Abstract:

The C-2 oilfield, located in the Bohai Bay Basin, is a fluvial-facies offshore oilfield primarily developed with horizontal wells. It is characterized by thin reservoir layers, vertically stacked multi-phase sandbodies, and rapid lateral facies transitions, leading to complex internal reservoir structures and connectivity. The combined effects of complex reservoir structures and well-seismic data make reservoir characterization challenging under sparse well patterns, hindering refined development. Conventional seismic inversion fails to meet the requirements for high-resolution prediction of thin reservoirs and detailed characterization of internal reservoir structures. To study the structural characteristics of braided river reservoirs in the oilfield, a stacking ensemble learning method based on seismic attributes was applied to predict the complex fluvial-facies reservoir structures. This approach significantly improved prediction accuracy compared to a single machine learning model. By integrating multi-dimensional information such as geology, geophysics, and reservoir dynamics, iterative optimization was conducted to further reduce the uncertainty in subsurface reservoir prediction and structural understanding. This enabled the precise characterization of the complex braided river reservoir structures in the study area, providing a basis for refined exploitation of remaining oil and potential sandbodies in the oilfield. The study demonstrates that the reservoir prediction method based on stacking ensemble learning not only enhances seismic vertical resolution, but also exhibits strong horizontal phase-control capabilities. The prediction results include sandbody stacking relationships and internal reservoir structures, making it more suitable for the prediction and fine characterization of continental fluvial sedimentary systems with rapid facies transitions and complex spatial architectural structures. This method can serve as a reference for the detailed characterization of fluvial-facies reservoir configurations during the middle and late development stages of offshore oilfields with sparse well patterns.

Key words: machine learning, reservoir prediction, braided river, reservoir architecture, horizontal well