An Innovative Model Preconditioning Approach for Fast and Reliable History Matching

Abstract

Abstract History matching is the process of calibrating a geological model to match the historical production, pressure and other reservoir characterization data. The methods developed in this work aims to address the history matching problem through an a priori integration of data-derived features into the static modeling phase, using a process we have termed pre-conditioned modelling. The methodology combines data mining techniques with intelligent visualizations to swiftly extract features from historical pressure and core permeability data, and integrate these features a priori into the static model. An anomaly detection algorithm was developed and used to detect data outliers and exclude anomalies from pressure data. A pattern recognition algorithm was developed, used for pressure cluster differentiation, and then to create reservoir regions containing wells within similar time-lapse pressure cluster, and the created regions were used as containers to guide permeability modeling and for applying sub-global permeability updates during history matching. The reservoir's internal architecture was derived from core permeability using a data mining algorithm we developed and termed fixed-window averaging. The intra-reservoir architecture was used to control the static model's vertical heterogeneity trends. This innovative approach to rapidly integrate historical pressure and core permeability dataset features into the static modeling process, resulted in a model that required no significant need for the traditional history-match process because the do-nothing match was excellent. A data analytics algorithm termed spatial Pythagorean search was developed and used to detect the causes of well productivity problems. An advanced visualization interface was developed which allows both time-based (e.g., water-cut, datum pressure) and depth-based (e.g., FTS, PNL, mobility, PTA) parameters to be displayed, permitting fully integrated decision making regarding necessary modifications. Leveraging holistic data visualization and data mining techniques, our methodology helps establish a reliable reservoir model by prior integration of data-derived features into the static model, significantly reducing the time required for history matching.