关键词: 强化学习, 智能化测绘, 关键路径, 防测绘, 对抗扰动

Abstract: The intelligent network mapping method based on Deep Reinforcement Learning (DRL) was developed to model the network mapping process as a Markov Decision Process (MDP) and train the attacking agent using trial-and-error learning to identify critical network paths and obtain network topology information. However, traditional network anti-mapping methods were usually based on fixed rules, making them ineffective against the dynamic behavioral strategies of DRL agents during the mapping process. To address this, a network anti-mapping method, Adaptive Interference Perturbation (AIP), was proposed to defend against intelligent network mapping attacks. First, the method predicted traffic conditions using historical traffic sequence data and calculated gradient information based on the differences from real traffic data. This gradient was used to generate adversarial perturbations, which were injected back into the original traffic samples to produce adversarial examples. Then, a feature reconstruction approach that combined traffic posture and routing state was adopted. Through an iterative process, the sparse dictionary was dynamically optimized, enabling sparse transformation of traffic data. Finally, this approach retained the impact of perturbations on critical traffic features while enhancing their stealthiness, further hindering the agent's decision-making process. Experimental results show that compared to standard perturbation defense methods such as Fast Gradient Sign Method (FGSM) defense and random manipulation defense, AIP significantly increases the attacker's susceptibility to perturbations when the network traffic intensity exceeds 25%, results in greater differences in the link weights of the network topology, and has a noticeable impact on network delay. Furthermore, compared with static and dynamic honeypot deployment strategies, AIP demonstrates continuous confusion of attackers, making it difficult to identify critical network paths, which ensures network delays remain within a controlled range and achieves certain improvements in defense efficiency and stability.

Key words: reinforcement learning, intelligent mapping, critical path, anti-mapping, adversarial perturbation