Predicting Oil Production Flow Rate Using Artificial Neural Networks - The Volve Field Case

Abstract

Abstract Accurate prediction of oil production flow rates helps production engineers to detect anomalous values which in turn will provide insights about any flaws in huge oil well systems. To aid this, oil flow rate is commonly estimated using empirical correlations. However, in some cases, significant error is inherent in application of this empirical correlation and will often yield inaccurate results. This present work aims to develop a machine learning algorithm based on an Artificial Neural Network to predict with (high accuracy) the oil production flow rate, using an open source data obtained from Volve production field in Norway. The Downhole Pressure and Temperature, Average Tubing and Annular Pressure Details, Onstream Hours, and Choke Details are used as the input parameters to the algorithm. The procedure can be considered a valid approach for its high accuracy and due to the wide acceptance of data-driven analytics in the industry today. To develop the model structure, 70% of the data was used the training dataset, and to further evaluate the performance, 30% of the data was used to derive the mean square error and determination coefficient. An error distribution histogram and the cross-plot between simulation data and verification data were drawn. These results show high predictability of the model and affirmed that ANN has the ability to handle large dataset and also will give a better prediction of oil flow rate when compared to the empirical correlations method. Therefore, equipping production engineers with the capacity to accurately predict oil flow rates from upstream pressure, choke size, and producing gas to oil ratio of a producing well rather than the use of empirical correlations.