This repository contains the Acropole model for aircraft fuel flow prediction and Python packages for aircraft trajectory processing and fuel flow enhancement.

For a trouble-free installation, creating a dedicated anaconda environment is recommended :

conda create -n acropole python=3.12 -c conda-forge

Activate the conda environment :

conda activate acropole

Install this library:

git clone https://github.com/DGAC/Acropole.git
cd Acropole
pip install .

Here is a minimal working example:

import pandas as pd
from acropole import FuelEstimator

fe = FuelEstimator()

flight = pd.DataFrame({
    "typecode": ["A320", "A320", "A320", "A320"],
    "groundspeed": [400, 410, 420, 430],
    "altitude": [10000, 11000, 12000, 13000],
    "vertical_rate": [2000, 1500, 1000, 500],

    # optional features:
    "second": [0.0, 1.0, 2.0, 3.0],
    "airspeed": [400, 410, 420, 430],
    "mass": [60000, 60000, 60000, 60000]
})

flight_fuel = fe.estimate(flight) #flight.data if traffic flight

Note:

• When the second column is provided, the fuel estimation is more accurate, especially due to derivatives of speeds (acceleration) used in the estimation.
• airspeed is optional. If not provided, it is assumed to be equal to groundspeed. However, accurate airspeed is recommended for better estimation.
• Expected sampling rate is 4 seconds, higher or lower sampling rate might induce noisier fuel flow. Resampling data before estimating fuel flow is recommanded.

For a more complete example, refer to examples/fuel_estimation.ipynb

Aircraft parameters from open data to feed the model are available in data/aircraft_params.csv and loaded by default. Model data is available in models/ and also loaded by default.

You can specify your own data and model file with the following initialization of FuelEstimator. You need to make sure the same column names are in your aircraft CSV file.

fe = FuelEstimator(
    aircraft_params_path="path/to/your/data.csv",
    model_path="path/to/your/SavedModel/",
)

The Acropole model is a neural network built using data from Quick Access Recorder (QAR) from different aircraft types. Evaluation of the model and list of aircraft is available at https://github.com/DGAC/Acropole/tree/main/evaluation/Dense_Acropole_FuelFlow_Scaling.

Comparison of different model performances per phase for 1000 test flights of A320-214 aircraft using real mass and true airspeed.

@inproceedings{jarry2024towards,
  title={Towards aircraft generic Quick Access Recorder fuel flow regression models for ADS-B data},
  author={Jarry, Gabriel and Delahaye, Daniel and Hurter, Christophe},
  booktitle={International Conference on Research in Air Transportation},
  year={2024}
}