In recent studies, dynamic capillary pressure has shown significant impacts on the flow behaviors in porous media under transient flow condition. However, the effect of dynamic capillary pressure effect on tight sandstone is still not very clear. Since lattice Boltzmann method (LBM) is a very promising and widely used method in analyzing flow behaviors, therefore, a two-phase D3Q27 LBM model is adopted in this paper to simulate the flow behaviors and analyze the dynamic capillary pressure effect in tight sandstone. Moreover, a new pore segmentation method for tight sandstone base on U-net deep learning model is implemented in this study to improve the pore boundary qualities of pore space, which is crucial for two-phase LBM simulation of tight sandstone. A total of 3800 3D sub-volume data sets extracted from computed tomography data of 19 tight sandstone samples are selected as ground truth data to train the network and segment the pore space afterward. The simulation results based on the segmented digital rock model, show that nonwetting phase fluid prefer the path with lower dynamic capillary pressure in the seepage process before breaking through the porous model. Furthermore, the increase of injection rate causes the saturation changes more quickly, injection rate also shows apparent positive correlation relationship with capillary pressure, which implies that dynamic capillary pressure effect also exists in tight sandstone, and LBM based two-phase flow simulation could be used to quantitatively analyze such effect in tight sandstone.

Permeability is one of the most important petrophysical properties of shale reservoirs, controlling the fluid flow from the shale matrix to artificial fracture networks, the production and ultimate recovery of shale oil/gas. Various methods have been used to measure this parameter in shales, but no method effectively estimates the permeability of all well intervals due to the complex and heterogeneous pore throat structure of shale. A hydraulic flow unit (HFU) is a correlatable and mappable zone within a reservoir, which is used to subdivide a reservoir into distinct layers based on hydraulic flow properties. From these units, correlations between permeability and porosity can be established. In this study, HFUs were identified and combined with a back propagation neural network to predict the permeability of shale reservoirs in the Dongying Depression, Bohai Bay Basin, China. Well data from three locations were used and subdivided into modeling and validation datasets. The modeling dataset was applied to identify HFUs in the study reservoirs and to train the back propagation neural network models to predict values of porosity and flow zone indicator. Next, a permeability prediction method was established, and its generalization capability was evaluated using the validation dataset. The results identified five HFUs in the shale reservoirs within the Dongying Depression. The correlation between porosity and permeability in each HFU is generally greater than the correlation between the two same variables in the overall core data. The permeability estimation method established in this study effectively and accurately predicts the permeability of shale reservoirs in both cored and un-cored wells. Predicted permeability curves effectively reveal favorable shale oil/gas seepage layers and thus are useful for the exploration and the development of hydrocarbon resources in the Dongying Depression.