Abstract:
The unsupervised non-intrusive load monitoring (NILM) method can realize load decomposition by analyzing the unlabeled aggregated load data directly. However, the corresponding appliance name cannot be automatically determined generally, which affects the interpretability of NILM results and hinders its scale application. In this paper, a novel method for autonomous labeling of unsupervised NILM results based on rough classification of appliances was proposed. Both operation characteristics and usage time distribution of appliances were analyzed and summarized. Three categories of operation characteristics as periodic operation, intensive fluctuation, and fixed-duration operation were defined separately. Correspondingly, an adaptive operation characteristic inference method based on cluster analysis of appliance operation data was provided. A method for identifying the appliance usage characteristics based on the correlation between the usage of appliances and the human activities in term of time distribution was proposed. At the same time, an index to mark the periods of strong/weak human activities was given based on the change of the load component. Furthermore, the appliances were roughly classified according to the defined operation and usage characteristics based on the rough set theory, and then a bilevel decision method was proposed to realize the NILM result labeling. Experiments in private and public datasets show that the proposed method could label NILM results of common appliances in different scenarios accurately. The proposed method can be integrated with other unsupervised NILM algorithm to form a fully unsupervised NILM solution.