aboulch/tec_prediction

TEC prediction using convolutional recurrent neural networks

Reference

When using this code, cite the related paper:

Ionospheric activity prediction using convolutional recurrent neural networks by Boulch Alexandre and Cherrier Noelie and Castaings Thibaut

@article{boulch2018ionosphere,
  title={Ionospheric activity prediction using convolutional recurrent neural networks},
  author={Boulch, Alexandre and Cherrier Noelie and Castaings Thibaut},
  journal={arXiv preprint arXiv:1810.13273},
  year={2018},
  url={https://arxiv.org/abs/1810.13273}
}

Please note that, the actual citation refers to a pre-print. The submission is under review at IEEE Transaction on Big Data. If the paper is accepted, please update your citation.

Project

This work is part the DELTA research project at ONERA, The French Aerospace Lab. Among its objectives are the development and the promotion of innovative machine learning based approaches for aerospace applications.

TEC prediction

Get the data

The data used for training and testing can retreived at:

ftp://igs.ensg.ign.fr/pub/igs/products/ionosphere/

Convert the data to Numpy

The file convert_raw_to_numpy.py provides utility functions to convert the previously downloaded TEC maps to numpy format.

In order to use it:

• Fill the root_dir with the path to the data directory
• Fill the dest_dir with the path where to put the processed data
• imsize is the image size, (72, 72) is default values
• apply_compensation apply Earth rotation compensation

Main file

Arguments

• seqLength: length of the total sequence (input + prediction)
• seqStart: length of the input sequence
• batchSize: batch size
• cuda: use cuda backend
• test: use network in test mode (training otherwise)
• model: which model to use (simple, unet, dilation121)
• diff: use residual prediction
• target: directory to save the results
• source: directory containing the data

Train a model

Test

License

The license is a dual license. For academic research, the code is released with LGPLv3 license. For commercial purpose, please contact the authors or ONERA. See the license.