Machine Learning for Transition-Metal-Based Hydrogen Generation Electrocatalysts

Abstract

As a pivotal technique of data mining and data analysis, machine learning (ML) is gradually revolutionizing how materials are discovered with the rise of big data. Catalysis informatics are rising from computational research in catalysis to liberate researchers from heavy trial-and-error investigations. However, ML remains at its early stage in hydrogen evolution reaction (HER) electrocatalysis with numerous underexplored opportunities. This Perspective focuses on the recent progress in ML-assisted identification of high-performance transition-metal (TM)-based HER electrocatalysts to arouse broader research ideas. The representative studies about feature engineering (descriptors) in ML and the potential applications of image identification with deep learning for TM-based HER electrocatalysts are highlighted. In addition to serving as a powerful means for discovery of electrocatalysts, these data-driven techniques also establish a deeper understanding for the relationships between intrinsic properties of TM-based materials and their electrocatalytic performance. Future perspectives on the development of ML-guided TM-based HER electrocatalysts are provided to give heuristics for further investigations in this field.

Theory‐Guided Machine Learning to Predict the Performance of Noble Metal Catalysts in the Water‐Gas Shift Reaction

“…Machine learning (ML) becomes an increasingly attractive method for catalyst discovery, [16–19] as state‐of‐the art ML models are capable of establishing the complex correlations between a large numbers of variables, which can be further exploited to meet designing requirements. In the context of machine learning on WGS reaction optimization, Odabaşi et al [20] .…”

“…While first principle based methods are commonly used in catalyst design, simplifying assumptions about the surface structure and reaction conditions are often made for computation tractability. [15] Machine learning (ML) becomes an increasingly attractive method for catalyst discovery, [16][17][18][19] as state-of-the art ML models are capable of establishing the complex correlations between a large numbers of variables, which can be further exploited to meet designing requirements. In the context of machine learning on WGS reaction optimization, Odabaşi et al [20] are the first to conduct a comprehensive study.…”

Ligand Enhanced Activity of In Situ Formed Nanoparticles for Photocatalytic Hydrogen Evolution

“…[2,3] In these systems, nanoparticles (NPs) are promising catalysts, [4] in particular because their physical and chemical properties can be tuned by changing their size, [5] shape, [6,7] surface chemistry, [8] and/or composition, [9] providing many handles by which their performance can be optimized. [10][11][12][13][14][15][16][17] Though NPs reduce the overpotential in these systems and thus catalyze the reaction, electrolysis of water still relies on electrical current, and electricity today is produced mainly from non-renewable resources. [18] An ideal HER would be solar-driven, which includes both photoelectrocatalytic and photocatalytic pathways.…”

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

“…142 On the other hand, ML is considered as a fast, accurate, inexpensive, 143 and supportive tool 144 to predict the properties of SACs towards their rational design. [145][146][147] As shown in Fig. 3, using ML, one can apply the available datasets from QM and DFT calculations to construct readily available databases applicable in the deep analysis and establishment of preparation-structure-activity relationships.…”