Applying the distance-based approach to model flow uncertainty in
Transcription
Applying the distance-based approach to model flow uncertainty in
Annual Meeting 2011 Stanford Center for Reservoir Forecasting Applying the distance‐based approach to model flow uncertainty in structurally complex reservoirs Aaditya Satija, Jef Caers Motivation • Structural Uncertainty remains a challenge – Computational complexity of Structural Modeling – Difficulty in automation • manual consistency checks needed – Multiple levels of uncertainty – Many independent parameters 2 Distance Kernel Methods Generation Generation of of multiple multiple models models Definition Definition of of aa distance distance Kernel Kernel Transformation Transformation Representation Representation of of Uncertainty Uncertainty MDS MDS Clustering Clustering Model Model Selection Selection Caers 2011 3 Applying DKM to Structural Models • Challenges – Constructing multiple structural models • Surface interactions: Consistency checks needed – Multiple uncertain input parameters • Fault Geometry – Orientation – Hierarchy • Fault Placement SCRF 2011 4 Elements of DKM 1. Model of structural uncertainty – Fault Geometry – Placement 2. Constructing multiple structural models – gOcad FaultMod 3. Fast flow simulation (3DSL) – Sensitive parameters to 3DSL response 4. Select models 5. Eclipse flow simulation on selected models – Sensitive parameters to Eclipse response SCRF 2011 5 Modeling Structural Uncertainty • Fault Geometry – Strike and Dip – Hierarchy – Well defined distance metrics • Fault placement – Not even a parameter – What to do with this one? SCRF 2011 6 Fault Placement SCRF 2011 7 Fault Placement Distances between fault origins Distance between fault models SCRF 2011 8 Fault Placement • From distance to parameter – distance matrix Æ MDS, Kernel transformation – Clusters Æ Categorical Parameter SCRF 2011 9 Fault Placement w.r.t wells SCRF 2011 10 Fault Placement w.r.t. wells Distance between Injector and Fault Origin SCRF 2011 11 Constructing multiple structural models • gOcad FaultMod – – – – Models fault surfaces as isovalues on continuous function Define dip and strike distributions, number of faults FaultMod generates 3D structural models Extract hierarchy from simulated model • gOcad SKUA – Define common volume of interest for all models – Make flow simulation grid • Layer‐cake model for petrophysical properties • Faults: transmissibility multiplier of 0.1 SCRF 2011 12 3DSL Simulation Range of Total Oil Production • Sensitivity of total oil production Fault strike angles SCRF 2011 13 3DSL Simulation • Sensitivity without Interactions SCRF 2011 14 3DSL Simulation • Sensitivity with Interaction SCRF 2011 15 DKM for Structural Models Total Oil Production MDS Simulation Time Step SCRF 2011 16 DKM for Structural Models MDS Clustering Kernel Transformation Selected Models SCRF 2011 17 Estimating Quantiles • Using difference in 3DSL response to select models for Eclipse simulation Total Oil Production True Estimate p90 p50 p10 SCRF 2011 18 Estimating Sensitivity • Eclipse Simulations – Around 8 days for 50 simulations – Not complete sensitivity picture in reality – Sensitivity Analysis using 10 selected models SCRF 2011 19 Reference Sensitivity • Sensitivity using a broader set of 50 models SCRF 2011 20 Exception Case • Horizontal faults • Close‐together wells SCRF 2011 21 Exception Case • Response – 3DSL water saturation in production well block • Multiway Sensitivity – Order of 10‐15 – Not sensitive • Why? – Conduit Effect SCRF 2011 22 Conclusions • Applied Distance Kernel workflow to structural models – Generated multiple structural models – Converted parameters (geometric and placement) to categorical variables • Analysed sensitivity of flow response to parameters • Estimated the uncertainty in flow response of structural models using DKM SCRF 2011 23 Acknowledgements • Nicolas Cherpeau – faultMod • Darryl Fenwick – StreamSim 3DSL SCRF 2011 24