Full text loading...
-
History Matching Methodology Using an Optimal Neural Network Proxy and a Global Optimization Method
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, Third EAGE Integrated Reservoir Modelling Conference, Dec 2016, Volume 2016, p.1 - 5
Abstract
History Matching is a key step in all reservoir engineering study. This inverse problem is known to provide not unique answers. To find out the optimal global solution, a global optimizer is required as gradient methods fails in finding the global solution often being trapped in local minimum. But using a global optimizer, thousands of reservoir simulation runs are required which is unpractical. It is here that comes in our innovation approach: replace the reservoir simulator by a proxy. This proxy is build using an Artificial Neural Network. It is the most efficient approach we found due to the no linear behaviour of the output again the parameters. Of course, several ANN are possible (number of layer and number of neurons per layer…). A methodology to find out the most predictive ANN is proposed. Coming back to the global optimizer, the paper will emphasize the advantages of using the covariance matrix adaptation evolution strategy (CMA-ES) through practical cases.