Machine Learning-Assisted Low-Dimensional Electrocatalysts Design for Hydrogen Evolution Reaction

Abstract

Efficient electrocatalysts are crucial for hydrogen generation from electrolyzing water. Nevertheless, the conventional "trial and error" method for producing advanced electrocatalysts is not only cost-ineffective but also time-consuming and labor-intensive. Fortunately, the advancement of machine learning brings new opportunities for electrocatalysts discovery and design. By analyzing experimental and theoretical data, machine learning can effectively predict their hydrogen evolution reaction (HER) performance. This review summarizes recent developments in machine learning for low-dimensional electrocatalysts, including zero-dimension nanoparticles and nanoclusters, one-dimensional nanotubes and nanowires, two-dimensional nanosheets, as well as other electrocatalysts. In particular, the effects of descriptors and algorithms on screening low-dimensional electrocatalysts and investigating their HER performance are highlighted. Finally, the future directions and perspectives for machine learning in electrocatalysis are discussed, emphasizing the potential for machine learning to accelerate electrocatalyst discovery, optimize their performance, and provide new insights into electrocatalytic mechanisms. Overall, this work offers an in-depth understanding of the current state of machine learning in electrocatalysis and its potential for future research.

Keywords: Algorithm; Descriptor; Hydrogen evolution reaction; Low-dimensional electrocatalyst; Machine learning.

Fig. 1

Process of machine learning for designing HER electrocatalysts

Fig. 2

Typical machine learning algorithms in electrocatalyst design

Fig. 3

a Histograms of the number and citation frequency of relevant articles were retrieved with the keywords "machine learning" and "hydrogen evolution reaction" from the Web of Science database. b Machine learning for designing various electrocatalysts including 0D electrocatalysts, 1D electrocatalysts, 2D electrocatalysts, and others

Fig. 4

Exploring MMA electrocatalysts via active learning and experiments. a Process of developing MMA electrocatalysts with small overpotential. b Overpotential and uncertainty of the Pt-Ru-Ni catalysts. At certain points during the iteration process, the changes in predictions, which were not influenced by adding any additional data, were symbolized by the red dotted circles. c Graph showing the overpotential of the top-five high-uncertainty points (THP). The corresponding results were marked by black and red circles, respectively. Any differences between the two results were highlighted by orange arrows to help provide a clear comparison. d Plot of the THP changes in overpotential. e Scatter-plotted ternary data points. Reproduced from Ref. [142] with permission from John Wiley and Sons

Fig. 5

Machine-learning-assisted investigation of E_Hads on bimetallic nanoclusters. a Cu₁₃Co₄₂ clusters include core–shell, segregated, ordered, and random structures. b A workflow outline exhibits the process from the formation of a cluster to the prediction of the E_Had distributions. c Learning curve of KRR, the inserted image shows the calculated versus predicted E_Had of 1767 DFT calculations. d Predicted E_Had distribution. e Evaluation of machine learning accuracy in the presence of adsorption site drift and surface reconstruction. Reproduced from Ref. [147] with permission from American Chemical Society

Fig. 6

Investigating jagged Pt nanowires via end-to-end simulation. a Flowchart for the jagged Pt nanowire using end-to-end simulation. b Identifying active sites with ΔrG_ads values towards the top, bridge, and hollow sites. c Plots of ΔrG_ads values versus coordination numbers. d Visualization of Pt nanowires with an optimal ΔrG_ads. The magnification indicates that the low coordination numbers of Pt atoms possess suitable ΔrG_ads values. Reproduced from Ref. [154] with permission from American Chemical Society

Fig. 7

Hydrogen adsorption on defective NCNTs was interpreted through machine learning. a Workflow utilized machine learning and SHAP analysis for defective NCNTs. b Unbiased generalization performance of the RF models. c The importance of ten features in the predictions of adsorption energy. d SHAP values for the important features. e Measuring SHAP strong interaction effects of the ten features. Partial descriptions for adsorption at f (8,8) graphitic, g (14,0) N₁V₁-pyridinic, h (14,0) N_1bSW-pyrrolic, and i (8,8) N₄V₂-pyridinic dopant configurations are illustrated. Pairwise SHAP interaction effects between j the dopant-adsorption site separation and the NCNT energy gap, k the dopant-adsorption site separation and the residual charge on the adsorption site, l the energy gap and the spin polarization on the adsorption site, as well as m the energy gap and the NCNT chirality. Reproduced from Ref. [156] with permission from American Chemical Society

Fig. 8

A descriptor for designing 2D MXene HER electrocatalysts. a Ti₂C structure containing top, fcc, and hcp, representing the O adsorbed sites. b Defective Ti₂CO₂ with Ti vacancy. c Doped model Ti₂CO₂-STM and TM = 3d, 4d, and 5d metals with single atom. S₀, S₁, and S₂ correspond to O positions for H adsorption. d Overall flow of the high throughput computation and machine learning. e Descriptor performance in the KRR. f R² of two important descriptors for KRR, and g other models. h Genetic programming processing. i, j The prediction performance of Ti₂CO₂-STM and Zr₂CO₂-STM with the new descriptor. k Fitting coefficient definition for Ta₂CO₂-STM. l Validation and new catalyst screening in Ta₂CO₂-STM. Reproduced from Ref. [165] with permission from Royal Society of Chemistry

Fig. 9

Application of machine learning in alloy electrocatalysts. a Illustration of (100) bimetallic alloys with the random sampling method. The red squares indicate the unique H adsorption environment created by a fourfold ensemble. b, c Feature analysis. Typical results of DFT calculated versus predicted E_Had, d, e The ratio is nine to one between training and testing, f, g eight to two, and h, i seven to three. Reproduced from Ref. [178] with permission from Royal Society of Chemistry

Fig. 10

Perspectives of machine learning for the HER