A GMDH-type neural network for prediction of water activity in glycol and Poly(ethylene glycol) solutions

Abstract

The objective of current work is to predict the activities of water in aqueous solution of glycols and polyethylene glycols (PEGs) by means of a group method of data handling (GMDH). The proposed model considers temperature, molecular weight of the polymer as well as weight fraction of water for input variables. A 387 experimental data set of activities of water in aqueous binary solution of different glycols and polyethylene glycols is used. Mean absolute error percent (MAE %) of the training and testing obtained were 2.45 and 2.46 respectively; the results when compared to those of an artificial neural network (ANN) model show a weaker correlation. However, a noticeably reduced runtime and more importantly an explicit general correlation function, typical of GMDH networks, offer superior advantages over an ANN model.