Professor Jung Kyu Kim’s research group (School of Chemical Engineering, Sungkyunkwan University), in collaboration with Professor Uk Sim’s group (School of Energy Technology, Korea Institute of Energy Technology, KENTECH) and Min-Cheol Kim’s group (Department of Chemistry, Sookmyung Women’s University) developed a facile experimental and systemical condition optimization by genetic algorithm based machine learning catalyst optimization and verification strategy for high efficient electrochemical hydrogen production system.

Electrochemical water splitting for hydrogen (H2) production in acidic media has attracted wide attention since its sufficient proton supply towards hydrogen evolution reaction (HER) induces much favourable reaction kinetics on the surface of catalysts. To develop the rational design of transition metal based electrocatalysts for electrochemical H2 production, high-throughput screening with density functional theory (DFT) simulations and machine-learning (ML) models trained on DFT data has been extensively used to predict candidate materials for high-efficiency catalysts. Although such data-aided screening strategies can be a useful tool for accelerating the design of efficient electrocatalysts, realizing the predicted candidates as practical electrocatalysts is challenging due to the over-simplification of DFT models and the complexity of electrochemical systems. Thus, for the rational design of practical and efficient electrocatalysts, it is critical to use a prediction model that can optimize various experimental factors, including the type of catalyst, which would be the type of an element as the TM dopant in the M@CQD catalyst. Unfortunately, the traditional methods to optimize such experimental conditions are not applicable to electrochemical experiments due to the dependency amongst different variables and poor scaling with the increasing number of optimization variables. Thus, utilizing ML techniques that are effective for such nonlinear multivariable problems is essential.

In this work, we present a simple and facile catalyst prediction and experimental condition optimization strategy that can be readily used in the rational design of electrocatalysts by combining the ML-based catalyst prediction and optimization step with experimental and theoretical verification processes. Due to the complexity of the electrochemical system, ML techniques were first used to predict catalytic properties of electrocatalysts to optimize the experimental conditions and find the systemical key factors influencing catalytic performance were considered as input data, including conductors, loading amount, electrode type, temperature, and pH of electrolyte. We used the genetic algorithm (GA), which is advantageous for multivariable optimization problems with a limited amount of prior knowledge, and the Naïve Bayes classifier during the selection process to accelerate the convergence. The prediction is validated by obtaining experimental data under the predicted optimized conditions. This resulted in the synthesized single atom Ni@CQD electrocatalysts presenting the lower overpotential of 151 mV at 10 mA cm−2 and Tafel slope of 52 mV dec−1 for HER, which demonstrates that the ML process can predict the results of unexplored experimental conditions with high accuracy and provide a complete picture of M@CQD HER performance within the parameter space.

This innovative research provides effective strategy of rational design and optimization process for M@CQD-based HER electrocatalysts by combining a BGA prediction model with verification steps of electrochemical experiments, thereby accelerating the development of efficient electrocatalytic models.

This research achievement was selected as the cover art of journal 'Carbon Energy' (DOI: https://doi.org/10.1002/cey2.70006) on July, 2025.

Bayesian Genetic Algorithm and Catalyst Material Synthesis Schematic

Development of carbon quantum dot synthesis technology incorporating Ni single atoms

Implementation of a high efficient electrochemical hydrogen production system integrated with PEMWE

Selected as the cover article of Carbon Energy (July 2025 issue)