Airspace-Encounter-Models/em-model-manned-bayes: October 2021

The Bayesian network encounter models are a collection of MATLAB scripts that produce random samples from models of how different aircraft behave, as previously documented in MIT Lincoln Laboratory technical reports. All these models were originally developed by MIT Lincoln Laboratory. Majority of these samples are of one independent aircraft track and a single sample is insufficient for a complete synthetic encounter.

Each manned aircraft model model is a Bayesian Network, a representation of a multivariate probability distribution as a directed acyclic graph. Models are trained using aircraft operational data derived from radar or other sensing system flight track data.

This release introduces object-oriented programming in MATLAB with a class for a generic encounter model and classes specific to the uncorrelated conventional aircraft and correlated terminal models. These classes enable the user to easily read in the ASCII parameter files with improved input handling. New example run scripts are included to demonstrate how to use the OOP classes. With a few lines of code, the uncorrelated and terminal model can be sampled or generate aircraft trajectories in a local Cartesian coordinate system. The uncorrelated model also supports generating tracks and translating them into a geodetic coordinate system. Rejection sampling based on speed has also been improved for the uncorrelated encounter model by defining minimum and maximum permitted speeds based on the individual probability distribution of the speed bins, rather than the minimum and maximum speeds of the overall model structure. Additional improvements to rejection sampling are scheduled for a future release.

Classes are not yet available for the RADES-based correlated, ETMS-based due regard, DFDR-based HAA, and most unconventional models.