We present the code used for the training and subsequent analysis of the recurrent neural network (RNN) model TAROQQO, used to forecast turbulence in free-space links. The results of our model, which considers free-space channels over the city of Ottawa, are featured in our preprint on arXiv (https://arxiv.org/abs/2406.14768).
- nn_architecture.py implements the RNN architectures to be trained.
- train.py prepares the dataset and trains the network.
- ForecastWindow.ipynb analyzes the model prediction on consecutive datasets.
- FeatureImportance.ipynb calculates the importance of features using the Permutation Feature Importance (PFI) technique.
Use of the train.py script requires configuration of the 'train.yaml' configuration file; it can be configured to train networks of arbitrary complexity, as well as to adjust the input series duration & output time resolution.
