Facies Classification via Iterative Matrix Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated systems for interpreting borehole log data rely on supervised machine learning methods but lack effective mechanisms to account for facies transitions and classification uncertainties, leading to suboptimal facies identification in subsurface geological formations.
Innovation Solution
A method that utilizes a facies transition matrix and confusion matrix to iteratively refine facies identifications by adjusting predicted classifications based on observed transitions, incorporating a user-adjustable parameter to balance predicted and observed probabilities, ensuring accurate facies modeling and uncertainty characterization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If supervised machine learning methods are used for automated facies interpretation, then automation extent is improved, but accuracy of facies identification deteriorates due to lack of transition and uncertainty modeling
Solution Approach 1:
The patent implements feedback mechanisms through iterative refinement processes where predicted facies classifications are continuously compared with observed transitions and uncertainty models. The system uses confusion matrices and facies transition matrices to feedback-correct predictions, allowing the automated system to learn from its own errors and improve accuracy over time while maintaining high automation levels.
Solution Approach 2:
The patent changes key parameters by introducing transition probability matrices and confusion matrices that capture facies transition patterns and classification uncertainties. These additional parameters allow the supervised learning system to account for spatial relationships and uncertainty, improving identification accuracy without reducing automation.
2Device complexity
If traditional supervised learning methods are used, then computational complexity is reduced, but reliability of facies classification deteriorates due to inability to model transitions and uncertainties
Solution Approach 1:
The patent segments the facies classification problem into multiple components: training data preparation, supervised learning model training, confusion matrix calculation, facies transition matrix calculation, and iterative refinement. This segmentation allows each component to be optimized independently, managing overall computational complexity while improving reliability through comprehensive modeling of transitions and uncertainties.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating confusion matrices and facies transition matrices from training data before final classification. These pre-computed matrices capture essential patterns and uncertainties, allowing the main classification process to focus on applying these pre-learned patterns to new data, thus improving reliability without proportionally increasing computational complexity.
3Ease of operation
If automated supervised learning is applied without transition modeling, then ease of operation is improved, but measurement precision of facies boundaries deteriorates
Solution Approach 1:
The patent introduces confusion matrices and facies transition matrices as intermediary structures that mediate between the automated learning process and the actual facies boundaries. These intermediaries capture the relationship between predicted and actual facies, allowing the system to automatically adjust boundary predictions to match observed transitions while maintaining ease of automated operation.
Data Source
AI summary
A method of automatically interpreting well log data indicative of physical attributes of a portion of a subterranean formation which include some portion of samples with known facies classification to be used as training data, dividing the training data into two subsets, a calibration set and a cross-validation set, using an automated supervised learning facies identification method to determine a preliminary identification of facies in the subterranean formation based on the calibration set, calculating a confusion matrix for the supervised learning facies identification method by comparing predicted and observed facies for the cross-validation set, calculating a facies transition matrix characterizing changes between contiguous facies, and using the preliminary identification, the facies transition matrix, and the confusion matrix, iteratively calculating updated facies identifications.