Architectures for neural networks as surrogates for dynamic systems in chemical engineering

Esche, Erik; Weigert, Joris; Rihm, Gerardo Brand; Göbel, Jan; Repke, Jens‐Uwe

doi:10.1016/j.cherd.2021.10.042

Cited by 18 publications

(9 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the BO framework, surrogate black-box functions are used to approximate the actual objective function, and subsequently, acquisition functions are utilized to sample the objective function probabilistically. The most widely used surrogate function in BO is GPR because of its ability to capture complex nonlinear characteristics and the uncertainty in prediction. , …”

Section: Methods: Gpr Modeling Shap and Multiobjective Bayesian Optim...mentioning

confidence: 99%

“…The most widely used surrogate function in BO is GPR because of its ability to capture complex nonlinear characteristics and the uncertainty in prediction. 74 , 75 …”

Section: Methods: Gpr Modeling Shap and Multiobjective Bayesian Optim...mentioning

confidence: 99%

“…The most widely used surrogate function in BO is GPR because of its ability to capture complex nonlinear characteristics and the uncertainty in prediction. 74,75 A key component of BO is the acquisition function. Expected improvement is one of the most extensively used acquisition functions in the BO framework.…”

Section: Fit the Weighted Linear Model By Minimizing The Loss L =mentioning

confidence: 99%

See 2 more Smart Citations

Multiobjective Bayesian Optimization Framework for the Synthesis of Methanol from Syngas Using Interpretable Gaussian Process Models

et al. 2022

View full text Add to dashboard Cite

Methanol production has gained considerable interest on the laboratory and industrial scale as it is a renewable fuel and an excellent hydrogen energy storehouse. The formation of synthesis gas (CO/H 2 ) and the conversion of synthesis gas to methanol are the two basic catalytic processes used in methanol production. Machine learning (ML) approaches have recently emerged as powerful tools in reaction informatics. Inspired by these, we employ Gaussian process regression (GPR) to the model conversion of carbon monoxide (CO) and selectivity of the methanol product using data sets obtained from experimental investigations to capture uncertainty in prediction values. The results indicate that the proposed GPR model can accurately predict CO conversion and methanol selectivity as compared to other ML models. Further, the factors that influence the predictions are identified from the best GPR model employing "Shapley Additive exPlanations" (SHAP). After interpretation, the essential input features are found to be the inlet mole fraction of CO (Y(CO, in)) and the net inlet flow rate (Fin(nL/min)) for our best prediction GPR models, irrespective of our data sets. These interpretable models are employed for Bayesian optimization in a weighted multiobjective framework to obtain the optimal operating points, namely, maximization of both selectivity and conversion.

show abstract

Section: Methods: Gpr Modeling Shap and Multiobjective Bayesian Optim...mentioning

confidence: 99%

“…The most widely used surrogate function in BO is GPR because of its ability to capture complex nonlinear characteristics and the uncertainty in prediction. 74 , 75 …”

Section: Methods: Gpr Modeling Shap and Multiobjective Bayesian Optim...mentioning

confidence: 99%

Section: Fit the Weighted Linear Model By Minimizing The Loss L =mentioning

confidence: 99%

See 1 more Smart Citation

Multiobjective Bayesian Optimization Framework for the Synthesis of Methanol from Syngas Using Interpretable Gaussian Process Models

et al. 2022

View full text Add to dashboard Cite

show abstract

“…In this work, DABNet is treated as one of the benchmark data-driven network-based models for evaluating the performance of the proposed architectures in this work. Furthermore, the proposed network structures are compared with a couple of other widely used data-driven models for system identification and the LSTM-type and the Gated Recurrent Unit (GRU)-type RNNs. , For three case studies considered in this work, performances of the proposed hybrid series and parallel network models have been compared with those of LSTMs, GRUs, and DABNet, in addition to the comparison with respect to the static-only and dynamic-only network architectures.…”

Section: Introductionmentioning

confidence: 99%

Hybrid Series/Parallel All-Nonlinear Dynamic-Static Neural Networks: Development, Training, and Application to Chemical Processes

Mukherjee

Bhattacharyya

2023

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

This paper presents the development of data-driven hybrid nonlinear static-nonlinear dynamic neural network models and addresses the challenges of optimal estimation of parameters for such hybrid networks. A parallel static-dynamic neural network and two variants of series networks, specifically, nonlinear static-nonlinear dynamic and nonlinear dynamic-nonlinear static networks, are investigated in this work. Performances of the proposed fully nonlinear hybrid series and parallel network models are compared with the existing state-of-the-art data-driven models like long−short-term memory networks and gated recurrent unit models as well as DABNet-type linear-dynamic-nonlinear-static neural network. Algorithms are developed for training the series and parallel hybrid networks where the static and dynamic networks can be trained independently by different algorithms. These algorithms offer flexibility for incorporating different types of static and dynamic network architectures and their training algorithms thus offering tradeoff between computational expense and accuracy for highly nonlinear systems. In addition to the typical training objective that minimizes some squared error, an objective function is considered that penalizes overfitting and uncertainty in parameter estimates. Performances of the proposed algorithms are evaluated for three nonlinear example problems�a two-tank pH neutralization reactor, the widely used Van de Vusse reactor, and a pilot plant for postcombustion CO 2 capture using monoethanolamine solvent. Computational expense and convergence performance of the proposed network architectures and their training algorithms are presented.

show abstract

“…Tripathy and Bilionis (2018) used a NN to create surrogate models for expensive high dimensional uncertainty quantification. Other recent applications of NN as surrogate models address chemical and process engineering (Cavalcanti et al, 2021;Esche et al, 2022) or materials science (Allotey et al, 2021).…”

mentioning

confidence: 99%

Accelerating models for multiphase chemical kinetics through machine learning with polynomial chaos expansion and neural networks

Berkemeier

Krüger²,

Feinberg³

et al. 2022

Preprint

View full text Add to dashboard Cite

Abstract. The heterogeneous chemistry of atmospheric aerosols involves multiphase chemical kinetics that can be described by kinetic multi-layer models (KM) explicitly resolving mass transport and chemical reaction. However, KM are computationally too expensive to be used as sub-modules in large-scale atmospheric models, and the computational costs also limit their utility in inverse modelling approaches commonly used to infer aerosol kinetic parameters from laboratory studies. In this study, we show how machine learning methods can generate inexpensive surrogate models based on the kinetic multi-layer model of aerosol surface and bulk chemistry (KM-SUB). We apply and compare two common and openly available methods for the generation of surrogate models, polynomial chaos expansion (PCE) with UQLab and neural networks (NN) through the Python package Keras. We show that the PCE method is well-suited to determine global sensitivity indices of the KM and demonstrate how inverse modelling applications can be enabled or accelerated with NN-suggested sampling. These qualities make them suitable supporting tools for laboratory work in the interpretation of data and design of future experiments. Overall, the KM surrogate models investigated in this study are fast, accurate, and robust, which suggests their applicability as sub-modules in large-scale atmospheric models.

show abstract

Architectures for neural networks as surrogates for dynamic systems in chemical engineering

Cited by 18 publications

References 40 publications

Multiobjective Bayesian Optimization Framework for the Synthesis of Methanol from Syngas Using Interpretable Gaussian Process Models

Multiobjective Bayesian Optimization Framework for the Synthesis of Methanol from Syngas Using Interpretable Gaussian Process Models

Hybrid Series/Parallel All-Nonlinear Dynamic-Static Neural Networks: Development, Training, and Application to Chemical Processes

Accelerating models for multiphase chemical kinetics through machine learning with polynomial chaos expansion and neural networks

Contact Info

Product

Resources

About