Automated Well-Top Correlation with Global-Local CNN Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual well-top correlation in wellbore interpretation is time-consuming and error-prone, especially for formations with subtle visual signatures in logs.
Innovation Solution
A method utilizing a soft attention-based Convolutional Neural Network (CNN) with both global and local models to automatically correlate well logs, incorporating a U-Net-based encode-decoder architecture with skip connections and supervised training to identify formation markers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual visual examination of logs is used for well-top correlation, then flexibility and adaptability are maintained, but time consumption and error rate increase significantly
Solution Approach 1:
The patent replaces the manual mechanical process of visual log examination with an automated machine learning system. The CNN-based model automatically processes well logs to identify formation tops, substituting human visual inspection with computational algorithms that can analyze multiple logs simultaneously and consistently, thereby reducing time consumption while maintaining or improving correlation accuracy.
Solution Approach 2:
The system enables self-service automation where the machine learning model independently performs well-top correlation without requiring manual intervention for each log analysis. The model trains on labeled data and then autonomously identifies formation tops in new logs, making the correlation process self-sufficient and eliminating the time-consuming manual examination step.
2Measurement precision
If manual visual examination is used, then human judgment can handle complex cases, but error rate increases for subtle visual signatures
Solution Approach 1:
The patent transforms the visual log data into numerical parameters and features that the CNN model can process. By converting visual signatures into quantitative measurements and feeding them to the machine learning algorithm, the system achieves consistent and precise marker identification without the variability and errors associated with manual visual interpretation of subtle features.
Solution Approach 2:
The patent divides the well log analysis into distinct segments or features that the CNN model can process independently. The model segments the log data into relevant characteristics and patterns, analyzing each segment to identify formation tops with high precision, thereby improving measurement accuracy for subtle visual signatures that manual examination might miss.
3Productivity
If automated machine learning models are used, then time consumption is reduced, but system complexity increases
Solution Approach 1:
The patent develops a universal machine learning system that can handle multiple well logs and various formation types with a single trained model. This multi-functional approach allows the system to process different log types and geological scenarios without requiring separate manual procedures for each case, thereby improving productivity while managing system complexity through a unified automated framework.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for correlating well logs includes receiving a well log as input to a first machine learning model that is configured to predict first markers in the well log based at least in part on a global factor of the well log, receiving the well log as input to a second machine learning model that is configured to predict second markers in the well log based at least in part on local factors of the well log, generating a set of predicted well markers by merging at least some of the first markers and at least some of the second markers, and aligning the well log with respect to one or more other well logs based at least in part on the set of predicted well markers.