Permeability prediction of considering organic matter distribution based on deep learning

At present, researchers predict permeability through core experiments that require specific experimental conditions and methods, which are difficult time-consuming. Conventional simulation methods for predicting considerable computational resources. Therefore, deep learning can be used as a pore-scale prediction method. In this study, we established workflow directly from images. Considering the mineral properties of nanopore wall shale oil have large influence on flow, dataset with organic distribution was constructed random circles, slip pores considered. From our dataset, found average 32.3% higher than without distribution. to simulate microscopic flow oil, considering differences in pore mechanisms different minerals is necessary. We designed convolutional network adopted structure SE-ResNet, added squeeze-and-excitation (SE) module double-layer residual ResNet18, combined characteristics SE block attention mechanism ResNet effectively obtain information between channels avoid problem gradient disappearance or explosion. Using SE-ResNet apparent images, accuracy test set reached 88.5%. The model had strong generalization ability, could map image permeability, approximately 100 times faster direct simulation.