Seismic Inversion and Machine Learning for High Resolution Rock Properties
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for improving well-planning, drilling, and production in unconventional reservoirs, such as the Permian basin, face limitations due to the lack of high-resolution, well-correlated subsurface geology data, with seismic inversion alone not providing sufficient temporal resolution and quality for optimal horizontal well placement.
Innovation Solution
A method combining pre-stack seismic inversion and machine learning (neural network processing) is used to enhance the quality and resolution of seismic data by tying seismic records with well logs, creating reflection coefficient volumes for accurate subsurface imaging and hydrocarbon exploration planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If seismic inversion alone is used to derive geomechanical properties, then the process is simple, but the temporal resolution and quality of the data are insufficient for optimal well placement
Solution Approach 1:
The patent combines seismic inversion with machine learning algorithms to integrate multiple data sources and processing methods. This merging approach allows the system to achieve high temporal resolution through the complementary strengths of both seismic data and machine learning patterns, while distributing the computational complexity across multiple modular components rather than requiring a single complex system.
Solution Approach 2:
The patent introduces machine learning models as an intermediary between raw seismic data and final geomechanical property derivation. This intermediary layer processes seismic signals to extract enhanced temporal resolution features and correlates them with well log data, effectively bridging the gap between seismic measurements and subsurface property characterization without requiring direct inversion of all parameters simultaneously.
2Reliability
If current seismic data methods are used, then the process is straightforward, but the data quality and correlation to actual depth geology are insufficient
Solution Approach 1:
The patent implements feedback loops where machine learning models are trained using correlated well log data and seismic data, then applied to enhance seismic interpretation. The system continuously refines its correlations by comparing predicted subsurface properties against actual well measurements, adjusting parameters to improve depth correlation accuracy while maintaining a systematic approach that manages complexity through iterative optimization.
Solution Approach 2:
The patent creates a composite data model that integrates seismic reflection data, well log measurements, and machine learning-derived predictions. This composite approach combines the depth information from well logs with the spatial coverage of seismic data, producing a hybrid dataset with enhanced reliability for well placement decisions while organizing the complexity into structured integration workflows.
Data Source
AI summary
A system and method combines model-based inversion and supervised neural networks to develop high resolution rock property volumes from surface seismic data. These volumes have higher frequency and are calibrated to fit well log data. In addition to rock volumes, a Reflection Coefficient (RC) volume is derived from the acoustic impedance volume. The RC volume has much higher frequency, better lateral continuity, and ties to the well logs better than conventional seismic or frequency enhanced data. By interpreting and mapping with this RC volume, a much more accurate depth model can be built, which allows for a horizontal well to be accurately drilled.