Hybrid dynamic modeling of an industrial reactor network with first-principles and data-driven approaches

Xie, Changrui; Yao, Runjie; Zhu, Lingyu; Gong, Han; Li, Hongyang; Chen, Xi

doi:10.1016/j.ces.2024.119852

Cited by 1 publication

(1 citation statement)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hybrid models were developed to optimize the styrene polymerization process . A hybrid dynamic modeling of industrial reactor networks has been carried out based on first-principles and data-driven approaches . Therefore, the model of mechanism-data fusion enables intelligent soft measurements of invisible parameters.…”

Section: Introductionmentioning

confidence: 99%

Intelligent Soft Measurement of Catalyst Activity during a Polyethylene Reaction Based on Mechanistic Modeling and Deep Learning

Tian,

Ji,

Wang

et al. 2024

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

The polymerization reaction is the core part of the polymerization process, where the catalyst has an important influence on the polymerization products. Catalyst activity, an essential operational indicator, is not directly monitored, resulting in the operator being unable to understand the operating conditions of the reactor fully. In this paper, an intelligent soft-measurement model based on mechanistic modeling and deep learning is proposed for predicting catalyst activity parameters in polyethylene reaction processes. First, a steady-state model of the polyethylene production process is established and utilized to collect data unavailable in the industry. A K-mean clustering method is used to cluster typical working conditions, and the collected data is used as inputs for mechanism modeling. Second, the primitive reactions in the polyethylene production process are studied to understand the mechanism of the polymerization process and to find the parameters that have the greatest impact on the catalyst activity. Then, the artificial neural network (ANN) deep learning method is selected, and its structure is optimized. The parameters and data provided by the mechanism model are integrated to constitute a complete data set. The catalyst activity characterized by the finger forward factor is predicted using the artificial neural network deep learning model. Finally, the self-built soft measurement model is validated with a maximum relative error of 0.44%. The validation results show that this method can reasonably predict the catalyst activity and guide the production process.

show abstract

Section: Introductionmentioning

confidence: 99%