基于深度信念网络的光通信网络数据异常识别研究  

Research on data anomaly recognition in optical communication network based on deep belief network

作　　者：王晓燕[1] 刘荷花[1] 徐国华[1] WANG Xiaoyan;LIU Hehua;XU Guohua(Taiyuan University Department of computer science and technology,Taiyuan 030032,China)

机构地区：[1]太原学院计算机科学与技术系,太原030032

出　　处：《激光杂志》2023年第2期149-153,共5页Laser Journal

基　　金：山西省科技创新计划项目(No.2020W275);山西省高等学校教学改革创新项目(No.J2020372);山西省高等学校哲学社会科学研究项目(No.2020W275)。

摘　　要：为了降低光通信网络被攻击的概率,保证光通信的安全顺畅,提出基于深度信念网络的光通信网络数据异常识别方法。利用时间-频率相结合的算法建立光通信信道模型,获取信道特征。根据信道特征密度设计数据异常特征的判断准则,利用数据挖掘聚类算法提取异常数据特征。融合BP网络和受限玻尔兹曼机网络,确立深度信念网络结构,结合隐藏层与可见层单元的概率分布情况构建数据异常识别模型,经过数据采集、特征归一化和模型微调等过程完成光通信网络数据异常识别。仿真实验表明,所提方法能够获取准确的光通信网络异常数据特征,光通信网络数据异常识别高和误报率低。In order to reduce the attack probability of optical communication network and ensure the safety and smoothness of optical communication, a data anomaly identification method of optical communication network based on deep belief network is proposed. The optical communication channel model is established by using the time-frequency algorithm to obtain the channel characteristics. According to the channel feature density, the judgment criteria of abnormal data features are designed, and the abnormal data features are extracted by data mining clustering algorithm.Combining BP network and restricted Boltzmann machine network, the depth belief network structure is established.Combined with the probability distribution of hidden layer and visible layer units, the data anomaly recognition model is constructed. The data anomaly recognition of optical communication network is completed through the processes of data acquisition, feature normalization and model fine-tuning. Simulation results show that the proposed method can obtain accurate abnormal data characteristics of optical communication network, and has high abnormal data recognition and low false positive rate.

关 键 词：深度信念网络 光通信网络 异常数据识别 挖掘聚类 受限玻尔兹曼机 

分 类 号：TN364[电子电信—物理电子学]