Research On Tight Oil Seepage Characteristics Based On Digital Rock

Tight oil is an important petroleum resource,and the pore structure of tight reservoirs is also the focus of the field of oil and gas research.Digital core technology has been gradually applied to the field of tight reservoir characterization in recent years because of efficient and accurate.However,due to the randomness of core sampling,the characterization results of digital cores lack representativeness.In order to solve this problem,this paper selects the tight sandstone of Gaotaizi in the Qijia-Longhupao area,and choose the CT image of the tight sandstone in this area as the data set.Combining the data set with deep learning theory,finally,this paper proposed a digital core reconstruction technology based on 3DCGAN-DR network.This method can not only reconstruct digital cores with different types of porosity and permeability characteristics,but also reflect the average status of pore structure and seepage characteristics of a certain type of reservoir as a whole.The specific work includes:According to the experimental results of the samples and the classification and evaluation criteria of tight reservoirs,this paper selects various types of representative samples,and threedimensional,cut and flip the CT images of these samples.The next task is to extract the pore structure of these digital rocks.The predecessors usually use the method of dividing by gray threshold.However,due to some uncontrollable factors in the experiment process,the pore gray threshold of digital rocks is not uniform,resulting in the segmentation result is not accurate.In order to solve this problem,this paper combines the U-net deep learning network to propose a new 3D digital rocks’ pore segmentation method.The U-net network takes into account the gray threshold of the rocks’ pore and the 3D topological structure during the training process.The purpose is to make the segmentation results more accurate.Next,based on the digital rocks’ pore structure,seepage simulation is performed.Among the many seepage simulation methods,this paper selects the lattice Boltzmann(LBM)method because of its high parallel efficiency.In order to further improve the parallel efficiency,this paper develops a GPU-based LBM-GPU seepage simulation program based on the serial D3Q19 model,and through the analysis and calculation of the seepage simulation results,the permeability and other seepage parameters are obtained.Finally,based on the above data,a3DCGAN-DR deep learning network model for digital cores was developed.On the basis of integrating the microscopic pore structure characteristics of conventional reservoirs,type I tight reservoirs,type II tight reservoirs and type III tight reservoirs.Finally,through continuous optimization of the program and training data set,a digital core with representative pore structure and seepage characteristics is reconstructedCombined with the reconstruction results of the digital rock,we finally get the following conclusions:(1)When the seepage simulation calculation of the same model is performed,due to the parallel processing of the LBM-GPU distribution function and collision terms,seepage simulation speed has been significantly improved.(2)When using U-net to extract the pore structure of digital rocks,during the training process,the network stores the local gray-scale changes of pores in the form of network coefficients such as convolution kernels,which enhances the accuracy of digital rock pore segmentation.(3)In the digital rock reconstruction training process,the deep learning network can continuously learn the pore throat distribution rules of different types of reservoirs,moreover,the 3DCGAN-DR model uses convolutionpooling to share parameters and parameter reductions,which can improve computational efficiency.The digital rock model reconstructed by the 3DCGAN-DR deep learning network can not only meet the requirements of pore and permeability parameters of different types of reservoirs,but also reflect the characteristics of different types of reservoirs in the overall pore structure and seepage characteristics,and the reconstruction results have strong representation.