Network Defense Strategy Optimization Based on Stochastic Gaming and A3C Deep Reinforcement Learning
The limitation of network resources and the dynamics of attack-defense confrontation make it difficult to select the optimal defense strategy.Therefore,the deep reinforcement learning is introduced into the field of attack and defense stochastic game modeling.By constructing the network attack-defense actor strategy network and critical value network,the stochastic gaming model is combined,the overall architecture of the game decision-making model for network attack-defense is constructed.On this basis,the asynchronous advantage actor-critical(A3C)agent learning framework is introduced to design the defense strategy selection algorithm.In view of the fact that the existing methods do not consider the collusion attacks among the attacker groups,the personality characteristics of group agents are introduced by establishing cooperation factor μ to describe the impact of attacker cooperation on the benefits of attack-defense strategies as well as that on the selection of defense strategies.Therefore,the constructed game decision-making model more conforms to the realistic attack-defense situation better.The experimental results show that strategy calculating speed of the proposed method is better than the existing method.At the same time,as the attack cooperation relationship is considered,which can be used to analyze the impact of the cooperation relationship among attacker groups on the decision-making of the defenders.The defense strategy selection is more targeted and the expected defense benefits are expected to be higher.
NETL
NSTL
万方数据
