Jaehoon Lee, Tapan Mukerji, Michael Tompkins Statistical Integration of Time-lapse Seismic and Electromagnetic Data for Reservoir Monitoring and Management Jaehoon Lee, Tapan Mukerji, Michael Tompkins
Motivation and Objective Joint integration of multidisciplinary geophysical data can provide complementary information not only in reservoir characterization but also in reservoir monitoring. Statistical Rock Physics Modeling (Avseth et al., 2001; Mukerji et al., 2001) Well logs Rock physics modeling Multivariate PDF To develop a statistical workflow to integrate time-lapse seismic and EM data with proper consideration of scale differences.
Example: 2-D Cross Section Stanford VI-E (Lee & Mukerji, 2012) A 3-layer fluvial channel system. 30 year simulation with 31 producers and 15 injectors . 150×200×200 cells. Dx = Dy = 25m, Dz =1m. Oil Saturation (30 years) Facies
Example: Impedances & Resistivity Acoustic Impedance Elastic Impedance (30°) Electrical Resistivity Born filtering Geometric moving average
Example: Well Logs Well 1 Well 2
Example: Well Log Data Extension Oil sand Shale Oil sand Brine sand Shale Brine sand Oil sand Brine sand Shale
Example: Well-scale Data PDF Well-log Data (Training Data) Field-scale Data (Classifying Attributes)
Example: Classification by Well-scale PDF True Facies Well-scale PDF Bad classification result caused by scale difference
Methodology: Analogous Reservoir Generation Field-scale Data Well-scale Data MP Geostatistics, SNESIM (Strebelle, 2000) Well-scale PDF
Example: Simulated Field-scale PDF Simulated Field-scale Data (Training Data) Field-scale Data (Classifying Attributes)
Example: Classification by Field-scale PDF Well-scale PDF True Facies Improved classification Field-scale PDF
Example: Classification by Field-scale PDF True Facies Only EM Seismic & EM Only seismic
Example: 2-D Cross Section (Time-lapse) Oil Saturation (20 years) Oil Saturation (30 years) Facies (20 years) Facies (30 years)
Methodology: Analogous Reservoir Generation Two different saturation profiles are simulated for each realization. Seismic and EM attributes are related to each other by porosity. Well-scale PDF
Methodology: Hexa-variate Gaussian mixture Hexa-variate Gaussian mixture models are used to incorporate time-lapse seismic and EM data into statistical integration. Acoustic Impedance Elastic Impedance Electrical Resistivity
Example: Classification with Time-lapse Data Gaussian mixture (with DAI, DEI, DR) True Facies Non-parametric (with DAI, DEI, DR) Gaussian mixture (with AIt-1, EIt-1, Rt-1)
Conclusions Developed analogous reservoir generation method to consider the scale differences between well logs, seismic and EM data. Consistent way with seismic and EM forward modeling and inversion. Filtering can be used to reduce computational time. A few realizations are enough. Realizations do not have to be as large as the target reservoir. Applied hexa-variate Gaussian mixture models to incorporate time-lapse data into field-scale joint PDFs with the analogous reservoir generation method.
Future Work The developed workflow will be applied to a three-dimensional synthetic reservoir and a real reservoir. Stanford VI-E layer 3 Norne field