Schöb, SamuelSamuelSchöbGünther, SebastianSebastianGünther0000-0002-8005-7546Regensburger, KarlKarlRegensburgerStaake, ThorstenThorstenStaake2019-09-192018-08-3120181865-2042https://fis.uni-bamberg.de/handle/uniba/44381In Europe and the US, hot water use accounts for 13–18% of the average home’s energy consumption, compared to just 4 and 6% for lighting and cooking, respectively. As water heating mostly relies on oil, gas, and electricity, hot water use has been identified as an important target of many carbon reduction programs. We propose and describe a system that—comparable to non-intrusive load monitoring for electricity—disaggregates water extractions from a central metering device. The system can be used to provide consumption feedback, feed information into energy management systems, and can help to identify excessive water and energy use. The system relies on event-detection techniques and adapted Random Forest classifiers. We have tested and validated the system in two households over four months. The system was able to detect 85% of the extraction events which we then classify (“Dishwasher”, “Shower”, “Tap”, “Toilet”, and “Washing machine”). Random Forest achieves an F-measure between 71 and 91%. The area under the curve is above 0.9 for each appliance. We conclude that appliances are predicted reliably.engNon-intrusive, Water disaggregation, Residential, Machine learningarticle10.1007/s00450-017-0353-8