An efficient supervised energy disaggregation scheme for power service in smart grid

Smart energy disaggregation is receiving increasing attention because it can be used to save energy and mine consumer's electricity privacy by decomposing aggregated meter readings. Many smart energy disaggregation schemes have been proposed; however, the accuracy and efficiency of these methods need to be improved. In this work, we consider a supervised energy disaggregation method which initially learns the power consumption of each appliance and then disaggregates meter readings using the previous learning result. In this study, we improved the fast search and find of density peaks clustering algorithm to cluster appliance power signals twice to learn appliance feature matrices. Additionally, we improved the max-min pruning matching optimization algorithm to decompose the aggregate power consumption into individual appliance. Experimental results obtained using the reference energy disaggregation dataset demonstrate that the proposed scheme achieves 81.9% accuracy and requires only 8 s to analyze 20-m readings for each sliding window. Thus, the proposed scheme exhibits better accuracy and efficiency compared with existing schemes.