Research On Key Technologies Of Data Privacy Protection And Anomaly Detection In Smart Grid

With the development of new technologies such as the Internet of Things and big data,smart grid has become the main trend of the future development of power system,and its security has attracted more and more researchers’ attention.On the one hand,frequent communication between various participants in smart grid will lead to higher risk of privacy disclosure;On the other hand,malicious attackers cause economic losses and security accidents to smart grid by injecting abnormal power data.In order to solve the problems of data privacy protection and anomaly detection in smart grid,this thesis proposes a smart grid data privacy protection scheme based on matrix completion and a smart grid abnormal data detection scheme based on unsupervised learning.The main work is as follows:(1)Aiming at the problem of data privacy protection between different trust domains in smart grid,a data privacy protection scheme for smart grid based on matrix completion is proposed.In this scheme,missing elements in the correlation matrix are repaired by matrix completion,and random disturbance is increased by adding noise with the same statistical characteristics as the original data to ensure data privacy.Analysis and simulation results show that this scheme has good privacy protection effect and data availability,and can effectively reduce communication overhead in smart grid.(2)Aiming at the problem that the power data has no labels and abnormal data lead to inaccurate principal component analysis,an abnormal power consumption detection scheme based on unsupervised learning is proposed.Firstly,power characteristics are extracted from each user’s power consumption pattern.Then,ROBPCA algorithm is used to classify users based on the extracted features.Finally,in order to eliminate the overlap problem of preliminary classification,an improved K-means clustering algorithm that removes outliers is used to optimize the results to achieve accurate detection of outliers.Analysis and experimental results show that the proposed scheme outperforms other anomaly detection methods in accuracy,recall and other indicators.