GAN_SR_wind_field

Super resolve 3D wind flow based on data from the HARMONIE-SIMRA model, https://asmedigitalcollection.asme.org/OMAE/proceedings-abstract/OMAE2017/57786/V010T09A051/282088?redirectedFrom=PDF.

Dependencies for the python 3.9 environment are listed in requirements.txt.

The model is trained, tested or used in a parameter seach by running the run.py file. Hyperparameters are set in the Config folder.

The model and test results, are thoroughly described in the following master thesis: [INSERT LINK WHEN AVAILABLE]

This work is besed on Vesterkjær 2019's implementation of ESRGAN, https://github.com/eirikeve/esrdgan/tree/master. It is therefore subject to APACHE licence, described in the LICENSE.txt file

Running the code

This code was developed using Python version 3.9. There will be dependency conflicts if other versions are used.

Installing everything

Clone the repository using:

git clone [email protected]:inductiva/GAN_SR_wind_field.git

Create a virtual environment to solve solve any clashes with the library versions used here and anything else that might be installed in your own system:

python3.9 -m venv .env
source .env/bin/activate

After creating and activating the virtual environment install all the requirements of the project using:

pip install -r requirements.txt

Downloading the training data

First modify the config files to specify the extend of the training data, by specifying the start and end date. Other training parameters and hyper-parameters can also be edited here.

Once the configuration file is ready, download the raw data and then create training, validation and test datasets from HARMONIE-SIMRA data using the following command.

python run.py --download

Train the model

Start model training using the following comand

python run.py --train

The trained model can be now tested and used by using --test and --use flags.

The z interpolation feature cannot be used as of now. It is being fixed.