Machine learning-based models for accessing thermal conductivity of liquids at different temperature conditions

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Renewable Energy

Background:

• Global warming necessitates reducing greenhouse gas emissions, with biomass biofuels as a key alternative to fossil fuels.
• Biomass conversion processes require accurate thermophysical property data, such as thermal conductivity, which is often lacking.
• Hydrocarbons and oxygenated compounds are essential components in biomass-derived energy sources.

Purpose of the Study:

• To develop predictive models for thermal conductivity of hydrocarbons and oxygenated compounds using chemoinformatics.
• To address the scarcity of experimental thermophysical data for biomass conversion processes.
• To enable accurate design and dimensioning of biofuel production technologies.

Main Methods:

• Compilation and curation of experimental thermal conductivity data to create a comprehensive database.
• Application of chemoinformatics methodologies for property prediction.
• Development and validation of a support vector regression (SVR) model using the curated dataset.

Main Results:

• The developed SVR model demonstrated good predictive abilities for thermal conductivity.
• The model successfully predicted thermal conductivity for compounds and temperatures not present in the training data.
• Validated the efficacy of chemoinformatics approaches in estimating thermophysical properties.

Conclusions:

• The SVR model provides a reliable method for predicting thermal conductivity of hydrocarbons and oxygenated compounds.
• This approach can fill data gaps for designing efficient biomass-to-energy conversion processes.
• Facilitates the advancement of sustainable biofuel technologies through accurate property prediction.