(185d) Machine Learning-Based Operational Modeling of an Electrochemical Reactor: Handling Data Variability and Improving Empirical Models

The electrochemical transformation of carbon dioxide (CO₂)into carbon-based fuels and chemicals has received growing interest in this century because of its potential to reduce CO₂ emissions and restore the energy produced from renewable resources [1]. The biggest challenge for research in this area is the difficulty in determining and quantifying the products that result from the reduction of CO₂. Specifically, the CO₂ reduction pathways constitute a complex web of reactions that result in the production of various alkanes, alkenes, and oxygenate species [2]. The complexity of the reaction mechanism and the stochastic nature of electrochemical reactions make it challenging to derive an explicit model, such as the first-principal model, to describe the underlying physical-chemical phenomena of the electrochemical reactor. With the development of open-source programming libraries and availability of advanced data storage technology, data-hungry methods such as machine learning (ML) and deep learning (DL) rose in popularity for chemical engineering research due to their superior agility that allows capturing the complex input-output relation of universal chemical process [3]. For example, in [4, 5, 6], various artificial neural network (ANN) models have demonstrated their ability to address regression and classification problems in chemical engineering. However, recent small-scale experiments on the electrochemical reactor show varying levels of experimental uncertainty, due to the minimum measurable limit of the sensors and other inevitable experimental errors, that can introduce a level of uncertainty into the data and increase the probability of over-fitting.

This work proposes a methodology to develop a feed-forward neural network (FNN) model to capture the input-output relationship of an experimental electrochemical reactor from experimental data that are obtained from easy-to-implement sensors. This FNN model is computationally efficient and can be used in real-time to determine energy-optimal reactor operating conditions. To further account for the uncertainty of the experimental data, the maximum likelihood estimation (MLE) method is adopted to construct a statistical FNN, which is demonstrated to be able to relieve the over-fitting problem. Additionally, by comparing the neural network with an empirical, first-principles (EFP) model, it is demonstrated that the neural network model achieves improved prediction accuracy with respect to experimentally-determined input-output operating conditions. Finally, the insights obtained from the FNN model, and the limitations identified of the EFP model are used to propose specific modifications to the EFP model to improve its prediction capability.

References:

[1] Morales-Guio, C.; Cave, E.; Nitopi, S.; Feaster, J.; Wang, L.; Kuhl, K.; Jackson, A.; John-
son, N.; Abram, D.; Hatsukade, T.; et al. Improved CO₂ reduction activity towards C₂₊ alcohols on a tandem gold on copper electrocatalyst. Nature Catalysis 2018, 1, 764–771.

[2] Nitopi, S.; Bertheussen, E.; Scott, S.; Liu, X.; Engstfeld, A.; Horch, S.; Seger, B.; Stephens, I.;
Chan, K.; Hahn, C.; et al., Progress and perspectives of electrochemical CO₂ reduction on copper in aqueous electrolyte. Chemical reviews 2019, 119, 7610–7672.

[3] Lu, L.; Jin, P.; Karniadakis, G.; Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators. arXiv preprint arXiv: 1910.03193.

[4] Shen, Q.; Jiang, B.; Shi, P.; Lim, C. Novel neural networks-based fault tolerant control scheme with fault alarm. IEEE Trans. Cybernetics 2014, 44, 2190−2201.

[5] Raccuglia, P.; Elbert, K.; Adler, P.; Falk, C.; Wenny, M.; Mollo, A.; Zeller, M.; Friedler, S.; Schrier, J.; Norquist, A. Machine-learning-assisted materials discovery using failed experiments. Nature 2016, 533, 73−76.

[6] Malek, A.; Wang, Q.; Baumann, S.; Guillon, O.; Eikerling, M.; Malek, K. A Data-Driven Framework for the Accelerated Discovery of CO₂ Reduction Electrocatalysts. Front. Energy Res. 2021, 9, 52.