Please use this identifier to cite or link to this item: https://hdl.handle.net/20.500.13087/3367
Title: Appliance-Level Anomaly Detection by Using Control Charts and Artificial Neural Networks with Power Profiles
Authors: Apaydın-Özkan, Hanife
Keywords: IoT
home appliance
power profile
Artificial Neural Network
Control Chart
anomaly detection
Energy Management
Load
Issue Date: 2022
Publisher: Mdpi
Abstract: Nowadays, the development of the Internet of Things (IoT) concept has increased the interest in some technologies, one of which is the detection of anomalies in home appliances before they occur. In this work, in order to contribute to the works that use appliance power profiles for anomaly detection, a novel Appliance Monitoring and Anomaly Detection System (AM-ADS) is presented. AM-ADS consists of a main controller, a database, IoT-based communication units, home appliances, and power measurement units (smart plugs or special measurement equipments) mounted on appliances. In AM-ADS, a new Control Chart (CC) based method, for the cases that a limited number of historical power profiles are available; and a new Artificial Neural Network (ANN) based method, for the cases that a sufficient number of historical power profiles of each anomaly free and anomalous situations are available, are used according to the developed rule-based AM-ADS procedure to maximize the advantages and to eliminate the disadvantages of these methods as much as possible. According to the AM-ADS procedure, power consumptions of appliances, which provide meaningful information about the health of appliances, are measured during their operations and the corresponding power profiles are created. Active power, power factor, and operation duration features of power profiles are considered as decisive control parameters and different characteristics of these parameters are used as inputs for CC and ANN-based methods. The efficiency and performance of AM-ADS are validated by application case studies, where the ability to detect anomalies varies between 94.56% and 99.03% when a limited number of historical data is available; and the ability to detect and classify anomalies varies between 96.34% and 99.45% when a sufficient number of historical data is available.
URI: https://doi.org/10.3390/s22176639
https://hdl.handle.net/20.500.13087/3367
ISSN: 1424-8220
Appears in Collections:Elektrik-Elektronik Mühendisliği Bölümü Koleksiyonu
PubMed İndeksli Yayınlar Koleksiyonu
Scopus İndeksli Yayınlar Koleksiyonu
WoS İndeksli Yayınlar Koleksiyonu

Show full item record

CORE Recommender

Google ScholarTM

Check

Altmetric


Items in GCRIS Repository are protected by copyright, with all rights reserved, unless otherwise indicated.