TU Darmstadt / ULB / TUbiblio

Fine-Granular Sensing of Power Consumption - A New Sensing Modality for an Accurate Detection, Prediction and Forecasting of Higher-Level Contextual Information in Smart Environments

Alhamoud, Alaa (2016)
Fine-Granular Sensing of Power Consumption - A New Sensing Modality for an Accurate Detection, Prediction and Forecasting of Higher-Level Contextual Information in Smart Environments.
Technische Universität Darmstadt
Ph.D. Thesis, Primary publication

Abstract

Investigating and extracting users' context has always been a main drive for research in the field of smart environments. Context is defined as the set of conditions that characterize a certain situation in which the user exists. An example of user's context is the activity currently performed by the user along with her/his current indoor location. Identifying users' context paves the road for the realization of a wide variety of context-aware services that improve the quality of life for individuals as well as societies. To extract users' context, researchers have resorted to the approach of deploying a huge number of sensors in the environment where users' context has to be extracted. Examples of these sensors are cameras, microphones, motion and contact sensors, state-change sensors and other types of sensors that monitor every aspect of users' context. However, this approach has always been criticized as causing a huge deployment and maintenance overhead for researchers. Moreover, it has been perceived as an intrusive approach by users because they feel themselves surrounded by all possible types of visible sensors.

Recent years have seen an increasing adoption of smart metering technologies along with the manufacturing of new appliance-level power sensors that are able to measure the fine-granular power consumption of individual devices in smart environments. As a result, fine-granular sensing of power consumption has emerged as a new sensing modality that avoids the afore-mentioned problems of other sensing modalities. One of the main goals of this thesis is to develop intelligent models that infer and predict several challenging and essential aspects of users' context only based on fine-granular sensing of power consumption.

Activities of daily living (ADL) represent an essential part of users' context that has always motivated researchers in the field of smart environments. Context-aware services such as energy conservation in smart environments and ambient assisted living can be realized based on the recognition of users' current activity. In this thesis, we develop SMARTENERGY.KOM, an intelligent hardware/software platform for recognizing activities of users in single-user environments. We build an activity recognition model and evaluate its performance based on a dataset we collect by deploying SMARTENERGY.KOM in two single-user apartments.

As an essential part of this thesis, we conduct an in-depth analytical study on the dataset collected by SMARTENERGY.KOM with three main contributions, namely modeling of user's daily behavior, indoor localization based on fine-granular power consumption data, and profiling of user's hourly power consumption. As users tend to follow a daily routine in performing their activities, identifying behavioral patterns of users helps to improve the predictive performance of activity recognition models. We develop an approach that identifies such patterns of user's behavior. Evaluation results show that feeding these patterns into the model of activity recognition leads to a significant improvement in its predictive performance. Indoor location represents another important aspect of users' context that has its potential benefits in realizing location-aware services. We develop a localization model that is able to determine the indoor location of users based on their fine-granular power consumption. Building a profile that characterizes average hourly power consumption of users has its potential benefits in increasing their awareness of the power they consume as well as in detecting abnormal consumption patterns. Driven by this motivation, we develop an approach that identifies and builds this profile based on SMARTENERGY.KOM dataset.

As more than one user tend to live, work and reside in one common place, activity recognition models need to cope with the fact that parallel and overlapping activities of several users have to be recognized and assigned to their respective users. This fact has always represented a great challenge for researchers in the field of activity recognition. In this thesis, we address this challenge by developing MLSMARTENERGY.KOM, our platform for activity recognition in multi-user environments. We develop our model for recognizing activities based on the concept of multi-label classification, which exploits label dependency and temporal relations between activities.

Forecasting fine-granular power consumption of individual consumers represents another aspect of users' context that has its potential benefits for consumers as well as electric utilities. In this thesis, we develop state-of-the-art forecasting models that are able to forecast hourly, daily, and monthly power consumption of individual buildings based on building characteristics, demographic features of residents, available appliances as well as historical power consumption values. We evaluate these models using a dataset collected by Commission for Energy Regulation (CER) in Ireland.

