This Github page hosts the code producing the results published in the paper "Predictive Joint Distribution of the Mass and Speed Profile to Improve Aircraft Climb Prediction".
The trajectory data are automatically downloaded by the script. They are hosted at https://opensky-network.org/datasets/publication-data/climbing-aircraft-dataset.
With this code, you can reproduce the Table 3 of the publication. In order to reproduce the Table, you must have computed the prediction with the GBM method using the repository of a previous publication https://github.com/richardalligier/trc2018.
If you have any problems using the provided code, please feel free to open an issue in this Github repository.
These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.
In order to run the Python3 scripts, you will need to install different packages. These packages can be installed with the command:
pip3 install pandas==0.25.3 numpy==1.17.4 scikit-learn==0.22 torch==1.3.1
You can also setup a conda environment named "aidaat2020" using the environment.yml file:
conda env create -f environment.yml
conda activate aidaat2020
In order to compile the OCaml binaries, you will need to install the OCaml compiler. Using Debian/Ubuntu, just type:
apt-get update
apt-get install ocaml ocaml-native-compilers
To install the project, you just have to clone or download this github repository. To clone this repository, just type:
git clone https://github.com/richardalligier/aidaat2020.git
Before running the scripts you might want to edit the file config. In this file, you can edit where the generated table will be created by modifying the variable TABLE_FOLDER. Likewise, you can edit DATA_PATH, this variable is the folder storing the trajectory data, the generated models and predictions. The trajectory data are automatically downloaded.
If you want to reproduce the Table, you must have computed the prediction with the GBM method using the repository of a previous publication https://github.com/richardalligier/trc2018. The GBM_RES variable in the config will specify the data folder used to reproduce the TRC2018 results.
The hyperparameter selection is done using random search. To speed-up
this process, you can use several GPUs to train several models in parallel. In config file NGPU is the number of
GPUs used and NWORKER_BY_GPU is the number of models trained in
parallel on one GPU.
As a reminder, the predicted distribution is specific, tailored, to each point of each flight. To test the script, you might want to compute it considering only the flights of a given aircraft type. For instance, if you want to compute for the DH8D flights, just type:
make MODELS="DH8D" METHODS="gmm"
To compute the predicted distributions for all the aircraft types, you only have to type (WARNING: Takes a lot of time!!) :
make
It takes several days (maybe a week depending on your computer).
Assuming that the folder GBM_RES contains the GBM predicted values using the publication https://github.com/richardalligier/trc2018, you just have to type:
make tables
- Richard Alligier
This project is licensed under the GPLv3 License - see the LICENSE file for details
- The OpenSky Network for providing and hosting the trajectory data
- FlightAirMap for providing data on Routes and ICAO codes
- World Aircraft Database for providing data on ICAO codes
- NVIDIA Corporation with the donation of the Titan Xp GPU used for this research
The data description is available here.