Optimal Sensor Selection and Placement for Perimeter Defense

This paper describes analytical methods for selecting and positioning sensors on terrain to detect, classify and locate events of interest. Our choice of sensors and their time-invariant locations is based on obtaining measured data that is most informative about the event. We adopt a novel system-based approachμminimize not simply the uncertainty of the sensor system's output, but rather minimize the risk of poor performance that arises from decisions made based on it. Our decision function combines the information provided by the sensors and is constructed by minimizing the system's posterior loss; it is therefore an optimal decision function. The sensor positioning problem presents several challenges for optimization methods arising from the intricate nature of sensor occlusion from terrain and the introduction of discrete variables such as the minimum number of sensors. This renders the optimal sensor positioning problem highly nonlinear, nonsmooth, and multi-modal. We will discuss several techniques, including metaheuristic methods in combination with "grid refinement" techniques, response-model-based methods, and novel modifications to problem objectives to handle the above optimization challenges.