(87a) Hybrid Modeling for Managing Process Changes in Chemical Processing Industries
AIChE Spring Meeting and Global Congress on Process Safety
2023
2023 Spring Meeting and 19th Global Congress on Process Safety
Industry 4.0 Topical Conference
Emerging Technologies in Data Analytics
Tuesday, March 14, 2023 - 1:30pm to 2:00pm
Managing process changes in CPI requires moving the process from its current operating region A to a new region B. In the first stage, models developed in region A are extrapolated directly to region B. Extrapolation performance is important to obtain reliable estimates of the quality metrics of interest. In the second stage, as the process continues operating in region B and data is collected in this region, model re-training can be performed to account for possible mismatches. Finally, there is a third stage where a large volume data on region B has been collected. Calling attention to the second stage, methodologies that efficiently utilize limited data are preferred. This second task is the focus of the field of transfer learning, allowing one to leverage information obtained in region A to region B.
In this work, we study the application of hybrid modeling for supporting extrapolation and transfer learning, both critical tasks when managing process changes. Hybrid modeling combines physics-based and data-driven models to achieve improved performance. Physics-based models are more challenging to develop and require significant a priori knowledge. In contrast, data-driven models are constrained to the data domain where the model is built, and their extrapolation capabilities are limited. Combining both types of modeling mitigates some of their disadvantages while improving predictions (Oliveira, 2004). In this work, we study different configurations of hybrid modeling (e.g., parallel, series) and compare them to benchmarks that include a physics-based model only and data-driven models only. The physics-based model considers simplified reaction kinetics. The set of data-driven methods includes partial least squares (PLS), least absolute shrinkage and selection operator (LASSO), random forest and boosting, support vector regression (SVR), and neural networks (NN).
We consider a simulated case study of biodiesel production (Fernandes et al., 2019), testing two setpoint changes in the biodiesel concentration. Modeling performance in the extrapolation and transfer learning tasks is assessed and guidelines are provided for selecting the most suitable methodologies that show robust performance. Hybrid modeling consistently shows improved results compared to using physics-based or data-driven models only. In particular, parallel hybrid approaches are preferred for the extrapolation task. Regarding the transfer learning task, hybrid modeling also shows advantages, requiring fewer samples than the other benchmarks. Therefore, the application of hybrid modeling to transfer learning, one of the novelties of this work, led to better results than the traditional alternatives.
References
Fernandes NCP, Romanenko A, Reis MS. Mechanistic Modeling and simulation for process data generation. Industrial & Engineering Chemistry Research. 2019;58.
Oliveira R. Combining first principles modelling and artificial neural networks: a general framework. Computers and Chemical Engineering. 2004;28.
Sansana J, Joswiak MN, Castillo I, Wang Z, Rendall R, Chiang LH, Reis MS. Recent trends on hybrid modeling for Industry 4.0. Computers and Chemical Engineering. 2021;151.
von Stosch M, Oliveira R, Peres J, Feyo de Azevedo S. Hybrid semi-parametric modeling in process systems engineering: Past, present and future. Computers and Chemical Engineering. 2014;60.