Options
2016
Conference Paper
Titel
Detecting anomalous electrical appliance behavior based on motif transition likelihood matrices
Abstract
With the rise of the Internet of Things, innumerable sensors and actuators are expected to find their way into homes, office spaces, and beyond. Electric power metering equipment, mostly present in the form of smart meters and smart plugs, can thus be anticipated to be installed widely. While smart plugs (i.e., individual power monitors attachable to wall outlets) primarily cater to the user comfort by enabling legacy devices to be controlled remotely, their power measurement capability also serves as an enabler for novel context-based services. We realize one such functionality in this paper, namely the recognition of unexpected behavior and appliance faults based on observed power consumption sequences. To this end, we present a system that extracts characteristic power consumption transitions and their temporal dependencies from previously collected measurements. When queried with a power data sequence collected from the appliance under consideration, it returns the likelihood that the collected power data indicate normal behavior of this appliance. We evaluate our system using real-world appliance-level consumption data. Our results confirm the individual nature of consumption patterns and show that the system can reliably detect errors introduced in the data within close temporal proximity to their time of occurrence.