A Computationally Efficient Gauss-Newton Method for Automatic History Matching

Abstract

ABSTRACT Methods for automatic history matching that have been previously published have either required excessive computational time or had linear convergence characteristics. This paper presents a modified nonlinear regression method, the Gauss-Newton method, that has quadratic convergence properties and yet is extremely computational efficient. The modifications to the method are described in full detail. The method has been implemented in a three-dimensional three-phase simulator. The simulator was applied to a case study using the Seventh SPE Comparative Problem. It was shown to be capable of estimating multiple parameters such as permeability and porosity. Selected matches from actual wells in a field study were also presented. The results of an actual well test were also matched. It was demonstrated that the automatic history matching simulator could identify the reservoir parameters that had significant effect on the observed data.