Bayesian Decision Model for Battle Damage Assessment

摘要

Battle damage assessment (BDA) is critical to success in any air campaign. However, Desert Storm highlighted numerous deficiencies in the BDA process, and operations since Desert Storm continue to point out weaknesses. We present a review of the Phase I BDA decision, or physical damage assessment, and model the decision process using a Bayesian belief network. Through subject matter expert (i.e., the targeteers) elicitation sessions, imagery was found to be critically important to the BDA process yet this information is generally not retained. This use of 'perfect information' is delineated in the BDA process models. We proposed a methodology based on Bayesian belief networks for incorporating this perfect information. We demonstrate the Bayesian belief network's capability to update conditional probability distributions using data generated in real world operations. This capability allows the network's conditional distributions to evolve, increasing model accuracy and reducing uncertainty in the decision.