Allocation of Air Resources Against and Intelligent Adversary

摘要

In a battlefield situation, the use of air assets can have a large impact upon the outcome. The problem we consider is allocating scarce resources among activities that conduct pre-strike Intelligence, Surveillance, and Reconnaissance (ISR), take strike actions against, or gather battle damage assessment (BDA) information about a set of targets in order to perform the targeting cycle. We explore methods that combine Partially Observable Markov Decision Processes (POMDPs), which prescribe strike and observation policies, and integer programing formulations, which pick the optimal set of policies given resource constraints. This work adds five major contributions beyond previous work on similar problems. The first improvement is the introduction of allocation decisions for ISR assets, which search out and identify new targets. Also included is a model of an intelligent adversary, specifically representations of regenerative and mobile targets. In addition to incorporating Cheng's Linear Support algorithm for solving two-dimensional targeting POMDPs, we incorporate the Incremental Pruning algorithm to solve higher dimensional POMDPs for target discovery and identification. Finally, we introduce a new initialization technique as well as two integer programming formulations of the targeting cycle problem. We demonstrate the computational benefits of this decomposition through a number of parameter variation tests and targeting cycle vignettes and discuss the qualitative characteristics of the solutions generated.