A smart agent-based combat modeling method using model and data
Simulation modeling,a classical model driven physics-based method,has always gained high priority in acquisition,design,and evaluation of various combat systems.In recent years,inspired by big data and artificial intelligence,more and more simulation modelers have paid attention to the combined use of data-driven data modeling methods.In this context,this research investigated the combat system modeling literature from model-driven,data-driven,and hybrid driven perspective,respectively.Thereafter,we proposed a model and data hybrid driven intelligent modeling approach in consideration of limitations of using simulation modeling or data modeling alone.Firstly,we designed a model and data two-wheel driven architecture.Secondly,we applied a novel behavioral modeling method,namely,the function decision tree(FDT),to represent combat behaviors properly.Thirdly,for decision points in a behavioral model,we used the deep reinforcement learning to train smart agents.As a proof of concept,we built a multi-targets assignment scenario of ballistic missile penetration,and the results revealed that the smart agent-embedded ballistic missile significantly increased the ratio of target hits when compared with the traditional rule-based behavioral model.
model and data drivenbehavioral modelingeffectiveness simulationdeep rein-forcement learning
NETL
NSTL
万方数据
维普
