Home // AMBIENT 2019, The Ninth International Conference on Ambient Computing, Applications, Services and Technologies // View article

Multivariate Event Detection for Non-intrusive Load Monitoring

Authors:
Alexander Gerka
Benjamin Cauchi
Andreas Hein

Keywords: NILM; Event Detection; Hotelling-T 2 statistic

Abstract:
Due to the growing interest for in-home activity monitoring, the tracking of appliances use, usually referred to as Non-Intrusive Load Monitoring (NILM), has to address new challenges. Indeed, as NILM has long been motivated by potential energy savings, most event detectors for NILM have focused on the detection of on- or off-switches of high power devices. On the contrary, in-home monitoring typically relies on the detection of events related to low-power devices from potentially noisy signals. Additionally, approaches that apply expert heuristics to a single-variate input, often favored for their low complexity and real-time applicability, can be overly sensitive to the choice of an arbitrary defined detection threshold. This paper aims at decreasing the sensitivity of a detector based on expert heuristic by applying it to the Hotelling-T2 statistic of a multivariate input, computed online from the current and voltage inputs. Focusing on realistic scenarios, the approach is evaluated on a dataset recorded in a real apartment using a commercially available smart-meter. The results, expressed in terms of precision, recall and F-score, show that the proposed approach can both yield higher performance and be less sensitive to the choice of the detection threshold.

Pages: 25 to 30

Copyright: Copyright (c) IARIA, 2019

Publication date: September 22, 2019

Published in: conference

ISSN: 2326-9324

ISBN: 978-1-61208-739-9

Location: Porto, Portugal

Dates: from September 22, 2019 to September 26, 2019