Use of artificial neural networks to study engineering factors for applying different soil amendments
2011
Boulos, D.S.M.
The main objective of this study is to use artificial neural network (ANN) models to predict some physico-engineering factors: bulk density, hydraulic conductivity, infiltration rate, soil penetration resistance and available water. Models also dealt with sorghum yield productivity, profit and WUE related to applying different soil amendments. The inputs for two ANN models included: Bitumen Emulsion (BE), Polyacrylamide (PAM), Organic Manure (OM), Sand (S), Silt (Si), Clay (C), Initial bulk density (IBd), Initial hydraulic conductivity (IKa), Initial infiltration rate (IIr), Initial soil penetration resistance (ISp) and Initial available water (IAW). The predicting outputs of one ANN model are: (Bd), (Ka), (Ir), (Sp) and (AW) and for a second ANN model are: productivity, profit and WUE. Multilayer feedforward ANN was trained using a backpropagation learning algorithm. The optimal configuration for the first ANN model consisted of 3 layers (11-15-5) with sigmoid transfer function at 100,000 training runs. The optimal configuration for the second ANN model consisted of 4 layers (11-20-10-3) with hyperbolic tangent transfer function at 50,000 training runs. During recall process, the results showed that the variation between observed and predicted outputs were very small and the correlation coefficients were 0.9850, 0.9903, 0.9946, 0.9987 and 0.9901 for Bd, Ka, Ir, Sp and AW respectively. Meanwhile, they were 0.9989, 0.9915 and 0.9856 for productivity, profit and WUE respectively .
显示更多 [+] 显示较少 [-]