Interpretable white-box modeling for nitrogen storage in metal-organic frameworks.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Efficient nitrogen (N₂) removal is vital for producing high-purity oxygen (O₂) and methane (CH₄) in industrial processes.
• Accurate prediction of nitrogen storage capacity in metal-organic frameworks (MOFs) is challenging but essential for optimizing separation processes.

Purpose of the Study:

• To develop user-friendly mathematical correlations for predicting the nitrogen storage capacity of MOFs.
• To compare the performance of three advanced modeling techniques: Group Method of Data Handling (GMDH), Gene Expression Programming (GEP), and Genetic Programming (GP).

Main Methods:

• Applied GMDH, GEP, and GP to a dataset of 3073 laboratory measurements for nitrogen storage in MOFs.
• Evaluated model accuracy and reliability using statistical metrics (MAE, R²) and graphical methods.
• Utilized correlation analyses (Pearson, Spearman, Kendall) to determine the influence of temperature and pressure on storage capacity.

Main Results:

• The GEP model demonstrated superior performance with high R² values (0.9703-0.9750) and low MAE (0.9924-1.0101) across training, testing, and overall datasets.
• All developed models accurately reflected the expected trend of N₂ storage under varying pressure conditions.
• Temperature was identified as the most significant factor influencing storage capacity, followed by nonlinear pressure interactions.

Conclusions:

• Gene Expression Programming (GEP) provides a highly reliable and accurate method for predicting nitrogen storage capacity in MOFs.
• The developed correlations are applicable and credible, validated by the leverage method showing >95% data within acceptable ranges.
• This work offers valuable predictive tools for optimizing MOF-based gas separation technologies.