Multi-Scale Electrical Appliance Load Signature for Non-Intrusive Load Monitoring Classification

Abstract

Non-intrusive Load Monitoring is a system that is able to monitor energy consumption and provide a detailed energy breakdown to the end consumer. This paper deals with one cycle of steady-state voltage and current used to construct a voltage-current (V-I) trajectory. After retrieving one cycle signals of voltage and current, the Fourier phase correction technique is applied on them to avoid low accuracy or mismatch in classification d ue to the starting point selection of one period signal. Then, Triangle Area Representation (TAR) at different side lengths is introduced to describe the V-I trajectory of each electrical load appliance. Due to the very high dimensional subspace of TAR signature, it is then beneficial to use principal component analysis for creating a low-dimensional space feature. Finally, the weighted k-Nearest neighbor is employed to classify each type of appliance according to the k-Nearest number. Plug-Load Appliance Identification Dataset is applied in the assessment of the proposed algorithm which shows good performance with high accuracy of 97.43%.