Data mining for abnormal power consumption pattern detection based on local matrix reconstruction

Abstract

Electricity theft is the main reason for non-technical losses (NTL) in distribution networks, which can lead to great economic losses in power supply enterprises. Efficient and accurate detection of abnormal power consumption patterns is a key part of demand side management. With the popular use of smart meters, it is more efficient and reliable to collect customers’ power consumption data, which make on-line monitoring of power consumption possible. In this paper, a detection algorithm based on local matrix reconstruction (LMR) is proposed and utilized to detect abnormal power consumption patterns in power systems. In this algorithm, five daily load characteristics are used to replace high-dimensional daily load curves to characterize power consumption patterns. Then, principal component analysis (PCA) is applied to calculate weighted reconstruction errors in a local scope. The reconstruction error of each sample is compared with its adjacent samples in order to calculate local outlier scores, which represent the abnormal degree of each load sample. Using two open source datasets, the detection performance of the proposed method is verified to be effective and efficient.