Geophysical Inversion via Convolutional Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current geophysical inversion methods face challenges such as high computational cost, non-convexity of the objective function, and non-uniqueness of solutions, making it difficult to accurately determine subsurface physical properties from geophysical data, especially in large-scale applications like full-wavefield inversion.
Innovation Solution
The use of deep convolutional neural networks (CNNs) is proposed to learn a map from geophysical data to subsurface models, incorporating geological prior information and synthetic data generated from acoustic or elastic wave equations, to efficiently extract subsurface physical properties, reducing computational time and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional geophysical inversion methods are used to determine subsurface physical properties, then measurement precision is improved, but productivity deteriorates due to high computational cost
Solution Approach 1:
The patent applies preliminary action by training the convolutional neural network in advance using synthetic seismic data generated from known subsurface models. This pre-training phase allows the network to learn the complex mapping between seismic data and subsurface properties before actual inversion, enabling rapid and accurate predictions during production without repeated heavy computational iterations
Solution Approach 2:
The patent substitutes the traditional mechanical optimization-based inversion system with a data-driven neural network system. Instead of iteratively minimizing objective functions using gradient-based methods, the system uses a trained CNN to directly predict subsurface physical properties from seismic data, replacing the computational mechanics of optimization with a learned statistical model
2Measurement precision
If conventional geophysical inversion methods are used, then measurement precision is improved, but loss of time worsens due to iterative optimization requirements
Solution Approach 1:
The network is pre-trained on synthetic data generated from the wave equation, performing the computationally intensive learning phase beforehand. During actual subsurface inversion, the pre-trained network provides rapid predictions without requiring repeated forward modeling and optimization iterations, significantly reducing the time loss in production scenarios
3Productivity
If deep convolutional neural networks are used to process geophysical data, then productivity is improved through faster processing, but device complexity increases
Solution Approach 1:
The patent uses synthetic copies of seismic data generated from known subsurface models through the wave equation to train the neural network. These synthetic training samples allow the network to learn from idealized scenarios before processing real data, reducing the complexity burden during actual inversion by pre-learning common geological patterns
Solution Approach 2:
The patent transforms the inversion problem from optimizing physical model parameters through iterative mathematical methods to predicting parameters directly from seismic data using neural network weights. This parameter transformation changes the problem from a computational optimization task to a pattern recognition task, improving productivity while managing complexity through appropriate network architecture selection
4Reliability
If conventional inversion methods are used, then reliability is maintained through established mathematical frameworks, but productivity deteriorates due to non-convexity and non-uniqueness issues
Solution Approach 1:
The patent replaces the mathematical optimization mechanics with a learned prediction mechanism. The neural network learns the mapping from seismic data to subsurface properties during training, avoiding the non-convex optimization landscape and non-uniqueness problems that plague conventional inversion methods, thereby maintaining reliability while dramatically improving productivity
Data Source
AI summary
A method including: storing, in a computer memory, geophysical data obtained from a survey of a subsurface region; and extracting, with a computer, a subsurface physical property model by processing the geophysical data with one or more convolutional neural networks, which are trained to relate the geophysical data to at least one subsurface physical property consistent with geological prior information.