基于机器学习的灵活动态光网络智能监测技术研究

High-speed and large-capacity optical networks are the cornerstone of national broadband information infrastructure. Optical performance monitoring (OPM) is crucial to ensure "high reliability" and "high efficiency" of optical networks. With the rapid development of "big data" service, optical networks are moving toward being high-speed, flexible and dynamic. The OPM technology face the technical challenges of dynamic reconfiguration of transmission links, flexible and diverse of signal formats and bandwidths, multiple damage interactions, complex whole network monitoring, and so forth. This project intends to focus on two key scientific issues: the low-cost performance monitoring mechanism and method of the whole network wavelength link; the mechanism and algorithm of end-to-end wavelength channel connection performance monitoring and damage diagnosis. Our team will carry out the basic research, innovative design and experiments on machine learning-based intelligent monitoring technologies of flexible and dynamic optical networks. The project strives to realize important breakthroughs in whole dynamic network monitoring mechanism and fault location strategy, low-cost and high-accuracy wavelength link monitoring scheme, and end-to-end wavelength links monitoring mechanism and scheme. This project aims to establish the theoretical and technical foundation for realizing low-cost intelligent monitoring technology of flexible dynamic optical network, provide guarantee for "high reliability" and "high efficiency" of optical networks, promote the development of big data-related industries and the improvement in national informatization level.

高速大容量光网络是国家信息基础设施宽带化的基石，光层监测对保障其“高可靠”和“高效率”的运行至关重要。随着“大数据”业务的迅猛发展，光网络正朝着“高速大容量、灵活动态”方向演进，光网络监测面临传输链路动态重构、信号格式和带宽灵活多样、多种损伤相互作用、全网监测复杂多变等技术挑战。本项目拟围绕全网波长链路的低成本监测机制与方法、端到端波长通道连接的性能监测和损伤诊断机制与算法两个关键科技问题，开展基于机器学习的灵活动态光网络智能监测的基础研究、创新设计与科学实验，拟在动态光网络全网监测机制和故障定位策略、低成本高准确度波长链路监测和具有多参数联合监测与损伤智能诊断功能的端到端波长连接监测机制与方法等方面实现重要突破。预期成果可为实现灵活动态光网络的智能监测提供低成本解决方案、奠定坚实理论与技术基础，为动态光网络“高可靠”“高效率”运行提供保障，并推动大数据相关产业发展以及国家信息化水平提升。

高速大容量光网络是国家信息基础设施宽带化的基石，光层监测对保障其“高可靠”和“高效率”的运行至关重要。本项目针对大容量灵活动态光网络监测所面临的传输链路动态重构、信号格式和带宽灵活多样、多种损伤相互作用、全网监测复杂多变等技术挑战，在“大容量灵活动态光网络波长链路与端到端波长通道连接智能监测”和“全网智能监测与故障定位”两大方面开展了基础理论和关键技术研究，取得了创新性研究成果。一、提出并验证了多种基于机器学习的灵活动态光网络波长链路监测机制与方法，实现了基于卷积神经网络和低带宽直接检测的低成本调制格式识别技术以及基于低带宽相干检测和随机森林的光性能监测方案。二、提出并验证了多种基于机器学习的端到端波长通道联合监测与损伤诊断方案，实现了具有低复杂度且高非线性容忍度的调制格式识别技术以及光信噪比估计与调制格式识别联合监测方案，利用迁移学习实现了眼图损伤多目标识别，可对强度调制信号的调制格式、信噪比、色散、非线性和光收发机损伤等进行智能分析与诊断。三、创新设计了基于实时泛在感知和灵活动态响应的数字孪生光网络架构，建立了灵活动态光网络中多源异构监测数据分布式采集系统，提出了基于多模态学习算法的多源异构监测数据融合处理与协同分析机制。四、提出并验证了基于机器学习和立体监测数据的光网络故障管理技术方案，实现了光网络设备故障提前预测以及故障原因诊断，从而保障光网络健康稳定运行。五、建立了光网络智能监测仿真平台及离线实验系统和包含卷积神经网络、循环神经网络、长短期记忆网络、支持向量机、随机森林和极度梯度提升树等多种机器学习算法在内的Python算法库。共发表论文49篇，其中高质量SCI检索论文23篇（包括中科院一区2篇，二区11篇）在OFC/ECOC/ACP等光通信国际知名学术会议发表EI检索26篇，申请国家发明专利6项，其中3项已授权。项目组主要完成人获得吴文俊人工智能科技进步二等奖和吴文俊人工智能优秀青年奖。