Several researchers have used Multi Linear Regressions (MLR) or Multi-Layer Perceptron (MLP) artificial neural networks to model the energy audit of agricultural production. A literature review showed that no previous analytical work has been reported on the comparison of MLR and ANN models to predict the energy output of fruit production. Therefore, the main goal of this research is to compare the MLR with MLP artificial neural networks modeling and select the best one to predict the energy output of peach production in Iran. For this purpose, the same data were used to train the MLR and MLP models and thus, 60, 70, 80 and 90% of data were selected to train the models. Levenberg–Marquardt learning algorithm was employed to train ANNs models. The results showed that 3.41 MJ of energy was consumed to produce one kilogram of peach in Iran. The application of the models highlighted that the differences between the actual and predicted values for the two models were not statistically significant. The performance indices such as coefficient of determination (R2), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Efficiency (EF) for the best ANN architecture were determined to be 0.96, 2297.75 kg, 11.79% and 0.96%, respectively. While, these indices for the best MLR model were 0.91, 3418.27 kg, 14.80% and 0.91%, respectively. Overall, it was concluded that the MLP models could better predict the energy output than those of MLR models and the performance of MLP highlighted that this model can be applicable to prognosticate the energy output of peach production.

Keywords