Item Type: Ph.D. Thesis
Erschienen: 2016
Creators: Alhamoud, Alaa
Type of entry: Primary publication
Title: Fine-Granular Sensing of Power Consumption - A New Sensing Modality for an Accurate Detection, Prediction and Forecasting of Higher-Level Contextual Information in Smart Environments
Language: English
Referees: Steinmetz, Prof. Dr. Ralf ; Wolf, Prof. Dr. Lars
Date: 15 December 2016
Place of Publication: Darmstadt
Refereed: 22 November 2016
URL / URN: http://tuprints.ulb.tu-darmstadt.de/5839
Abstract:

Investigating and extracting users' context has always been a main drive for research in the field of smart environments. Context is defined as the set of conditions that characterize a certain situation in which the user exists. An example of user's context is the activity currently performed by the user along with her/his current indoor location. Identifying users' context paves the road for the realization of a wide variety of context-aware services that improve the quality of life for individuals as well as societies. To extract users' context, researchers have resorted to the approach of deploying a huge number of sensors in the environment where users' context has to be extracted. Examples of these sensors are cameras, microphones, motion and contact sensors, state-change sensors and other types of sensors that monitor every aspect of users' context. However, this approach has always been criticized as causing a huge deployment and maintenance overhead for researchers. Moreover, it has been perceived as an intrusive approach by users because they feel themselves surrounded by all possible types of visible sensors.

Recent years have seen an increasing adoption of smart metering technologies along with the manufacturing of new appliance-level power sensors that are able to measure the fine-granular power consumption of individual devices in smart environments. As a result, fine-granular sensing of power consumption has emerged as a new sensing modality that avoids the afore-mentioned problems of other sensing modalities. One of the main goals of this thesis is to develop intelligent models that infer and predict several challenging and essential aspects of users' context only based on fine-granular sensing of power consumption.

Activities of daily living (ADL) represent an essential part of users' context that has always motivated researchers in the field of smart environments. Context-aware services such as energy conservation in smart environments and ambient assisted living can be realized based on the recognition of users' current activity. In this thesis, we develop SMARTENERGY.KOM, an intelligent hardware/software platform for recognizing activities of users in single-user environments. We build an activity recognition model and evaluate its performance based on a dataset we collect by deploying SMARTENERGY.KOM in two single-user apartments.

As an essential part of this thesis, we conduct an in-depth analytical study on the dataset collected by SMARTENERGY.KOM with three main contributions, namely modeling of user's daily behavior, indoor localization based on fine-granular power consumption data, and profiling of user's hourly power consumption. As users tend to follow a daily routine in performing their activities, identifying behavioral patterns of users helps to improve the predictive performance of activity recognition models. We develop an approach that identifies such patterns of user's behavior. Evaluation results show that feeding these patterns into the model of activity recognition leads to a significant improvement in its predictive performance. Indoor location represents another important aspect of users' context that has its potential benefits in realizing location-aware services. We develop a localization model that is able to determine the indoor location of users based on their fine-granular power consumption. Building a profile that characterizes average hourly power consumption of users has its potential benefits in increasing their awareness of the power they consume as well as in detecting abnormal consumption patterns. Driven by this motivation, we develop an approach that identifies and builds this profile based on SMARTENERGY.KOM dataset.

As more than one user tend to live, work and reside in one common place, activity recognition models need to cope with the fact that parallel and overlapping activities of several users have to be recognized and assigned to their respective users. This fact has always represented a great challenge for researchers in the field of activity recognition. In this thesis, we address this challenge by developing MLSMARTENERGY.KOM, our platform for activity recognition in multi-user environments. We develop our model for recognizing activities based on the concept of multi-label classification, which exploits label dependency and temporal relations between activities.

Forecasting fine-granular power consumption of individual consumers represents another aspect of users' context that has its potential benefits for consumers as well as electric utilities. In this thesis, we develop state-of-the-art forecasting models that are able to forecast hourly, daily, and monthly power consumption of individual buildings based on building characteristics, demographic features of residents, available appliances as well as historical power consumption values. We evaluate these models using a dataset collected by Commission for Energy Regulation (CER) in Ireland.

