Predicting and Optimizing Tillage Draft using Artificial Network Technique

Abstract

Tillage as one of the agricultural practices consumes the largest amount of energy, which reflects on the total production cost. The artificial neural network (ANN) technique was utilized in the current study to optimize the performance of the tillage process. The ANN-modeled multilayer perceptron network with a backpropagation learning algorithm and momentum term was used by the PYTHON program. The ANN inputs were: the implement type, soil texture, moisture, bulk density, width, speed, and depth. The draught was the output (kN). Five layers composed the ANN model's optimal configuration (13-64-16-4-1). The linear and rectified linear units (ReLU) functions were utilized with hidden layers and the output layer, respectively. Momentum term and learning rate were 0.00003 and 0.9 respectively. The iterations number were 1000 epochs and stopped at 290 epochs. The coefficient of determination in the test datasets was high (0.92) while, the difference between actual and predicted output was low (2.08). Bulk density and depth were the main determinants of the draft. The evaluation of the developed model for chisel, moldboard, and disk plow gave satisfactory results of 0.985, 0.924, and 0.917. In comparison to the ANNs, the regression model's correlation coefficient for predicting draught force was the lowest (0.373).