Machine learning accelerates design of bilayer-modified graphene hydrogen storage materials

Abstract

To accelerate the exploration of modified graphene hydrogen storage materials, this paper proposes a Data-Driven Exploration Framework (DDEF) based on the best-fitting machine learning (ML) algorithms (Random Forest: Recall = 0.83, Accuracy = 0.83, Gradient boosted regression: R2 = 0.90, RMSE = 0.06) and density functional theory (DFT). The hydrogen storage capacity of bilayer double-deficient graphene (BDG[Li]) doped with B atoms and modified with Li and Ti atoms was predicted using ML, which shortens the design cycle of hydrogen storage materials. The interlayer composite hydrogen storage mechanism of the BDG[Li] structure is revealed by electronic property analysis. For the first time, the calculation method of the interlayer electrostatic force energy (Ej) is proposed, and the equation for calculating the interlayer H2 adsorption energy is optimized. The hydrogen storage data of BDG[Li] and the construction of Ej provide new feature data for subsequent ML calculations. Both DDEF and Ej are computationally verified to have high accuracy and generalizability, which is of research significance for accelerating the design of hydrogen storage materials.