Solution Error Models: A New Approach for Coarse Grid History Matching

摘要

This paper presents a new technique - solution error models - and applies it to the problem of history matching using simple coarse grid models. The solution error model technique is based on building a statistical model that describes the errors introduced in the simplified solution. The difference between observations and the coarse grid solution is measured by the "goodness of fit" (or misfit) term. The results show that a standard least squares approach to quantifying the misfit function yields significantly biased estimates of parameters, if the underlying model is biased. The bias can be almost completely corrected through use of the solution error model. The solution error model also introduces the use of time dependent covariance, instead of the usual single valued variance. This improves confidence in estimates of reservoir parameters.