Evaluation of a Data-Driven, Machine Learning Approach for Identifying Potential Candidates for Environmental Catalysts: From Database Development to Prediction

Abstract

Data-driven, machine learning approaches are increasingly used for the discovery and development of catalytic materials in the area of material science and engineering. In this paper, the approach was evaluated with respect to its applicability in identifying potential environmental catalysts (ECs) using the selective catalytic reduction (SCR) of the air pollutant NO x as an example. The detailed procedures including database assemblage, the training and testing of a machine learning model, the validation and prediction of the model, and model uncertainties are all provided. The results indicated that there is a significant amount of data accumulated in environmental catalysts that can be exploited for accelerating the exploration and optimization of ECs for specific applications. The results also indicated that the approach is powerful for identifying new ECs and optimizing conditions for ECs synthesis and applications. However, the reported data in the literature are often incomplete, which limits the application potentials of the data. With limited data, the simulated results from the model contained uncertainties, especially in the prediction of unknown ECs. Repeated predictions and ensemble averaging were then proposed as an approach to find conditions for synthesizing and applying promising ECs.

Iterative Approach of Experiment–Machine Learning for Efficient Optimization of Environmental Catalysts: An Example of NOx Selective Reduction Catalysts

“…The number of hidden layers and neurons was adjusted to optimize the performance of the ANN model based on the values of the correlation coefficient R and root means square error (RMSE) (eqs and ), respectively. As demonstrated previously, a three hidden-layer structure with 6, 4, and 2 neurons for each layer (Figure S1 of the Supporting Information) was an optimal combination for this problem RMSE = ∑ 1 N ( P mesu − P pred ) 2 N where P mesu is the measured value and P pred is the predicted value. N is the total amount of data R = ∑ 1 N false( P mesu − P̅ mesu false) false( P pred − P̅ pred false) ∑ 1 N ( P mesu − P̅ mesu ) 2 ( P pred − P̅ pred ) 2 where P mesu , P pred , P̅ mesu , and P̅ pred are the measured, predicted, average of measured, and average of predicted values, respectively.…”

“…This is also expected because additional data would improve the prediction as shown in a previous study. 18 This study demonstrated an iterative approach of the MLbased model and experiment to identify new catalysts that typically have spare data in the literature. In the approach, the ML-based model was first trained from relevant data in the literature and was then used to screen candidate catalysts to guide the experimental step, and the experiment step is used to synthesize the candidate catalysts to evaluate the performance predicted from the ML model and to provide new data to retrain the ML model.…”

“…The database development starts with collecting relevant data from the literature as described previously. 18 In this study, the database was iteratively updated using the experimental data generated during this research. For the case of SCR NO x catalysts, five categories of data comprising composition, structure, morphology, preparation method, and reaction condition that are commonly provided in the literature were collected to establish a raw database.…”

Machine learning for design principles for single atom catalysts towards electrochemical reactions

“…Also, the loading amount of Zr was found to play an important role due to the fact that the Cr 5+ species can reduce as the Zr loading amount increases, which can subsequently lower the NO x conversion efficiency. 88 In addition, a ML algorithm along with 27 descriptors was applied to 2228 experimental data obtained from the literature 89 to predict the activity of heterogeneous catalysts, which reveals that temperature is the most important descriptor for the water-gas shi reaction. 90 Moreover, learning from a large database in nanoscience can be used for rapid design and discovery of new heterogeneous catalysts using ML.…”

Recommendations for Advancing FAIR and Open Data Standards in the Water Treatment Community

“…Synthesis-based methods, such as the meta-analyses or systematic reviews widely performed in the medical research community, minimize these challenges by evaluating entire bodies of research . Data synthesizability in water treatment research domains is often limited by the paucity of observations and inadequate data documentation, which hinders broader use of data for materials informatics research, uncertainty quantification efforts, and plant control schema. − Strong synthesis research requires access to underlying study data, facile recombination with other interoperable study datasets, and holistic evaluation of the synthesized data. These methods help to elucidate the robustness of conclusions, pinpoint and manage discrepancies across findings, and identify key knowledge gaps warranting further inquiry …”