Estimating the solubility of HFC/HFO in ionic liquids from molecular structure using machine learning method

Abstract

Hydrofluorocarbon (HFC) + ionic liquid (IL) and hydrofluoroolefin (HFO) + IL are two new types of working pairs developed for absorption refrigeration system. In this paper, aiming to provide a guide for screening the optimal one from many candidates, a model based on group contribution (GC) method and artificial neuron network (ANN) is presented to estimating the solubility of HFC/HFO in ILs from molecular structure. The input variables of the ANN-GC model are temperature, pressure, and the number of various groups. A dataset containing 1693 solubility data for 18 HFC/HFO in ILs consisting of 10 cations and 17 anions at temperature from 273.13 K to 413.30 K and pressure from 0.99 kPa to 41,000 kPa were established to train the model. The ANN-GC model has great regression ability indicated by the average relative deviation of 8.8 % from experimental data. Besides, the case study on predicting [HMIM][BF4] - R134a working pairs solubility shows that our model has great prediction ability on new substances. The sensitivity analysis points out the groups influence on the solubility and give a guideline for designing high solubility working pairs. We also used Leverage approach to find the outlier data and high leverage data.