Predictive analysis of chopping length of a forage machine using Artificial Neural Network (ANN) in MatLab

Abstract

In this work, Chopping Length (CL) of a multi-forage machine was predicted using Artificial Neural Network (ANN), in order to improve on the overall performance of the Forage Machine (FM).

Data was imported from the experiment carried out at the Department of Agricultural Engineering, University of Ibadan on FM after chopping Guinea Grass (GG), Siam Weed (SW) and Maize Stover (MS). The data were cleaned to remove monotonous (similar) numbers and outliers. The data were split into two sets (Training and testing sets) and ANN algorithm was used to create the model. Network building (Multi-Layered Perceptron), training, predictions, evaluations and improvement were done using MatLab of R2013a model.

The results revealed that GG, SW and MS has a predictive CL of 3.1, 2.5 and 3.5 cm respectively. The coefficients of determination (R²) for training (43.9, 95.9 & 29.88 %), validation (99.79, 18.8 & 92.14 %) and all data (31.78, 66.09 & 29.27 %) gave a low error of ANN model. The Mean Square Errors (MSE) obtained are 0.84, 2.06 and 0.076 for GG, SW and MS respectively. But the MSE of MS gave the smallest error and the best fitting pattern whose average experimental data was the same with the predictive data obtained from ANN model. The ANN performance for GG, SW and MS are 3.72, 0.47 and 0.47 with epoch number of 2, 6 and 7 respectively.

The CL obtained using ANN were within the international chopping length standard of 2 – 4 cm for FM.

Keywords: ANN, forage machine, machine learning, MatLab.