Ann Proxy Model for Seismic Forward Modelling; a 4D Seismic History Matching Case Study

Time and computation issues have always been inseparable issues for studies such as 4D seismic history matching (4D SHM) as they need numerous reservoir flow simulation runs and seismic forward modelling executions. One way to mitigate these issues is replacing the conventional tools and algorithms with faster and cheaper proxy models. In line with this research trend, we trained an artificial neural network (ANN) as a proxy for conventional seismic forward modelling. This proxy directly transforms the maps of changes in pressure and saturation to 4D seismic quadrature. After training and validating the proxy model, we used it for 4D SHM of a real filed in the North Sea. The results showed the proxy’s capability to speed-up the seismic forward modelling part of the 4D SHM by a factor of 8, while still maintaining accuracy levels comparable to those achieved through traditional forward modelling.