Enhancement of dynamic reservoir interpretation by correlating multiple 4D seismic monitors to well behavior

Abstract

We have found that dynamic reservoir interpretation can be enhanced by directly correlating the seismic amplitudes from many repeated 4D seismic monitors to the field production and injection history from wells. This “well2seis” crosscorrelation was achieved by defining a linear relationship between the 4D seismic signals and changes in the cumulative fluid volumes at the wells. We also found that the distribution of the well2seis correlation attribute can reveal key reservoir connectivity features, such as the seal of faults, fluid pathways, and communication between neighboring compartments. It can therefore enhance dynamic reservoir description. Based on this enhanced interpretation, we have developed a workflow to close the loop between 4D seismic and reservoir engineering data. First, the reservoir model was directly updated using quantitative information extracted from multiple surveys, by positioning and placing known barriers or conduits to flow. After this process, a seismic-assisted history matching was applied using the well2seis attribute to honor data from the seismic and engineering domains, while remaining consistent with the fault interpretation. Compared to traditional history matching, that attempts to match individual seismic time-lapse amplitudes and production data, our approach used an attribute that condensed available data to effectively enhance the signal. In addition, the approach was observed to improve the history-matching efficiency as well as model predictability. The proposed methodology was applied to a North Sea-field, the production of which was controlled by fault compartmentalization. It successfully detected the communication pathways and sealing property of key faults that are known to be major factors in influencing reservoir development. After history matching, the desired loops were closed by efficiently updating the reservoir simulation model, and this was indicated by a 90% reduction in the misfit errors and 89% lowering of the corresponding uncertainty bounds.