We present a new algorithm for automatically history matching production or well test data. The algorithm is designed to be robust and efficient for history matching problems which have moderate to small numbers of unknown parameters.

The algorithm is based on Marquardt's modification of the Gauss-Newton method for minimization of least-squares functions. The implementation of the algorithm differs significantly from that of Marquardt. The performance of the algorithm is evaluated by history matching actual production data using an analytical dual-porosity model. Its performance is compared to the Gauss-Newton and steepest descent algorithms. We have found that our algorithm is very effective for these problems, but that Gauss-Newton and steepest descent are not suitable.

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