Configuration of artificial neural network for prognosis the production of eucalyptus clonal stands

Abstract

The objective of this study was to define appropriate configurations of Artificial Neural Networks (ANN) for prognosis of forest production of eucalyptus plantations at the stand level. Data were obtained from continuous forest inventory and were evaluated different settings of ANN for the number of neurons in the hidden layer activation function, number of cycles and learning algorithms with their parameters. The training of network was held at Neuroforest system. The evaluation of the estimates was performed using the correlation coefficient between observed and estimated values, the root mean square error (RMSE%) and graphical analysis of waste. Satisfactory results are obtained with simple configurations of ANN containing only 03 neurons in the hidden layer. All activation functions tested (hyperbolic tangent, sigmoid, identity, log, linear, sine) may be used. The training of RNA may be made with 500 cycles. The algorithms Resilient Propagation, Scaled Conjugate Gradient and Quick Propagation are efficient for the modeling of forest prognosis. The prognosis of production of eucalyptus clonal stands may be modeled using several ANN configurations.