KNI-DRL: Key Nodes Identification Method Based on Deep Reinforcement Learning in Inter-domain Routing Networks

The inter-domain routing system plays a crucial role in ensuring the security of the Internet. Identifying and safe-guarding key nodes within the inter-domain routing network is essential to maintain its security and stability. Currently, existing methods for identifying key nodes in the inter-domain routing network fail to strike the optimal balance between node accuracy and speed. Moreover, almost all machine learning-based methods for identifying key nodes require supervised learning, which is not well-suited for accurately defining labels in inter-domain routing networks. A novel approach, KNI-DRL, is proposed in this paper to address the aforementioned issues by identifying key nodes in inter-domain routing networks. The fundamental concept of this approach involves the utilization of an unsupervised deep reinforcement learning model, which can be trained purely on small synthetic networks and then applied to large-scale node inter-domain routing networks. Effective path change is proposed as a reward function to evaluate the importance of nodes in the inter-domain routing network based on their business relations and routing rules. Experiments conducted on inter-domain routing networks of varying scales have demonstrated that KNI-DRL achieves an average node identification accuracy of 80%. In an inter-domain routing network comprising over 10,000 nodes, this method can accomplish the evaluation within a mere 5 seconds, which is nearly 1/40 of the speed of the current optimal method.