Modelagem da produção industrial de celulose Kraft com modelos aditivos generalizados e redes neurais

• Ano de defesa: 2010
• Autor(a) principal: Stein, Fabiano da Rocha
• Orientador(a): Leite, Hélio Garcia
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado Profissionalizante em Tecnologia de Celulose e Papel
• Departamento: Qualidade da madeira; Tecnologia de celulose e papel
• País: BR
• Palavras-chave em Português: Modelagem de produção industrial; Celulose Kraft; Modelos aditivos generalizados; Redes neurais
• Palavras-chave em Inglês: Modeling of industrial production; Kraft pulp; Generalized additive models; Neural networks
• Área do conhecimento CNPq: CNPQ::CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL::TECNOLOGIA E UTILIZACAO DE PRODUTOS FLORESTAIS::TECNOLOGIA DE CELULOSE E PAPEL
• Link de acesso: http://locus.ufv.br/handle/123456789/5899

In this study, data collected on an industrial scale for some years, underwent modeling using generalized additive models (GAM) and artificial neural networks (ANN) as tools to evaluate the influence of some variables, timber and process on production and digester alkali charge. Generalized additive models were fitted using the software R (R Development Core Team, 2010), through the library "mgcv" (Wood, 2006), specific settings for generalized additive models. Significance tests were applied for each model set. In order to supplement the data were analyzed using artificial neural networks.One hundred RNA were adjusted to relate to production and the digester alkali charge, and variables such as wood density, age, precipitation, dry content of wood chips, bulk density of chips, the wood basic density, pulp viscosity and kappa. In this step we employed the software Statistica (Statsoft, Inc., 2007). The results show that the Generalized Additive Model (GAM) is a good choice to represent the phenomena of the pulp industry, where the variables are highly variable and there is strict control, unlike what happens on data from experimental designs. Should the use of RNA to estimate the output from the digester alkali charge and also proved a useful tool, since the correlations between actual and estimated data were above 88% and 60% respectively. Several variables associated with the raw material and the pulping process that were studied showed similar behavior and / or equal to what the majority of experimental studies have found.