الفهرس 
REVIEW OF LITERATUREOnly 14 pages are availabe for public view
 |  | from | 138 |  |  |
| | | from | 138 | | |
Abstract
Tillage is considered one of the agricultural practices that consume the largest amount of energy, which reflects on the total production cost. The easiest approach to define how much energy a tool needs to operate is to determine the draft needed to pull it. This force is determined by the following variables: soil conditions, tool parameters, and operating parameters.
In recent years, a variety of fields have often employed machine learning techniques due to the development of high-performance computing. Artificial Neural Networks (ANNs), one of the many machine learning techniques, are particularly good for nonlinear mapping, adaptive learning, and environmental problem prediction since they don’t rely on statistical assumptions about the distribution of the data. Therefore, they have done well in recent studies.
The objectives of this research are to utilize an ANN approach to expect the tillage draft force and calculate the required energy according to the input parameters, to obtain the optimum required conditions to optimize the performance of the plowing process and to compare the performance of the ANN technique with regression models in the prediction of draft force.
A huge quantity of data related to draft requirements for tillage tools under various operating and soil conditions was gathered to create the models. This data included 735 data points that rely on field experiments.
The inputs to the ANN-modeled multilayer perceptron with a backpropagation learning technique and momentum term were implement type (chisel, moldboard, disk, rotary, and subsoiler plow), the particle size distribution of soil (sand, silt, and clay %), the moisture content (%), the bulk density (g/cm3), depth (cm), speed (km/h), and width (m). The output was the draft (kN). The collected data were split randomly into 3 sets: a training, a validation, and a testing sample each comprising 60%, 20%, and 20%, respectively by using PYTHON software.
The results showed that the optimal architecture to ANN was (13-64-16-4-1) consisting of 5 layers, The rectified linear unit (ReLU) was utilized in the layers hidden and the linear function in the layer of output, and the learning rate and MT were 0.00004 and 0.9 respectively and the number of iterations were 1000 epochs and stopped at 760 epochs by using early stopping.
The efficiency of the ML regression algorithms was assessed using the mean square error (MSE), the root mean square error (RMSE), and the determination coefficient (R2) among the actual and predicted values. The ANN gave the highest performance (R2=0.944) and minimum error (MSE=1.73).
After training and testing the ANN, the optimum conditions for optimizing the performance of the tillage process were obtained which gave a minimum draft value. The performance of the developed ANN model was evaluated, and the results demonstrated that the developed ANN model gave results for ”R” ^2 higher than 92% for plows. Then, the energy requirements were calculated for plows under various conditions.
Finally, the performance of ANN was compared with regression models, the outcomes illustrated that the ANN achieved the highest performance compared with these models.
RECOMMENDATION
The ANN model can be relied on to predict the draft and calculate the amount of energy required to optimize the performance of the tillage process compared to regression models.
The optimum conditions are the moisture content is 15% , the bulk density is 1.5 g/cm3, the depth is 10 cm, the speed is 2.5 km/h, the width of the plowing is 1.75 m which gave a minimum draft value of 1.7 kN/tool by using the Disk plow, and the soil type is clay loam.