A Knowledge-Data Dual-Driven Framework for Predicting the Molecular Properties of Rechargeable Battery Electrolytes

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Growing demand for rechargeable batteries necessitates wider operating temperature ranges and enhanced safety.
• Accurate prediction of electrolyte molecular properties like melting point (MP), boiling point (BP), and flash point (FP) is crucial for efficient battery development.

Purpose of the Study:

• To develop a knowledge-data dual-driven framework for predicting electrolyte molecular properties.
• To accelerate the discovery of novel electrolyte molecules for high-performance batteries.

Main Methods:

• The Knowledge-based electrolyte Property prediction Integration (KPI) framework collects and structures molecular data.
• Interpretable machine learning models analyze structure-property relationships.
• Discovered knowledge is embedded into property prediction models.

Main Results:

• Achieved low mean absolute errors of 10.4 K (MP), 4.6 K (BP), and 4.8 K (FP).
• Reached state-of-the-art performance on 18 out of 20 datasets.
• Identified 15 and 14 promising molecules for wide-temperature-range and high-safety batteries using high-throughput screening.

Conclusions:

• The KPI framework accurately predicts molecular properties and enhances understanding of structure-property relationships.
• KPI efficiently integrates artificial intelligence with domain knowledge for battery material discovery.
• This approach accelerates the development of safer, high-performance rechargeable batteries.