Abstract
Three-dimensional reservoir models are an essential decision-making tool for geoscientists to understand the by-passed oil in multilayer fluvial systems in San Jorge Basin. The dedicated reservoir teams spend significant amounts of time to build three-dimensional models to decrease risk in the implementation of chemical EOR projects, particularly in the assessment of reservoir connectivity. Traditional 3D static reservoir modelling requires important effort to construct the structural models, interwell correlation and facies modeling. The objective of this work is to present the implementation of a computational algorithm that generates three-dimensional reservoir models. We successfully tested its capability to adequately represent the structural model, interwell correlation, and facies propagation from the well log data without any previous interpretation.
This work is based on the application of label propagation/ label spreading on the graph network using the three-dimensional percolation theory with considerations of the phase transition of the system. This methodology is powerful, since it does not consider the distance between the input data, but rather its distance in a graph network.
The results of the novel unsupervised algorithm show a good match between the existing 3D geological models made by experienced geologists and the model made by the computational algorithm implemented. Finally, we performed several sensitivity analyses to better understand the tuning parameters.
This novel approach for 3D reservoir modelling is characterized by the percolation in complex networks. One of the keys is that the generated 3D reservoir models have an unbiased geological interpretation and have the capability to quickly downscale/upscale the 3D grid with tuning parameters.