Data-Driven Electronic Structure Analysis for Investigating Intermolecular Interactions in Asphalt

Intermolecular interactions in asphalt are challenging to investigate due to the complex molecular composition. To better understand these interactions, it is essential to analyze the electronic structure of asphalt molecules. In this paper, a data-driven framework was developed using machine learning to predict the electronic properties of asphalt. The model was trained using chemical composition features to predict electrostatic potentials calculated via density functional theory, with previous grey relational analysis employed to identify key features for improved prediction accuracy. The correlations between electrostatic potentials and intermolecular interaction energy were then investigated. The findings demonstrated that the models predicting electrostatics potentials achieved goodness-of-fit values exceeding 0.82, indicating their reliability. Results further showed that stronger intermolecular interaction energy was associated with a higher electrostatic potential minimum and a lower potential maximum. Based on this qualitative relationship, two electrostatic potential parameters were proposed as indicators for evaluating molecular interactions, enabling straightforward prediction of asphalt intermolecular interactions at the electronic level. These findings are expected to facilitate rapid, theory-guided prediction of asphalt intermolecular interactions without extensive quantum calculations, supporting deeper insight into the microscopic basis of asphalt performance.