Prediction and evaluation of fault lateral sealing ability: A new approach to integrating comprehensive controlling factors based on BP neural network

This study presents a novel approach to evaluating fault seal using Back Propagation (BP) neural network method that assesses the impact of several sealing factors and solves for the most permissible based on observed trapped hydrocarbon columns. The methodology is applied to assess trapped oil and gas within the Shuangtaizi structure in China. Subsequently， oil-water units are conducted in the established 3D working area using FAPSeal＿3D software， followed by calculations of fault sealing factors based on various required parameters. The key fault sealing factors were extracted including clay content (CC)， fault throw (FT)， effective normal stress (ENS)， fault strike (FS)， dip angles (DA)， shear strain (SS)， longitudinal strain (LS)， dip slip gradient (DSG)， and surface gradient (SG). These factors along with hydrocarbon column height (HCH) serve as characteristic values for constructing a dataset used in the fault lateral sealing evaluation model based on BP neural network. The newly proposed BP evaluation approach was developed to enhance prediction accuracy， while a traditional shale gouge ratio (SGR) model was constructed for comparative validation. Contrary to traditional models， the results reveal that ENS predominantly govern sealing ability in this structural setting， while CC parameters show secondary influence. This indicates the existence of multiple influencing factors regarding fault sealing ability. The BP model enables prediction of oil-gas ranges in undrilled traps while facilitating failure analysis of unsuccessful wells along with forecasting potential areas for exploration. This will provide more scientifically informed decision-making support for future oil and gas exploration activities， including optimizing well locations.