Artificial neural networks for basic wood density estimation

The Wood density is the property which correlates most closely with the other physical and mechanical properties and is used to set the appropriate use of wood industrially. Several computational and mathematical tools are used to model the basic density of stands, highlighting the Artificial Neural Networks (ANN). Data came from 2247 trees of 20 clones of Eucalyptus spp. in 6 spatial arrangements, aged between 3 and 6 years. The RNA's were trained in order to estimate the density at the cutting age (6 years), with input variables such as age (years), basal area (m(2) / ha), mean annual increment (m(3) / ha / year) of measuring. age, total height (m), diameter at 1.3 m from the soil surface (Dap) (cm), stems number per hectare (n / ha) and the ratio of dbh, total height and interaction between variables. Training algorithms were: error backpropagation, resilient propagation, Manhattan update rule, scaled conjugate gradient, Levenberg Marquardt, quick propagation, simulated annealing, and genetic algorithms. The estimates were evaluated according to the correlation coefficients with the observed values, square root of the average percentage error (RMSE), graphic analysis of residues (percentage error) and histogram percentage frequency of the percentage errors. The best settings were 4 neurons in the hidden layer and hyperbolic tangent activation function in the hidden sigmoid layer and the output layer and 8 neurons in the hidden layer, hyperbolic tangent activation function in the hidden layer and the output layer sigmoid.