PIANN-FWI for estimating the Marmousi velocity model

In this repository, I implemented the physics-informed neural network (PINN) for full-waveform inversion. This PINN can be implemented with or without attention block. The architecture of their study is shown in the following figure.

.. figure:: /readme_files/architecture.png :alt: architecture

For running the code, you should use this notebook <https://github.com/AmirMardan/pinn_fwi/blob/main/pinn_fwi.ipynb>. The required parameters for running this notebook should be set in this config <https://github.com/AmirMardan/pinn_fwi/blob/main/config.py> file.

Note: I have commented cell 3 in this notebook, you should run this cell whenever you change an acquisition parameter (and for the first time using the codes).

Note: Please use the requirements <https://github.com/AmirMardan/pinn_fwi/blob/main/requirements.txt>__ file (written in the jupyter file) to install the packages with specified versions to be sure everything works.

.. code:: console

pip install -r requirements.txt

In this repo, there are four scripts for running FWI: | 1. pinn_fwi.py <https://github.com/AmirMardan/pinn_fwi/blob/main/pinn_fwi.py>__ for performing PINN- or PIANN-FWI. | 2. original_fwi.py <https://github.com/AmirMardan/pinn_fwi/blob/main/original_fwi.py>__ for running the conventional FWI (Not available). | 3. pinn_for_init.py <https://github.com/AmirMardan/pinn_fwi/blob/main/pinn_for_init.py>__ for performing PINN- or PIANN-FWI to create an initial model and use that to perform the conventional FWI (Not available). | 4. pinn_fwi.ipynb <https://github.com/AmirMardan/pinn_fwi/blob/main/pinn_fwi.ipynb>__ for performing PINN- or PIANN-FWI, but this notebook might not be updated.

The result of running this code for 22 shots with 2500 epochs on the Marmousi model is shown in the following figures.

|res| For a faster convergence (300 epochs), I considered geophones around the model and the results are |with_init| where the hybrid method is using the PIANN-FWI for creating only initial model.

Reference:

::

@inproceedings{mardan2024piann_eage, title = {Physics-informed attention-based neural networks for full-waveform inversion}, author = {Mardan, Amir and Fabien-Ouellet, Gabriel}, year = {2024}, booktitle = {85$^{th}$ {EAGE} Annual Conference & Exhibition}, publisher = {European Association of Geoscientists & Engineers}, pages = {1-5}, doi = {} }

.. |res| image:: /readme_files/marmousi_clean.png .. |with_init| image:: /readme_files/image2024_marmousi_clean.png