计算机科学
调度(生产过程)
强化学习
接头(建筑物)
计算机网络
分布式计算
作业车间调度
布线(电子设计自动化)
人工智能
数学优化
缩小
实时计算
作者
Yiqin Lu,Lizhuo Lin,Weiqiang Pan,Jiancheng Qin
出处
期刊:
日期:2026-05-15
卷期号:: 533-540
标识
DOI:10.1109/ainit70033.2026.11558111
摘要
Joint routing and scheduling of time-triggered traffic in time-sensitive networking (TSN) is challenging due to strict timing constraints and dynamic resource contention under high traffic load. This paper proposes a load-aware Dijkstra-guided deep reinforcement learning (DRL) framework for joint routing and slot scheduling in TSN. Unlike conventional shortest-path guidance based on static hop count, the proposed method dynamically adjusts edge weights according to link utilization and slot availability, thereby providing congestion-sensitive path priors for decision making. In the proposed framework, the DRL agent selects the next legal edge, while time-slot allocation is performed by a heuristic based on a Conflict-Urgency Score (CU_score). The load-aware Dijkstra guidance is integrated into both state construction and action processing, where it serves as a distance prior for edge evaluation and as a fallback safeguard for invalid actions. Experimental results on a typical TSN topology with 100–500 TT flows show that the proposed method consistently outperforms DRLS and heuristic baselines in scheduling success rate, end-to-end delay, network utilization, and load balancing, while incurring only marginal runtime overhead (within milliseconds per flow). Further evaluation on a more complex topology confirms that the method maintains strong performance without significant degradation, demonstrating its robustness across different network structures.
科研通智能强力驱动
Strongly Powered by AbleSci AI