Deep Learning Sedimentary Facies Prediction Constrained by 4D Seismic Information

Deep learning has achieved success in the field of seismic data processing, such as denoising. In the field of seismic interpretation, deep learning has been used to identify typical faults and horizon interpretation. However, in the field of prediction of sedimentary facies, deep learning has not yet exerted great power, possibly due to the dissimilarity between samples and real data, as well as insufficient information constraints. Taking deep-water channel sandstone prediction as an example, we improved the Alluvsim sedimentation simulation algorithm, used seismic forward modeling to obtain samples, and added 4D seismic as important constraint information to deep learning training to obtain a reasonable channel sandstone model. Finally, the effectiveness of the model was verified through reservoir fluid numerical simulations.