Modeling of ionic liquids viscosity via advanced white-box machine learning

Area of Science:

• Physical Chemistry
• Chemical Engineering
• Computational Chemistry

Background:

• Ionic liquids (ILs) are increasingly vital in industry, necessitating accurate models for their physicochemical properties.
• Predicting IL viscosity is challenging due to complex intermolecular interactions and lack of a universal theoretical framework.
• Process optimization relies on reliable viscosity data, driving the need for advanced predictive models.

Purpose of the Study:

• To develop and evaluate white-box machine learning models for predicting the viscosity of pure ionic liquids.
• To compare the performance of genetic programming (GP) and group method of data handling (GMDH) models against existing theoretical and empirical approaches.
• To identify key parameters influencing IL viscosity and assess the applicability domain of the developed models.

Main Methods:

• Utilized genetic programming (GP) and group method of data handling (GMDH) as machine learning techniques.
• Developed models using a comprehensive dataset of 2813 experimental viscosity values for 45 ionic liquids across various temperatures and pressures.
• Investigated models with five, six, and seven input parameters to balance accuracy and formula simplicity.

Main Results:

• The GMDH model with seven inputs achieved the highest accuracy, with an average absolute relative deviation (AARD) of 8.14% and R² of 0.98.
• All seven-input models demonstrated superior accuracy compared to five- and six-input models, which offered simpler formulas.
• The proposed GMDH and GP models significantly outperformed existing theoretical and empirical models in predicting IL viscosity.
• Statistical analysis confirmed the high quality and applicability domain of the experimental data for both GMDH and GP models.

Conclusions:

• The developed GMDH and GP models provide accurate and reliable predictions of ionic liquid viscosity.
• These machine learning approaches offer a powerful alternative to time-consuming and expensive experimental viscosity measurements.
• Temperature was identified as the most influential factor affecting ionic liquid viscosity, as indicated by relevancy factor analysis.