Machine Learning Relationships Between Nanoporous Structures and Electrochemical Performance in MOF Supercapacitors

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry

Background:

• Supercapacitor development is limited by complex relationships between electrode structure and electrochemical performance.
• Machine learning (ML) applications are hindered by a lack of comprehensive, unified databases for nanoporous materials.

Purpose of the Study:

• To construct a unified database of metal-organic framework (MOF) electrodes using constant-potential molecular simulation.
• To develop ML models for predicting supercapacitor performance metrics like capacitance and charging rate.
• To elucidate the structure-property relationships governing supercapacitor performance.

Main Methods:

• Constant-potential molecular simulation to generate a database of hundreds of MOF electrodes.
• Development and application of decision-tree-based ML models for performance prediction.
• Experimental validation of ML predictions and SHAP (SHapley Additive exPlanations) analysis for feature importance.

Main Results:

• ML models accurately predict capacitance and charging rate, validated experimentally.
• Specific surface area (SSA) is the primary determinant of gravimetric capacitance.
• Porosity and pore dimensionality significantly influence volumetric capacitance and charging rate, especially in 3D-pore MOFs.

Conclusions:

• SSA and porosity are critical descriptors for designing high-performance supercapacitor electrodes.
• Pore dimensionality plays a crucial role in optimizing volumetric capacitance.
• Understanding these structure-property relationships enables rational design of advanced energy storage materials.