基于数据特征融合的管网信息物理异常诊断方法

Cyber-physical Abnormity Diagnosis Method Using Data Feature Fusion for Pipeline Network

随着管网物理空间和信息网络的深度融合,系统面临着物理和信息空间异常带来的运行风险.本文根据管网系统数据量大、耦合性强的特点,提出一种基于数据特征融合的信息物理异常诊断方法.首先通过站场信息数据构建信息增维矩阵并且通过矩阵预分析实现信息传输中断异常的判断.然后基于不同站场信息构建的信息增维协方差矩阵,通过矩阵特征值分布的变化情况对物理异常以及信息传输错误异常进行区分.在此基础上,为了对管网物理异常分类实现系统运行状态的有效分析,将管网信息增维协方差矩阵最大特征向量映射的二维图像作为输入,采用卷积神经网络进行研究,进而实现对物理异常的准确判断.最后通过某实际管网数据进行仿真分析,验证所提方法的有效性.

With deep fusion between pipeline physical network and cyber network,the system is facing operational risks caused by cyber-physical anomalies.According to pipe network features of big data and strong coupling,this paper proposes a cyber-physical abnormity diagnosis method using data feature fusion for pipeline network.Firstly,a cyber augmented matrix can be built by station cyber data,and cyber interrupt is diagnosed by matrix pre-analysis.Furthermore,the cyber augmented covariance matrix is established in light of the different station cyber,and physical anomaly and cyber transmission error anomaly are distinguished from each other by the variation condition of the matrix eigenvalue distribution.To effectively analyze the operating state of the physical network anomaly classiˉcation,the two-dimensional images which are mapped by the maximum eigenvectors of the cyber augmented covariance matrix of the pipeline network are regarded as the input signals,meanwhile,a convolutional neural network is utilized to carry out the analysis,thus the accurate judgment of the physical anomaly is realized.Eventually,the effectiveness of the proposed method is demonstrated through the time-domain simulation result obtained on a practical pipeline network.