Well production in oil fields is a dynamic and complex activity. The patterns and characteristics inherent to the well, such as pressures and flow rates, are changing based on production time and the fluid composition – a complex multiphase mixture composed of oil, water, and gas. Thus, it is necessary to evaluate well behavior with periodic production tests. This paper proposes an automatic tool based on machine learning models to assist the production tests validation process in a quick manner. The developed methodology was applied to 13 representative wells of a Brazilian offshore oil field. For each examined well, a dataset is created with operation variables obtained from valid and invalid production tests. Six classification algorithms are analyzed, Logistic Regression, Naïve Bayes classifier, K-Nearest Neighbor, Decision Tree, Random Forest and Support Vector Machine (SVM) in reason to automatically label a new production test as valid or invalid, according to production historical data for the well. The dataset was divided into training and validation sets. The training set was used to perform feature selection, to calibrate and choose the proper model. The validation set was then used at the end of the procedure to evaluate obtained results, by comparing the model’s output with real test labels. From the results obtained in the case study, it was possible to identify that IGLR (Injection Gas/Liquid Ratio), oil flow rate and the pressure loss between wellhead and platform were representatives for most of the wells, which implies that these variables have a huge influence at the production well test validation. Furthermore, the validation set indicates that SVM and logistic regression were the models with the best performance. Besides that, accurate results were achieved, since the model correctly classified at least 5 of the 6 tests in 70% of wells analyzed, and for the remaining wells, 4 of 6 production tests.

No abstract

No abstract