Global Optimization for Resource Allocation in a Networked-Surveillance Radar

In this paper, Bayesian optimization is compared to a genetic algorithm as applied to a resource allocation problem in a networked surveillance radar setting. In the chosen setting, the requirement is to optimize the conflicting objectives of both improved detection and improved tracking. As such, the objective function is a weighted sum of two parts. The first part represents the number of undetected targets and the second part represents the sum of tracked target covariance matrix determinants. The Bayesian optimization algorithm outperforms the genetic algorithm, by producing better solutions with the same number of objective function evaluations. Additionally, for a specific choice of objective function weights, better tracking performance and similar detection performance is achieved compared to the case with a uniform scan time over all radar sectors.