This is a python library to identify the air traffic flows in a ATC sector/airspace by applying a progressive 2D/3D trajectory clustering method based on DBSCAN.
[1] L. Basora, V. Courchelle, J. Bedouet and T. Dubot. Occupancy Peak Estimation from Sector Geometry and Traffic Flow. Proceedings of the SESAR Innovation Days, 2018
The clustering method is described in this paper as part of a larger framework to estimate the occupancy peak (capacity) of a ATC sector based on its traffic flow and geometric characteristics.
[2] X. Olive and L. Basora. Identifying Anomalies in past en-route Trajectories with Clustering and Anomaly Detection Methods. Proceedings of the 13th Air Traffic Management R&D Seminar, 2019
The same clustering method has been also used in this ATM Seminar paper in order to identify the 3D flows in a sector as a preliminary step to detect trajectory anomalies in the flows.
Although sectflow is not directly dependent on the traffic library, it has been designed to work closely integrated with it. See traffic installation instructions here.
The only necessary dependencies are Numpy and scikit-learn.
To install sectflow and its dependencies run:
python setup.py install
The demo_clustering notebook illustrates how to use the library. Also, it is important to read the clustering documentation in the traffic library to understand how both libraries work together.
The traffic features to be used for clustering are specified in the features parameter of both the TrajClust constructor and the clustering method of the Traffic class.
Thus, traffic flows can be identified in 2D if only features x an y (or latitude and longitude) are provided. Also, 3D flows can be identified by including altitude (or log_altitude) in the parameter features.
Please note that in any case x an y are mandatory features, even if latitude and longitude are also provided. This is because the coordinate projections are needed for internal flow computation. If the traffic dataset does not contain the projected x an y coordinates, they will be automatically computed by the traffic library before calling sectflow fit method.
These are only some examples of 2D/3D plots of the cluster centroids and trajectories you can generated from the library.
