A Novel Method of 3D Multipoint Geostatistical Inversion Using 2D Training Images

Abstract

The seismic inversion method combined with multipoint geostatistics theory has begun to receive attention, but the acquisition accuracy and calculation efficiency of 3D training image still need more optimization. This paper presents a novel method of 3D multipoint geostatistical inversion based on 2D training images directly. The 2D training image was scanned by the data template to acquire the multipoint statistical probability in 2D direction. The probability fusion method is used to fuse the 2D multipoint probability into 3D multipoint probability. The rock facies types and patterns of the simulated points are obtained by random sampling. On this basis, the elastic parameters are extracted from the statistical rock physics model, and the seismic records are convoluted. Then, the synthetic records and the actual records were compared under a given threshold. If the error exceeds the given threshold, the iterative adaptive spatial sampling method will be used to repeat the process above-mentioned, so as to ensure that the error is below the threshold. Because the 2D training image is easy to obtain and evaluate, the demand problem of 3D training image is solved. The 2D training image scanning, probability storage and access are more convenient, and the adaptive spatial sampling method is more efficient than the reject sampling, so as to ensure the operation efficiency. The model from the Stanford Center for Reservoir Forecasting is selected to test the effectiveness of this newly designed method.

Definition of Spatial Copula Based Dependence Using a Family of Non‐Gaussian Spatial Random Fields

“…According to Journel (2003), the major source of uncertainty in MPS is the choice of the training image and not the fluctuations between multiple realizations. Further 3 dimensional training images are not available thus 3D simulations require different treatment for example, a combination of 2D images (Huang et al, 2022). extensive summary of the methods and application of random fields, therefore the reader interested in this topic should refer to that publication.…”

“…According to Journel (2003), the major source of uncertainty in MPS is the choice of the training image and not the fluctuations between multiple realizations. Further 3 dimensional training images are not available thus 3D simulations require different treatment for example, a combination of 2D images (Huang et al, 2022).Recently in Papalexiou et al (2021) the authors published interesting random field constructions related to complex natural processes. Their simulations are based on Gaussian fields and subsequent geometrical transformations which are used to simulate complex patterns and motion, for example, representing advection of rainfall,…”