Machine learning-enabled performance prediction and optimization for iron-chromium redox flow batteries.

Area of Science:

• Materials Science
• Electrochemistry
• Data Science

Background:

• Iron-chromium flow batteries (ICRFBs) show significant potential for large-scale energy storage.
• Scaling up ICRFBs from lab to industry is challenging due to complex performance factors.

Purpose of the Study:

• To develop a data-driven methodology for precise prediction of ICRFB system performance.
• To optimize ICRFB design by considering operational conditions and material selection.

Main Methods:

• Utilized active learning and multitask machine learning (ML) models trained on literature data.
• Applied Shapley additive explanations (SHAP) for ML model interpretability.
• Validated ML predictions with experimental results.

Main Results:

• Achieved high prediction accuracy (R² > 0.92) for energy efficiency, coulombic efficiency, and capacity.
• Identified critical descriptors: current density, cycle number, and electrode type significantly impact efficiency; electrode size affects capacity.
• Active learning identified optimized operational cases for maximum energy efficiency and capacity.

Conclusions:

• The data-driven approach accurately predicts ICRFB performance and provides insights into critical property-performance relationships.
• ML model interpretability reveals key factors influencing battery efficiency and capacity.
• This work accelerates the rational design of next-generation ICRFBs.