BC10 O-North multi-attributes seismic inversion for saturation changes using ML

Abstract

Time lapse (4D) seismic is widely deployed in offshore operations to monitor reservoir changes, especially those utilizing improved oil recovery methods including water flooding. But its value for well and reservoir management (WRM) is not fully realized due to the long cycle time required transferring the 4D seismic into reservoir property changes, such as water/gas saturation changes and pressure changes. To shorten the cycle time, we designed a machine learning based agile workflow to invert for reservoir property changes directly from a variety of 4D seismic attribute maps and reservoir static property maps. We use reservoir simulation, calibrated rock physics model and the real seismic data well tied wavelet to generate synthetic 4D seismic data and their attribute maps. As the reservoir property changes are related not only to the 4D attribute maps, but also to the reservoir static parameters, each training dataset is composed of multi-4D attributes and static parameters from reservoir simulation as input, and the reservoir property changes from the reservoir simulator as output. A shallow neural net (NNet) is applied to train on the synthetic datasets demonstrating satisfactory convergence within a few minutes, allowing prediction of property changes from the real 4D seismic attribute maps with seconds of time.This ML based agile workflow is applied to BC10 O-North field to predict water and gas saturation changes (dSw and dSg) simultaneously. The turnaround time can be reduced from weeks to days, allowing early engagements with reservoir engineers to enhance integration and removing a deterrent to the acquisition of frequent 4D surveys.