Threat assessment using Bayesian networks

Abstract

While threat assessment for air defence can be implemented using several methods, Bayes nets have advantages over other methods (e.g., logical or fuzzy) because they employ consistent reasoning and have representations of uncertainty that are compatible with the more efficient tracking and data fusion algorithms. In this paper we present a Bayes net based threat assessment algorithm that uses soft decisions (target state estimates and their measures of uncertainty) from a tracking and data fusion module to evaluate the threat posed by a given intruder on a specified asset. The threat level is represented by a continuous variable between zero (no threat) and one (maximum threat) that takes into account the intent and capability of the intruding air target. Following the development of the threat variable, this paper presents the Bayesian network and the linear Gaussian approximations necessary for its implementation using Murphy's Bayes Net Toolbox. Numerical results showing the be-havior of each node variable are shown for a selected asset-intruder pair. Because all continuous nodes are assumed Gaussian, the results are highly conservative and can therefore be improved upon. © Commonwealth of Australia 2003.