基于Nash-Q的网络信息体系对抗仿真技术

摘要

武器装备体系作战仿真研究隶属于复杂系统研究范畴,首次对基于Nash-Q的网络信息体系(network information system-of-systems,NISoS)对抗认知决策行为进行探索研究.Nash-Q算法与联合Q-learning算法具有类似的形式,其区别在于联合策略的计算,对于零和博弈体系作战模型,由于Nash-Q不需要其他Agent的历史信息即可通过Nash均衡的求解而获得混合策略,因此更易于实现也更加高效.建立了战役层次零和作战动态博弈模型,在不需要其他Agent的完全信息时,给出了Nash均衡的求解方法.此外,采用高斯径向基神经网络对Q表进行离散,使得算法具有更好的离散效果以及泛化能力.最后,通过NISoS作战仿真实验验证了算法的有效性以及相比基于Q-learning算法以及Rule-based决策算法具有更高的收益,并且在离线决策中表现优异.%Battle simulation for weapon equipment sysem-of-systems (SoS) belongs to the research category of complex system and the confrontation cognition of network information system-of-systems (NISoS) based on Nash-Q technology is researched.The form of the Nash-Q is similar with the union Q-learning except the obtaining of the union policy.For the zero-sum game model of the SoS battle simulation,the realization and solution of the Nash-Q model is more effective since the Nash-Q does not need the history action messages of other Agents.The zero-sum game command model for the battle simulation of the tactical command level is built and the solving process of Nash-equilibrium is introduced through the complete information of other Agents is not known.The Gauss radial basis function neural network is used to discrete the Q-table to improve the discrete performance and generalization ability of Nash-Q.Finally,the effectiveness of the algorithm is validated through battle simulation of NISoS.Compared with Q-learning and Rule-based algorithm,the proposed algorithm has higher gains and can be used to off-line decision.