Validação de testes de produção de poços de petróleo baseada em mineração de dados

Abstract

During production of petroleum fields, production tests are frequently conducted in each well to identify the current conditions of well production. After tests are done, they are evaluated by the responsible team and, according to the information obtained, can be validated or not. The objective of this work is to create production validation tools based on data mining to assist real-time validation process. The proposed methodology is divided into three main stages. In the first one, a preprocessing is done to identify anomalous data, using the Local Outlier Factor (LOF), modified Z-score and average distances. After that, in the second step, predictive models of classification are analyzed to characterize production test as valid and not valid, according to information of the production history of the well. In third step, regression models are applied to predict the oil, water and gas flow variables. In this part, a prediction interval for each variable is constructed using the bootstrap sampling technique. The proposed methodology was applied in 13 representative wells of a Brazilian oil field. The developed techniques collaborate with the decision-making process in oil production activities.