Multiple-point geostatistical modeling for fault-controlled tight sandstone reservoirs based on probability fusion of permanence of ratios: a tight sandstone oil reservoir in the southern margin of the ordos basin

Unlike conventional sandstone reservoirs, which store hydrocarbons in sandstone pores, fault-controlled tight sandstone reservoirs are unconventional, primarily storing oil or/and gas in fault zones. While these reservoirs have significant reserves, their highly heterogeneous fault zones structures, including fault core and damage zone, pose challenges for geological modeling and precise development. Traditional two-point geostatistics (TPG) struggle to reproduce strike-slip fault zones patterns, and object-based methods have difficulty statistically quantifying their structural parameters. Deterministic methods, truncated by seismic data threshold, often misalign with well data, reducing accuracy in representing fault zone details. To overcome these challenges, we propose a new modeling workflow for fault-controlled tight sandstone reservoirs based on multi-sources information-constrained multiple-point geostatistics (MPG). First, a deep neural network (DNNs) is used to correlate conventional logging curves with fracture density (FD) to obtain well-interpreted facies data. Next, inter-well factors like brittleness index, shale content, and fault proximity are used to construct four single-sources probability bodies. These are combined into a multi-source probability body using the Permanence of Ratios (PR) method, which effectively integrates the contributions of different sources for greater constraint. Finally, the multiple-point geostatistical direct sampling (DS) algorithm generates a three-dimensional (3-D) geological model that captures the reservoir's geological features, while satisfying the multi-source information constraints. The results shows that the proposed method effectively reduces model uncertainty and improves spatial prediction of the reservoir, achieving over 85% accuracy when compared with field production data. This workflow offers a promising approach for fine-scale modeling of fault-controlled tight sandstone reservoirs, with broad potential for similar reservoir development and management.