取后的信息关联到产品收率。为了验证所提方法的有效性和实用性,将其应用于实际的催化裂化生产过程,分别预测5种产品的收率,并与其他方法进行比较。结果表明,本文方法具有最高的预测精度,其决定系数(R 2)均超过0.95,说明该方法能很好地适应于不同产品收率的预测场景,可为催化裂化工艺的在线控制与优化提供支撑。Considering the complex temporal and spatial characteristics of fluid catalytic cracking(FCC)process,a deep learning modeling method combining temporal convolution network(TCN),attention mechanism and back propagation neural network(BPNN)is proposed to predict the yield of FCC products.Firstly,the information is deeply extracted from the process data using TCN.Secondly,the time weight in the timing process is adaptively captured using the attention mechanism to enhance the information extraction ability of the model.Finally,the extracted information is correlated to the product yield using BPNN.In order to verify the effectiveness and practicability of the proposed method,it was applied to the actual FCC process to predict the yields of 5 products respectively,and compared with other methods.The results show that the proposed method has the highest prediction accuracy,and its determination coefficient(R 2)exceeds 0.95,indicating that the proposed method can be well adapted to the prediction of different product yields,and can provide support for online control and optimization of FCC process.

关 键 词：催化裂化 深度学习 时间卷积网络 注意力机制 过程建模 预测 

分 类 号：TE65[石油与天然气工程—油气加工工程]