Security-Aware Routing Optimization in TSNs via GCN-based Deep Reinforcement Learning

Driven by Industry 4.0 advancements, industrial networks face heightened demands for communication determinism and security. Time-Sensitive Networking (TSN), with its low-latency and highly reliable transmission capabilities, has emerged as a core enabling technology for next-generation industrial systems. However, traditional routing protocols fail to guarantee real-time scheduling performance, and encryption delays introduced by TSN security mechanisms may violate transmission deadlines. To address these challenges, this paper proposes a security-aware routing optimization framework integrating Graph Convolutional Networks (GCNs) and Deep Reinforcement Learning (DRL). Our approach employs a flow-priority-driven encryption strategy to dynamically assign differentiated encryption strengths to High Time-Sensitive (HTS) and Audio-Video Bridging (AVB) flows, while the GCN-DRL model jointly learns topological features and dynamic routing policies under latency constraints to achieve secure and controllable end-to-end latency minimization. Experimental results demonstrate 38.6% higher scheduling success rates and 19.3%/12.7% reductions in HTS/AVB flow delays within multi-priority mixed-traffic scenarios, effectively synergizing real-time performance with security enforcement.