Design of Adaptive Network Defense Mechanism Combining Reinforcement Learning
Pengyu Cui · 2025
In the context of increasingly severe cybersecurity threats, traditional defense mechanisms struggle against dynamic and unknown attacks. This study proposes an adaptive network defense mechanism integrating a Deep Q-Network (DQN) with RL. The DQN model processes real-time network traffic features to detect attacks via Q-value updates, while RL dynamically optimizes defense strategies through environmental interaction and reward mechanisms. Experiments simulated diverse attacks in a virtual enterprise network. Results demonstrated high detection accuracy (average F1-score >0.90), rapid response (≤2 seconds), and defense success rates of 89–93%. Notably, the system achieved adaptability indices of 0.82 –0.88 against unknown attacks, outperforming static methods. The mechanism’s self-learning capability enables real-time adjustments to emerging threats, significantly enhancing defense flexibility. This work advances intelligent cybersecurity by merging deep learning with RL, offering a robust solution for dynamic network environments. Its success in balancing computational efficiency and defense efficacy highlights practical applicability in high-security sectors like finance and enterprise networks.