Prediction of Global Warming Potential for Gases Based on Group Contribution Method and Chemical Activity Descriptor

Area of Science:

• Environmental Science
• Computational Chemistry
• Materials Science

Background:

• Accurate prediction of global warming potential (GWP) is vital for evaluating SF₆ substitute gases.
• Existing methods for GWP prediction require robust computational models.
• Machine learning approaches offer a promising avenue for developing predictive environmental performance models.

Purpose of the Study:

• To develop and compare machine learning models for predicting the global warming potential of SF₆ substitute gases.
• To identify key chemical descriptors that influence global warming potential.
• To determine the most effective modeling approach for environmental performance assessment.

Main Methods:

• Construction of machine learning models using a group contribution method for 165 molecules.
• Calculation of 58 chemical activity descriptors using M06-2X/def2-TZVP.
• Identification of key descriptors via Pearson correlation coefficient and subsequent model building.
• Comparative analysis of group-based and descriptor-based models, including Artificial Neural Network, Random Forest, Gradient Boosted Decision Trees, and Support Vector Machines.

Main Results:

• Descriptor-based machine learning models significantly outperformed group-based models.
• The descriptor-based Random Forest model achieved the highest predictive accuracy (R²=0.82) on the test set.
• The best model demonstrated excellent performance metrics: MSE=0.015, RMSE=0.024, MAE=0.09.
• Descriptor-based models exhibited superior stability and robustness over 1000 training iterations.

Conclusions:

• Machine learning models based on chemical activity descriptors are highly effective for predicting the global warming potential of SF₆ substitute gases.
• The Random Forest algorithm, utilizing descriptor-based features, provides a stable and accurate method for environmental performance assessment.
• This study offers a valuable tool for the development and selection of environmentally friendly SF₆ alternatives.