Spatial Zonation of the Field by One-Dimensional Signal Clustering in Order to Improve Quality of History Matching

The main goal of this work is development of methodology for spatial zonation of an oilfield area and modification of these zones honoring the well data in order to achieve more accurate history matching. The production data consistent history matching zonation criteria will be proposed in order to reproduce the history data, thus increasing model prediction reliability. The history matching results of zoned and non-zoned reservoir models are compared. Unsupervised machine learning was used to cluster production profiles. The dynamic time warping (DTW) algorithm was used as a metric for clustering. The assisted history matching was used in this research to tune the models, namely particle swarm optimization algorithm. in this research, a technique with its detailed workflow has been developed and applied to perform AHM. This technique was applied and validated on a synthetic case and good matching results were obtained. There is an excellent agreement in cluster location using production log data. The analysis of simulated data shows that the history matching quality of FOPR and FWPR of zoned model higher by 9% and 34% respectively in comparison with non-zoned model.