Seismic Data Inversion Using Bayesian Uncertainty Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing seismic data inversion methods, particularly amplitude variation with offset (AVO) inversion, face challenges in accuracy and reliability due to reliance on deterministic approaches that do not capture uncertainties, leading to reduced precision in reservoir characterization.
Innovation Solution
Implementing a probabilistic Bayesian approach with techniques like Markov Chain Monte Carlo (MCMC) or optimization processes, such as Stein Variational Gradient Descent (SVGD), to generate a distribution of outcomes and quantify uncertainties, improving the accuracy of subsurface elastic parameter estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deterministic AVO inversion methods are used, then computational simplicity is maintained, but measurement precision and reliability of subsurface parameter estimation deteriorate due to inability to capture uncertainties
Solution Approach 1:
The patent transforms the inversion approach from deterministic to probabilistic by changing the fundamental parameter representation. Instead of estimating single deterministic values for elastic parameters, the method estimates probability distributions characterized by mean values and uncertainties. This parameter change enables simultaneous capture of both the estimated parameters and their uncertainties, directly improving measurement precision while maintaining computational tractability through efficient sampling algorithms.
Solution Approach 2:
The patent replaces traditional deterministic mechanical inversion algorithms with probabilistic sampling methods (MCMC and SVGD). This substitution introduces stochastic elements into the inversion process, allowing the system to explore the parameter space and quantify uncertainties. The probabilistic framework substitutes the single-point estimation mechanism with a distribution-based approach, improving reliability by capturing the inherent uncertainties in seismic data interpretation.
2Reliability
If probabilistic Bayesian approaches with MCMC or SVGD are implemented, then reliability and accuracy of reservoir characterization is improved through uncertainty quantification, but computational complexity and processing time increase
Solution Approach 1:
The patent employs dynamic sampling strategies where the sampling process adapts during computation. The SVGD algorithm dynamically adjusts particle positions and weights based on the likelihood function and prior information, allowing the system to converge more efficiently to the posterior distribution. This dynamic approach improves reliability by thoroughly exploring the parameter space while reducing processing time compared to static or exhaustive sampling methods.
Solution Approach 2:
The patent applies local optimization within the global probabilistic framework. Both MCMC and SVGD methods focus computational effort on regions of parameter space that are more likely to contain the true values, based on the seismic data likelihood. This local quality approach concentrates computational resources where they are most needed, improving reliability through targeted uncertainty quantification while reducing overall processing time by avoiding exhaustive sampling of unlikely parameter combinations.
3Loss of information
If deterministic inversion methods are used, then processing speed is maintained, but the ability to capture and quantify uncertainties in subsurface properties is lost
Solution Approach 1:
The patent adds an uncertainty dimension to the traditional deterministic inversion output. Instead of producing single-value estimates for elastic parameters, the probabilistic approach generates probability distributions that include both mean values and uncertainty measures (standard deviations, confidence intervals). This dimensional expansion captures previously lost information about parameter reliability while maintaining productivity through efficient computational algorithms that can handle the increased information content.
Solution Approach 2:
The patent creates a composite estimation that combines multiple sources of information: seismic data likelihood, prior geological knowledge, and uncertainty quantification. The probabilistic framework integrates these diverse information sources into a unified posterior distribution, preserving valuable information about uncertainties while improving the overall efficiency and reliability of reservoir characterization. This composite approach leverages both data-driven and knowledge-driven constraints.
Data Source
AI summary
Techniques to allow for increases in the accuracy of the determination of properties of a formation (e.g., a formation's fluid content, porosity, density, etc.) or seismic velocity, shear wave information, etc. are disclosed herein. The techniques include generating initial input data comprising based at least in part on initial seismic data, modeling the initial input data to generate synthetic seismic data based on different combinations of the initial input data, iteratively updating a value of each particle of a set of particles utilizing the synthetic seismic data to generate a final set of particles and outputting the final set of particles as a target distribution.