Evolving optimal and diversified military operational plans for computational red teaming

Abstract

Computational Red teaming (CRT) is a simulation-based optimization application utilized by defense analysts to uncover vulnerabilities of operational plans. In CRT, agent-based simulation models of military scenarios are automatically analyzed and modeled using evolutionary computation techniques. The CRT optimization process aims at identifying simulation models which exhibit emergent system behaviors of interest, e.g., when the adversary (called “Red”) breaks the defensive (“Blue”) strategies. Numerous multiobjective evolutionary algorithms (MOEAs) have been applied to CRT; however, the elitist and converging nature of these Pareto-based optimization algorithms typically leads to the generation of optimal, with respect to the Pareto front, but poorly diversified adversarial operational plans. As a result, the near-optimal alternative strategies are omitted; this considerably limits the applicability of CRT when considering the decision makers point of view. We propose a diversity enhancement scheme for MOEAs which uses the diversity contribution of individual solutions in the aggregated (combining both the objective and decision variable spaces) space to compute the fitness assignment. This feature enables both the exploitation of Pareto-optimal solutions whilst promoting diversification of the solutions in the decision variable space. Our experimental results indicate that this diversity enhancement mechanism can effectively resolve the diversification issue and, ultimately, enhance CRT to assist decision making.