Automated Workflow for Unstructured Grid Reservoir Simulation

Abstract

Abstract Machine learning based intelligent automation is developed by extending a prior workflow of unstructured grid reservoir simulation (Ding et al. 2015; Fung et al. 2014, 2013). Reservoir heterogeneity, geological internal boundary features and well geometry complexity are being taken into account to automatically detect well zone and focusing reservoir area by calculating the region-of-interests in the model and define cell spacing for grid coarsening and refinement in the reservoir. Automated workflow is demonstrated by using an unstructured grid reservoir simulation example. Previously, an unstructured grid reservoir simulation workflow is introduced (Ding et al. 2015; Fung et al. 2014, 2013). One major component of the workflow is the near-well unstructured grid modeling framework which consists of a 2.5D unstructured PEBI grid engine and its input criteria, such as the locations of reservoir where grid coarsening and refinement are being applied in, along with respective cell spacing being allocated. The creation of such regions of interest and selection of cell spacing involve user's manual interaction, which is user-experience dependent and not intended to serve as a long term solution in the simulation workflow. This paper enhances this process by automatically detecting the target area in the reservoir by computing the convex hull of the well dataset or modified convex hull if concave exists in the dataset. The convex hull is used as the basis for reservoir polygon with defined cell spacing computed by cell density control scheme. The automation considers the heterogeneity and complexity of the reservoir, such as geological internal boundaries and complicated well geometry. The targeted locations in the reservoir cover necessary area for grid refinement with high density grids to capture the accurate flow dynamics near the well, whereas unimportant area in the reservoir are detected for being coarsened to avoid extreme large model size and long simulation runtime. The result of this work enhances the unstructured grid modeling process by automatically computing the local reservoir areas for grid coarsening and refinement with respective grid density on the multi-level hierarchical grids, it avoids user's manual interaction, which is neither efficient nor user friendly. The automated workflow improves unstructured gridding efficiency and enhances user's simulation experience. Utilizing emerging technology breakthrough such as machine learning is important towards a successful era of the Fourth Industrial Revolution (4IR), this work of workflow automation is an example of using machine learning for enhancing problem solving in reservoir simulation.