Bernard, TimoTimoBernardVerbunt, MartinMartinVerbuntBögel, Gerd vomGerd vomBögelWellmann, ThorstenThorstenWellmann2022-03-132022-03-132018https://publica.fraunhofer.de/handle/publica/40078510.1109/ICSGCE.2018.8556735Energy savings are an important building block for the clean energy transition. Studies show that the consideration of overall load profiles is not sufficient to identify significant saving potentials - as is the case with smart meters. Nonintrusive Load Monitoring enables a device specific consumption disaggregation in a cost effective way. Our work focuses on the fusion of low, mid and high frequency features which can enhance the disaggregation performance. Furthermore our suggested approach consists of an unsupervised machine learning technique which enables novelty detection, a small training phase and live processing. We conclude this paper with the algorithm evaluation on household and industrial datasets.endevice identificationenergy consumptionenergy efficiencyhigh frequency electrical featureload disaggregationNonintrusive Load Monitoring (NILM)smart meteringunsupervised machine learningload management621conference paper