摘要
由于在工业过程中产生的时序数据本身具有高度非线性和动态性,导致对聚乙烯关键指标的准确预测和生产优化指导变得困难。因此,提出一种基于新型时空双注意力模型的聚乙烯树脂密度软测量建模方法,旨在解决聚乙烯工业过程中的关键指标预测和能源结构优化问题。方法中引入了图注意力网络(GAT)和长短期时间序列网络(LSTNet),分别用于捕获复杂的时空关系以及提取时间相关特征,并将它们融合到一个统一的框架中,即时空融合模块(GLST),以实现自适应控制和准确预测。在GLST模型中,多头GAT模型被用于建模变量间显式的非线性关系,充分利用其信息聚合能力来提取时序数据的空间特征。同时,LSTNet模型有助于捕捉潜在的时间相关特征,从而更好地理解时序数据的动态性。GLST的引入使得能够将采集到的时空交互特征有效融合,从而实现对聚乙烯树脂密度的准确预测。为了验证方法的有效性,将该方法应用于实际工业生产聚乙烯树脂密度软测量建模中,结果表明:该方法不仅在与其他方法的比较中表现出显著的优越性,而且能够为实际聚乙烯生产工艺提供最佳的能源结构优化方案。
Considering the highly nonlinear and dynamic nature of temporal data in industrial process which resulting in the difficulty in accurately predicting the key indicators of polyethylene and guiding the production optimization,a novel soft measurement modeling approach for polyethylene resin density based on a new spatial-temporal dual-attention model was proposed to solve key indicator’s prediction and energy structure optimization in the polyethylene industry.In which,the graph attention network(GAT)and long short-term temporal network(LSTNet)was introduced to capture complex spatial-temporal relationships and extract spatial features of the temporal data and then have them integrated into spatio-temporal fusion module(GLST)to achieve adaptive control and accurate prediction.In the GLST model,the multi-head GAT model was used to model the explicit nonlinear relationship between variables,and make full use of its information aggregation ability to extract the spatial characteristics of time series data.At the same time,the LSTNet model helps to capture potential time-dependent features to better understand dynamics of time series data.The introduction of GLST makes it possible to effectively fuse the spatio-temporal interaction features collected so as to realize accurate prediction of the polyethylene resin density.For purpose of verifying effectiveness of the proposed method,the proposed method was applied to the actual industrial production of polyethylene resin density soft sensor modeling.The results show that,the proposed method has significant superiority in comparison with other methods and it can provide the best energy structure optimization scheme for actual polyethylene production process.
作者
李俊杰
马军鹏
马春雷
贺海波
安东玲
李子辉
陈志伟
LI Jun-jie;MA Jun-peng;MA Chun-lei;HE Hai-bo;AN Dong-ling;LI Zi-hui;CHEN Zhi-wei(China Coal Shaanxi Energy&Chemical Group Co.,Ltd.;College of Information Science and Technology,Beijing University of Chemical Technology;Engineering Research Center of Intelligent PSE,Ministry of Education of China)
出处
《化工自动化及仪表》
CAS
2024年第5期900-906,共7页
Control and Instruments in Chemical Industry
