Bayesian Search Under Dynamic Disaster Scenarios

Abstract

Search and Rescue (SAR) is a hard decision making context where there is available a limited amount of resources that should be strategically allocated over the search region in order to find missing people opportunely. In this thesis, we consider those SAR scenarios where the search region is being affected by some type of dynamic threat such as a wilder or a hurricane. In spite of the large amount of SAR missions that consistently take place under these circumstances, and being Search Theory a research area dating back from more than a half century, to the best of our knowledge, this kind of search problem has not being considered in any previous research. Here we propose a bi-objective mathematical optimization model and three solution methods for the problem: (1) Epsilon-constraint; (2) Lexicographic; and (3) Ant Colony based heuristic. One of the objectives of our model pursues the allocation of resources in riskiest zones. This objective attempts to find victims located at the closest regions to the threat, presenting a high risk of being reached by the disaster. In contrast, the second objective is oriented to allocate resources in regions where it is more likely to find the victim. Furthermore, we implemented a receding horizon approach oriented to provide our planning methodology with the ability to adapt to disaster's behavior based on updated information gathered during the mission. All our products were validated through computational experiments.