Prediction of CO2 sorption in poly(ionic liquid)s using ANN-GC and ANFIS-GC models

In the present work, group contribution method (GC) together with the Artificial Neural Network (ANN) and Artificial Neuro-Fuzzy Inference System (ANFIS) models have been used to predict the carbon dioxide sorption in several Poly Ionic Liquids (PILs). A number of PILs based on ammonium and imidazolium by different anion and cation parts have been investigated based on a dataset containing 35 PILs with 350 data points. 70% of data points have been used for training the network, 25% for testing and 5% for validation. The model is a multilayer Feed Forward Artificial Neural Network (FFANN) with Levenberg-Marquardt as a function for optimizing error. The number of optimum neurons in the hidden layer is equal to 20. To distinguish PILs from each other, their structure has been defined by group contribution method. In addition of type and number of the chemical structures, pressure (bar) and temperature (K) have been also fed as input parameters to the network. Amount of carbon dioxide sorption in terms of concentration has been defined as the output of the network. The Absolute Average Deviation (AAD%) indicates that the ANN-GC and ANFIS-GC models are appropriate tools to predict carbon dioxide solubility where ANN-GC model shows sign of some superiority.