Sign in Subscribe
Hydrogen

Machine Learning Predicts Hydrogen Storage Capacities with High Accuracy

Strategic Insights

— 1 min read

Researchers at Guilin University of Electronic Technology have utilized machine learning models to predict hydrogen adsorption capacities of Metal Organic Frameworks (MOFs) with high accuracy. The study employed classical density functional theory (cDFT) calculations and trained Gradient Boosting Regression (GBR), Random Forest (RF), XGBoost, and Support Vector Regression (SVR) models on five geometric features. The results demonstrated strong consistency with cDFT calculations, confirming the reliability of XGBoost in estimating hydrogen storage performance across diverse MOFs.

Key Takeaways:

  • The researchers developed and evaluated machine learning models to predict hydrogen adsorption capacities of 1000 MOFs from the CoRE database at 298 K and 6.5 MPa.
  • The XGBoost model showed the highest accuracy among the four machine learning models used, achieving strong consistency with cDFT calculations.
  • The study assessed the transferability of the XGBoost model on 4000 additional CoRE MOFs and 24,214 hypothetical MOFs (hMOFs), demonstrating its reliability in estimating hydrogen storage performance across diverse MOFs.
  • The research was funded by the National Key Research & Development Program of China, National Natural Science Foundation of Guangxi Province, National Natural Science Foundation of China (NSFC), and Guilin Lijiang Scholar Foundation.
  • The study has potential applications in the development of hydrogen storage materials for various industries.

Statistics:

  • The researchers trained the machine learning models on 1000 MOFs from the CoRE database.
  • The XGBoost model achieved an accuracy of 97.5% on the training data.
  • The study assessed the transferability of the XGBoost model on 4000 additional CoRE MOFs and 24,214 hypothetical MOFs (hMOFs).
  • The research concluded that the XGBoost model is robust and reliable in estimating hydrogen storage performance across diverse MOFs.

Sources:

  • "Prediction of Hydrogen Storage Capacities In Metal Organic Frameworks By Using Machine Learning Methods With a Small Dataset." Chemical Physics Letters, 2025; 877. (Elsevier - www.elsevier.com; Chemical Physics Letters - www.journals.elsevier.com/chemical-physics-letters/)
  • Guilman University of Electronic Technology, Guangxi Collaborat Innovat Ctr Struct & Property N, School of Materials Science and Engineering, Guangxi Key Lab Informat Mat, Guilin 541004, People's Republic of China (contact: Yongpeng Xia)
  • National Key Research & Development Program of China, National Natural Science Foundation of Guangxi Province, National Natural Science Foundation of China (NSFC), and Guilin Lijiang Scholar Foundation (funders)

Read more