BGP Anomaly Events Detection Method Based on Deep Reinforcement Learning

BGP is the most important protocol for inter-domain routing network connectivity, and plays a key role in the global network interconnection. However, the design flaws of BGP make it vulnerable to attacks or failures, leading to network fluctuations or even outages. Rapid detection of BGP anomalies and timely implementation of effective countermeasures are essential for maintaining internet security and stability. Current research on BGP anomaly detection has many problems such as poor generalization ability and high training cost, which makes it difficult to apply in real-world scenarios. To address theses problems, a semi-supervised BGP anomaly event detection method based on deep reinforcement learning, BAD-DRL, is proposed in this paper. This method trains an agent for anomaly detection scenarios, which autonomously learns the latent features of unlabeled data through an internally designed reward function and optimizes the agent's behavioral strategies through an externally designed reward function. Experiments show that the proposed method achieves an improvement in average accuracy of at least 1.35% on datasets from the same source and at least 7.57<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> on datasets from different sources.