Feature extraction and classification using power demand information

Electrical load monitoring, by means of a smart meter, is getting more and more popular these days. Power demand information from smart meters is drawing attention among researchers, since it could be applied for power demand control. Providing attractive services with smart meters encourage electricity retailers to utilize demand side management, which could be a solution for energy-related problems in our society. In this paper, a novel service is proposed by classifying private information from the household electricity usage. The private information is estimated using feature vectors extracted from time series analysis of power demand information. In order to extract feature vectors effectively, two extraction methods were proposed: simple statistical method, and Discrete Fourier Transform (DFT) based extraction method. Then, Support Vector Machines (SVMs) classifier is carried out after the optimization of hyper-parameters. As the estimated information, both family structure and floor space were selected. The classification result is evaluated using F-measure and accuracy. As a result, the accuracy of DFT-based classification was superior to the statistical method for detecting the floor space in a house.