Vaporization enthalpy prediction of ionic liquids based on back-propagation artificial neural network

Vaporization enthalpy (Δ vap H ) is a fundamental thermodynamic property of ionic liquids (ILs). Accurate prediction of vaporization enthalpy relies on appropriate mathematical models grounded in precise experimental measurements. The quantitative structure-property relationship (QSPR) model, a key semi-empirical approach, could predict physicochemical properties based on the molecular structure of a substance. However, accurately predicting vaporization enthalpy and adequately describing the molecular structure of ILs remain significant challenges for this model. In this study, we used the cavity volume and charge density distribution area at specific intervals, derived from the conductor-like screening model for segment activity coefficient (COSMO-SAC) method, as molecular descriptors. Utilizing the developed descriptors, we constructed an improved QSPR model (Δ vap H -ANN) to predict the vaporization enthalpy of ILs across a broad temperature range, employing the back-propagation artificial neural network (BP-ANN) algorithm. The dataset for our model consists of 3150 data points for 148 ILs within a temperature range of 298–631.86 K. Overall, the results show that the proposed Δ vap H -ANN model, which treats ILs as “ion pairs”, can accurately predict the Δ vap H of ILs across various temperatures. • An improved Δ vap H -QSPR model with BP-ANN algorithm was proposed with a high R 2 . • The Δ vap H can be predicted over a wide temperature range, not just at 298 K. • The descriptors obtained from COSMO-SAC yield good performance in describing ILs. • More IL species and data points were considered compared to other models.