Tins paper discusses the use of an automatic reservoir history matching process using the method of gradients on a work station. We illustrate the method's robustness and performance using a simple field example.

Traditional gradient methods have utilised perturbation methods to compute the gradients which require n + 1 simulations for n parameters and could consequently be quite expensive. To overcome the expense of making several runs of the simulator we have implemented into our simulator a method which can compute derivatives of many parameters simultaneously. We shall refer to this simulator as GRADSIM for the rest of the article. We show by numerical experiments that the additional cost of computing derivatives is highest for the first derivative, after which the cost rises by about 8 percent for each additional derivative.

We then use the gradients to minimise a so-called objective function which is a measure of the difference between the historical and simulated reservoir data. This is accomplished by passing the gradients to a constrained minimisation algorithm which calculates the changes required in various user-defined reservoir parameters to achieve a better history match. The reservoir description is then updated and the reservoir simulation and gradient calculations are repeated. The process continues in an iterative manner until the best match has been achieved.

This simple example of the application of the method of gradients shows that considerable savings in time could be obtained by the engineer for both history matching and proper parameter identification.

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