Anomalous Power Usage Pattern Detection Based on Markov Model Clustering

Cluster analysis of the features of the load profile has the problem of difficult feature construction. In the face of a large amount of smart meter data, this paperuses a data dimensionality reduction method based on symbolic aggregation approximation (SAX), and proposes a new dynamic clustering method for electricity consumption behaviour. The term dynamic refers to the transition trend between electricity consumption states in neighbouring periods. The method is applied to the detection of anomalous electricity consumption patterns, which improves the detection rate and accuracy of the anomalies.