Seismic Noise Attenuation via DNN Signal Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional seismic data processing algorithms are laborious and often inadequate in accurately removing seismic interference noise, leading to signal loss or residual noise, especially when noise sources have varying angles and amplitudes, hindering effective detection of underground resources.
Innovation Solution
A deep neural network (DNN) is trained using a seismic interference model and simulated random noise to attenuate noise, allowing for the generation of a training dataset that enables accurate separation of signal from noise, thereby improving seismic data processing efficiency and quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional algorithms are used for SI noise attenuation, then processing can be performed with established methods, but the process is laborious and requires manual parameter testing and selection
Solution Approach 1:
The patent replaces manual mechanical parameter adjustment with an automated neural network system. The neural network learns optimal parameter settings through training on labeled data, automatically adapting to different noise characteristics without requiring manual intervention for parameter selection and testing.
Solution Approach 2:
The neural network performs self-adjustment by automatically learning the relationship between input seismic data and optimal processing parameters during training. The system serves itself by autonomously selecting parameters based on learned patterns, eliminating the need for continuous manual parameter optimization.
2Object-affected harmful factors
If conventional filtering methods are applied to remove SI noise, then noise reduction can be achieved, but signal loss or residual noise occurs especially when noise sources have varying angles and amplitudes
Solution Approach 1:
The patent transforms the seismic data into different parameter domains (frequency-wavenumber domain) where noise and signal can be more effectively distinguished. The neural network learns to identify and separate components based on their distinct parameter characteristics, enabling effective noise removal while preserving signal integrity across varying noise conditions.
Solution Approach 2:
The patent extends the processing from traditional time-space domain to additional dimensions including frequency and wavenumber domains. This multi-dimensional approach allows the neural network to exploit differences in noise and signal characteristics across multiple parameter spaces, improving separation accuracy for varying noise angles and amplitudes.
3Productivity
If DNN approaches are used for seismic data processing, then processing time and labor cost can be reduced, but the DNN can easily fail to accurately extract the signal from noise
Solution Approach 1:
The patent performs preliminary data transformation and feature extraction before feeding data to the neural network. By pre-processing the seismic data into appropriate domains and formats, the network receives optimized input that enhances its ability to accurately distinguish signal from noise, ensuring both efficiency and precision in the final extraction.
Data Source
AI summary
Seismic exploration methods and data processing apparatuses employ a deep neural network to remove seismic interference (SI) noise. Training data is generated by combining an SI model extracted using a conventional model from a subset of the seismic data, with SI free shots and simulated random noise. The trained DNN is used to process the entire seismic data thereby generating an image of subsurface formation for detecting presence and/or location of sought-after natural resources.