Since non-intrusive appliance load monitoring (NILM) gained interest, numerous researches have been focusing on the identification and disaggregation of household and industrial electrical devices by using a variety of different signatures. Steady-state analysis, transient analysis, and hybrid approaches are used in order to form an unique appliance signature and recent researches have come up with a systematic selection of a set of features as well. As there are an enormous number of electrical and electronic devices, proper categorization of appliances for building a well-organized load signature database is a paramount requirement for more systematically maintained database and effective training for NILM purpose.
In this paper, a review of electrical characteristics for an effective load identification in NILM systems is provided. Then systematic selection of a set of features for effective appliance classification is discussed. Furthermore, a detailed analysis of existing taxonomies of appliances is presented by means of comprehensive literature research. On top of this, novel taxonomy of appliances is derived by applying renowned clustering algorithms to the optimal set of features extracted from the BLUED dataset. Finally, taxonomies derived by our experiments are compared with each other and further compared with some of the existing systematic taxonomies of appliances. This paper is exploiting the optimal set of electrical characteristics to derive taxonomy of appliances for NILM purpose for the first time, and exhibited meaningful implication for the derivation of appliance taxonomy.