Comparison of flow simulation proxies for model selection in reservoir uncertainty management.

Gaetan Bardy and Pierre Biver and Guillaume Caumon and Philippe Renard and Peter R. King. ( 2014 )
in: Proc. 34th Gocad Meeting, Nancy

Abstract

To study the impact of subsurface uncertainties on oil recovery, it is common to build a large set of models which cover these uncertainties. Despite increase of computational capabilities, as models become more complex, it is not possible to perform full physic flow simulation for all the generated models. This is why stochastic reservoir model sets are often decimated to assess the impact of static uncertainties on dynamic reservoir performance. This contribution will focus on the use of proxy to perform this model reduction. A lot of different proxies have been developed with varying degrees of accuracy so it is difficult to choose the appropriate one according to a particular goal. We present several criteria to compare the quality of proxies in assessing uncertainties about oil recovery. A first criterion will be based on the relation which may exists between the model distances computed on the proxy responses and those compute on the actual flow responses. Another criterion is the speed factor and simplification provided by the proxy as compared to the full physic simulator. These two criteria are very simple and can be applied early on to avoid deploying time consuming proxies which won’t provide accurate information. Last, we suggest to compute confidence intervals around probabilistic reservoir production forecasts computed on a small representative subset of models. This provides some quantification about a possible bias created by a proxy and the remaining uncertainties on oil recovery. We apply this methodology compare widely different proxy responses on a real dataset. This gives us some keys to choose a proxy which is a good compromise between accuracy and easy to handle methodology.

Download / Links

BibTeX Reference

@inproceedings{BardyGM2014,
 abstract = { To study the impact of subsurface uncertainties on oil recovery, it is common to build a large set of models which cover these uncertainties. Despite increase of computational capabilities, as models become more complex, it is not possible to perform full physic flow simulation for all the generated models. This is why stochastic reservoir model sets are often decimated to assess the impact of static uncertainties on dynamic reservoir performance. This contribution will focus on the use of proxy to perform this model reduction. A lot of different proxies have been developed with varying degrees of accuracy so it is difficult to choose the appropriate one according to a particular goal. We present several criteria to compare the quality of proxies in assessing uncertainties about oil recovery. A first criterion will be based on the relation which may exists between the model distances computed on the proxy responses and those compute on the actual flow responses. Another criterion is the speed factor and simplification provided by the proxy as compared to the full physic simulator. These two criteria are very simple and can be applied early on to avoid deploying time consuming proxies which won’t provide accurate information. Last, we suggest to compute confidence intervals around probabilistic reservoir production forecasts computed on a small representative subset of models. This provides some quantification about a possible bias created by a proxy and the remaining uncertainties on oil recovery. We apply this methodology compare widely different proxy responses on a real dataset. This gives us some keys to choose a proxy which is a good compromise between accuracy and easy to handle methodology. },
 author = { Bardy, Gaetan AND Biver, Pierre AND Caumon, Guillaume AND Renard, Philippe AND King, Peter R. },
 booktitle = { Proc. 34th Gocad Meeting, Nancy },
 title = { Comparison of flow simulation proxies for model selection in reservoir uncertainty management. },
 year = { 2014 }
}