History matching studies can involve adjusting a high number of reservoir uncertain parameters. Sensitivity analysis can help reservoir engineers focusing on the most influential parameters to prioritize the effort thus reducing sensibly the history matching time. However, due to nonlinear and interactions effects and depending on the method used, this exercise can be misleading or in some cases very time consuming. In this work, several statistical nonparametric methods from the most recent to the most classical have been implemented and tested on some typical oil reservoir applications. Different tests are made using different problems of different dimension. Most of the nonparametric methods investigated are based on response surfaces such as kriging, polynomial or smoothing splines responses with different types of penalization (LASSO, COSSO,...), and some others are issued from the recent statistical literature. The response surface methods were tested using space filling experimental design such as maximin latin hypercube. In the comparisons the computational cpu time for each method is also reported because even if these are generally negligible respect to large simulation time, they can be considerable for large dimensional input problems. To assess the quality of the methods several criteria have been used such as the prediction accuracy of the response surfaces, but also other criteria such as the number of times the method fails to detect an influential parameter (type 1 errors) or the number of times it indicates a non influential parameter as influential (type 2 errors). Numerical tests were made on different outputs (objective functions) of realistic synthetic reservoir models.


Article metrics loading...

Loading full text...

Full text loading...

This is a required field
Please enter a valid email address
Approval was a Success
Invalid data
An Error Occurred
Approval was partially successful, following selected items could not be processed due to error