Stratigraphic Forward Modeling for Synthetic AI Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for integrating stratigraphic processes in 3D modeling are tedious, manual, and time-consuming, and require large amounts of multiscale training datasets for AI and ML algorithms, which are not feasible with real interpreted well logs or 3D seismic data.
Innovation Solution
A synthetic data set is generated using process mimicking algorithms at the reservoir scale, creating labeled 3D stratigraphic models with various geological and petrophysical properties, which can be used to train AI and ML algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional manual workflows are used for stratigraphic interpretation from well logs and seismic data, then the process can be performed with existing tools, but the workflow becomes tedious, manual, and time-consuming
Solution Approach 1:
The patent creates synthetic copies of real well logs, seismic data, and facies models through forward stratigraphic modeling. These synthetic datasets replicate the characteristics of real subsurface data while being generated from known ground truth models, enabling automated AI/ML training without requiring extensive manual interpretation of actual field data
Solution Approach 2:
The system performs preliminary stratigraphic modeling and data synthesis before actual interpretation tasks. By pre-generating comprehensive synthetic datasets with known answers and diverse geological scenarios, the system prepares training materials in advance that enable rapid automated interpretation without time-consuming manual workflows during actual operations
2Productivity
If AI and ML algorithms are used to automate stratigraphic processes, then productivity increases, but large amounts of multiscale training datasets are required which are not feasible with real interpreted data
Solution Approach 1:
The patent generates synthetic copies of multiscale subsurface data including well logs, seismic volumes, and facies models through forward stratigraphic modeling. These synthetic datasets provide the large volumes of multiscale training data required for AI/ML algorithms while avoiding the limitations of real interpreted data, enabling efficient automated stratigraphic interpretation
Solution Approach 2:
The system varies multiple parameters in the forward stratigraphic modeling process including geological time steps, depositional environments, facies properties, and structural configurations. This parameter variation generates diverse synthetic datasets covering multiple scales and geological scenarios, providing comprehensive training data for AI/ML algorithms without requiring equivalent volumes of real interpreted data
3Reliability
If real interpreted well logs and 3D seismic data are used for training AIML engines, then the training data reflects actual subsurface conditions, but the datasets are limited by sampling bias and user bias
Solution Approach 1:
The patent creates synthetic copies of subsurface data through forward modeling that replicate real geological processes and data characteristics while eliminating biases present in real interpreted data. These synthetic datasets maintain realistic relationships between geological features and measurement data while providing complete coverage without sampling limitations
Solution Approach 2:
Instead of interpreting real data to create training sets (forward approach), the system inverts the process by first generating known ground truth stratigraphic models, then simulating the acquisition of well logs and seismic data from these models. This backward approach ensures complete knowledge of the underlying geology and eliminates interpretation biases inherent in traditional workflows
4Ease of operation
If traditional manual stratigraphic modeling methods are used, then the process can be performed with existing expertise, but the workflow is isolated and time-consuming
Solution Approach 1:
The patent merges forward stratigraphic modeling, synthetic data generation, and AI/ML training into an integrated automated workflow. By combining these previously separate processes into a unified system that automatically generates training data and trains interpretation models, the system eliminates the isolated, time-consuming nature of traditional workflows while maintaining operational simplicity through automation
Data Source
AI summary
A process mimicking forward modeler with deposition and erosion at each specific geological time step. The 3D derived properties are high resolution depositional environments and rock properties that are used to generate multiscale labelled synthetic data. These synthetic data can range from 1D logs such as grain size, gamma ray, density, and velocity, to 3D synthetic seismic, and are used directly as training data for various AIML applications.