In reservoir simulation, history matching has long been performed in an empirical way. Applying inverse problem methods to reservoir simulation requires the calculation of gradients of observations at producing wells with respect to reservoir characteristics, which leads to a local linearisation of the full non linear reservoir simulation problem.

This paper describes a rigorous analytical method for the computation of gradients. The gradient calculation has been implemented in the fully implicit, 3-phase, 3-dimensional reservoir simulator, SCORE, in a way which leads to significantly reduced computation time compared to other methods.

The second part of the paper reports on a preliminary attempt at computer aided history matching, based upon a meta simulator which uses the calculated values of the gradients and a minimisation procedure together with several extrapolation laws. Two applications of the method are presented. They serve to demonstrate that the experience of the reservoir engineer cannot be replaced by such a method, used in an automatic way, because the system cannot decide which parameters should be changed. However, once an appropriate choice had been made, the meta-simulator provided an accurate estimate of the modified parameters which had to be used in order to simulate precisely the observed results.

Application of this method increased the accuracy of the history match whilst significantly reducing both the number of computer runs and the total processing time required compared to other methods.

You can access this article if you purchase or spend a download.