Coupling Sequential-self Calibration and Genetic Algorithms to Integrate Production Data in Geostatistical Reservoir Modeling

The sequential-self calibration (SSC) method is a geostatistical-based inverse technique that allows fast integration of dynamic production data into geostatistical models. In this paper, we replace the gradient-based optimization in SSC by genetic algorithms (GA). GA, without requiring sensitivity, searches for global minimum. Although GA is computationally intensive, it provides significant flexibility to study parameters whose sensitivities are difficult to compute, e.g., master point locations. A steady-state GA is implemented under the SSC framework for searching the optimal master point locations, as well as the associated optimal perturbations that match the observed pressure, water cut and saturation data. We demonstrate that GA is easy to implement and results are robust. We examine different approaches of selecting master point locations including fixed, stratified random, and purely random methods. Results from this study demonstrate that there are not clear preferential master point locations that are best suited for matching production data for the given well pattern and for the given initial model. This is consistent with the early findings that master point locations can be randomly selected with the stratified random method yielding the best results due to its flexibility and good control for the overall model.