A detection model of scaling attacks considering consumption pattern diversity in AMI

As an important branch of the Internet of Things, the smart grid has become a crucial field of modern information technology. It realizes the two-way information flow and power flow by integrating the advanced metering infrastructure (AMI) and distributed energy resources, which greatly improves users’ participation. However, owing to smart meters, the most critical components of AMI, are deployed in an open network environment, AMI is a potential target for data integrity attacks. Among various attack types, the scaling attack is the most typical one, because it can be used as a general expression for most of other ones. By launching a scaling attack, adversaries can randomly reduce hourly reported values in smart meters, thereby causing economic losses. A number of research efforts have been devoted to detecting data integrity attacks. Nonetheless, most of the existing investigations focus on all attack types, and little attention has been paid to a detection strategy specially designed for scaling attacks. Our contribution addresses this issue in this paper and hence, developing a detection model of scaling attacks considering consumption pattern diversity (SA2CPD), to ensure that scaling attacks can be effectively detected when users have multiple consumption patterns. To be specific, we leverage Kmeans to distinguish different consumption patterns, and then the consumption intervals can be extracted to binarize the data. We divide time periods in every day into two categories based on the binarization values, and use one of them with the greatest information gain to construct a decision tree for judgment. Both theoretical and simulation results based on the GEFCom2012 dataset show that our SA2CPD model has a higher F1 score than the decision tree model without considering consumption pattern diversity, the KNN model and the Naive Bayes model.