Base de dados e benchmarks para prognóstico de anomalias em sistemas de elevação de petróleo

• Ano de defesa: 2019
• Autor(a) principal: Vargas, Ricardo Emanuel Vaz
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Doutorado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Detecção e diagnóstico de falhas; Monitoramento de poços de petróleo.; Gerenciamento de eventos anormais; Classificação de séries temporais multivariáveis; Fault detection and diagnosis; Oil well monitoring; Abnormal event management.; Multivariate time series classification; subject.br-rjbn; Engenharia Elétrica
• Link de acesso: http://repositorio.ufes.br/handle/10/13455

The oil industry considers that prognosis of anomalies in oil-producing wells can help to reduce production losses, environmental accidents, and human casualties and reduce maintenance costs. An oil well refers to a set of sensors and mechanical, pneumatic, and hydraulic systems. As in virtually any industrial process, several types of anomalies also occur in the process of oil lifting and flow assurance. This thesis formulates and evaluates the hypothesis that anomalies in naturally flowing wells can be detected with Machine Learning and that the use of expert hand-drawn and simulated instances is a feasible solution for the training of rare actual anomalies' detectors. The scarcity of measurements in such processes is a drawback due to the low reliability of instrumentation in such hostile environments. Another issue is the absence of anomalies' data – in quantity, quality, and adequately structured – in naturally flowing wells. To contribute to Machine Learning-based approaches to the prognosis of this type of anomaly, this work prepared and made public an original and realistic dataset with instances of eight types of anomalies characterized by eight process variables. Many hours of working together with experts from Petróleo Brasileiro S.A. were required to validate historical instances and to produce simulated and hand-drawn instances. The methodology developed and used in this preparation is detailed. Specific challenges that researchers can explore with the published dataset are defined. Experimental results related to these challenges suggest that the formulated hypotheses are true. This work has resulted in two relevant contributions. A challenging public dataset that can be used as a benchmark for the development of (i) machine learning techniques related to inherent difficulties of actual data, and (ii) methods for specific tasks associated with anomalies' classification in naturally flowing wells. The other contribution is the proposal of the defined benchmarks.