Multidimensional smart meter data analytics based on sparse representation technique

Abstract

Customer's electricity consumption data analysis is very helpful for distribution system operators and electricity retailers in many aspects such as demand response and load forecasting. Dictionary learning and sparse representation techniques can extract important features of a load curve, which has not been properly applied in the current literature. In this paper, a sparse representation-basedmultidimensionalframeworkforanalyzinguserbehaviorsisproposedfromtheaspectsofbehaviormodeling, variability modeling, and preference modeling. Concretely, behavior modeling includes basic characteristic analysis, spectrumbased periodic pattern analysis, and working/off day pattern analysis. Variability is modeled through entropy analysis of user patterns. Finally, customer's preferences are modeled from the distribution of partial usage patterns. Numerical experiments prove the effectiveness of the analysis.