Transferring Elastic Low Frequency Extrapolation from Synthetic to Field Data

Ovcharenko, Oleg; Kazei, Vladimir; Peter, Daniel; Alkhalifah, Tariq

doi:10.3997/2214-4609.202112949

Cited by 4 publications

(4 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To mimic realistic noise, we extract the noise imprint directly from the field data, which turned out to be a simple and efficient way to focus training on removing a particular noise pattern from the predicted data. Instead of training the network to cope with diverse random synthetic noise, we train the network in a semisynthetic framework [87,88]. This implies blending synthetic data on waveforms with the real-world noise specific to the target dataset.…”

Section: B Generation Of Semi-synthetic Datasetmentioning

confidence: 99%

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Ovcharenko

Kazei

Alkhalifah

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Low-frequency signal content in seismic data as well as a realistic initial model are key ingredients for robust and efficient full-waveform inversions. However, acquiring low-frequency data is challenging in practice for active seismic surveys. Data-driven solutions show promise to extrapolate low-frequency data given a high-frequency counterpart. While being established for synthetic acoustic examples, the application of bandwidth extrapolation to field datasets remains non-trivial. Rather than aiming to reach superior accuracy in bandwidth extrapolation, we propose to jointly reconstruct low-frequency data and a smooth background subsurface model within a multi-task deep learning framework. We automatically balance data, model and trace-wise correlation loss terms in the objective functional and show that this approach improves the extrapolation capability of the network. We also design a pipeline for generating synthetic data suitable for field data applications. Finally, we apply the same trained network to synthetic and real marine streamer datasets and run an elastic full-waveform inversion from the extrapolated dataset.

show abstract

Section: B Generation Of Semi-synthetic Datasetmentioning

confidence: 99%

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Ovcharenko

Kazei

Alkhalifah

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

show abstract

“…A similar approach was used in Ovcharenko et al . (2021) to improve low frequency content recovery in full‐waveform inversion. This augmentation process yields a training dataset of 36,000 images in total.…”

Section: Data Preparationmentioning

confidence: 99%

“…Therefore, the random noises in the training dataset are likely different from the noises in the total 346 records of microseismic events, which helps avoid the potential problem of the CNN learning based on similar noises instead of the guided waves. A similar approach was used in Ovcharenko et al (2021) to improve low frequency content recovery in full-waveform inversion.…”

Section: Training Data and Its Augmentationmentioning

confidence: 99%

Convolutional neural network‐based classification of microseismic events originating in a stimulated reservoir from distributed acoustic sensing data

Liu

Huff

Luo

et al. 2022

Geophysical Prospecting

View full text Add to dashboard Cite

This study presents a workflow of using a convolutional neural network to automatically classify microseismic events originating from a more productive oil and gasbearing formation (the Eagle Ford Shale), as compared to events originating in less productive formations (the Austin Chalk and Buda Limestone). These microseismic events occur due to hydraulic stimulation and are recorded by fibre optic-distributed acoustic sensing measurements from a horizontal monitoring well. The convolutional neural network is trained to recognize guided wave energy in distributed acoustic sensing seismograms, since microseismic events originating within or close to a lowvelocity reservoir (such as the Eagle Ford) generate significant guided wave energy. The training of convolutional neural network is conducted using synthetic seismograms overlain with real noise profiles from field data. Field events with guided waves are then classified by the convolutional neural network as occurring within or close to the Eagle Ford, while events without guided waves are classified as occurring far outside the Eagle Ford. Noise attenuation steps (including a bandpass filter, median filter and non-local means filter) are implemented to increase the signal-to-noise ratio of the field data and improve the classification accuracy. The accuracy of the convolutional neural network is measured by comparison with labels of the events determined by human inspection of guided wave presence. We also evaluate the impact of different network architectures and noise attenuation methods on classification accuracy. The accuracy and F1-score of the final classification are both 0.85 when tested on a high signal-to-noise ratio subset of the field data. On the complete dataset including low signal-to-noise ratio events, an accuracy and F1-score of 0.80 are achieved. These results demonstrate the high effectiveness of the trained convolutional neural network on guided wave detection and classification of microseismic events inside and outside the Eagle Ford formation.

show abstract

“…We test the approach on synthetic data modeled from 973 realistically, but randomly, generated velocity models (Ovcharenko et al, 2021). Three shot gathers were generated for every model for a total of shot gathers.…”

Section: Pre-training and Fine-tuningmentioning

confidence: 99%

Machine-Learning Seismic Processing Tasks by Fine Tuning a Pre-Trained Attention-Based Neural Network: Storseismic

Alkhalifah

Harsuko

2022

83rd EAGE Annual Conference &Amp; Exhibition Workshop Programme

View full text Add to dashboard Cite

Transferring Elastic Low Frequency Extrapolation from Synthetic to Field Data

Cited by 4 publications

References 13 publications

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Convolutional neural network‐based classification of microseismic events originating in a stimulated reservoir from distributed acoustic sensing data

Machine-Learning Seismic Processing Tasks by Fine Tuning a Pre-Trained Attention-Based Neural Network: Storseismic

Contact Info

Product

Resources

About