关键词: 杀伤链, 复杂网络, 关键边, Q-learning

Abstract: This paper proposes an adaptive key edge detection framework based on deep reinforcement learning to address the challenge of identifying key edges in air based kill chain networks. This method first uses complex networks to model the air based kill chain; Secondly, a template kill chain search method is proposed by introducing the idea of backtracking search; Once again, based on deep reinforcement learning, a key edge exploration paradigm was constructed that combines a layered experience replay mechanism with a dynamic ε - greedy strategy. Through multi-dimensional state representation, precise localization of key edges was achieved. The experiment shows that in the simulation of air based kill chain network testing, the Top-10 recognition accuracy of this method reaches 85%, which is 89.5% higher than the traditional betweenness centrality method and 19.7% higher than the benchmark of deep Q-network (DQN); In terms of network robustness indicators, the global efficiency reduction rate η is 37.4% higher than the edge betweenness centrality, and the prediction error of the maximum connected component retention rate is controlled within 5%.

Key words: kill chain, complex network, key edge, Q-learning