Home // SIGNAL 2018, The Third International Conference on Advances in Signal, Image and Video Processing // View article

HELD1: Home Equipment Laboratory Dataset for Non-Intrusive Load Monitoring

Authors:
Pirmin Held
Steffen Mauch
Alaa Saleh
Djaffar Ould Abdeslam
Dirk Benyoucef

Keywords: dataset; feature extraction; feed forward neural net; supervised classification

Abstract:
Non-Intrusive Load Monitoring (NILM) can be split into event detection, classification and energy tracking. Different algorithms have already been proposed for the respective tasks. Each algorithm has been verified based on publicly available data sets to assess its performance. The two types of data sets that currently exist can be distinguished into two types: laboratory measurements and data sets from real world environment. In general, the available laboratory measurements provide data of individual devices; these are only of limited use for overall benchmark tests. Measurements, in which several devices have been active simultaneously, only exist in real scenario datasets. Nevertheless, the assignment of reference data in real scenarios is somehow problematic: issues are, for example, the synchronization between reference data and measured data, absence or excess of events and the number of on and off cycles of each device respectively. Furthermore, the probability distribution of the devices, as well as long measurement cycles with correspondingly large amounts of data, but low number of events, are challenging. Therefore, it is very difficult to compare the current NILM algorithms. Home Equipment Laboratory Dataset (HELD1) has multiple switching on and off events of several devices acting individually and/or simultaneously. Since the individual devices can be controlled separately, the reference data is available in a very high quality. Thus, high number of events can be generated within a short measuring time. In addition, the dataset contains different complex scenarios of various numbers of appliances. The objective of this data set is to offer a better basis to enhance the comparability between the individual NILM approaches.

Pages: 15 to 20

Copyright: Copyright (c) IARIA, 2018

Publication date: May 20, 2018

Published in: conference

ISSN: 2519-8432

ISBN: 978-1-61208-638-5

Location: Nice, France

Dates: from May 20, 2018 to May 24, 2018