A Multi-Scale Path for the Characterization of Heterogeneous Karst Carbonates: How Log-to-Seismic Machine Learning Can Optimize Hydrocarbon Production
- Francesco Bigoni (Eni S.p.A.) | Marco Pirrone (Eni S.p.A.) | Fabio Pinelli (Eni S.p.A.) | Gianluca Trombin (Eni S.p.A.) | Fabio Vinci (Eni S.p.A.)
- Document ID
- Society of Petrophysicists and Well-Log Analysts
- SPWLA 60th Annual Logging Symposium, 15-19 June, The Woodlands, Texas, USA
- Publication Date
- Document Type
- Conference Paper
- 2019. held jointly by the Society of Petrophysicists and Well Log Analysts (SPWLA) and the submitting authors
- 5 in the last 30 days
- 180 since 2007
- Show more detail
The accurate modeling of carbonate reservoirs is a longstanding challenging task due to the difficulties in capturing and characterizing their intrinsic multi-scale heterogeneities. Unpredictable variations in pore size distribution, in pore geometry and connectivity strongly influence both reservoir rock properties and fluid-flow behavior. This fact often results in discrepancies between the petrophysical characteristics estimated from cores and/or logs and distributed in the 3D model, and the actual dynamic performances of wells and reservoir.
This paper introduces a novel fully-integrated machine learning workflow, based on a multi-dimensional/multiscale approach, aimed at obtaining a robust static and dynamic characterization of carbonate reservoirs. In detail, open-hole logs, micro-resistivity images, production logging and pressure transient analyses are used to simultaneously depict the rock characteristics and the associated dynamic behavior at well scale. Then, a supervised classification allows to build a link between the aforementioned reservoir properties and particular seismic attributes extrapolated at the well locations. Finally, the interesting carbonate features are distributed into the entire 3D reservoir model through a fit-for-purpose neural network algorithm that exploits the full seismic cube.
The complete methodology is presented by means of a study performed on an oil-bearing carbonate reservoir characterized by an extremely high heterogeneity due to diagenetic processes, in particular to karstification. These are responsible of important permeability enhancements in low porosity intervals that are critical for production optimization and reservoir management purposes. At well scale, karst features are characterized by an advanced image log interpretation mainly focused on the quantification of connected vugs. Moreover, multi-rate production logging and well test analyses are accomplished in order to evaluate the proper permeability values in such karst intervals where corecalibrations and log-based predictions are not reliable. Next, karst-related vug densities, flow-calibrated permeabilities and selected seismic parameters such as lineaments (from continuity and curvature attributes), and the outputs of spectral decomposition have been used in a neural network process as sets for log-to-seismic learning and validation phases. In the end, the network is run to distribute the karst-related permeability enhancements into the 3D reservoir model according to the driving seismic attributes.
The final outcomes of the workflow are karst probability maps that are deemed fundamental to guide new wells location and trajectory. As a matter of fact, four proof of concept case histories have demonstrated the reliability of the approach. The newly drilled wells with paths guided by these prediction-maps have intercepted the desired karst intervals as per the subsequent image log interpretation. The latter has been also used to define the proper perforation strategy including low porosity intervals but with high vug density. Well tests and multirate production logging interpretations have proven the outstanding well performances associated with permeability values in the order of the Darcy right through the karstified rock. Based on these successful results, the ongoing drilling and perforation campaign of several other wells is built upon this comprehensive methodology.
|File Size||1 MB||Number of Pages||9|