基于聚类神经网络的光纤网络节点异常识别算法  

Fiber optic network node anomaly recognition algorithm based on clustering neural network

作　　者：原娇杰[1] 焦梦甜 赵杰文[1] YUAN Jiao-jie;JIAO Meng-tian;ZHAO Jie-wen(School of Information Engineering,Jiaozuo University,Jiaozuo 454003,China)

机构地区：[1]焦作大学信息工程学院,河南焦作454003

出　　处：《激光与红外》2025年第3期466-471,共6页Laser & Infrared

基　　金：河南省科技攻关计划项目(No.172102210390)资助。

摘　　要：为了提高光纤网络节点异常的识别准确率与识别速度,提出了一种基于聚类神经网络的节点异常识别算法。通过聚类计算完成输入数据的预分类,解决传统分类识别算法容易陷入局部最优的问题。将预分类后的测试数据分组作为输入层,并将聚类权值和聚类度作为隐藏层的加权系数,提高异常信号的识别度。实验对光纤网络中64个FBG节点进行测试,分别采用温度递变、重物撞击及周期振动模拟异常信号。对比实验结果显示,三种异常信号均存在的混叠条件下,本算法的识别准确率为80.3%、92.8%和91.6%,比不进行预分类的神经网络算法提升了约20%。在四种测试情况下,本算法的测试结果最优。对相同数据量测试时,本算法的速度仅为SVM算法的1/2,验证了本算法具有更好的时效性。To improve the accuracy and speed of identifying node anomalies in fiber optic networks,a node anomaly identification algorithm based on clustering neural networks is proposed in this paper.Firstly,the preclassification of input data is achieved through clustering calculation,which solves the problem of traditional classification and recognition algorithms easily falling into local optima.Then,the test data grouped after preclassification is used as the input layer,and the clustering weights and clustering degrees are used as the weighting coefficients of the hidden layer to improve the recognition of abnormal signals.Experiments are conducted on 64 FBG nodes in a fiber optic network,and the temperature increment,heavy impact and periodic vibration are used to simulate the anomalous signals,respectively.The results of the comparison experiments show that the recognition accuracy of this algorithm is 80.3%,92.8%,and 91.6%under the condition of aliasing where all three types of abnormal signals exist,which is an improvement of about 20%over the neural network algorithm without preclassification.Therefore,the present algorithm has the optimal test results in the four test cases.For the same data volume test,the speed of this algorithm is only half of that of SVM algorithm,which verifies that this algorithm has better timeliness.

关 键 词：聚类神经网络 预分类处理 聚类度 异常信号识别 

分 类 号：TN929.11[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]