A key issue of waterflooding is to locate the injectors so as to optimize the production from producers. To resolve the issue in the face of geologically variation, we need to understand how the production is influenced by the injection, in other words, the connectivity between the producers and injectors. It would be very useful for us to build a reservoir scale network based on the connectivity and perform reservoir analysis using such a network representation, without reference to a reservoir simulation model that is obliged to make assumptions about the geology.
The work shown in this paper was organized into three parts. First, a modified Pearson's correlation coefficient was proposed as a new connectivity definition based on the correlation between injection and production histories. Next, machine learning based multiwell testing was applied to learn the correlation between producer rate features and injector pressures, which provided a validation to the connectivity estimated by modified Pearson's correlation coefficient. Last, a connectivity based network was proposed as an abstract reservoir representation. A series of network analyses can be performed on the established network, e.g. reservoir compartmentalization by detecting the strongly connected clusters in the network.
The proposed methodologies were applied on both synthetic and real data sets. The connectivity was shown to be able to capture the injectors' influence on producers, and was verified by comparing with the true reservoir model and results from tracer tests. The target injectors with high influence on production were identified easily as the ones connected to multiple producers, information that can be used to guide a better waterflooding operation.