Using artificial neural network to predict the ternary electrical conductivity of ionic liquid systems

作者：Hezave Ali Zeinolabedini; Lashkarbolooki Mostafa; Raeissi Sona^*

来源：Fluid Phase Equilibria, 2012, 314: 128-133.

DOI：10.1016/j.fluid.2011.10.028

• 1-Butyl-3-methylimidazolium
• hexafluorophosphate
• [bmim][PF6]
• Ionic liquid
• Water
• Ethanol
• Acetone
• Ternary systems
• Physical property estimation
• Electrical conductivity
• Modeling
• ANN

The unique physical properties of ionic liquids play a decisive part in many of their applications. Therefore, the ability to predict the physical properties of ionic liquids is extremely important for the rational design of proper ionic liquids with specific properties. In practice, the processes involving ionic liquids usually contain other components, in addition to the ionic liquids. Therefore, in addition to pure component properties, knowledge of the physical properties of mixtures are also crucial for various applications. In the present study, the feasibility of using a feed-forward multi-layer perceptron neural network (MLPNN) model was investigated to predict the electrical conductivity of the ternary mixtures of 1-butyl-3-methylimidazolium hexafluorophosphate ([bmim][PF6]) + water + ethanol and [bmim][PF6] + water + acetone in the temperature range from 288.15K to 308.15K, consisting of 104 data points. Not only were different networks, namely the linear and the hyperbolic tangent sigmoid transfer functions, considered in this study, but also the effects of the number of hidden layers, hidden neurons and the training algorithm were investigated on the accuracy of the results using 78 data points as training data to minimize the average absolute relative deviation percent (AARD%), mean square error (MSE) and correlation coefficient (R-2). Among the various cases studies, statistical analyses indicated the best configuration of the network to include one hidden layer and seven neurons in the hidden layer. The optimum network was then validated using 26 data points (test data) not used in the training stage which indicated the good interpolative ability of the trained network with AARD%= 1.44, MSE= 2.87 x 10(-8) and R-2 (=) 0.9981.

• 出版日期2012-1-25