A MPS-based novel method of reconstructing 3D reservoir models from 2D images using seismic constraints

Abstract

While multi-point geostatistics (MPS) be used to model complex underground reservoirs, this method relies heavily on high-quality three-dimensional (3D) training images that may be difficult to acquire in real field studies. Furthermore, the original MPS technique does not include seismic data, which may noticeably improve underground reservoir predictions. Our proposed model, Se3DRCS, is a novel method to reconstruct 3D geological models from two-dimensional (2D) images and seismic information. In Se3DRCS, we are able to perform MPS-based modeling using 2D training images, rather than 3D images. These 2D images are produced by analyzing well profiles and sedimentary facies planes. After using the Se3DRCS method to obtain an initial geological model, we then generate elastic parameters and synthetic seismograms via direct sampling and convolution with the extracted seismic wavelets, respectively. By comparing the observed and synthetic seismic records, we refine the simulation results using the adaptive sampling method. The initial geological model is then updated with conditional data after each subsequent simulation iteration. When the error between the observed and synthetic seismograms falls below a reasonable threshold value, we have arrived at the best-fitting geological model. Our results indicate that the Se3DRCS method is capable of predicting the data elicited from wells with an accuracy of up to 82.88%, providing constraints on the distribution of branch channels in a delta reservoir, and reproducing local complex non-stationary geological features.