A Method for Evaluating Reservoir Permeability Based on Machine Learning Flow Unit Index

Abstract The H formation of the Y gas field in the X depression belongs to a low-permeability tight sandstone reservoir affected by sedimentation, diagenesis, and cementation. The lithology and pore structure of the target layer are complex, with strong physical heterogeneity and complex pore-permeability relationships. Conventional core pore permeability regression and nuclear magnetic resonance software-defined radio methods do not satisfy the requirements for precise evaluation in terms of permeability calculation accuracy. Based on the principle of the flow zone index (FZI) method, this study analyzed the influence of pore structure on permeability and extracted three pore structure characterization parameters, namely, the maximum pore throat radius (Rmax), displacement pressure (Pd), and average throat radius (R), from the mercury injection capillary pressure curve. The relationship between the FZI and pore structure is clarified. Therefore, the FZI in this area can characterize the permeability differences within different flow units. Based on the flow unit theory, a refined evaluation model for three types of reservoirs was established in the study area. By analyzing the response characteristics and correlation of conventional logging curves using machine learning, three optimization combination curves were selected, and a multiparameter fitting equation for the FZI was established, which was applied to predict the permeability of new wells. The results showed that the calculated permeability was highly consistent with the core analysis results, thereby providing a theoretical basis for the precise evaluation of low-permeability tight reservoirs.