Landscape Dynamics in Multi-agent Simulation Combat Systems

摘要

Traditionally optimization of defence operations are based on the findings of human-based war gaming. However, this approach is extremely expensive and does not enable analysts to explore the problem space properly. Recent research shows that both computer simulations of multi-agent systems and evolutionary computation are valuable tools for optimizing defence operations. A potential maneuver strategy is generated by the evolutionary method then gets evaluated by calling the multi-agent simulation module to simulate the system behavior. The optimization problem in this context is known as a black box optimization problem, where the function that is being optimized is hidden and the only information we have access to is through the value(s) returned from the simulation for a given input set. However, to design efficient search algorithms, it is necessary to understand the properties of this search space; thus unfolding some characteristics of the black box. Without doing so, one cannot estimate how good the results are, neither can we design competent algorithms that are able to deal with the problem properly. In this paper, we provide a first attempt at understanding the characteristics and properties of the search space of complex adaptive combat systems.