Abstract
The primary objective of this paper was to develop an artificial neural
network (ANN) simulation environment and mathematical models for predicting with
high accuracy soil compression parameters. The experiments were conducted at the
College of Agriculture - University of Basra, located at Garmat Ali, the soil was silty
clay loam. The factors that were investigated are moisture content (14 and 24%), tillage
depths (0, 15, 30, 45, and 50 cm) forward speeds (0.57, 0.94, and 1.34 m.s-1) and tire
pressures (50, 100, and 150 kPa). ANN environment was developed with the back
propagation algorithm using MATLAB software with various structures and training
algorithms. Design Expert software utilized to evaluate the studied parameters and
produce mathematical models. The results showed that all studied parameters had a
significant effect on soil physical properties including bulk density and cone index. The
effects of the studied factors on bulk density were depth > moisture content > forward
speed, > tire pressure (6% 4%, 2.4%, 2%, respectively). Whereas, the order of the
investigated factors based on their effects on cone index were depth > moisture content
> tire pressure > forward speed (6%, 4%, 2.4% and 2%, respectively). The best model
for predicting the bulk density under different field conditions was the 4-8-1 architecture.
Levenberg-Marquardt (Trainlm) produced outstanding performance with an MSE of
0.00226 and R2 of 0.986. Moreover, this performance was occurring at an epoch of 100.
For predicting cone index, the best performance was achieved by Levenberg-Marquardt
(trainlm) in 85 epochs, giving minimum MSE equal to 0.005112 and greater (R2) equal
to 0.967 during the training process. Thus, the optimal structure for predicting cone index
was 4-7-1
network (ANN) simulation environment and mathematical models for predicting with
high accuracy soil compression parameters. The experiments were conducted at the
College of Agriculture - University of Basra, located at Garmat Ali, the soil was silty
clay loam. The factors that were investigated are moisture content (14 and 24%), tillage
depths (0, 15, 30, 45, and 50 cm) forward speeds (0.57, 0.94, and 1.34 m.s-1) and tire
pressures (50, 100, and 150 kPa). ANN environment was developed with the back
propagation algorithm using MATLAB software with various structures and training
algorithms. Design Expert software utilized to evaluate the studied parameters and
produce mathematical models. The results showed that all studied parameters had a
significant effect on soil physical properties including bulk density and cone index. The
effects of the studied factors on bulk density were depth > moisture content > forward
speed, > tire pressure (6% 4%, 2.4%, 2%, respectively). Whereas, the order of the
investigated factors based on their effects on cone index were depth > moisture content
> tire pressure > forward speed (6%, 4%, 2.4% and 2%, respectively). The best model
for predicting the bulk density under different field conditions was the 4-8-1 architecture.
Levenberg-Marquardt (Trainlm) produced outstanding performance with an MSE of
0.00226 and R2 of 0.986. Moreover, this performance was occurring at an epoch of 100.
For predicting cone index, the best performance was achieved by Levenberg-Marquardt
(trainlm) in 85 epochs, giving minimum MSE equal to 0.005112 and greater (R2) equal
to 0.967 during the training process. Thus, the optimal structure for predicting cone index
was 4-7-1
Keywords
ANN
Bulk Density
Cone index
Design-Expert software
Keywords
ﻟا : Design Expert ﺔﻓﺎ ، .قا ﺧ