Estimation of Heat Capacity of Ionic Liquids Using Sσ-profile Molecular Descriptors

Abstract

In order to estimate the heat capacity of ionic liquids (ILs), statistical models have been proposed using the quantum-chemical based charge distribution area (Sσ-profile) as the molecular descriptors for two different mathematical algorithms: multiple linear regression (MLR) and extreme learning machine (ELM). A total of 2416 experimental data points, belonging to 46 ILs over a wide temperature range (223.1–663 K) at atmospheric pressure, have been utilized to carry out validation. The average absolute relative deviation (AARD %) of the whole data set of the MLR and ELM is 2.72% and 0.60%, respectively. Although both algorithms are able to estimate the heat capacity of ILs well, the nonlinear model (ELM) shows more accuracy, due to its capacity of determining a complex nonlinear relationship. Moreover, the derived models can shed some light onto structural features that are related to the heat capacity and be a suitable option to decrease trial-and-error experiments.