PhD dissertation: SEISMIC SOURCE AND ELASTIC FULL-WAVEFORM INVERSION USING DISTRIBUTED ACOUSTIC SENSING AND PERFORATION SHOTS IN UNCONVENTIONAL RESERVOIRS

• Description
• Installation with Docker
• Installation with Docker with GPU enabled
• Running jupyter notebook
• Reproducing thesis results
• Getting the input field data

Description

This project aims at reproducing the results in the PhD dissertation "SEISMIC SOURCE AND ELASTIC FULL-WAVEFORM INVERSION USING DISTRIBUTED ACOUSTIC SENSING AND PERFORATION SHOTS IN UNCONVENTIONAL RESERVOIRS" published by Stanford University in 2023 and authored by Milad Bader. The dissertation can be accessed here.

Installation with Docker

Build the docker image (it should take about 15 minutes)

docker build -f Dockerfile -t phd23 .

Then run a container

docker run -it -p 8080:8080 phd23

By default a bash shell will be opened at /home inside the container.

Installation with Docker with GPU enabled

FWI (in 2D) will run much faster if a CUDA enabled GPU is available. Repeat the process above after replacing 'Dockerfile' with 'Dockerfile_gpu'.

Running jupyter notebook

Run jupyter notebook from within the container

jupyter notebook --ip 0.0.0.0 --port 8080 --no-browser --allow-root &

Open the browser at localhost:8080/​ and use the printed token to authenticate.

Reproducing thesis results

To reproduce the field data results figuring in the Phd dissertation, the starting DAS data and well logs must be copied into /home/thesis/input_data (see next section). Otherwise, only the synthetic data results and corresponding figures in Chapters 2, 3, and 5 can be reproduced. In all cases, it is best (and even necessary) to reproduce the results sequentially (Chapter 2 then 3 then 4 etc.).

• Chapter 2

Go to /home/thesis/chapters/ch2/notebooks and run both notebooks (in whichever order).

• Chapter 3

Go to /home/thesis/chapters/ch3/notebooks and run the notebook.

• Chapter 4

Go first to /home/thesis/chapters/ch0/notebooks and run the notebook. This requires the input_data directory. Then, go to /home/thesis/chapters/ch4 and run

mkdir -p dat fig
make all

from the terminal. Peek into the Makefile to see what is happening. Green's functions may take a while when using limited computing power. Finally, go to /home/thesis/chapters/ch4/notebooks and run the notebooks Radiation_pattern, MT_inversion_unstimulated, MT_inversion_stimulated, and MT_inversion_3d.

• Chapter 5 (and Appendix D)

Go to /home/thesis/chapters/ch5. To generate synthetic FWI results and figures, run

mkdir -p dat fig
make all_synthetic

then go to the notebooks directory and run the notebook Synthetic_figures. For the field data results and figures, assuming input_data is available and Chapter 4 results have been generated, follow these steps:

• run make prepare_data
• run the notebook Data_selection
• run make prepare_avo
• run the notebook AVO_estimation
• run make fwi_unstimulated and make fwi_stimulated (it will take a while)
• run the notebooks Dispersion_analysis, Field_figures_thesis, and ST_effects

• Chapter 6

The computation and results in this chapter were performed using Stanford HPC cluster Sherlock and could take hours to days to complete. Go to /home/thesis/chapters/ch6, run mkdir -p dat fig, then follow the instructions in the Makefile. For the FWI results, it may be better to submit each computation job manually using Slurm (check for the corresponding targets in the make files). Slurm script examples are provided.

• Appendix E

Go to /home/thesis/chapters/ch99/notebooks and run the notebook Attenuation_modeling.

Getting the input field data

The field data is stored in a separate git repository as a .tar lfs file. If permissions are granted, follow these steps to copy the data into the appropriate location:

cd /home
git clone https://premonition.Stanford.EDU/nmbader/phd_dissertation.git
cd /home/phd_dissertation
git lfs pull --include="input_data.tar"
tar -xvf input_data.tar
mv input_data /home/thesis/
cd /home
rm -r /home/phd_dissertation