Big Data Approach for Assessing Hydraulic Interference Between Wells in Not-Controled Systems
- Victor Costa Da Silva (Petrobras)
- Document ID
- Offshore Technology Conference
- Offshore Technology Conference Brasil, 29-31 October, Rio de Janeiro, Brazil
- Publication Date
- Document Type
- Conference Paper
- 2019. Offshore Technology Conference
- Data Driven Methods, Big Data, Reservoir Management
- 1 in the last 30 days
- 67 since 2007
- Show more detail
- View rights & permissions
When we are evaluating reservoirs of very high hydraulic communication, as in the case of several Brazilian pre-salt fields, the identification of the effects of a well, be it source or sink, in other observer wells becomes very complex to be observed. It becomes even more difficult when we do not have control of the volumes that are injected in each zone of interest, uncertainty in the reported flows (mainly of the producers) and difficulty to define a perfect observer point. This work proposes to use the large volume of pressure and flow data that we have available to, through a linear optimization process, identify the hydraulic communication index of each well (producer or injector) at each point of observation.
To achieve this objective the author resorts to physical-based data-driven methods, and through linear optimization, reach hydraulical interference coefficients between wells. Those coefficients may delivers relevant, and even unexpected information on how wells are communicated, if there are fractures or vuges unseen by geological methods, and allow the reservoir managing team to anticipates water and/or gas breakthrough, a well is more responsive to which other, etc. Furthermore the methodology may give important information to subsidize the history matching process.
The paper shows that the methodology is widely applicable in reservoirs where either the hydraulical communication or the wells densification is high enough to avoid any conclusive assessments from usual methods and has as greatest advantage a strong physical background behind it, unlike several machine learning data driven methods. It will be presented through several examples, applying both in controlled (obtained by synthetically generated data from reservoir flow models) and uncontrolled systems (hard data obtained from Brazilian pre-salt reservoirs).
|File Size||1 MB||Number of Pages||14|