Modelos de previsão do conteúdo de silício no ferro-gusa usando redes neurais artificiais

• Ano de defesa: 2018
• Autor(a) principal: Diniz, Ana Paula Miranda
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Blast Furnace; Silicon; Forecasting; Artificial Neural Networks; Pruning Algorithm; Wavelet Transform; Algoritmo de poda (Computação); Redes Neurais Artificiais; Transformada Wavelet; Altos-fornos; Silício; Previsão; Redes neurais (Computação); Wavelets (Matemática); Eletrônica Industrial, Sistemas e Controles Eletrônicos; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/9572

The silicon content present in hot metalhas been used over the years as one of the most representative indices of the thermal state of a blast furnace, as well as the quality of pig iron produced. Previous work has demonstrated the efficiency of Artificial Neural Networks (ANN) in terms of prediction of silicon content. Based on this premise, this work proposes the use of neural models to predict the elements of the time series of the silicon content in the hot metal, which due to delays in updating the silicon content measurements will be used forecast horizons of at least 3 hours ahead and a maximum of 8 hours ahead.Models with and without exogenous entrances will be tested, using for their selection, the expertise of the blast furnace operators and optimization techniques such as the Pruning Algorithm. In addition, in order to improve prediction performance, the signal of the silicon series will be decomposed into additive components using the Maximal Overlap Discrete Wavelet Packet Transform (MODWPT), and later these subseries will be appliedas inputs to neural models.The results indicate that the hybrid algorithm presents superior performanceto the algorithm using only ANN. The results obtained with real data of a steel industry indicate that the hybrid algorithm presents superiorresults to the models using ANNonly with exogenous entrances.Therefore, the results obtained by this work can anticipate control actions during the production process, contributing not only to the quality of the final product but alsofor the reduction of the costs associated with its production.