Hybrid Iterative Algorithm for Non-Intrusive Load Disaggregation

Abstract

Non-intrusive load disaggregation techniques are low cost and they allow the determination of power consumption of individual appliances from the aggregated readings provided by the smart meters using computational algorithms. The available non- intrusive load disaggregation algorithms do not provide consistently high detection accuracy for all type of appliances in a typical household. Some algorithms perform better for one set of appliances, while others perform well for different set of appliances. In this paper, we develop a hybrid iterative algorithm to achieve a better overall detection accuracy for all type of appliances in a household. We model the load disaggregation problem as a linear integer program and a data clustering problem. The results provided by our linear integer program and clustering algorithms are matched in each iteration. All those appliances for which the detection results of both the algorithms are similar are declared to be correctly detected and removed from the appliance set. Linear integer program and clustering algorithms are re-applied on the reduced appliance set. In this way, the number of appliances that are required to be detected are successively decreased in every iteration. In our proposed algorithm, a design parameter is also provided that can be tuned for better convergence. This parameter can also be adjusted to improve the overall detection accuracy. Simulation results using REDD data-set of six houses show the superiority of our hybrid iterative algorithm as compared to other benchmark algorithms.