A Classification Method for Ultra-Deep Reservoirs Based on Multidimensional Well-Logging Information Fusion

Ultra-deep clastic reservoirs in the NY block of the Junggar Basin, China, are generally characterized by ultra-low porosity, extremely low permeability, the development of micro-fractures, and weak fluid-response signals. In addition, the lack of prior information such as rock-physics data limits the applicability of traditional reservoir classification methods based on logging-derived petrophysical parameters. To address these challenges, this study proposes a reservoir classification method for ultra-deep formations based on multidimensional well-logging feature fusion and multivariate statistical analysis. First, multiple identification factors are constructed using conventional logs, array acoustic logs, and nuclear magnetic resonance (NMR) logs. Second, principal component analysis is applied to reduce the dimensionality of the identification factors and extract principal components that reflect overall reservoir characteristics, fluid properties, and invasion features. Finally, linear discriminant analysis is employed to establish a reservoir classification function using well-testing results as labels. Field application demonstrates that the proposed method can effectively classify ultra-deep reservoirs. Reservoir classification was conducted for 10 wells in the study block, covering a total of 16 tested intervals, achieving an identification accuracy of 81%, which indicates that the method has strong practical applicability and good potential for wider application.