Combining several distinct electrical features to enhance nonintrusive load monitoring

Smart meters are state of the art for electricity measurement in domestic and commercial buildings. So far they are only able to track the overall electricity consumption, though appliance specific feedback can lead to substantial higher energy savings. One promising option to reach appliance specific consumption information is nonintrusive load monitoring (NILM), in which this information is gained by disaggregating the overall load profile from a single-point measurement. To improve the accuracy of NILM, in this paper we investigate several distinct electrical features and combine them in an unsupervised learning algorithm. Our algorithm evaluation shows promising results for this method.