Intelligent fusion of seismic spectral-decomposed attributes and thin sand body prediction method based on Stacking ensemble learning: Taking Qingcheng south district of Ordos Basin as an example

Gravity-flow deposits in lacustrine faulted depressions of continental basins represent favorable reservoirs for tight oil and shale oil accumulation, typically characterized by thin interbedded sand-mud layers, which pose significant challenges for conventional reservoir prediction methods. In response to this issue, this study proposes a thin-layer sandbody prediction technique based on frequency-divided intelligent inversion and intelligent fusion of frequency-divided attributes, using the interbedded tight sandstone of the Chang 7 member in the Qingcheng oilfield, Ordos Basin as a case study. First, wavelet-based spectral decomposition is employed to divide the seismic volume into multiple frequency bands, from which a set of attributes that effectively characterize sandbody thickness is selected. Subsequently, frequency-divided intelligent inversion is performed using a limited number of wells and seismic data to generate inversion volumes reflecting lithological properties. Finally, a Stacking ensemble learning framework is constructed, using well-log-interpreted sandbody thickness as the supervisory label, to intelligently fuse the selected frequency-divided and inversion attributes. This approach improves the correlation between predicted results and validation well data from 0.375 to 0.678. The proposed method provides a valuable reference for the fine-scale characterization of thin interbedded sand layers and holds broad application potential.