基于XGBoost-Informer模型及多源数据的铁路通信QoS告警机制  

Railway Communication QoS Alarm Mechanism Based on XGBoost-Informer Model and Multi-source Data

作　　者：王允琪 蔺伟 梁轶群 杨居丰 李岸宁 刁洪宝 WANG Yunqi;LIN Wei;LIANG Yiqun;YANG Jufeng;LI Anning;DIAO Hongbao(Signal and Communication Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;National Research Center of Railway Intelligence Transportation System Engineering Technology,Beijing 100081,China;Infrastructure Inspection Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司通信信号研究所,北京100081 [2]国家铁路智能运输系统工程技术研究中心,北京100081 [3]中国铁道科学研究院集团有限公司基础设施检测研究所,北京100081

出　　处：《铁道学报》2024年第10期86-96,共11页Journal of the China Railway Society

基　　金：中国国家铁路集团有限公司科技研究开发计划(P2021G012);中国铁道科学研究院集团有限公司科技研究开发计划(2023YJ107)。

摘　　要：针对当前铁路通信业务实时监测数据量大、维度广,且缺少有效智能监测和告警机制,探究如何对多源铁路监测数据进行高效分析,并建立一种更为智能的网络服务质量预测告警机制,已成为支持铁路高效运营和安全保障的迫切需要。将多源铁路监测数据进行数据对齐、清洗并进行特征优选,建立XGBoost-Informer融合模型进行训练,用XGBoost模型找寻数据特征与目标值间的横向分类关系进行粗分类预测,再结合Informer模型对数据前后文间的纵向时序关系进行修正,得到列车运行过程中的时延、场强和服务质量优劣的预期结果,以在预期不良时提前触发告警,并根据所预测的网络特征进行告警分类。使用本文算法模型对多源铁路监测数据粗分类准确率为96.7%,经过修正后准确率为98.7%;在对比其他分类模型时,其均方根误差为0.113,平均绝对误差为0.003,均优于其他模型并有显著提升,模型拟合效果良好。该机制在保证铁路运营安全的情况下提高了对监测数据的处理效率并对网络状态趋势进行预测,对提升下一代铁路在途监测和智能运维具有指导意义。Addressing the challenges of massive real-time monitoring data and the lack of effective intelligent monitoring and alarm mechanisms in railway communication services,it is crucial to explore efficient analysis of multi-source railway monitoring data and establish an intelligent network service quality prediction and alarm mechanism to support efficient railway operations and safety.In this paper,by performing data alignment,cleaning,and feature selection from multi-source railway monitoring data,an XGBoost-Informer fusion model was established for training.The XGBoost model was employed to identify the cross-sectional classification relationship between data features and target values for coarse-grained prediction.Then,the Informer model was used to refine the longitudinal temporal relationships between preceding and subsequent data,resulting in predictions of latency,signal strength,and service quality during train operation,so as to trigger early alarms for anticipated anomalies,and classify alarms based on the predicted network features.The proposed algorithm model achieves a coarse-grained classification accuracy of 96.7%and an accuracy of 98.7%after refinement when applied to multi-source railway monitoring data.Comparative analysis with other classification models shows that the proposed model has a root mean square error of 0.113 and a mean absolute error of 0.003,outperforming other models with significant improvements and demonstrating a good fitting effect.This mechanism enhances the processing efficiency of monitoring data while ensuring railway operation safety,and provides predictive insights into network status trends,which are highly valuable for enhancing on-the-way monitoring and intelligent maintenance in the next generation of railways.

关 键 词：铁路通信 智能运维 监测 XGBoost INFORMER 预测告警 

分 类 号：U285.2[交通运输工程—交通信息工程及控制]