Seismic Data Bandwidth Extension via Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Seismic datasets in oil and gas exploration have limited temporal frequency bandwidth, resulting in low resolution and inability to accurately depict subtle geological features and thin beds, which hinders the interpretation of subterranean regions.
Innovation Solution
A method involving the generation of synthetic seismic datasets with broader bandwidth, using machine-learned models like convolutional neural networks to enhance the bandwidth of acquired seismic data, allowing for better resolution and identification of hydrocarbon reservoirs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional seismic survey methods are used, then the acquisition process is straightforward, but the temporal frequency bandwidth is limited resulting in low resolution
Solution Approach 1:
The patent creates synthetic copies of seismic data through neural network training. A deep neural network is trained on synthetic seismic datasets where the network learns to map low-bandwidth input data to high-bandwidth output data. The trained model then generates extended bandwidth seismic datasets by processing actual seismic data, effectively creating a high-resolution copy from low-resolution input without requiring additional physical measurements
Solution Approach 2:
The patent transforms the bandwidth parameter of seismic data through machine learning. The neural network is trained to change the frequency content parameters of the seismic signal, extending the temporal frequency bandwidth from a limited range to an extended range. This parameter transformation allows the system to produce high-resolution seismic images with enhanced bandwidth characteristics from standard acquisition data
2Measurement precision
If the bandwidth of seismic datasets is increased to improve resolution, then subsurface features are better resolved, but the complexity of data processing increases
Solution Approach 1:
The patent replaces traditional mechanical signal processing methods with an intelligent system based on deep neural networks. Instead of using conventional filtering, spectral expansion, or deconvolution techniques that require complex parameter tuning and multiple processing stages, the system uses a trained neural network model that automatically performs bandwidth extension through learned transformations, simplifying the processing workflow while achieving superior results
Solution Approach 2:
The patent performs preliminary training of the neural network model using synthetic datasets before applying it to actual seismic data. During this offline training phase, the network learns the complex mappings between low and high bandwidth seismic characteristics. Once trained, the model can be rapidly applied to production data without requiring complex real-time processing, as the heavy computational work was already performed during the preliminary training stage
Data Source
AI summary
A method to extend the bandwidth of seismic data. The method includes collecting a seismic dataset from a seismic survey conducted over a subterranean region of interest and generating a plurality of synthetic datasets, where each synthetic dataset comprises an input seismic dataset with a first bandwidth and an associated target seismic dataset with a second bandwidth. The second bandwidth is broader than the first bandwidth. Further, each synthetic dataset is generated directly in a time domain and includes an event, where the event includes a geometric shape and a wavelet. The method further includes splitting the plurality of synthetic datasets into a training set, selecting a machine-learned model, training the machine-learned model using the training set, and using the machine-learned model to produce an extended bandwidth seismic dataset from the seismic dataset. The method further includes determining a location of a hydrocarbon reservoir using the extended bandwidth seismic dataset.