Summary Shale reservoirs are characterized by complex mineral compositions and pore structures, strong heterogeneity, low porosity, and ultralow permeability. Permeability serves as a critical parameter for assessing fluid flow capacity and mobility. However, traditional permeability prediction methods from well logs are not accurate enough to meet the demands of shale reservoir evaluation. With this study, we propose an intelligent permeability prediction method for shale reservoirs that integrates physics-informed neural networks (PINN) with attention mechanisms. A physics-based permeability model of shale based on pore structure parameters was developed and integrated as a constraint into the PINN framework. Feature attention and self-attention mechanisms were incorporated to achieve feature screening and correlation modeling. A multiloss function combined with Bayesian optimization (BO) strategy was used to enable efficient model training on small-sample data sets. Based on the four-component porosity division from nuclear magnetic resonance (NMR) logging T2 distribution, a well-log interpretation method for pore structure parameters was established. The case study in the WXN Sag shows that the predicted permeability of the shale oil reservoir is in good agreement with core measurements, the proposed PINN model achieves a mean logarithmic error (MLE) of 0.35, representing a 47.0% improvement over the traditional physical model (MLE = 0.66) and a 37.5% improvement over the pure data-driven neural network (MLE = 0.56). This study provides a novel technical approach for permeability evaluation in shale oil reservoirs.
Bodong et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: