Artificial Intelligence for Electrical Percolation of AOT-based Microemulsions Prediction

Different Artificial Neural Network architectures have been assayed to predict percolation temperature of AOT/i-C-8/H2O microemulsions. A Perceptron Multi layer Artificial Neural Network with five entrance variables (W value of the microemulsions, additive concentration, molecular weight of the additive, atomic radii and ionic radii of the salt components) was used. Best ANN architecture was formed by five input neurons, two middle layers (with eleven and seven neurons respectively) and one output neuron. Root Mean Square Errors (RMSEs) are 0.18 degrees C (R = 0.9994) for the training set and 0.64 degrees C (R = 0.9789) for the prediction set.