Calculation of hydrogen dispersion in cushion gases using machine learning

Area of Science:

• Energy storage
• Geological engineering
• Computational science

Background:

• Underground Hydrogen Storage (UHS) is vital for sustainable energy, but hydrogen dispersion in cushion gases reduces purity and increases costs.
• Current methods for predicting hydrogen dispersion coefficients (K_L) are often inaccurate, costly, and time-consuming under dynamic reservoir conditions.

Purpose of the Study:

• To develop a more accurate and cost-effective method for predicting hydrogen dispersion coefficients (K_L) in UHS.
• To leverage machine learning (ML) to model K_L based on experimental data and operational parameters.

Main Methods:

• Integration of experimental data with various machine learning models, including Random Forest (RF), Least Squares Boosting (LSBoost), Bayesian Regression, Linear Regression (LR), Artificial Neural Networks (ANNs), and Support Vector Machines (SVMs).
• Quantification of K_L as a function of pressure (P) and displacement velocity (U_m).

Main Results:

• The Random Forest (RF) model demonstrated superior performance, achieving a high R² of 0.9965 for test data and 0.9999 for training data.
• RF model yielded low Root Mean Square Error (RMSE) values of 0.023 for test data and 0.001 for training data, indicating high predictive accuracy.
• ML models, particularly RF, significantly improved the prediction accuracy of K_L compared to traditional methods.

Conclusions:

• Machine learning offers a powerful and efficient approach for accurately predicting hydrogen dispersion in UHS.
• Optimized prediction of K_L can enhance UHS operational efficiency, reduce hydrogen contamination risks, and lower purification costs.
• This ML-driven methodology supports the advancement of large-scale hydrogen storage for a sustainable energy future.