2025-AGILE-AircraftTrajectoryPrediction

Code supporting the experiements presented on the paper "Experiments on Geospatial Data Modelling for Long-Term Trajectory Prediction of Aircrafts". The paper was accepted on the AGILE 2025 conference. The aim of the project was to explore geospatial data modeling, specially the H3 index, for trajectory predicitons of Aircraft using a CVAE-LSTM model.

Pre-requisites

• Create a virtual environment with Python 3.11.10
• Create an account in Opensky (optional, for using the download_raw.py)
  • store the credentials inside secrets.conf

Installation

For installing all the dependencies, you can use the installation script on this repository. To do so, please execute the following code on a terminal inside the folder of this project on a virtual environment with Python 3.11.10:

chmod +x install.sh
bash install.sh

Reproduce the Experiments

To reproduce the experiments of the paper, you can use the already downloaded data inside src/data/raw/. However, more data, or data from different airports can be download using the code provided in the download_raw.py file. To download the exact same dates used on the experiments for the paper, you can use the code as it is by executing the following commands:

source .venv/bin/activate
cd ./src/data/
python3 download_raw.py 

Experiments

All the experiments were executed using the code on experiments.ipynb. To try different configurations, you can change the parameters in the Experiment Setup section.

Different H3 Resolutions

Here, we fixed IN_SEQ_LEN, TGT_SEQ_LEN, and SAMPLING_TIME, while trying H3_RESOLUTION between 5 and 11. To use different resolutions, change the H3_RESOLUTION parameter, to try different geodata representations, change the GEODATA_REPRESENTATION parameter.

Example of Experiment Setup:

    GEODATA_REPRESENTATION : Literal["h3", "xy", "latlon"] = "h3"
    H3_RESOLUTION : Literal["5", "6", "7", "8", "9", "10", "11"] = "5"
    IN_SEQ_LEN = 5 #Fixed
    TGT_SEQ_LEN = 60 #Fixed
    SAMPLING_TIME = 60 #Fixed

Different Long-term Prediction Ranges

Here, we fixed IN_SEQ_LEN, and H3_RESOLUTION, while trying different TGT_SEQ_LEN, and SAMPLING_TIME. To use different geodata representations, change the GEODATA_REPRESENTATION parameter.

Example of Experiment Setup:

GEODATA_REPRESENTATION : Literal["h3", "xy", "latlon"] = "h3"
H3_RESOLUTION : Literal["5", "6", "7", "8", "9", "10", "11"] = "5" #Fixed
IN_SEQ_LEN = 2 #Fixed
TGT_SEQ_LEN = 60
SAMPLING_TIME = 60

The different combinations we used for TGT_SEQ_LEN and SAMPLING_TIME are shown in the table bellow:

TGT_SEQ_LEN	SAMPLING_TIME
60	60
30	60
120	30
60	30
180	20
90	20
360	10
180	10
720	5
360	5