True-3D AVA Seismic Inversion for Reservoir Characterization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing seismic amplitude analysis technologies rely heavily on 1D earth models and fail to effectively separate and distill the relationships between well-log and seismic data, leading to less accurate hydrocarbon reservoir identification and characterization.
Innovation Solution
A method involving the generation of synthetic seismograms with known amplitude variation with angle (AVA), seismic migration, true-3D AVA modeling, and inversion, along with windowed deconvolution and rock property prediction using machine-learning models, to create a more accurate digital seismic image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If 1D earth models with horizontal layers are used for AVO/AVA modeling and inversion, then the processing complexity is reduced and computation is simplified, but the accuracy of hydrocarbon reservoir identification and characterization deteriorates
Solution Approach 1:
The patent transitions from 1D earth models to true 3D AVA modeling by incorporating dip fields and spatially varying parameters. The method uses 3D velocity tomography and full-azimuth 3D prestack depth migration to create three-dimensional representations of subsurface properties, allowing accurate characterization of complex geological structures with dipping reflectors and lateral variations that cannot be captured by 1D horizontal layer assumptions.
2Ease of operation
If ML algorithms are applied directly to mismatched well-log and seismic data without separating deterministic and stochastic sub-problems, then the workflow is simplified, but the effectiveness of rock property prediction deteriorates
Solution Approach 1:
The patent segments the seismic inversion problem into deterministic and stochastic sub-problems. Deterministic components include 3D velocity tomography, dip field extraction, and AVA basis function calculation. Stochastic components include non-stationary spatial variations in wavelet and migration aperture effects. This segmentation allows appropriate ML algorithms to be applied to each sub-problem with proper preprocessing, improving prediction effectiveness.
Solution Approach 2:
The patent performs preliminary processing steps including 3D velocity tomography, full-azimuth 3D prestack depth migration, and dip field extraction before applying ML algorithms. Well-log and seismic data are preprocessed to separate deterministic relationships from stochastic variations, creating properly matched datasets that are suitable for training ML models and improving rock property prediction effectiveness.
3Device complexity
If post-migration seismic trace samples are used without correcting for acquisition footprint and migration aperture effects, then the data processing is simplified, but the spatial coherence between seismic and well data deteriorates
Solution Approach 1:
The patent applies local quality corrections by estimating and removing acquisition footprint and migration aperture effects separately at each spatial location. Dip fields are extracted to account for local geometric variations, and wavelet deconvolution is performed with location-specific parameters. This localized processing restores spatial coherence between seismic and well data by correcting position-dependent distortions that vary from point to point in the subsurface.
Data Source
AI summary
A method is described for seismic amplitude analysis that uses a set of artificial and individually separable reflectors consistent with dip fields in the subsurface volume of interest to define an AVA basis functions; the AVA basis functions are used in true-3D AVA modeling and true-3D AVA inversion. The inversion result and well logs representative of the subsurface volume of interest are used to train a model to create a rock property prediction model. The method may apply the rock property prediction model to a second seismic image to generate a rock property volume. The method is executed by a computer system.