Alternative Abstract:
Alternative abstract Language

Die Erfassung und Analyse von Nutzerkontext ist ein großer Bestandteil der Forschung auf dem Gebiet der intelligenten Umgebungen. Kontext wird als die Menge von Bedingungen definiert, die eine bestimmte Situation charakterisieren, in der der Nutzer existiert. Ein Beispiel für Kontext ist die Aktivität, die gerade durch den Nutzer an seinem aktuellen Standort durchgeführt wird. Die Identifizierung des Nutzerkontexts ebnet den Weg für die Realisierung einer Vielzahl von kontextsensitiven Diensten, die die Lebensqualität des Einzelnen als auch ganzer Gesellschaften verbessern können. Um den Kontext des Nutzers zu extrahieren, wird in verwandten Arbeiten der Ansatz verfolgt, eine große Anzahl von Sensoren in der Umgebung einzusetzen, in der der Nutzerkontext erfasst werden soll. Beispiele für solche Sensoren sind Kameras, Mikrofone, Bewegungsmelder, Kontaktsensoren, Zustandsänderungssensoren und andere Arten von Sensoren, die jeden Aspekt des Nutzerkontexts überwachen. Allerdings wurde dieser Ansatz immer kritisiert, weil er großen Einsatz- und Wartungsaufwand verursacht. Darüber hinaus wird dieser Ansatz als invasive Methode von den Nutzern wahrgenommen, weil sie sich umgebend von allen möglichen Arten von sichtbaren Sensoren unwohl fühlen.

Die Einführung von Smart-Metering-Technologien einhergehend mit der Innovation von Stromsensoren, die in der Lage sind den fein-granularen Stromverbrauch einzelner Geräte in intelligenten Umgebungen zu messen, hat in den letzten Jahren stark zugenommen. Als Ergebnis meiner Arbeit hat sich die fein-granulare Erfassung des Stromverbrauchs als neue Erfassungsmodalität herausgestellt, die die oben genannten Probleme der anderen Erfassungsmodalitäten vermeidet. Eines der wichtigsten Ziele dieser Arbeit war es, intelligente Modelle zu entwickeln, die in der Lage sind, herausfordernde und wesentliche Aspekte des Nutzerkontexts - sowohl für einzelne Nutzer als auch für mehrere Nutzer und sowohl für einzeln als auch parallel durchgeführte Aktivitäten - anhand der fein-granularen Erfassung des Stromverbrauchs zu erkennen und vorherzusagen.

Aktivitäten des täglichen Lebens stellen einen wesentlichen Teil des Nutzerkontexts dar. Kontextbewusste Dienste wie beispielsweise zur Energieeinsparung in intelligenten Umgebungen und umgebungsunterstütztes Leben können auf Basis der Erkennung der aktuellen Nutzeraktivität realisiert werden. In dieser Arbeit wird SMARTENERGY.KOM konzipiert und entwickelt, eine intelligente Hardware/Software Plattform für die Erkennung von Nutzeraktivitäten in Einzelnutzer-Umgebungen. Es wird ein Aktivitätserkennungsmodell erstellt und basierend auf einem Datensatz, der durch den Einsatz von SMARTENERGY.KOM in zwei Einzelnutzer-Wohnungen gesammelt wird, wird seine prädiktive Performanz bewertet. Als ein wesentlicher Teil dieser Arbeit wird eine gründliche analytische Studie über den SMARTENERGY.KOM-Datensatz mit drei wichtigen Beiträgen durchgeführt, das sind die Modellierung des täglichen Benutzerverhaltens, Lokalisierung in Gebäuden basierend auf fein-granularen Stromverbrauchsdaten und die Profilierung des stündlichen Stromverbrauchs. Menschen neigen dazu, eine tägliche Routine bei der Durchführung ihrer Aktivitäten zu verfolgen, die Identifizierung von Verhaltensmustern hilft dabei, die prädiktive Performanz der Aktivitätserkennungsmodelle zu verbessern. Daher wird einen Ansatz entwickelt, der solche Verhaltensmuster identifizieren kann. Die Auswertungsergebnisse zeigen, dass die Integration dieser Muster in das Aktivitätserkennungsmodell zu einer deutlichen Verbesserung der prädiktiven Performanz dieses Modells führt. Lokalisierung in Gebäuden ist ein weiterer wichtiger Aspekt des Nutzerkontexts, der die Realisierung von standortbezogenen Dienste ermöglicht. Es wird ein Lokalisierungsmodell entwickelt, welches in der Lage ist, den Aufenthaltsort der Nutzer in Gebäuden basierend auf ihrem fein-granularen Stromverbrauch zu bestimmen. Der Aufbau eines Profils, das den durchschnittlichen, stündlichen Stromverbrauch charakterisiert, ermöglicht es, das Bewusstsein für den Stromverbrauch zu erhöhen, sowie ungewöhnliche Konsummuster zu entdecken. Angetrieben von dieser Motivation, wird in dieser Arbeit der Ansatz verfolgt, dieses Profil auf Basis von SMARTENERGY.KOM-Datensatz zu identifizieren und zu erstellen.

