Prediction of SF₆ Replacement Gases Based on Machine Learning

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Materials Science

Background:

• Global Warming Potential (GWP) is critical for evaluating SF₆ alternatives.
• Radiative efficiency (RE) and atmospheric lifetime (τ) determine a gas's GWP.
• High-throughput screening is needed for identifying suitable SF₆ replacement gases.

Purpose of the Study:

• To develop and optimize machine learning (ML) models for predicting GWP, τ, and RE.
• To explore correlations between molecular descriptors and GWP parameters.
• To screen potential SF₆ replacement gases using developed ML models.

Main Methods:

• Utilized six modified machine learning methods, including Histogram Gradient Boosting Regression, Gradient Boosting Regression, and Extreme Tree.
• Employed molecular descriptors to analyze the relationships influencing τ and RE.
• Applied trained models to screen 853 molecules from the QuanDB dataset for GWP₁₀₀.

Main Results:

• Optimal ML models achieved high accuracy (R² > 0.90) for predicting GWP, τ, and RE.
• Identified highest occupied molecular orbitals and their energy gap as key molecular descriptors influencing τ and RE.
• Screened 853 molecules, identifying six promising SF₆ replacement candidates with desirable properties.

Conclusions:

• Developed accurate ML models for predicting GWP and its key parameters, facilitating efficient screening of SF₆ alternatives.
• Provided insights into the molecular-level factors governing the environmental impact of gases.
• Identified specific low-GWP SF₆ alternatives with favorable physical and electrical properties.