Previsão de produção de óleo em unidades de fluxo hidráulico com rede neural perceptron de múltiplas camadas

• Ano de defesa: 2021
• Autor(a) principal: Melo Neto, Edvaldo Francisco de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Informática; UFPB
• Palavras-chave em Português: Previsão de produção; Redes neurais artificiais; Perceptron de múltiplas camadas; Unidades de fluxo hidráulico; Production forecast; Artificial neural networks; Multilayer perceptron; Hydraulic flow units; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/22103

Identifying wells that return maximum oil production is a crucial point for reservoir planning and management, as drilling a well involves a lot of investment. Commercial reservoir simulators are able to predict production curves with high confidence, but in many cases they require a lot of time and computational effort. To overcome this difficulty, this work proposes a methodology based on the Multilayer Perceptron (MLP) neural network to predict cumulative oil production in reservoir wells that cross Hydraulic Flow Units (HFUs), which are volumes with good flowability. Each well was drilled vertically from points of maximum closeness within the HFUs. The database of this work, divided into training, validation and testing, was obtained from the UNISIM-I-D synthetic reservoir model, which has similar characteristics to the Namorado field, located in the Campos Basin, Brazil. The production results were presented from two perspectives: the original MLP and its post-processed version, in which both were compared with the oil production curves generated by the Computer Modelling Group Ltd (CMG) simulator. The performance was measured using the metrics: Root Mean Squared E (RMSE) and Mean Absolute Scaled Error (MASE), where the post-processed MLP overcame the results of its original version with an average of 0.0620 and 12.9053 for the RMSE and MASE, respectively, with respect to the test data.