Molecular-based artificial neural network for predicting the electrical conductivity of deep eutectic solvents

Abstract

Due to their unique features, deep eutectic solvents (DESs) are well-known as promising and environmentally friendly solvents. Their use in various processes has recently become the focus of several research groups. However, designing DESs with optimal properties for a particular application requires many resources and is time-consuming. Therefore, it is crucial to develop predictive models to estimate the properties of DESs, which will save resources and time. Electrical conductivity is one of the most critical factors for the design, control and optimization of electrochemical processes. In this work, a model capable of estimating the electrical conductivity of DESs is presented. The model combines the Quantitative Structure-Property Relationships (QSPR) approach with artificial neural networks (ANNs) and COSMO-RS-based molecular parameters known as Sσprofiles.. The QSPR-ANN training set consists of 2,266 data points from 191 DES mixtures with 334 compositions prepared from 8 anions, 26 cations, and 73 hydrogen bond donors (HBDs) measured at various temperatures ranging from 218 to 403 K. The coefficient of determination (R2) for the QSPR-ANN developed was 0.993 in training and 0.984 in testing. In conclusion, the proposed approach can reliably estimate the electrical conductivity of DESs and can be used to determine appropriate DESs with the desired electrical conductivity for various electrochemical applications.