基于DLPP-LOF的信息物理系统异常诊断方法

DLPP-LOF-based Cyber Physical System Anomaly Diagnosis Method

为了准确诊断信息物理系统的异常类型,提出了一种新的基于动态局部保持投影-局部离群因子(dynamic locality preserving projections-local outlier factor,DLPP-LOF)的方法。首先,采用数据增广策略在判别模型中考虑自相关性,进而利用对数据分布没有要求的流形学习方法——局部保持投影(locality preserving projections,LPP)提取特征。其次,计算测试数据特征相对于训练数据集各类别特征的局部离群因子(local outlier factor,LOF),将具有最小离群因子的类作为测试数据的类别。确定了异常类别后,在已建立的历史异常数据及相应决策方案库中搜索制定应急响应预案。最后,将所提出的DLPP-LOF方法在典型信息物理系统上进行测试,验证了其有效性及优越性。

In order to accurately diagnose anomaly types for cyber physical system,a novel method based on dynamic locality preserving projections-local outlier factor(DLPP-LOF)is proposed.First,the data augmentation strategy is used to consider the autocorrelation in the discriminant model,and then the manifold learning method that does not require data distribution-locality preserving projections(LPP)is used to extract features.Secondly,calculate the local outlier factor(LOF)of the features of the testing data relative to the features of each category of the training dataset,and use the class with the smallest outlier factor as the category of the testing data.Once the anomaly category is determined,search and formulate emergency response plans in the established historical anomaly data and corresponding decision-making plan database.Finally,the DLPP-LOF method is tested on a typical cyber-physical system to demonstrate its effectiveness and superiority.