基于立体监测数据与深度学习算法的动态光网络中光层故障分析关键技术研究

作为信息基础设施宽带化的基石，光网络容量和规模不断扩大，并朝着动态、可靠、智能的方向演进。光层故障分析对避免重大损失、保障服务质量至关重要。动态光网络中光传输链路性能灵活多变、节点设备状态参数复杂多样，但传统的故障分析技术对动态的光层故障分析不够全面及时准确。.本项目将通信网络技术、人工智能领域的深度学习算法、应用数学中的统计学方法交叉融合，通过深度学习强大的挖掘和推理能力来解决光层故障的多目标参数分析问题，通过统计学方法辅助进行趋势预测和量化分析；提出并实现轻量化多目标的光传输链路性能参数监测算法、自学习高灵敏的光网络节点设备状态参数预测算法等，在立体监测数据集和深度学习算法的基础上，实现高效、准确、智能的光层故障预测、诊断和定位；并依据实测数据和现网日志进行实验验证。为动态可靠智能的光网络技术进步提供有效的故障分析模型、多维标签化数据集、综合分析算法方案、紧密结合现网的实验依据。

As the foundation of broadband information infrastructure, the capacity and scale of optical networks in China are expanding considerably in China. And the optical networks are evolving in dynamic, reliable and intelligent direction. The fault analysis in optical layer is thus very important to prevent network failure, avoid great loss and guarantee the QoS. However, the traditional static fault analysis technology is unfit for the dynamic optical networks, where coexist flexible optical transmission links, complex equipment operating parameters and so on. .This project integrates organically the communication networking technologies, the deep learning algorithms and the statistical methods. We take advantage of strong ability of deep learning in information mining during the multi-target monitoring. We also use the statistical method to assist the trend projecting and quantitative analyzing. A scheme of lightweight multi-target parameter monitoring for optical transmission is to be proposed, as well as a scheme of self-learning and highly sensitive prediction technique for operating parameters of optical layer equipment. Therefore, based on the stereoscopic monitoring data and the deep learning algorithms, we make efforts to realize a series of efficient, accurate and intelligent fault prediction, fault diagnosis and fault location. According to the measured data and the actual network logs, we will evaluate the validity of the proposed schemes experimentally. Consequently, we aim to provide a fault analyzing model, a stereoscopic monitoring data set, adaptive synthetic analysis algorithms and helpful convincible experiment evidences in promoting the progress of optical networks.