Extended Look-Ahead Sensor Management for Missile Defense with a Bayes Net Battle Model

Our Phase I effort explored using Dynamic Bayesian Networks (DBNs) to predict results of sensor tasking in missile defense Sensor Resource Management (SRM). The DBN output was coupled with engagement planning analysis to take advantage of two proven approaches to improving missile defense SRM; integrating sensor allocation with weapon-target pairing and using extended look-ahead to make optimal use of defense-in-depth tactics. Further, the coupling reduces computational requirements by several orders of magnitude. Testing showed the benefits, individually and in combination, of extending look-ahead horizons and integrating planning tasks, and demonstrated the viability of the approach. The Phase II effort will develop abilities to automatically translate a Hercules System Model into a predictive DBN `battle model'. The battle model topology will minimize the computational cost of predicting sensor results, making it more efficient for this purpose than the Hercules System Model, which is designed to infer target type based on collected evidence. Battle model predictions will be fed into an engagement planner algorithm to rank sensor plans by the Value of Information (VOI) of the data they will collect. Significant effort will be devoted to producing and integrating a series of prototype systems into the Hercules Decision Architecture.