Assessment of energy consumption and modeling of output energy for wheat production by neural network (MLP and RBF) and Gaussian process regression (GPR) models

The objective of this study was to predict the irrigated and rainfed wheat output energy with three soft computing models include Artificial Neural Network (MLP and RBF models) and Gaussian Process Regression (GPR) for the first time, in Shahreza city, Isfahan province, Iran. Data were collected from an extensive research on wheat farms, including 120 irrigated and 90 rainfed wheat farms (totally 210 questionnares) at three levels (small: < 2ha, medium: 2-4 ha and large: > 4ha) using with face to face questionnaire method. Results of energy analysis showed that diesel fuel was the most influential factor on energy consumption in irrigated wheat production and also for medium and large lands of rainfed wheat production, but for small rainfed lands, total of fertilizers and poisons had the highest impact on total energy consumption. Results of output modeling showed that ANN-RBF model is more accurate than MLP-ANN and GPR models. RMSE and MAPE for irrigated and rainfed output modeling for ANN-RBF were 63.12-72.30 MJ and 0.05-0.14%, respectively. The results of selecting the best spread factor (one of the best parameters on RBF model performance) showed that for irrigated wheat with LM algorithm and at training phase (irrigated-LM-training) and irrigated wheat with BR algorithm and at training phase (irrigated-BR-training), this factor is equal to 7 and 4, respectively. The ANN-RBF model developed was capable of predicting irrigated and rainfed wheat output energy under different land size and using input energies. (C) 2017 Elsevier Ltd. All rights reserved.