Perpustakaan Digital ITB

Production forecast holds an important role in the oil and gas industry. Overestimated or underestimated reserve through future production forecast could lead to a serious matter for oil and gas companies. The utilization of the widely known Arps’ Decline Curve Analysis in future production forecasts raises the question of whether this method is still relevant to be used these days. Aside from its computational efficiency compared to simulation models, the subjectivity matter during the DCA process has been issued over time. Through the manual outlier detection and removal, interpretation of missing measurements, and data fitting through production trend selection can vary from different engineers. Moreover, the underlying assumption for conventional DCA that there is no change in well operation settings has increased the uncertainty of this technique. With the help of the Machine Learning approach, uncertainty and subjectivity matter can decrease throughout appropriate data processing. This study will propose an enhanced methodology for data clustering in production well using DBSCAN (Density- Based Spatial Clustering of Application with Noise) by utilizing various other well properties including production rate, choke opening, wellhead pressure, and bottom-hole pressure and taking into account to choose the optimum epsilon parameter value for the available dataset. This study will use a new approach by utilizing a dimensionless dataset that will be inputted into the DBSCAN algorithm to obtain the best clustering for the well’s parameters. A series of study cases have been done using Volve Oil Field to test the reliability of the new enhanced method. Overall, the reserve estimation using the proposed methodology serves a more fitted trend that can capture the well’s performance.