Modelagem do crescimento e produção utilizando máquina de vetor de suporte e redes neurais artificiais

• Ano de defesa: 2020
• Autor(a) principal: Cordeiro, Márcio Assis
• Orientador(a): Arce, Julio Eduardo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Centro-Oeste
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciências Florestais
• Departamento: Unicentro::Departamento de Ciências Florestais
• País: Brasil
• Palavras-chave em Português: Modelagem do crescimento e produção; máquina de vetor de suporte; redes neurais artificiais
• Palavras-chave em Inglês: Growth and production modeling; support vector machine; artificial neural networks
• Área do conhecimento CNPq: CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL; RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL::MANEJO FLORESTAL
• Link de acesso: http://tede.unicentro.br:8080/jspui/handle/jspui/1312

This study aimed to evaluate the performance of artificial neural networks (ANNs) and support vector machines (SVM) in the modeling of dendrometric variables in eucalyptus stands. The data used come from non-thinned commercial plantations, located in four municipalities located in the southern mesoregion of the state of Amapá and were made available by the company Amcel-Amapá florestal e celulose S / A. They come from permanent plots, temporary plots and pre-cut inventory, with ages varying between 22 and 88 months. Hypsometric, volumetric and growth and production models established in the literature were adjusted, and compared with the support vector machine technique and artificial neural networks. For each type of modeling, the data were randomly divided into two groups, 80% of the data for adjustment / training and 20% for validation / generalization. The same dendrometric variables used by the regression models were used by the MVS and ANNs. For the training and generalization of support vector machines (SVM), four configurations were used, formed from two error functions and two kernel functions. For configuration, training and generalization of the ANNs, Neuro 4.0 software was used, in which configurations of networks of the Adaline type (Adaptive Linear Element) were used; Multilayer Perceptron (MLP) and Radial Basis Functions (Radial Basis Function-RBF). Prior to the growth and production modeling, the site curves were adjusted and the production capacity was classified using the guide curve method. For that, two non-linear models were evaluated and then the stability of the site curves in the plots that had more than three measurements was evaluated. In modeling growth and production, the site index estimated by the selected equation was used. The quality of the adjustments of the regression models, and of the methodologies using ANNs and SVM, were evaluated using the correlation coefficient between the observed and estimated values (ryŷ), the square root of the mean error, expressed as a percentage of the mean (RMSE %), graphical analysis of residues (Res%). Support vector machines and artificial neural networks performed well in the estimates of height, individual volume and in the basal area and volume per hectare projections, proving to be promising techniques for applications in the area of measurement and forest management.