Bayesian Seismic Fault Detection With Uncertainty Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing seismic fault detection methods in hydrocarbon exploration lack accuracy and fail to provide uncertainty values, leading to suboptimal well placement and increased drilling risks.
Innovation Solution
Employing a Bayesian deep learning neural network data processing model that generates predictions of faults and fractures with associated uncertainty values, trained using paired seismic images and corresponding fault or fracture labels, to improve fault detection and well placement accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If machine learning methods are used for fast and automatic fault detection, then processing speed is improved, but measurement precision of fault detection deteriorates
Solution Approach 1:
The patent replaces traditional deterministic machine learning classification with a Bayesian probabilistic framework. This substitution allows the system to maintain fast automatic processing while improving measurement precision by quantifying uncertainty in fault detections through posterior distributions and evidence lower bound optimization.
Solution Approach 2:
The patent changes the fundamental parameter representation from fixed weight values to probability distributions over weights. By modeling weights as random variables with prior distributions and updating them to posterior distributions through Bayesian inference, the system achieves both speed and precision by leveraging probabilistic reasoning and uncertainty quantification.
2Measurement precision
If deep neural networks are trained to improve fault detection accuracy, then measurement precision is improved, but overfitting increases
Solution Approach 1:
The patent implements feedback through the evidence lower bound (ELB) optimization process. The ELB serves as a feedback mechanism that balances fitting the training data while maintaining model generalization by incorporating prior distributions and likelihood functions, preventing overfitting through probabilistic regularization.
Solution Approach 2:
The patent applies beforehand cushioning by introducing prior distributions over network weights before training. These priors act as a cushion against overfitting by constraining the weight space to plausible regions, allowing the model to generalize better to unseen data while maintaining high detection accuracy.
3Device complexity
If traditional classification models are used, then device complexity is reduced, but loss of information occurs due to lack of uncertainty values
Solution Approach 1:
The patent adds another dimension to the output by providing not just classification labels but also uncertainty estimates through posterior distributions. This dimensional expansion from single-value predictions to probability distributions preserves information about detection confidence without substantially increasing operational complexity.
Data Source
AI summary
Systems and methods are configured for identifying faults or fractures in a subsurface region for performing hydrocarbon extraction. The systems and methods execute a Bayesian neural network on seismic images to generate, for locations in the seismic images, predictions of the presence or absence of faults and fractures in the subsurface region. The predictions are associated with uncertainty values. The locations of wells are selected for drilling based on the predictions and the associated uncertainty values.