Prediction of optimized pretreatment process parameters for biodiesel production using ANN and GA

An artificial neural network (ANN) based program coupled with genetic algorithm (GA) was developed on MATLAB platform for predicting the optimized process parameters required for reducing high free fatty acids (FFA) of any vegetable oils for successful transesterification. The developed ANN was a feed forward back propagation network (4-7-13-1) with one input, two hidden and one output layers. The input parameters for the ANN to generalize the pretreatment process were initial acid value of vegetable oil (IAV), methanol-to-oil ratio (M), catalyst concentration (C) and reaction time (T) and the output parameter was final acid value (FAV) of oil. The developed ANN was trained with the experimental data obtained for jatropha, mahua, simaruoba and rice bran oils with acid value more than 14 mg KOH/g-oil. The trained ANN was tested with separate set of data generated from pretreatment of mahua oil using response surface methodology (RSM) based on central composite rotatable design (CCRD) and found to predict the input pretreatment process parameters with low mean square error (MSE) and relative percent deviation (RPD). The well trained ANN synaptic joint weights and threshold values were used by GA to evaluate the fitness (to get FAV of oil less than 2 after pretreatment) of individuals (combinations of M, C and T) for optimization. The optimized process parameters predicted by the developed ANN–GA technique for sunflower oil with IAV 28 ± 1 mg KOH/g-oil were experimentally verified and the FAV was measured to be 2 ± 0.2 mg KOH/g-oil against the predicted value of 2 mg KOH/g-oil.