Identification of Key Wells for Waterflood Optimization Considering Geologic Uncertainty

Abstract

AbstractStreamline-based rate optimization is an iterative process, requiring several simulation runs. Though efficient for a single realization, it can be prohibitively expensive while considering geologic uncertainty involving large number of realizations. Moreover, the optimal schedule based on one individual geologic model may not necessarily result in favorable outcomes for the real field due to the geologic inconsistencies between the real field and the model. This paper proposes a workflow that integrates unsupervised machine learning and streamline techniques to select representative geologic realizations based on their flow features. The proposed approach generates a distribution of optimal rates for each well, and this in turn is used to identify key wells for which we may advise rate change with high certainty.Given a set of historical production and injection data, firstly, an ensemble of Nreal history-matched geologic realizations is generated using ensemble-smoother with multiple data assimilation (ESMDA). Subsequently, the streamline time-of-flight (TOF) and principal component analysis (PCA) are used to extract the flow feature of all realizations, based on which k-means clustering algorithm generates a subset of Nclust realizations representing the whole ensemble. The rate optimization is performed on each of the representative realizations using a streamline-based rate optimization algorithm that seeks to maximize the oil production during the optimization period. The distribution of optimal schedules obtained by optimizing the representative realizations is shown to be in high correspondence with that obtained by optimizing the full ensemble. Using the optimal schedule distribution, the key wells are identified, for which rate change is advised with high certainty. The workflow is tested on a synthetic 2D reservoir model as well as a 3D field-scale benchmark reservoir model (SAIGUP model).The novelty of this work is the combination of the streamline-extracted flow features and unsupervised machine learning methods to formulate an efficient workflow for uncertainty analysis of optimal well schedules. The proposed approach ensures quality and rigor of uncertainty analysis with significantly reduced number of geologic realizations and thus, makes the approach well-suited for large scale field applications.