Seismic Interpretation Confidence Mapping for Fault-Aware Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Seismic interpretation is uncertain due to low confidence measures, often leading to indiscriminate filtering of valuable data near faults, which can degrade the accuracy of subsurface modeling.
Innovation Solution
A system utilizing machine learning (ML) to compute relative confidence in seismic interpretation by processing geometric and seismic attribute volumes, iteratively training a model, and applying localized filtering based on confidence values to improve subsurface framework quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static filtering is applied to seismic interpretation data near fault locations, then the likelihood of generating modeling artifacts is reduced, but high confidence and valuable information is undesirably filtered out
Solution Approach 1:
The patent applies different filtering operations to different regions of the seismic data based on local characteristics. High confidence regions are processed with minimal or no filtering, while low confidence regions near faults receive stronger filtering. This localized approach preserves valuable information in high confidence areas while still preventing artifacts in problematic areas.
Solution Approach 2:
The patent transitions from static filtering (uniform across all data) to dynamic filtering where the filtering strength is adjusted based on confidence measures. The filtering operation adapts to each region's characteristics, applying appropriate levels of filtering based on local confidence assessments rather than a uniform static approach.
2Ease of manufacture
If coarse and static operators are applied to filter seismic interpretation data, then processing simplicity is maintained, but the ability to preserve high confidence information near faults is reduced
Solution Approach 1:
The patent changes the parameters of the filtering operation dynamically based on confidence measures. Instead of using fixed coarse operators, the system adjusts filtering parameters (such as filter strength, kernel size, or threshold values) according to the local confidence assessment, enabling precise control over what information is preserved or filtered.
Solution Approach 2:
The patent introduces a feedback mechanism where confidence measures are computed from the seismic interpretation data, and these confidence values are used to adjust the filtering operation. The system continuously assesses confidence and uses this information to modulate the filtering strength, creating a closed-loop process that adapts to data quality variations.
3Stability of the object's composition
If indiscriminate filtering is applied to reduce modeling artifacts, then model robustness is improved, but data quality and interpretation accuracy are degraded
Solution Approach 1:
The patent applies quality-aware filtering that treats different regions differently based on their confidence characteristics. High confidence regions maintain their original quality and detail, while low confidence regions receive filtering to improve stability. This localized quality preservation maintains interpretation accuracy where it exists while improving robustness where it is needed.
Solution Approach 2:
The patent dynamically changes filtering parameters based on local confidence assessments. In high confidence regions, filtering parameters are set to preserve detail and accuracy, while in low confidence regions, parameters are adjusted to prioritize stability and artifact reduction. This adaptive parameter adjustment resolves the contradiction between robustness and precision.
Data Source
AI summary
A system and method in accordance with the present disclosure include a workflow applied to identify areas of confidence in seismic interpretations that meets a pre-selected threshold by utilizing machine-learning (ML) techniques. The ML techniques enable one or more filters to be applied, based on a range of confidence values, to areas of seismic interpretation both in the vicinity of the faults but also away from fault locations. The contribution of the various inputs is evaluated and weighted by the ML model, which reduces the time to prepare input seismic interpretation data and to obtain results from executing the model.