基于双重图注意力网络的光路传输质量智能预测技术研究
Research on Optical Transmission Quality Prediction Based on Hierarchical Graph Attention Network
闫欢 1谷志群 1纪越峰1
作者信息
- 1. 北京邮电大学信息与通信工程学院，北京 100876
- 折叠
摘要
弹性光网络中传输质量（QoT）的快速精准估算,是实现资源高效分配与智能管控的关键。传统解析模型计算成本高昂,常规机器学习方法又易忽略网络拓扑的空间相关性与非线性损伤效应,难以适配动态网络需求。针对上述问题,本文提出一种基于双层图注意力网络（HGAT）的 QoT 在线估算模型：将物理光网络映射为业务依赖图,设计光路内与光路间双重注意力机制,实现对单光路物理劣化瓶颈和跨光路非线性串扰的自适应捕获。基于 JNet 与 NSFnet 骨干网拓扑的仿真实验表明,与 ANN、基础 GAT 等基线模型相比,HGAT降低了 RMSE、MAE 等预测误差,在高负载强串扰场景下仍保持较好的鲁棒性；且模型单样本在线推理时间降至毫秒级,满足光网络动态管控的低延迟要求。该模型为弹性光网络传输质量智能预测提供了高效可行的技术方案
Abstract
Fast and accurate estimation of Quality of Transmission (QoT) in elastic optical networks (EONs) is crucial for efficient resource allocation and intelligent management. Traditional analytical models suffer from high computational costs, while conventional machine learning methods tend to overlook the spatial correlation of network topologies and nonlinear impairment effects, making them inadequate for dynamic network demands. To address these issues, this paper proposes an online QoT estimation model based on a Hierarchical Graph Attention Network (HGAT). By mapping the physical optical network into a service dependency graph, the model designs a dual attention mechanism at both intra-lightpath and inter-lightpath levels to adaptively capture the physical degradation bottlenecks of a single lightpath and the nonlinear crosstalk across lightpaths. Simulation results based on JNet and NSFnet backbone topologies demonstrate that, compared with baseline models such as ANN and basic GAT, HGAT reduces prediction errors including RMSE and MAE, and maintains good robustness under high-load and strong-crosstalk scenarios. Furthermore, the single-sample online inference time of the proposed model is reduced to the sub-millisecond level, satisfying the low-latency requirements for dynamic management of optical networks. This model provides an efficient and feasible technical solution for intelligent QoT prediction in elastic optical networks.关键词
信息与通信系统/传输质量估算/双层图注意力网络/弹性光网络/非线性串扰Key words
Information and Communication Systems/Quality of Transmission Estimation/Hierarchical Graph Attention Network/Elastic Optical Networks/Nonlinear Crosstalk引用本文复制引用
闫欢,谷志群,纪越峰.基于双重图注意力网络的光路传输质量智能预测技术研究[EB/OL].(2026-03-17)[2026-03-20].http://www.paper.edu.cn/releasepaper/content/202603-152.学科分类
光电子技术
评论