Autoencoder Seismic Velocity Inversion Without Labeled Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning seismic velocity inversion methods rely on supervised or semi-supervised learning, requiring labeled data and specific observation system setups, making them difficult to apply in practical engineering due to high non-linearity, reliance on accurate priori information, and inability to perceive location information in seismic data from different setups.
Innovation Solution
A multi-scale unsupervised seismic velocity inversion method using an autoencoder that extracts large-scale information from observation data, embeds location codes into the network, and constructs a convolutional-fully connected network to invert seismic data without real geological models, allowing for the extraction of global information and reducing non-linearity by forming a multi-scale unsupervised loss function.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised or semi-supervised learning is used for seismic velocity inversion, then the inversion accuracy can be improved through labeled data guidance, but the method becomes difficult to apply in practical engineering due to the difficulty of obtaining real velocity models as labels
Solution Approach 1:
The patent applies self-service by using the seismic observation data itself to guide the inversion process without requiring external labeled velocity models. The loss function is constructed directly from the seismic data, enabling the network to learn and invert velocity models autonomously without human-provided labels or complex priori information.
2Productivity
If deep learning networks are used for seismic velocity inversion, then the inversion speed and automation can be improved, but the networks cannot perceive location information of shot points and receiver points, making them strict to observation system layouts
Solution Approach 1:
The patent applies preliminary action by adding location codes to the seismic observation data before inputting it into the neural network. This preprocessing step embeds spatial information about shot points and receiver points into the data, enabling the network to perceive location information and adapt to different observation system layouts without modifying the network structure.
3Reliability
If traditional multi-scale inversion is used to reduce non-linearity, then the dependence on initial models can be reduced, but the process of extracting low-frequency information through filtering algorithms is complex and time-consuming
Solution Approach 1:
The patent replaces the mechanical filtering algorithm system with a deep learning-based autoencoder system. Instead of using traditional signal processing filters to extract low-frequency information, the patent uses an autoencoder network to automatically learn and extract multi-scale features from the seismic data, significantly reducing processing time while maintaining the ability to reduce non-linearity dependence.
4Ease of manufacture
If unsupervised inversion with background velocity models is used, then the need for labeled data is reduced, but the method still requires accurate apriori information and complex processes to obtain background velocity models
Solution Approach 1:
The patent extracts only the essential low-frequency information needed for inversion from the seismic data using the autoencoder, removing the need for complex background velocity model construction processes. By focusing on extracting critical features rather than requiring complete background models, the method simplifies both data requirements and processing complexity while maintaining inversion effectiveness.
Data Source
AI summary
A multi-scale unsupervised seismic velocity inversion method based on an autoencoder for observation data. Large-scale information is extracted by the autoencoder, which is used for guiding an inversion network to complete the recovery of different-scale features in a velocity model, thereby reducing the non-linearity degree of inversion. A trained encoder part is embedded into the network to complete the extraction of seismic observation data information at the front end, so it can better analyze the information contained in seismic data, the mapping relationship between the data and velocity model is established better, then the inversion method is unsupervised, and location codes are added to the observation data to assist the network in perceiving the layout form of an observation system, which facilitates practical engineering application. Thus a relatively accurate inversion result of the seismic velocity model when no real geological model serves as a network training label can be achieved.