Seismic Feature Extraction for Sparse Well Log Property Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating subsurface properties from seismic data are unreliable due to the sparsity of well logs, leading to inaccurate predictions and overfitting, especially when integrating seismic data with well logs using one-dimensional machine learning models.
Innovation Solution
Employing two machine learning models, a first model to extract seismic features and a second model to integrate seismic data with well logs, reducing the risk of overfitting and improving lateral consistency in subsurface property estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If one-dimensional machine learning models are used to integrate seismic data with well logs, then the mapping function can be trained, but the estimation becomes unreliable throughout the entire seismic survey area
Solution Approach 1:
The patent transitions from one-dimensional mapping functions to two-dimensional mapping functions that simultaneously process both inline and crossline seismic directions. This dimensional expansion allows the model to capture lateral variations in subsurface properties across the entire survey area, resolving the unreliability issue while maintaining manageable complexity through structured multi-dimensional processing
2Adaptability or versatility
If well logs are down-sampled to seismic scale, then the data can be integrated, but the estimation remains valid only around training wells
Solution Approach 1:
The patent develops a mapping function that serves multiple purposes: it accurately represents subsurface properties around training wells while simultaneously providing reliable predictions across the entire survey area. The two-dimensional approach enables the model to generalize well beyond the immediate well locations, achieving universal applicability without sacrificing precision
3Quantity of substance
If seismic data is integrated with sparse well logs, then subsurface properties can be estimated, but overfitting occurs and predictions become inaccurate
Solution Approach 1:
By expanding from one-dimensional to two-dimensional mapping, the patent effectively increases the information capacity of the training data utilization. This allows the model to learn more robust patterns from the sparse well log data while maintaining generalization capability, preventing overfitting and improving prediction accuracy across the entire survey area
Data Source
Figure 1A~1D
Figure 2
Figure 3A
AI summary
A method for seismic processing includes extracting, using a first machine learning model, one or more seismic features from seismic data representing a subsurface domain, receiving one or more well logs representing one or more subsurface properties in the subsurface domain, and predicting, using a second machine learning model, the one or more subsurface properties in the subsurface domain at a location that does not correspond to an existing well based on the seismic data, the one or more well logs, and the one or more seismic features that were extracted from the seismic data.