Seismic horizon picking is the foundation of structural interpretation,seismic inversion,and reservoir prediction.Current intelligent seismic horizon picking methods have not utilized the inherent characteristics of seismic data and their attributes to supervise the crucial steps,including sample selection,network modeling,and result analysis.Therefore,an intelligent seismic horizon picking method based on sample selection and multiple quality control measures is proposed.The method adopts a novel sample selection strategy,which considers the generation direction of the data set,the size of training sample,and the degree of labeling errors to realize intelligent horizon picking using a U-shaped neural network.A variety of geophysical quality control methods including horizon flattening,horizon tie,t0 contour,geosteering phase slice,and root mean square amplitude slice are introduced to verify the rationality of the sample selection strategy and guide the establishment of the optimal intelligent picking model.The optimal horizon-picking model can allow the simultaneous tracking of multiple target horizons and ensure the reliability and rationality of the picked horizons.Conclusions can be drawn from the tests of 3D fracture physical simulation data and 3D field data.①The intelligent horizon interpretation model established by seismic samples and labels extracted along the spatial direction with complex geological structure features can describe the details related to small-scale faults and fractures.②The neural networks trained by large-size seismic profiles can characterize horizons with comparatively more detail and reduce the occurrence of geological artifacts.③The horizon labels have 20% random errors within ±15m around the correct horizons,and the U-shaped neural network can identify target horizons with high precision.④The sample size has a greater influence on the generalization of the intelligent horizon-picking model than the dataset generation direction.
论文详情