A Reinforcement Learning Model to Adaptive Strategy Determination for Dynamic Defense

To combat advanced unknown cyber-attacks, dynamic defense technology is constantly being studied. In this research, aiming at the balance of defense benefits and costs in dynamic defense under heterogeneous redundancy architecture, we propose a dynamic defense strategy model based on reinforcement learning, which can adaptively select defense strategies based on adversary attributes. The model is verified by the third-party attack dataset Blackenergy attack sample, which shows that the model can achieve adaptive defense against unknown attack events. For the contribution of this research, it provides the foundation for future research of intelligent defense technology.