Research On The Reconstruction Of Porous Media Based On Multiple-point Geostatistics

The mechanism of fluid flow in porous media is widely involved in many engineering fields.Darcy's law,based on the macroscopic statistical theories,is of great importance in the mechanism of fluid flow in porous media.With the development of mechanism of fluid flow in porous media,many researchers found that a large amount of phenomena that did not according with Darcy's law.Then some corrections were made to them,which failed to reflect the rule of transportation in non-Darcy situation because those corrections also relied on the macroscopic statistical hypotheses.The object studied in the mechanism of fluid flow in porous media is the rules of fluid flow,which are not only dependent on the properties of fluid,but also associated with the characteristics of porous media.Fluid properties are determined by the transportation characteristics of porous media,which are also dependent on the geometrical and topological structure of pore space.Therefore,it is very important to describe the pore space and connectivity for the study of micro-scale mechanism of fluid flow in porous media.It is impossible to use only several parameters to describe the porous media due to their high irregularity and complicated topology.Therefore,extracting and copying structural features from the real pore space can reconstruct better pore space similar to the real condition,which is also the key problem for the study of micro-scale mechanism of fluid flow in porous media.The 2D and 3D data of porous media were respectively obtained by scanning electron microscopy microtomography and synchrotron microtomography. Reconstruction of porous media was made based on MPS(multiple-point geostatistics) and those data taken as training images.The detailed research is as follows:1.A 2D reconstruction method of porous media based on images from scanning electron microscopy microtomography and MPS is proposed.2D data scanned by scanning electron microscopy microtomography are used as a training image which will be scanned by data templates.A training image is the numerical representation of a prior geological model that contains the patterns believed to exist in realistic porous media under study.By reproducing high-order statistics,MPS can capture complex features from the training image and regenerate them in reconstructed images.The reconstructed 2D image has the similar characteristics with the real porous media by comparing their variogram curves in the X and Y directons. 2.A 3D reconstruction method of porous media based on 2D images and MPS is proposed.The information in the Z direction should be added if 3D structure of pore space is reconstructed based on only 2D images.Sample points drawn from the training image are used as conditional data.MPS regenerate the next layer of original training image.New reconstructed images are used as new training images repeatedly to reconstruct their next layers.Also,sample points should be extracted from each new training image.At last,a group of 2D pore space images are obtained.Stack these images sequentially,and then the 3D pore structure can be achieved.3.The reconstructed results from MPS and two-point geostatistics are compared.Volume data of porous media,obtained from synchrotron microtomography devices,are used as training image.Patterns existing in the training image are extracted and "copied" to the reconstructed region. These reconstructed results of porous media from MPS have similar structural characteristics with the real volume data.Compared with the results reconstructed by two-point geostatistics,the results from MPS are better.4.Servosystem can control the target distribution well in reconstructed images. Although the reconstructed results of porous media are stochastic,they still have similar structural characteristics and close porosity.5.A method using multiple grids in reconstruction is presented.During reconstruction,the data search neighborhood should not be taken too small; otherwise,large-scale structure of the training image cannot be reproduced. On the other hand,if the search neighborhood is too large,the associated data template will include a large number of grid nodes,which will increase the CPU-time and raise the memory demand.One solution to capture large-scale structure while considering a data template with a reasonably small number of grid nodes is provided by the multiple-grid method. Experimental results prove the method is practical.6.A method integrating soft data with hard data in porous reconstruction is proposed.In many fields,there are two types of data:hard data and soft data. Soft data typically provide an extensive coverage of the field under study although with low resolution.Sometimes,the hard data we can obtain are quite little,but soft data are abundant.It is necessary to condition the reconstructed models to all these different types of data to improve the accuracy of reconstructed porous media,which will be better than the unconditional reconstruction and the reconstruction using hard data only,but the cost of CPU and memory won't increase largely.7.A method to reconstruct the continuous variables in porous media is proposed.Filtersim,based on filters,can reconstruct the continuous categories in porous media,such as porosity and permeability.Filters can reduce the dimensions of training images to form "filter score" space,which enhances the efficiency of reconstruction."Filter score" space is partitioned to classify the patterns existing in training images and forms a "database" of patterns.The closet pattern will be found by comparing the patterns in that "database" and current data event.Then this pattern will be patched to the nodes to be simulated.The porosity simulation is made by using filtersim and matches the training image well.