Correlation and prediction of surface tension of highly non-ideal hydrous binary mixtures using artificial neural network

摘要

Prediction of surface tension of highly non-ideal binary aqueous-organic mixtures is crucial for interpreting the interaction between the molecules. In this regard, a multi-layer perceptron (MLP) artificial neural network (ANN) model is developed to predict the binary aqueous-organic surface tension as a function of mixture composition and temperature while the organic compounds are very dissimilar in size and type. To correlate the binary surface tension, gathered experimental surface tension data consisted of 30 binary mixtures containing 2271 data points in the wide temperature range of 273-471.15 K are randomly divided into three different subsets namely training (70 % of total data), validation (15 % of total data) and testing (15 % of total data) subsets. Different input variables are examined and the number of hidden neurons is optimized. The obtained results revealed that it is possible to correlate the binary surface tension with the best MLP network with 27 neurons in the hidden layer and inputs variables of temperature, mole fraction, molecular weight and critical pressure of non-water component with the average absolute relative deviation (AARD %) of lower than 1.43 %. Comparison of accuracy of the MLP model with several common models such as Jouyban-Acree model, Wilson equation, Paquette and Rasmussen areas and several equations of state including SRK, PR and CPA revealed more accuracy of the proposed MLP based model.