Neben überwiegend einzeln genutzten Wohnungen gibt es viele Orte, an denen gemeinsam gelebt bzw. gearbeitet wird. Damit müssen Aktivitätserkennungsmodelle auch überlappende Aktivitäten mehrerer Nutzer erkennen und ihrem jeweiligen Nutzern zuordnen. Dies stellt eine große Herausforderung auf dem Gebiet der Aktivitätserkennung dar. Diese Arbeit befasst sich daher mit dieser Herausforderung durch die Entwicklung von MLSMARTENERGY.KOM, unsere Plattform für die Aktivitätserkennung in Multi-Nutzer-Umgebungen. Das Aktivitätserkennungsmodell wird basierend auf dem Konzept der Multi-Label-Klassifikation entwickelt, das die Abhängigkeit von Klassen und die zeitlichen Beziehungen zwischen den Aktivitäten ausnutzt.

Die Vorhersage des fein-granularen Stromverbrauchs einzelner Verbraucher stellt einen weiteren Aspekt des Nutzerkontexts dar, der verschiedene potenzielle Vorteile für die Verbraucher als auch Stromversorger bietet. In dieser Arbeit werden Vorhersagemodelle entwickelt, die in der Lage sind, stündliche, tägliche, und monatliche Stromverbräuche der einzelnen Gebäude auf Basis von Gebäudeeigenschaften, demographischen Merkmalen der Bewohner, verfügbaren Geräten sowie historischen Stromverbrauchswerten vorherzusagen. Diese Modelle werden auf Basis des von der Commission for Energy Regulation (CER) in Irland gesammelten Datensatzes evaluiert.

German
Uncontrolled Keywords: Activity Recognition, Multi-Label Classification, Smart Environments, Power Forecasting, Behavioral Modeling
URN: urn:nbn:de:tuda-tuprints-58396
Classification DDC: 000 Generalities, computers, information > 004 Computer science
600 Technology, medicine, applied sciences > 620 Engineering and machine engineering
Divisions: 18 Department of Electrical Engineering and Information Technology > Institute of Computer Engineering > Multimedia Communications
18 Department of Electrical Engineering and Information Technology > Institute of Computer Engineering
18 Department of Electrical Engineering and Information Technology
Date Deposited: 18 Dec 2016 20:55
Last Modified: 18 Dec 2016 20:55
PPN:
Referees: Steinmetz, Prof. Dr. Ralf ; Wolf, Prof. Dr. Lars
Refereed / Verteidigung / mdl. Prüfung: 22 November 2016
Export:
Suche nach Titel in: TUfind oder in Google
Send an inquiry Send an inquiry

Options (only for editors)
Show editorial Details Show editorial Details