Sorting Reservoir Models According to Flow Criteria: A Methodology, Using Fast Marching Methods and Multi-Dimensional Scaling

In this paper, we will present a two-step methodology to efficiently select a subset of models representative of the variety of dynamic behaviors for a large set of models. We will use different proxies to quickly approximate the flow simulator responses. A suggested first proxy is based on volumes connected to the wells. Another suggested proxy is based on the Fast Marching algorithm. It computes a front propagation from the wells to the rest of the reservoir.The methodology also uses Multi-Dimensional Scaling and kernel clustering to select a small number of models for real flow simulator. Finally, we compare Q10, Q50 and Q90 for the subset and for the entire set of models in order to validate the proxy and the model selection.