Training Dataset Generation for Moment Tensor Inversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating training data for machine learning models processing microseismic data during hydraulic fracturing face challenges such as low signal-to-noise ratio and limited angle coverage, leading to inaccurate deterministic inversion predictions of moment tensor components, which are crucial for understanding subsurface structures and optimizing fracking operations.
Innovation Solution
A data processing system is configured to efficiently generate a training dataset for machine learning models by simulating seismic wave propagation from optimally distributed seismic sources, reducing the number of simulations required while maintaining accuracy, and using this data to train artificial neural networks for moment tensor component estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deterministic inversion methods are used to estimate moment tensor components from microseismic data, then the process provides interpretable physical models, but the accuracy deteriorates due to low signal-to-noise ratio and limited angle coverage
Solution Approach 1:
The patent replaces traditional deterministic mechanical inversion methods with machine learning models (neural networks) that learn optimal solutions from training data, substituting physics-based iterative solving with data-driven pattern recognition that is more robust to noise and limited coverage
Solution Approach 2:
The patent performs preliminary action by generating synthetic training data through forward modeling of seismic wave propagation before the actual inversion task, allowing the machine learning model to learn from pre-computed examples that cover various noise conditions and geometric configurations
2Measurement precision
If a large number of seismic sources are simulated to generate comprehensive training data, then the model accuracy improves, but the computational cost increases significantly
Solution Approach 1:
The patent applies self-service by implementing an adaptive sampling strategy where the machine learning model itself identifies which training examples are most informative, allowing the system to automatically optimize the training set composition without external intervention and reducing redundant computations
Solution Approach 2:
The patent uses synthetic seismogram data generated through efficient forward modeling as disposable training samples that can be rapidly created and discarded, replacing the need for extensive field measurements or expensive repeated simulations
3Productivity
If traditional deterministic methods are used for microseismic analysis, then the process is computationally efficient, but the prediction accuracy deteriorates in complex heterogeneous velocity models
Solution Approach 1:
The patent changes the fundamental approach parameter from deterministic physics-based solving to probabilistic data-driven learning, allowing the system to achieve high accuracy in heterogeneous velocity models by learning from diverse training examples that capture complex wave propagation effects
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach significantly reduces computational costs and improves the accuracy of moment tensor component estimation, enabling better subsurface structure determination and enhanced control over fracking operations, with predictions being practically instantaneous after training.
Implementation Method 1
The data processing system executes a forward modeling code to simulate microseismic data generated by known values of the moment tensor components at the source location
Data Source
AI summary
Methods and systems for training a machine learning model to process microseismic data recorded during fracturing of a subterranean geological formation are configured for selecting a volume in the subterranean geological formation, the volume comprising a set of vertices and a center, the set of vertices defining a first dimension; determining seismogram data for sources at the vertices of the volume and at the center of the volume; generating training data from the seismogram data, the training data relating values of seismogram data to values of moment tensor components; training a machine learning model using the training data; and determining, based on the trained machine learning model, a second dimension defined for the set of vertices, the second dimension being a maximum value enabling an accuracy for outputs of the trained machine learning model that satisfies a threshold.