Deep Learning Seismic Velocity Estimation Framework
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current seismic velocity modeling techniques, such as full-waveform inversion, face challenges with non-uniqueness and sensitivity to data acquisition methods, leading to inaccurate subsurface structure identification and hydrocarbon reservoir mapping.
Innovation Solution
A deep learning-based framework that perturbs an initial velocity model to generate multiple synthetic models, simulates seismic data using a forward modeling process, and transforms data into the wavenumber-time domain for training a machine-learned model to predict accurate velocity models robust to seismic survey configurations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional seismic velocity modeling techniques (e.g., full-waveform inversion) are used, then velocity models can be obtained, but the results suffer from non-uniqueness and sensitivity to data acquisition methods
Solution Approach 1:
The patent transforms seismic data from the time-domain to the wavenumber-domain, fundamentally changing the parameter space in which the data is represented. This transformation allows the machine learning model to learn velocity models in a domain that is less sensitive to acquisition configuration variations, thereby improving robustness while maintaining accuracy
Solution Approach 2:
The patent replaces conventional iterative inversion algorithms (mechanical/mathematical optimization systems) with a machine learning-based system. The neural network is trained to directly map transformed seismic data to velocity models, eliminating the non-uniqueness and sensitivity issues inherent in traditional inversion methods
2Adaptability or versatility
If machine learning models are trained on transformed seismic data, then generalization power improves and re-training needs are reduced, but the framework complexity increases
Solution Approach 1:
The patent applies a forward modeling step and data transformation to the training data before feeding it to the machine learning model. This preliminary processing prepares the data in an optimal format that enhances the model's ability to generalize across different survey configurations, reducing the need for re-training
Solution Approach 2:
The patent introduces a forward model as an intermediary component between the velocity models and the seismic data. This forward model generates synthetic seismic data that is then transformed and used for training, creating a controlled training environment that improves model generalization without requiring complex adaptations
Data Source
AI summary
A method which includes obtaining an initial velocity model and perturbing the initial velocity model to form a first plurality of velocity models. The method includes using a forward model to simulate seismic data sets from the first plurality of velocity models and transforming the seismic data sets to the wavenumber-time domain. The method includes training a machine-learned model using the first plurality of velocity models and the transformed seismic data sets, wherein the machine-learned model is configured to accept transformed seismic data. The method includes obtaining a second seismic data set for a subsurface region of interest, wherein the second seismic data set is acquired according to a second survey configuration and transforming the second seismic data set to the wavenumber-time domain. The method further includes processing the second transformed data set with the trained machine-learned model to predict a second velocity model for the subsurface region of interest.