Research On Abnormal Detection Of Data Of Industrial Internet Platform

The industrial Internet platform is a new generation of industrial production platform generated under the trend of"Internet+",connecting industrial control system,Internet and industrial cloud platform to realize the comprehensive perception of industrial data.However,due to equipment layer failures and some threats of intrusion attacks,anomaly detection and analysis of industrial Internet platforms are needed.At the same time,due to human errors,noise interference and other reasons,there exists that data is missing,and the collected data is high dimensional and contains noise,which reduces the timeliness and accuracy of anomaly detection.Therefore,before anomaly detection,the problem of data missing and dimensionality reduction of high-dimensional noise data should be solved first.Thesis mainly studies the problem of filling in missing data,dimensionality reduction of high-dimensional data and anomaly detection of industrial Internet platform.Specific research contents are as follows:1.Based on the data filling scheme of MDKM-IEW,thesis uses k-means method based on Mahalanobis distance to cluster the data,then calculates the similarity between missing data and complete data in the class,and fills missing data by weighting the complete data through information entropy.The experimental results show that the filling method proposed in thesis is better than the k-means algorithm based on Euclidean distance.2.Based on the data dimension reduction scheme of RFD-PCAFE,the random forest model is used to select the features of the data and remove the noise in the data.After that,principal component analysis(PCA)is used to extract features from noise-free data,and the data is transformed into a new low-dimensional feature space to achieve dimensionality reduction.The experimental results show that the dimension reduction effect of the model combined with random forest and PCA is better than that of PCA model.3.Based on the CIFIF-DCGD anomaly detection model,the continuous and discrete features of data are detected separately,the isolated forest model is trained by the continuous features,the Gauss distribution model is trained by the discrete features,and then the two models are combined to detect the anomaly of industrial Internet platform data.The experimental results show that compared with the single continuous feature detection model and the single discrete feature detection model,the integrated detection model in thesis has better detection effect.