Prediction of Composite Supercapacitor Performance Through Combining Machine Learning with Novel Binder-Related...

Area of Science:

• Materials Science
• Electrochemistry
• Machine Learning

Background:

• Composite electrode performance in supercapacitors is limited by complex interactions between active materials, conductive agents, and binders.
• Traditional design methods struggle with non-linear relationships, hindering optimization of specific capacitance.
• Accurate prediction requires accounting for electronic percolation, ion accessibility, and interfacial contact.

Purpose of the Study:

• To establish a physics-guided and interpretable machine learning (ML) framework for predicting composite electrode specific capacitance.
• To develop novel descriptors, including Binder-to-Conductive Ratio (BCR) and Specific Binder Loading (SBL), to better represent binder influence.
• To optimize ML model selection and hyperparameter tuning for accurate performance prediction.

Main Methods:

• Constructed a feature space of ten descriptors, including new binder-related proxies.
• Systematically evaluated 17 ML algorithms on a high-fidelity dataset.
• Employed Bayesian optimization for hyperparameter tuning and SHAP for interpretability analysis.

Main Results:

• Identified XGBoost, optimized via Bayesian optimization, as the best predictor (R² = 0.981, MAPE = 14.49%).
• SHAP analysis revealed that high BCR suppresses capacitance via an insulating barrier effect.
• Lattice distortion in filler materials was found to promote ion transport.

Conclusions:

• The developed ML framework offers a robust, data-driven approach for optimizing composite electrode performance.
• Interpretable ML models demonstrate significant potential for the rational design of advanced energy-storage materials.
• This work provides physical insights into factors governing supercapacitor performance.