To support home energy management, users or operators prefer appliance-level energy consumption information than the house monthly electricity bill report. Two methods exist for appliance energy usages recognition: Non-intrusive Load Monitoring (NILM) and Intrusive Load Monitoring (ILM). Both have not been widely used due to either insufficient performance or high cost. This paper proposed a practical socket-level non-intrusive load monitoring method. First, through socket submeters, the load disaggregation accuracy can be improved by reducing occurrences of indistinguishable appliances when using simple power features; Second, by involving users’ feedback, the load classification accuracy can be enhanced by feature registration and match. An unsupervised hierarchical clustering algorithm was used for load disaggregation, and the dynamic time wrapping algorithm was used for appliance feature match. This method was validated through a public dataset and showed a great promise.
Keywords house NILM, socket NILM, load disaggregation, hierarchical clustering, load classification, feature registration