(317h) Nonlinear Model Predictive Control with Online Model Re-Identification for Improved Process Sustainability

The chemical industry has rapidly improved human life by transforming ecological goods and services into more valuable products through chemical and physical processes. However, these transformations are often not sustainable, as industry may not consider factors such as renewability, scarcity, and environmental impact when using resources from nature. As a result, practices that rely on critical materials and substances have been developed, leading to negative environmental and societal impacts from the release of toxic and hazardous substances. To minimize the impact of such processes, in this work, an advanced control framework is proposed for improved process sustainability. As typically chemical and biological processes are nonlinear, a data-driven model for the controller that can be re-identified online depending on the process operating point is considered.

Specifically, the proposed framework is based on a Model Predictive Control (MPC) method with online system re-identification. The nonlinear system re-identification procedure generates a Gaussian process (GP) model to be coupled with MPC as the predictive model. The proposed structure re-identifies the process model using GP with the nonlinear autoregressive with exogenous input regression (NARX) technique [1] when model prediction capabilities are lacking. To activate the model re-identification, two triggers are considered. The first trigger activates an evolving re-identification that takes place when the model prediction is no longer accurate given by the standard deviation of the GP model around the current operating region [2]. In parallel, the second trigger is activated based on the integrated error between the overall process behavior and model prediction. As results from the first or second triggers, a point is added to the dataset to be re-identified, or a new dataset is used for re-identification, respectively.

To demonstrate the framework, a fermentation reactor to produce bioethanol [3,4] is considered. In this process, the ethanol concentration is controlled while monitoring the sustainability of the process using GREENSCOPE indicators [5] based on economic, efficiency, environmental, energy use, and social impact measures. As the process has multiple steady states [5], the sustainability index is employed to drive the process control to the optimal steady state in conjunction with system re-identification for stable and sustainable operations around the selected steady state.

[1] Kocijan, J. (2016). Modelling and control of dynamic systems using Gaussian process models (pp. 33-38). Cham: Springer International Publishing.

[2] Maiworm M., Limon D., Findeisen R. Online learning-based model predictive control with Gaussian process models and stability guarantees. Int J Robust Nonlinear Control. 2021;31:8785–8812. https://doi.org/10.1002/rnc.5361.

[3] Li S., Mirlekar G., Ruiz-Mercado G.J., Lima F.V. Development of chemical process design and control for sustainability. Processes. 2016 Jul 25;4(3):23.

[4] Lima, F.V.; Li, S.; Mirlekar, G.V.; Sridhar, L.N.; Ruiz-Mercado, G.J. Modeling and Advanced Control for Sustainable Process Systems. In Sustainability in the Design, Synthesis and Analysis of Chemical Engineering Processes; Ruiz-Mercado, G.J., Cabezas, H., Eds.; Elsevier: Cambridge, MA, USA, 2016.

[5] Smith, R.L.; Ruiz-Mercado, G.J.; Gonzalez, M.A. Using Greenscope Indicators for Sustainable Computer-Aided Process Evaluation and Design. Comput. Chem. Eng. 2015, 81, 272–277.