Corn crop production prediction using artificial neural network

This investigation aimed to evaluate the performance of artificial neural networks to predict the corn grain yield in the city of Jaguari, Central region of Rio Grande do Sul, based on morphological characteristics of this culture. It was used the data published by SOARES (2010) for training the neural networks. Several multilayer perceptron neural networks with backpropagation-optimized algorithm (Levenberg-Marquardt) were tested. The input layer variables used were leaf area index, total green matter, plant height and number of plants m^-2 and the output layer: corn grain yield. Each architecture was trained 10 times, picking up at the end of training the one with lower mean relative error and less variance for data validation. efficiency of the networks was analyzed by means of statistical indicators. Among many architectures trained, the network with 35 neurons in the hidden layer had the lowest error in training and validation processes. In this way, the neural network with architecture 4-35-1 presents a good performance, being efficient to estimate grain production, considering the region covered by the experiment.

modeling; artificial intelligence; production; Zea mays L.