Dynamic Surrogate Modeling for Multistep-ahead Prediction of Multivariate Nonlinear Chemical Processes

Abstract

This work proposes a methodology for multivariate dynamic modeling and multistep-ahead prediction of nonlinear systems using surrogate models for the application to nonlinear chemical processes. The methodology provides a systematic and robust procedure for the development of data-driven dynamic models capable of predicting the process outputs over long time horizons. It is based on using surrogate models to construct several Nonlinear AutoRegressive eXogenous models (NARX), each one approximating the future behavior of one process output as a function of the current and previous process inputs and outputs. The developed dynamic models are employed in a recursive schema to predict the process future outputs over several time steps (multistep-ahead prediction). The methodology is able to manage two different scenarios: 1) one in which a set of input-output signals collected from the process is only available for training, and 2) another in which a mathematical model of the process is available and can be used to generate specific datasets for training. With respect to the latter, the proposed methodology includes a specific procedure for the selection of training data in dynamic modeling based on Design Of Computer Experiment (DOCE) techniques. The proposed methodology is applied to case studies from the process industry presented in the literature. The results show very high prediction accuracies over long time horizons. Also, thanks to the flexibility, robustness and computational efficiency of surrogate modeling, the methodology allows dealing with a wide range of situations, which would be difficult to address using first principle models.

Exploiting Domain Partition in Response Function-Based Dynamic Surrogate Modeling: A Continuous Crystallizer Study

“…Although the system to be modeled is complex or could be composed of several sub-systems, in a series or parallel coupling layout, sometimes the correlation between the system outlet and the input variables can be simplified by low-order differential equations, without the need of a multiple sub-model combination. Examples of the parametrical differential form of a first-and second-order response function are reported in Equations ( 11) and (12), respectively:…”

“…In order to mitigate these issues, alternative approaches based on constrained sampling or subsets have been proposed and have demonstrated good results, but they have not yet achieved thorough and general validation [11]. Moreover, the vast majority of the proposed studies usually refer to steady-state applications, while the process dynamics domain can only be found in a few works, mostly related to process unit or energy system optimization, that are usually based on explicit analytical or statistical functions with case-specific applications [12][13][14]. If we consider that computationally efficient digital twins for dynamic modeling would substantially increase the effectiveness of tools such as RTDO and optimal control by enabling real-time scheduling and process parameter adjustments, it is evident that the process dynamics field deserves more attention.…”

Critical analysis of the use of white-box versus black-box models for multi-objective optimisation of small-scale biorefineries

“…Contrarily, black-box models do not require prior insight knowledge on the considered system but instead rely on a welldesigned and -spread dataset containing input/output data of the concerned system's aspect. Combined with a well-defined experimental design, black-box models only rely on the provided input/output data to define the model framework, structure, and parameters (Gernaey et al, 2004;Almquist et al, 2014;Lo-Thong et al, 2020;Shokry et al, 2020;Mora-Mariano and Flores-Tlacuahuac, 2022). Surrogate models could be considered as a special type of blackbox models, bridging the gap between both black-box and white-box models: Surrogate models are in essence black-box in the sense that they map the relationship between a certain process input and output whitout much attention to system's inner workings.…”

A Deep Learning Hybrid Framework Combining an Efficient Evolutionary Algorithm for Complex Many-Objective Optimization of Sustainable Triple CO2 Feed Methanol Production

“…The simulation comprises a network of interconnected neurons that exhibit strong data processing capabilities and the ability to adapt and learn. This makes it well-suited for addressing diverse and intricate pattern recognition and prediction challenges. − Different ANNs have demonstrated their exceptional performance with applications to various industrial practices. − Wang et al built an ANN surrogate model to represent the vacuum pressure swing adsorption process under consideration. The research employed four meta-heuristic algorithms based on the built model, and these algorithms were evaluated to enhance the efficiency of the process.…”