摘要
通信链路在高密集业务场景中承担关键传输任务。其运行状态易受拓扑结构复杂性与节点干扰行为影响,导致传统规则诊断方法难以实现精准定位。文章构建基于图卷积神经网络的故障定位方法,设计邻接矩阵与状态张量融合结构,结合边权调节机制与卷积传播策略提取异常特征,在分类层引入Focal Loss函数压制易判样本影响,形成具备路径感知能力的定位输出方式。测试表明,该方法在误判率抑制、响应延迟压缩及传播路径覆盖方面均实现了显著提升。
Communication links undertake critical transmission tasks in high-density business scenarios,and their operational status is easily affected by the complexity of topology and node interference behavior,making it difficult for traditional rule diagnosis methods to achieve accurate positioning.The article constructs a fault localization method based on graph convolutional neural networks,designs a fusion structure of adjacency matrix and state tensor,combines edge weight adjustment mechanism and convolution propagation strategy to extract abnormal features,introduces Focal Loss function in the classification layer to suppress the influence of easily judged samples,and forms a localization output method with path perception capability.Tests have shown that this method achieves significant improvements in false positive rate suppression,response delay compression,and propagation path coverage.
作者
杨开元
YANG Kaiyuan(Xi’an Zhonghe Nuclear Instrument Co.,Ltd.,Xi’an 710061,China)
出处
《通信电源技术》
2025年第18期203-205,共3页
Telecom Power Technology
关键词
通信链路
图卷积神经网络
故障定位
communication link
graph convolutional neural network
fault location
