Redes neurais artificiais para prognose da produção de povoamentos não desbastados de eucalipto

• Autor(a) principal: Binoti, Mayra Luiza Marques da Silva
• Data de Publicação: 2010
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/3030
• topic: Redes neurais; Prognose; Eucalipto; Neural networks; Prognosis; Eucalyptus; CNPQ::CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL::MANEJO FLORESTAL

The aim of this study was to train, implement and evaluate the efficiency of artificial neural networks (ANN) to perform the prognosis of yield of even-aged stands of clones of eucalyptus. The data were from stands of eucalyptus located in south of Bahia, totaling about 2,000 hectares of forest. Were used numerical variables such as age, basal area, dbh, total height, volume and categorical variables such as class of soil, texture, spacing, topography, project and clone. The levels of detail of modeling were at the level of stand and individual tree. The data were randomly divided into two groups: training (80%) and generalization (20%). The training data were used to create neural networks, while the generalization data were used to evaluate the capacity of network toprovide accurate results for the unknown data. Were trained three types of networks: perceptron, multilayer perceptron and radial basis function.The ANN that presents the best performances in training and generalization were selected to perform the prognosis with data from the first inventory. Concluded that the ANN showed satisfactory results for total stand and individual tree, proving the potential and applicability of the technique in solving problems of forest measurement and management.