Neural Network Geophysical Data Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional seismic data acquisition methods, such as towed streamer surveys, face limitations in capturing low and high frequency ranges due to noise and attenuation, which affects the accuracy of subsurface property inversion and imaging, particularly in petroleum exploration.
Innovation Solution
A computer-implemented method that utilizes a neural network to reconstruct missing frequency bands in geophysical datasets by training the network with data from a second, differently acquired dataset, such as ocean bottom node data, to enhance the first dataset, thereby improving the signal-to-noise ratio and extending the frequency range.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional towed streamer acquisition is used, then the data can be acquired with standard equipment and procedures, but the frequency bandwidth is limited (4-60 Hz) and ultra-low frequencies are unreliable due to ghosts and noise
Solution Approach 1:
The patent uses an intermediary neural network model that learns the relationship between high-frequency and low-frequency seismic data. The network acts as a mediator to reconstruct missing ultra-low frequency components (1-4 Hz) from available high-frequency towed streamer data, avoiding the need to directly acquire unreliable low-frequency data while preserving the beneficial ghost-free characteristics of high-frequency acquisition.
Solution Approach 2:
The patent creates a synthetic copy of the missing low-frequency data by training a neural network to generate realistic ultra-low frequency seismic signals that mimic what would be obtained from specialized acquisitions. This allows the system to work with copied low-frequency information derived from high-frequency data, achieving frequency bandwidth extension without actual low-frequency acquisition.
2Strength
If signal boosting is applied to enhance ultra-low frequencies, then the signal strength increases, but the noise is also amplified due to poor signal-to-noise ratio
Solution Approach 1:
The patent extracts only the useful signal information from the high-frequency data and uses the trained neural network to generate the corresponding low-frequency signal components. By separating the signal extraction from noise through the learning-based approach, the system enhances signal strength in the ultra-low frequency range without amplifying the noise that would be present in direct low-frequency acquisition or boosting.
Solution Approach 2:
The patent replaces the mechanical signal boosting approach with a computational neural network-based signal generation approach. Instead of physically amplifying the weak low-frequency signals (which also amplifies noise), the system uses learned patterns from high-frequency data to computationally reconstruct the low-frequency signals, achieving signal enhancement without the harmful noise amplification effect.
3Reliability
If specialized acquisition methods (OBN, ultra-shallow tow) are used to obtain ultra-low frequencies, then the signal-to-noise ratio improves, but the acquisition cost increases significantly
Solution Approach 1:
The patent makes the high-frequency towed streamer acquisition method multi-functional by enabling it to produce both high-frequency and reconstructed low-frequency data through the neural network. This universal approach allows a single acquisition method to serve multiple frequency ranges, eliminating the need for separate specialized acquisitions and significantly reducing costs while maintaining reliability.
Solution Approach 2:
The patent changes the parameter of frequency bandwidth through computational processing rather than physical acquisition changes. By applying a trained neural network to transform high-frequency data into multi-frequency output, the system achieves the effect of acquiring low-frequency data without changing the physical acquisition parameters, thereby avoiding the high costs associated with OBN or ultra-shallow tow deployments.
4Measurement precision
If high frequencies are boosted to improve resolution, then the imaging quality improves, but the frequencies become overwhelmed by noise factors
Solution Approach 1:
The patent converts the limitation of having only high-frequency data (which normally lacks low-frequency information needed for full bandwidth imaging) into a benefit by using the neural network to generate the missing low-frequency components. The high-frequency data serves as a reliable foundation, and the network adds the complementary low-frequency information, achieving full bandwidth imaging without the noise problems of direct low-frequency acquisition.
Data Source
AI summary
A method for enhancing properties of geophysical data with deep learning networks. Geophysical data may be acquired by positioning a source of sound waves at a chosen shot location, and measuring back-scattered energy generated by the source using receivers placed at selected locations. For example, seismic data may be collected using towed streamer acquisition in order to derive subsurface properties or to form images of the subsurface. However, towed streamer data may be deficient in one or more properties (e.g., at low frequencies). To compensate for the deficiencies, another survey (such as an Ocean Bottom Nodes (OBN) survey) may be sparsely acquired in order to train a neural network. The trained neural network may then be used to compensate for the towed streamer deficient properties, such as by using the trained neural network to extend the towed streamer data to the low frequencies.