Modelagem matemática da produtividade do corte florestal mecanizado

• Ano de defesa: 2017
• Autor(a) principal: Gonçalves, Saulo Boldrini
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Doutorado em Ciências Florestais; Centro de Ciências Agrárias e Engenharias; UFES; Programa de Pós-Graduação em Ciências Florestais
• Palavras-chave em Português: Forestry techniques and operations; Forest mechanization; Forestry planning; Machine productivity; Técnicas e operações florestais; Planejamento florestal; Produtividade de máquinas; Mecanização florestal; Máquinas florestais; Florestas - Inovações tecnológica; Projetos de desenvolvimento florestal; Produtividade florestal; Recursos Florestais e Engenharia Florestal; 630
• Link de acesso: http://repositorio.ufes.br/handle/10/7710

The The productivity of the wood harvesting operations is one of the main viability variables of the forest enterprise, being directly influenced by the characteristics of the land, the stand and the operational planning. The variables that can affect the productivity of harvesting machines are, for the most part, indirectly identifiable, difficult to measure, and present complex relationships, making it difficult to predict the productivity of operations. The objective of this study was to generate a model through artificial neural networks (RNA) and linear regression to estimate harvester productivity as a function of terrain, settling and operational planning variables. For this purpose, a database was used, from a forest company, containing information on mechanized forestry cutting operations with harvester. RNA input variables for modeling harvester productivity were (individual mean volume of trees, timber volume, cutting age, spacing, operator experience and management regime). Data were randomly divided to be used the network training (70%) and generalization (30%) were used. The networks training was also performed with combinations of the input variables, in order to verify the influence of each variable on harvester productivity. Using only the variables that showed a significant linear correlation with the productivity, according to Pearson correlation coefficient matrix, by the test ta 5% and 1% of probability. Both modeling techniques were evaluated by means of statistics and graphical analysis of the residues. The artificial neural networks selected in the training and validation for estimating harvester productivity presented correlation coefficient values above 0.89 and less than 11.91, indicating strong correlation and high accuracy between the estimates and the observed values. The combination of the input variables of the network that presented the best result was the one that used all six variables evaluated in the study. The multiple linear regression with all variables of significant correlation was the one that had the best fit for harvester productivity, correlation coefficient 0.83 and RMSE% 14.5. Among the variables evaluated in the model, the one that explains the productivity estimated by the linear regression is the individual mean volume. Both modeling techniques were efficient in predicting harvester productivity in mechanized xii forest cutting, but RNA presented more accurate estimates and could be indicated instead of the traditional multiple linear regression model.