S-GEMS-UQ: An Uncertainty Quantification Toolkit for SGEMS Annual Meeting 2013 Stanford Center for Reservoir Forecasting S-GEMS-UQ: An Uncertainty Quantification Toolkit for SGEMS Lewis Li, Alex Boucher, & Jef Caers

Geostatistical Modeling of Uncertainty Geostatistical Realizations Forward model Analysis of Responses Spatial Model parameters Forward model parameters This figure represented a generalized workflow for uncertainty quantification. We first have field data that can be for instance, seismic data or well logs. From this data, we can abstract parameters that are used for spatial modeling. These are then passed into some geostatisical algorithm; be it variogram based or mult point statistics. The result realizations are then put through a forward model, (this could be either a full flow simulator or some proxy response). We then would analyze the response data to quantify uncertainty or do sensitivity analysis on the model parameters. Field Data SCRF 2013

Software Challenges Nearly every step of workflow is performed in separate programs Needs extensive modification for different scenarios Linking to external programs Post processing is cumbersome due to parameters, models, responses, etc. being fragmented across different programs Large size of data sets Often need to generate hundreds or thousands of models, system needs to be responsive enough to accommodate Ease of visualizing No user friendly and intuitive interfaces SCRF 2013

SGEMS-UQ Goals Compatible: Reduce/eliminate ad-hoc linking of codes Flexible: Can be adapted regardless of: Which geostatistical algorithms are generated Which proxy models are used Which simulation parameters are specified How distances are computed How users which to interpret results and perform sensitivity analysis Extensible: Framework for data mining Scalable: Be scalable for large datasets User Friendly: Provide intuitive GUI SCRF 2013

Goal: Compatibility Link with SGEMS 3.0 Develop as a plugin Link with external flow simulators Results are usually comma separated values Use XML (Extensible Markup Language) SCRF 2013

Response XML Responses can be of a variety of formats Vector Scalar Time series Probability distribution Defined a specified XML format with appropriate tags Program detects response type and selects appropriate data structure and links to corresponding object Provide Python script to convert exported files into this format

Goal: Extensible Modular Design XML combined with SQL database for storing parameters Allows for a data mining framework

Data Manipulation Language SELECT ... FROM ... WHERE ... INSERT INTO ... VALUES ... UPDATE ... SET ... WHERE ... DELETE FROM ... WHERE ... Ex: SELECT * FROM SGSIM WHERE TrainingImage = TI1 All we need to know is which algorithm each realization was generated with Use a hash table to store this in memory

Data Mining SQL SQL Parameter A Parameter B Parameter C … Parameter A

Distance Based Uncertainty Quantification

SGeMS-UQ Workflow Access SGeMS main program workspace Read, store and manage simulation parameters Read proxy responses from third party simulators Display responses Compute MDS Perform k-medoid clustering Export clustering results Data mine the parameters and perform sensitivity analysis

SCRF 2011