|
NILM-Eval: Disaggregation of Real-World Electricity Consumption Data (M)Status: AbgeschlossenBackgroundThis student project is a part of our our ongoing project Innovative Services based on Smart Meters, in which we collaborate with two Swiss energy providers. Within the last year, we instrumented multiple test households with different sensors, measuring values such as the electricity consumption of each household, occupancy, or consumption of individual appliances. The underlying technology has been developed in previous student projects: "Smart meters in the field - A sensor framework for a real world deployment" and "Integrating submeters to individually monitor appliances".ChallengeEnergy providers are highly interested in providing enhanced feedback about their customer's electricity consumption as a premium service, for example as a part of an energy efficiency program. In its best case, such enhanced feedback contains a disaggregation of the overall electricity consumption of a household - listing the contribution of each individual appliance to the overall bill. Measuring the consumption for each appliance, however, would be both too costly and cumbersome to deploy for the customers. To this end, many researchers investigate obtaining such a consumption breakdown just by measuring and analyzing the overall electricity consumption of a household - so-called non-intrusive load monitoring (NILM). While NILM is a widely established research field, extensive data sets have been missing (so far) to thoroughly evaluate the recently developed algorithms in real world environments. In our deployment described above we performed a data collection that lead to one of the largest data sets worldwide: our data set contains high granularity whole house meter readings of several households over the time frame of multiple months - as well as plug load information serving as ground truth to evaluate our algorithms. The goal of this thesis is to investigate 2-3 cutting-edge load monitoring algorithms and evaluate their performance on our data set. Hereby we want to learn (a) the overall accuracy of NILM based on real world data, (b) the type of devices that can be detected with high accuracy, and (c) (optional) to what extent additional instrumentation or information can be employed to improve performance.RequirementsInterested students should...
Contact/Ansprechpartner: Christian Beckel |
  |
