Configuration of artificial neural network for estimation of taper of trees eucalyptus

Abstract

The aim of this work was to define appropriate configurations of Artificial Neural Networks (ANN) to model the taper of eucalyptus trees. were used cubage data of eucalyptus plantations located in southern Bahia. Several ANN configurations were evaluated differing in the number of neurons in the hidden layer, activation function, number of cycles and learning algorithms with their parameters. ANN were trained in Neuroforest system, and estimates were evaluated using the correlation coefficient between observed and estimated values, the root mean square error (RMSE%) and graphical analysis of waste. Simple configurations, with only 04 hidden neurons, have provided satisfactory results. All activation functions tested (hyperbolic tangent, sigmoid, identity, log, linear, sine) may be used, wherein functions linear and identities are appropriate for the output layer of the ANN. The training of ANN may be done with 2000 cycles. The algorithms Resilient Propagation and Quick Propagation are efficient to applications of taper. Several ANN configurations may be used to applications of taper.