Extreme gradient boosting machine for modeling hydrogen gas storage in carbon slit pores from molecular simulation data

Abstract

To accelerate the computation of hydrogen storage capacity in uniform slit-shaped porous carbon determined by molecular simulation, the traditional Gradient Boosting and XGBoost algorithms are introduced to create the predictive models, and evaluate their prediction effectiveness and accuracy. The resultant models are tuned their hyperparameters by the random grid search method. From the comparison among the obtained models, it is found that the XGBoost model with optimized hyperparameters shows superior performance in predicting the hydrogen storage capacity in simulated carbon pores. According to the comparison of the results, the XGBoost model with optimized hyperparameters outperforms the other models in forecasting hydrogen storage capacity in simulated carbon pores as a function of pressure and pore sizes. Furthermore, the predicted results are in the best agreement with the pristine target dataset as measured by various evaluation metrics. Note that other models yield reasonable performance metrics, but they are unable to forecast high-pressure storage capacity in the ultramicropore region (less than 1 nm). The developed model could be applied for precisely and rapidly searching and comprehending the temperature-dependent optimal pore size for high-capacity hydrogen-storage systems in vehicular applications.

Keywords

• Hydrogen storage
• Data prediction
• Gradient Boosting
• XGBoost
• Molecular simulation