Mostrar registro simples

dc.contributor.advisorSteiner, Maria Teresinha Arns, 1957-pt_BR
dc.contributor.otherDyminski, Andrea Sellpt_BR
dc.contributor.otherSiqueira, Paulo Henrique, 1976-pt_BR
dc.contributor.otherUniversidade Federal do Paraná. Setor de Tecnologia. Programa de Pós-Graduação em Métodos Numéricos em Engenhariapt_BR
dc.creatorVillwock, Rosangelapt_BR
dc.date.accessioned2024-05-20T18:24:37Z
dc.date.available2024-05-20T18:24:37Z
dc.date.issued2009pt_BR
dc.identifier.urihttps://hdl.handle.net/1884/20898
dc.descriptionOrientador: Profª. Dra. Maria Teresinha Arns Steinerpt_BR
dc.descriptionCoorientadores: Profª. Dra. Andréa Sell Dyminski e Prof. Dr. Paulo Henrique Siqueirapt_BR
dc.descriptionTese (doutorado) - Universidade Federal do Paraná, Setor de Ciências Exatas e Setor de Tecnologia, Programa de Pós-Graduação em Métodos Numéricos em Engenharia. Defesa: Curitiba, 29/07/2009pt_BR
dc.descriptionInclui bibliografiapt_BR
dc.descriptionÁrea de concentração: Programação matemáticapt_BR
dc.description.abstractResumo: O monitoramento da estrutura de uma barragem, de importância bem conhecida, pode gerar uma enorme massa de dados, definidos em domínios multidimensionais,cuja análise e interpretação nem sempre são triviais. É importante selecionar asinformações que melhor "expliquem" o comportamento da barragem, permitindo a previsão e a resolução de eventuais problemas que possam ocorrer. A Usina Hidrelétrica de Itaipu, maior geradora de hidroeletricidade do mundo, possui mais de2.200 instrumentos que monitoram seu comportamento geotécnico e estrutural, os quais possuem leituras armazenadas em um banco de dados há mais de 30 anos.Assim sendo, o objetivo principal deste trabalho é apresentar uma metodologia,enquadrada na área de KDD, "Descoberta de Conhecimento em Bases de Dados", com o intuito de realizar a hierarquização de instrumentos de monitoramento de barragens, maximizando a eficácia e eficiência das análises das leituras, através da identificação de grupos de instrumentos semelhantes e, também, detectando osprincipais instrumentos. A metodologia foi aplicada à 30 extensômetros localizados em diferentes blocos do trecho F da referida barragem que, com uma, duas ou três hastes, totalizam 72 medidas de deslocamentos, das quais 24 foram automatizadaspela empresa. Para a fase de pré-processamento dos dados, do processo KDD, identificou-se que para a maioria dos instrumentos tem-se uma leitura mensal, porém, alguns deles, apresentam mais de uma leitura por mês sendo que, nestes casos, foi obtida a média mensal. Por outro lado, alguns instrumentos apresentaram leituras faltantes e, nestas situações, foram realizadas interpolações por sériestemporais garantindo, desta forma, que todos os instrumentos tivessem exatamente 120 leituras (10 anos). Já para a fase de Mineração de Dados, do processo KDD, a tarefa é o agrupamento de padrões e, para isso, foram utilizados os seguintesmétodos: da área de Análise Estatística Multivariada (Ligação Simples, Média, Completa e Método Ward); da área de Redes Neurais Artificiais (Redes Neurais de Kohonen) e da área de Metaheurísticas (foi proposto um Algoritmo de Agrupamento Baseado em Formigas). Em relação ao algoritmo proposto, este foi testado em trêsbases de dados reais (IRIS, WINE e PIMA Indians Diabetes) e em duas bases de dados reais de séries temporais (GUN e LIGHTNING-2), sendo que o seu desempenho foi comparado com o de outros dois métodos (Método Ward e RedesNeurais de Kohonen). Na aplicação da Análise de Agrupamento (pelo Método Ward) aos dados de instrumentação geotécnica-estrutural da Itaipu, mostrou-se que é possível encontrar justificativas técnicas para a formação dos grupos, inclusive identificando um grupo de hastes de maior importância. Já a aplicação da Análise Fatorial aos referidos dados, mostrou-se bastante eficaz para realizar a hierarquização das hastes de extensômetros, com base nas comunalidades. No algoritmo proposto, as principais modificações em relação ao algoritmo básico proposto por Deneubourg et al. (1991, apud Handl, Knowles e Dorigo, 2006), foram: a introdução de uma comparação da probabilidade de descarregar um padrão na posição escolhida aleatoriamente com a probabilidade de descarregar este padrão em sua posição atual; a introdução de uma avaliação da probabilidade de uma posição vizinha, quando a decisão de descarregar um padrão for positiva e a célula em que o padrão deveria ser descarregado estiver ocupada; e a substituição do padrão carregado por uma formiga, caso este padrão não seja descarregado em 100 iterações consecutivas. O algoritmo proposto apresentou resultados satisfatórios em relação aos resultados de Boryczka (2008) para as bases de dados reais e, quando aplicado aos dados de instrumentação geotécnica-estrutural da Itaipu, o mesmo foi capaz de identificar o grupo de hastes de maior importância.pt_BR
