Accelerated discovery of hydrogen storage hydride perovskites: A combined machine learning and first-principles approach

Qi Zhou, Minming Jiang, Jiang Xu, Zong-Han Xie, Paul Munroe,
Accelerated discovery of hydrogen storage hydride perovskites: A combined machine learning and first-principles approach,
International Journal of Hydrogen Energy,
Volume 179,
2025,
151697,
ISSN 0360-3199,
https://doi.org/10.1016/j.ijhydene.2025.151697.
(https://www.sciencedirect.com/science/article/pii/S0360319925046993)
Abstract: The efficient discovery of novel hydrogenated perovskite materials (ABH) with superior hydrogen storage capacity presents a significant challenge, primarily due to the limitations of traditional trial-and-error methods and the vast unexplored compositional space. First-principles approaches, while accurate in calculating molecular-scale electronic properties and guiding experimental material design, are time-intensive for exploring all possible perovskite combinations. To meet this challenge, we developed a streamlined and efficient machine-learning framework for high-throughput screening, enabling the rational design of ABH materials with enhanced hydrogen storage capacity. Our methodology begins with the construction of a comprehensive database of material properties, followed by feature engineering to identify key attributes that influence hydrogen storage capacity. By introducing novel descriptors, such as the d-band center, which require low-cost computations, we significantly reduce the required volume of training data. The RF and GBDT algorithms exhibited superior performance, achieving R2 values of 0.86 and 0.78, respectively, underscoring the reliability of our predictive model. Furthermore, we employed Shapley additive explanations (SHAP) to enhance model interpretability, revealing that the d-band center is a critical determinant of hydrogen storage performance. This study addresses the lack of efficient design strategies for ABH materials by integrating machine learning with first-principles calculations into a unified screening framework. It not only provides a rapid screening approach for high-performance perovskites and other nanomaterials, but also offers a cost-effective strategy to enhance the accuracy of machine learning models in material science. Our framework establishes a strong foundation for accelerating the design and discovery of advanced hydrogen storage materials.
Keywords: Hydrogen storage; Hydride perovskites; Materials design; First-principles; Machine learning