Carbonate reservoirs characterization based on frequency Bayesian principal component analysis

Carbonate reservoirs, as critically important exploration targets, possess significant geological resource potential. Within a southern Chinese basin, many karst fissure-cave reservoirs are observed at the top of Lower-Permian strata. However, these reservoirs are coincidentally situated beneath low-velocity, low-density mudstone layers from the base of the Upper-Permian strata. These mudstone layers exhibit notable impedance differences with underlying carbonate formations, resulting in low-frequency, continuous and robust seismic peak reflection at the interface. This occurrence imparts obvious shielding effect on the reflections of the reservoirs developed at the top of the Lower-Permian strata, and complicates the identification of reservoir responses, ultimately reducing reservoir identification accuracy. Addressing this, the study proposes frequency Bayesian principal component analysis (FBPCA) technique to reconstruct initial seismic data, aiming to suppress the strong interface reflection and alleviate their masking effect on the reservoirs responses. FBPCA introduces Bayesian principles, constructing a hierarchical model based on probabilistic principal component analysis (PPCA) and incorporating continuous hyper-parameters to automatically determine the number of principal components. FBPCA also undertakes frequency-based calculation to address overlapping and interference among the reflection frequency bands of both the reservoir and interface. Seismic forward modeling initially validates the attenuation effect of FBPCA on intense interface reflections. Subsequently, seismic reflection features are analyzed for data before and after the FBPCA-based reconstruction in the actual working area. Amplitude attributes are then extracted to predict the extent of reservoirs development. Comparative analysis indicates closer alignment with actual drilling conditions using FBPCA-reconstructed data. Evidently, FBPCA enhances carbonate karst reservoir prediction precision.