dc.description.abstractAbstract: The monitoring of the dam structures, of known importance, can generate an enormous mass of data, defined in multidimensional domains, which analysis and interpretation are not trivial. It is important to select the information that best"explains" the behavior of the dam, allowing the forecast and the resolution of eventual problems that can happen. The Hydroelectric Power Plant of Itaipu, the largest hydro electrical power producer of the world, has more than 2.200 instruments to monitor its geotechnical and structural behavior, which has readings stored in a database for more than 30 years. In this way, the main goal of this work isto present a methodology, framed in the KDD area, "Knowledge Discovery in Databases", in order to carry out the ranking of instruments of monitoring of dams, maximizing the effectiveness and the efficiency of the readings analyses, through the identification of groups of similar instruments and, also, detecting the main instruments. The methodology was applied to 30 extensometers located in different blocks of the sector F of the referred dam which, with one, two or three rod, totalized 72 measures of displacements, of which 24 were automated by the company. For the phase of preprocessing of the data, of the KDD process, it was identified that the majority of the instruments had a monthly reading, however, some of them, presented more than a reading by month and, in these cases, it was obtained the monthly average. In the other hand, some instruments presented failed readings and,in these situations, interpolations were carried out by time series assuring, in this way, that all of the instruments had exactly 120 readings (10 years). In the Data Mining phase, of the process KDD, the task is to group the patterns and, for that, the following methods were used: of the Multivariate Statistical Analysis area (Single Linkage, Average Linkage, Complete Linkage and Ward Method); of the Artificial Neural Networks area (Kohonen Maps) and of the Metaheuristics area (it wasproposed an Ant Based Clustering Algorithm). In relation to the proposed algorithm, it was tested in three real databases (IRIS, WINE and PIMA Indians Diabetes) and in two time series real databases (GUN and LIGHTNING-2), and their performanceswere compared with other two methods (Ward Method and Kohonen Maps). In the application of the Clustering Analysis (by Method Ward) at the data ofinstrumentation geotechnical and structural of the Itaipu, it was shown that it is possible to find technical justification for the formation of the groups and, also,identifying a group of rods of greatest importance. The application of the FactorialAnalysis to the referred data showed to be effective to realize the extensometer rods ranking, based in the communality. In the proposed algorithm, the main modifications in relation to the basic algorithm proposed by Deneubourg et al. (1991, apud Handl,Knowles e Dorigo, 2006), were: the introduction of a comparison of the probability of drop a pattern in a random chosen position with the probability of drop this pattern in the current position; the introduction of a evaluation of the probability of a neighboring position, when the decision of dropping a pattern is positive and the cell in which the pattern should be dropped is busied; and the replacement of the carried pattern by an ant, in case this pattern is not dropped in 100 consecutive iterations. The proposed algorithm presented satisfactory results compared with Boryczka (2008)’s results for the real databases and, when applied to the data of instrumentation geotechnical and structural of the Itaipu, the same was able to identify the group of rods of greatest importance.pt_BR
dc.format.extent124f. : il., grafs., tabs.pt_BR
dc.format.mimetypeapplication/pdfpt_BR
dc.languagePortuguêspt_BR
dc.relationDisponível em formato digitalpt_BR
dc.subjectRedes neurais (Computação)pt_BR
dc.subjectAnálise multivariadapt_BR
dc.subjectBarragens e açudes - Medidas de segurançapt_BR
dc.subjectUsinas hidrelétricaspt_BR
dc.subjectAnálise numéricapt_BR
dc.titleTécnicas de agrupamento e de hierarquização no contexto de KDD - aplicação a dados temporais de instrumentação geotécnica-estrutural da usina hidrelétrica de Itaipu.pt_BR
dc.typeTesept_BR


Arquivos deste item

Thumbnail

Este item aparece na(s) seguinte(s) coleção(s)

Mostrar registro simples