Real-time Post-Processing Method to Enhance Multiple-Point Statistics Simulation

Multiple-point statistics (MPS) simulation has gained increasing application to reservoir modeling as an effective facies modeling tool, providing a way to simulate complex geological feature by honoring well and seismic data. The geological pattern we desire to reproduce is provided as a training image. It is often recommended to use a large enough training image to accurately reproduce the desired pattern; otherwise, MPS simulation tends to fail to honor multiple-point statistical information, resulting in anomalies in the simulated realizations (e.g. disconnection of channel sands). However, the use of such a large training image can be extremely computational time demanding. <br>This paper proposes a new method to improve the modeling accuracy of MPS simulation. Our method is designed in the context of multi-grid sequential simulation. The idea is to detect the failure of conditioning to the given training image during the sequential simulation, and as soon as detecting the failure, repair the previously simulated facies pattern by maximizing the conditioning to the training image. This processing is implemented simultaneously with the sequential simulation and applied only in the early stage of multi-scale simulation, reducing the computational cost but yet obtaining effective improvement of pattern reproduction accuracy. <br>