太阳集团网tyc9728(百度)有限公司

The development of oil and gas is constrained by difficulties in dynamically characterizing pore structures. Traditional methods inadequately represent the complex interactions between mineral dissolution, precipitation, and fluid flow. This study addresses these gaps by introducing a Transformer U-Neural Network (TransUNet) for computed tomography (CT) image segmentation. The integrated workflow combines conventional CT (Resolution of 5.4 μm) and synchrotron radiation CT (Resolution of 0.8 μm) for dynamic flooding, imaging, segmentation, and precise 3D pore network extraction, overcoming resolution limits. TransUNet's strong global attention and feature extraction reduce overfitting and deliver high-accuracy segmentation of minerals, pores, and argillaceous microporous networks (AMN), achieving 74.92% intersection over union (IoU) for AMN. A porosity correction method improves conventional CT porosity accuracy to 94% of gas-measured values. Alkaline flooding experiments reveal: (1) initial clay swelling reduces small pore size by ∼50% as alkaline ions destabilize clay; (2) mineral dissolution, such as dolomite, creates secondary pores, increasing 80 μm pores by 1.8 times; (3) silicate dissolution increases porosity and leads to a 93.7% rise in permeability. Clay reorganization enhances the AMN by 46.1%. The pore size distribution shifts to log-normal at steady state, and throat connectivity improves flow capacity. This work pioneers Transformer-based CT image segmentation, introduces cross-resolution prediction, and clarifies pore regulation by mineral phase changes, establishing a new paradigm for chemical flooding in sandstone reservoirs.