Data Driven Predictive Modelling Assists in Proactive Water Cut Management in a North Kuwait Heavy Oilfield
- Arun Kharghoria (Kuwait Oil Company) | Abdulrahman Alshammari (Kuwait Oil Company) | Santiago Gonzalez (Kuwait Oil Company) | Ayodele Olusegun Sanwoolu (Kuwait Oil Company) | Abdullah Abdul Karim Al-Rabah (Kuwait Oil Company) | Jacobo Montero (Shell) | Gregorio Gonzalez (Shell)
- Document ID
- Society of Petroleum Engineers
- SPE Kuwait Oil & Gas Show and Conference, 13-16 October, Mishref, Kuwait
- Publication Date
- Document Type
- Conference Paper
- 2019. Society of Petroleum Engineers
- Water Cut, Heavy Oil, Predictive Model, Multiple Linear Regression, Data Driven
- 1 in the last 30 days
- 54 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 5.00|
|SPE Non-Member Price:||USD 28.00|
This study presents the development of predictive models (for a single or a group of wells) to proactively manage water production in a heavy oil field in North Kuwait (average 16 °API crude, ~100 cp at 100 °F). Each resulting model is based on reservoir and time-dependent wellbore mechanical parameters that were obtained through several well tests. The primary goal is to be able to predict the water cut based on a well's current as well as "possible near-future" operating conditions.
The exercise involves using advanced predictive methodologies. Standard parameters (referred to as features) include water cut as the response variable along with pump intake pressure, total liquid production, pump speed, and surface unit power consumption among others as response variables. Random ForestsTM (RF), Alternating Conditional Expectations (ACE) and Multiple Linear Regression (MLR) algorithms were tested. Features were primarily grouped based on clustering. The predicted results are compared to history. The resulting models are further extended to generate type curves for future water cut prediction for different combinations of the operating parameters.
MLR and ACE based predictive models showed promising results. These algorithms also provided a way to characterize the influence of different features on the model outcome (i. e. predicting historical water production). The model quality significantly improves when only time-dependent features are used. The static parameters (e.g. permeability, perforation stand-off to oil-water contact, pump size) do not appear to have a big impact on improving the model score. Also, improvements were observed by introduction of principal component analysis (PCA) followed by clustering. The generated type curves allow the prediction of the near-future water production and to proactively adjust a well's operating parameters.
A predictive model is a time-saving tool compared to full range dynamic simulation. This provides a means to the engineers to establish an operating envelope for a well (or a group of wells) to manage water production. This is also a precursor to a full-scale machine learning process that could be implemented for automatically updating models when the quantity of data is expected to be massive in the coming years.
|File Size||1 MB||Number of Pages||14|
O. Alomair, A. Malallah, A. Elsharkawy, M. Iqbal : ‘Predicting CO2 Minimum Miscibility Pressure (MMP) Using Alternating Conditional Expectation (ACE) Algorithm, HAL Id: hal-01931395, https://hal.archives-ouvertes.fr/hal-01931395.
J. B. Ceballos, O. A. Vivas, "Mathematical model of controllers for progressive cavity pumps," Rev. UIS Ing., vol. 18, no. 2, pp. 17-30, 2019. doi: 10.18273/revuin.v18n2-2019002
ACEPACK Fortran Code by Phil Spector, Jerome Friedman, Robert Tibshirani, Thomas Lumley, Shawn Garbett, Jonathan Baron (https://github.com/cran/acepack)
Random Forests Fortran Code by Leo Breiman and Adele Cutler, https://www.stat.berkeley.edu/~breiman/RandomForests/cc_software.htm.
Miller, Alan J. "Algorithm AS 274: Least Squares Routines to Supplement Those of Gentleman." Journal of the Royal Statistical Society. Series C (Applied Statistics) 41, no. 2 (1992): 458-78. doi: 10.2307/2347583.
Clarke, M. R. B. "Algorithm AS 163: A Givens Algorithm for Moving from One Linear Model to Another Without Going Back to the Data." Journal of the Royal Statistical Society. Series C (Applied Statistics) 30, no. 2 (1981): 198-203. doi: 10.2307/2346398. (Fortran code by John F. Monahan at https://www4.stat.ncsu.edu/~monahan/aug00/chap5/flyreg.f95)