A New Approach for Predicting Maximum Allowable Annulus Surface Pressure (MAASP) with Supervised Machine Learning
Amega Yasutra1, Ganesha Rinku Darmawan2*, Ardhi Hakim Lumban Gaol1, Stephen Salomo1

1Department of Petroleum Engineering, Faculty of Mining & Petroleum Engineering, Bandung Institute of Technology, jl. Ganesha 10, Bandung, Jawa Barat, Indonesia
2Department of Petroleum Engineering, Faculty of Design & Technology, Bandung Institute of Science Technology, Kota Deltamas CBD, Ganesha Boulevard, Cikarang Pusat, Bekasi, Jawa Barat, Indonesia.
*Corresponding Author: ganesharinkudarmawan[at]gmail.com

Abstract

Several well integrity issues may arise when a well commences production. Those integrity issues can create deviations from the expected annular pressure behavior indicating the presence of Sustained Casing Pressure (SCP). Hence, a pressure threshold applied in the annulus should be established, referred to as the Maximum Allowable Annulus Surface Operating Pressure (MAASP). This study based on 43 wells in a field that was compiled, studied to predict the MAASP using supervised machine learning application. The formulas used for MAASP calculations developed in ISO 16530-1 (2017) method. Out of 43 wells data, 31 wells data will be used in the training phase to build the model and 12 data in the model validation phase.
Supervised machine learning model used is predictive modeling as it enables the operator to develop a model using historical data to make a prediction on the new unanswered data. Random Forest, AdaBoost, and KNN Model was used to develop the model. The validation result for A-annulus, B-annulus, and C-annulus shows a strong resemblance proved by the precision index (R-value and R^2 value) for AdaBoost model. By using this approved model, MAASP prediction of three new sample data with and without smoothing action is performed. Furthermore, to test the extent of this supervised machine learning capability, the input data for prediction purposes is decreased to 75%, 63%, 50%, and lastly 25%. The study shows that though some data has been diminished, the supervised machine learning is presumed to have completed its purpose as it is still able to grasp the limited input and produce good results in MAASP predictions. This new approach could be used in a develop fields with typically same construction design

Keywords: Please Just algorithm, supervised machine learning, MAASP, predictive modeling, prediction, Try to Submit This Sample Abstract

Topic: Engineering of Drilling, Production, and production surface and facility