ISSN: 2578-4846
Authors: Ebere F*, Minou R, Hui P, Vamegh R, Fadairo A, Adu-Mensah D and Abderraouf C
In the oil and gas industry, there is a major challenge to accurately predict the crude oil production due to the complexity and sophistication of the subsurface conditions. Production forecasting is highly limited by the non-linearity between hydrocarbon production and any relevant petrophysical parameter. Trying to use just the conventional mathematical approaches might give inaccurate result because of the numerous assumptions employed by this approach. Therefore, there is a huge need to develop a reliable prediction model of hydrocarbon production. This will surely assist Petroleum Engineers to have a better understanding of the entire reservoir behavior to solve, evaluate, and optimize its overall performance. Utilizing data driven models which is the machine learning techniques can help to predict crude oil production with much more acceptable accuracy. In this paper, Python-Support Vector Regression and Orange-Linear Regression have been implemented to build the models that predict the daily oil production of a well in Bakken-Three Forks Formations. The statistical data for the Bakken-Three Forks formation oil production was from North Dakota Industrial Commission (NDIC) website. An open-source visual programmingbased data mining software Orange was used to train a multi-linear regression model of 817 datasets with addition of 200 lines of code algorithm written in Python which is a high-level programming language. Combination of these two software models gave a more robust and accurate predictions compared to the conventional method of using just a software model by others. The models developed can practically estimate the Daily oil production of a well in Bakken-Three Forks Formations. The R2 obtained is 0.98 from the low performance value of 0.35, the MAE became 10.593 and RMSE is 16.593 for SVR and linear regression with a cross validation of 10 folds for the 70% train dataset and 30 % test dataset shows MSE value of 2.826, RMSE of 1.681, MAE of 1.045 and R2 value of 0.998. The performance of this SVR model indicate that this developed model can be used to predict the Daily oil produced per well accurately with the supervised algorithm. The values obtained from the Orange-Linear regression show better performance when compared with the SVR and validates the values obtained from the Python Support Vector Regression from the model criteria evaluation of the results.
Keywords: Crude Oil Production; Support Vector Regression; The Bakken and Three Forks; formations; Machine Learning
