摘要
MCX系统(MCPTT、MCData及MCVideo的统称)是5G-R宽带集群通信的核心组成部分,其结构复杂、功能繁多。目前,MCX系统通信故障时采用传统的人工故障排查方法,效率低下,缺乏智能化的识别与分析手段,对此,提出一种基于DNN(deep neural network,DNN)超参数优化的5G-R宽带集群通信故障识别方法。首先,通过多维特征融合提取特征,构建用于故障分类模型训练和测试的样本数据集;其次,针对大规模且分布不均的MCX系统通信故障数据,提出反馈驱动的自适应超参数优化(feedback-driven adaptive hyperparameter optimization,FDAHO)算法,优化数据采样和处理方法,改进超参数优化算法;最后,利用DNN结合FDAHO算法构建故障分类模型,采用公共数据集和全实物平台测试处理后所得的实测数据集,将所构建的模型分别与CNN(convolutional neural networks,CNN)结合FDAHO算法的模型、传统贝叶斯优化算法下的模型进行对比。实验结果表明:FDAHO算法在提高分类准确率和F1分数的同时,显著减少了超参数优化时间,提升了模型在资源受限环境下的实用性,且与DNN网络的结合具有更高的优越性和稳定性;实测数据集下所提模型在分类准确率、F1分数上相较于无超参数优化算法的模型分别提高了10.734%、11.328%;相较于DNN结合传统贝叶斯优化模型分别提高了0.342%、0.365%,且超参数优化时间减少了2152 s,可实现对MCX系统通信故障的高准确率和高效分类。研究结果为5G-R宽带集群通信的故障智能检测及监测提供参考。
The MCX system is a core component of 5G-R broadband trunking communication,with a complex structure and diverse functions.Current fault diagnosis in MCX systems relies on manual troubleshooting methods,which are inefficient and lack intelligent identification and analysis capabilities.To address these limitations,a 5G-R broadband trunking communication fault identification method based on DNN hyperparameter optimization was proposed.First,features were extracted through multidimensional feature fusion to construct the sample dataset for training and testing fault classification models.Second,for large-scale and unevenly distributed MCX system communication fault data,a feedback-driven adaptive hyperparameter optimization(FDAHO)algorithm was proposed.This algorithm optimized data sampling and processing while enhancing the hyperparameter optimization process.Finally,a fault classification model was constructed using DNN combined with the FDAHO algorithm.The constructed model was compared with the CNN combined with the FDAHO algorithm model and the traditional Bayesian optimization-based model using two public datasets and the actual measurement dataset obtained after testing and processing on a full physical platform.The experimental results demonstrate that the FDAHO algorithm improves classification accuracy and F1 score while significantly reducing hyperparameter optimization time,enhancing the model’s practicality in resourceconstrained environments.Its integration with DNNs exhibits superior performance and stability.On the measured dataset,the proposed model achieves improvements of 10.734%in classification accuracy and 11.328%in F1 score compared to the model without hyperparameter optimization.Compared to the DNN combined with traditional Bayesian optimization,it enhances accuracy and F1 score by 0.342%and 0.365%,respectively,while reducing hyperparameter optimization time by 2152 seconds.This enables high-accuracy and efficient classification of communication faults in the MCX system.The research findings provide a reference for intelligent fault detection and monitoring in 5G-R broadband trunking communication.
作者
乔婉淇
丁建文
郭强亮
孙斌
王玮
QIAO Wanqi;DING Jianwen;GUO Qiangliang;SUN Bin;WANG Wei(State Key Laboratory of Advanced Rail Autonomous Operation,Beijing Jiaotong University,Beijing 100044,China;Signal and Communication Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Key Laboratory of Railway Industry of Broadband Mobile Information Communications,Beijing Jiaotong University,Beijing 100044,China;School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China;Beijing Engineering Research Center of High-speed Railway Broadband Mobile Communications,Beijing Jiaotong University,Beijing 100044,China;Wireless and Computing Product Operation Division,ZTE Corporation,Xi’an 710114,China)
出处
《铁道科学与工程学报》
北大核心
2025年第10期4749-4760,共12页
Journal of Railway Science and Engineering
基金
国家自然科学基金资助项目(62171021)
中国铁道科学研究院集团有限公司科研项目重点项目(2024YJ182)
先进轨道交通自主运行全国重点实验室自主研究课题(RAO2023ZZ004)
中国国家铁路集团有限公司科技研究计划科研专项(J2023G006)
中国国家铁路集团有限公司科技研究开发计划项目(K2023G006)
中国铁路青藏集团有限公司重大专项(2023QZzhtl2101)
中兴产学研合作项目(ZTE-BJ-TU-2501)。
