The aim of this thesis is to propose a model that explains the relationship between customer satisfaction, retention and loyalty based on service quality attributes. The three elements of satisfaction, retention and loyalty towards products represent ongoing challenges for the corporate financial performance. Customer behaviour analysis (known as business intelligence or customer relationship management or customer experience management) has become a major factor in the corporate decision making and strategic planning processes. Prevailing logic dictates that by improving service attributes one should expect better customer satisfaction levels. Consequently, improved satisfaction levels should increase the probability of customer retention and degree of loyalty. Substantial research work has been dedicated to explain the importance of customer behaviour measurement for industry. However, there is little evidence that there has been an overall integrating empirical research that relates the three elements of satisfaction, retention and loyalty with respect to service quality attributes. Empirical data collected from the UK mobile telecommunication for this research shows that such an objective model that is capable of capturing this three dimensional relationship will contribute towards more robust decision making and better strategic planning. The proposed thesis extracts the data about key service attributes from a combination of literature review, surveys, and interviews from the UK mobile telecommunication industry. Responses were analysed using multiple regression, regression analysis with dummy variables, logistic regression, logistic regression with dummy variables and structural equation modelling (SEM) to test variables and their interrelationships. This study makes a step forward and contributes to the body of knowledge as it: (a) highlights the role of service attribute performance towards customer satisfaction, consequently identifies attributes that affect satisfaction and dissatisfaction of customers, (b) maps the relationship between attribute importance and attribute performance, (c) optimise resource allocation process using importance-performance analysis (IPA), (d) classifies customers with respect to the role and length of relationship they have with the company (switching probability), and (e) describes the interrelationship between customer satisfaction, retention and loyalty. The novelty of the research lies in: (a) establishment of a framework that links service attribute performance to customer satisfaction and then to customer future intentions (customer retention and customer loyalty), and (b) provision of a model that could assist key decision makers in prudent usage of resources for maximum profitability. This dissertation presents a novel approach methodology and modelling construct for customer behaviour analysis. For proof of concept it presents a case study in the mobile telecommunication industry. It is worth noting that in this research work Customer Retention is interpreted as probability of switching between service providers. Customer Loyalty is interpreted as referral (word-of-mouth) activity by existing customers.

Malgré le progrès en recherche et développement dans le domaine de système autonome, de tels systèmes nécessitent l’intervention humaine pour résoudre les problèmes imprévus durant l’exécution des tâches par l’utilisateur.Il est donc nécessaire, malgré cette autonomie, de tenir compte du comportement du conducteur et il est difficile d’ignorer l’effet de l’intervention humaine dans le cadre de l’évolution continue de l’environnement et des préférences de l’utilisateur. Afin d’exécuter les opérations selon les attentes de l’opérateur, il est nécessaire d’incorporer dans la commande les besoins de l’utilisateur.Dans les travaux présentés dans cette thèse un modèle comportemental de l’utilisateur est développé et intégré dans la boucle de commande afin d’adapter la commande à l’utilisateur. Ceci est appliqué à la commande des fauteuils électrique et assiste dans la navigation du fauteuil dans un milieu encombré.Le développement du modèle comportemental est basé sur la méthode de potentielles orientées et la détection des obstacles et le comportement du conducteur vs de ces obstacles par l’adaptation duL’étude contribue également au développement d’un modèle dynamique du fauteuil utilisable dans des situations normales et exceptionnelles telle que le dérapage. Ce modèle est développé pour un le cas le plus courant des fauteuil avec roues arrière conductrices utilisant le formalisme Euler Lagrange avec les forces gravitationnelles et sur des surfaces inclinées.Dans la formulation de la commande, le modèle du conducteur est introduit dans la boucle de commande. L’optimalité de la performance est assurée par l’utilisation du commande prédictif généralisé pour le système en temps continue. Les résultats de la simulation démontrent l’efficacité de l’approche proposée pour l’adaptation de la commande au comportement du conducteur / Although the progressive research and development of autonomous systems is fairly evident, such systems still require human interventions to solve the unforeseen complexities, and clear the uncertainties encountered in the execution of user-tasks. Thus, in spite of the system's autonomy, it may not be possible to absolutely disregard the operator's role. Human intervention, particularly in the control of auto-mobiles, may as well be hard to ignore because of the constantly changing operational context and the evolving nature of the drivers' needs and preferences. In order to execute the autonomous operations in conformity with the operator's expectations, it may be necessary to incorporate the advancing needs and behaviour of the operator in the design. This thesis formulates an operator behaviour model, and integrates the model in the control loop to adapt the functionality of a human-machine system to the operator's behaviour. The study focuses on a powered wheelchair, and contributes to the advancement of steering performance, through background assistance by modelling, empirical estimation and incorporation of the driver's steering behaviour into the control system. The formulation of the steering behaviour model is based on two fundamentals: the general empirical knowledge of wheelchair steering, and the experimental steering data captured by a standard powered wheelchair, on both virtual and real environments. The study considers a reactive directed potential field (DPF) method in the modelling of drivers' risk detection and avoidance behaviour, and applies the ordinary least square procedure in the identification of best-fitting driver parameters. The study also contributes to the development of a dynamic model of the wheelchair, usable under normal and non-normal conditions, by taking into consideration the conventional differential drive wheelchair structure with two front castor wheels. Derivation of the dynamic model, based on the Euler Lagrange formalism, is carried out in two folds: initially by considering the gravitational forces subjected to the wheelchair on inclined configurations with no slipping situations, and finally by incorporating slipping parameters into the model. Determination of the slipping parameters is approached from the geometric perspective, by considering the non-holonomic motions of the wheelchair in the Euclidean space. In the closed-loop model, the input-output feedback controller is proposed for the tracking of user inputs by torque compensation. The optimality of the resulting minimum-phase closed-loop system is then ensured through the performance index of the non-linear continuous-time generalised predictive control (GPC). Simulation results demonstrate the expected behaviour of the wheelchair dynamic model, the steering behaviour model and the assistive capability of the closed-loop system

The residential sector in England is often identified as having the largest potential for emissions reduction at some of the lowest costs when compared against other sectors. In spite of this, decarbonisation within the residential sector has not materialised. This thesis explores the complexities of decarbonising the residential sector in England using a whole systems approach. It is only when the interaction between social, psychological, regulatory, technical, material and economic factors are considered together that the behaviour of the system emerges and the relationships between different system components can be explained giving insight into the underlying issues of decarbonisation. Building regulations, assessments and certification standards are critical for motivating and driving innovation towards decarbonising the building stock. Many existing building performance and evaluation tools are shown to be ineffective and confound different policy objectives. Not only is the existing UK SAP standard shown to be a poor predictor of dwelling level energy demand but it perversely incentivises households to increase CO2 emissions. At the dwelling level, a structural equation model is developed to quantify direct, indirect and total effects on residential energy demand. Interestingly, building efficiency is shown to have reciprocal causality with a household’s propensity to consume energy. That is, dwellings with high-energy efficiency consume less energy, but homes with a propensity to consume more energy are also more likely to have higher energy efficiency. Internal dwelling temperature is one of the most important parameters for explaining residential energy demand over a heterogeneous building stock. Yet bottom up energy demand models inadequately incorporate internal temperature as a function of human behaviour. A panel model is developed to predict daily mean internal temperatures from individual dwellings. In this model, socio-demographic, behavioural, physical and environmental variables are combined to estimate the daily fluctuations of mean internal temperature demand. The internal temperature prediction model is then incorporated in a bottom-up engineering simulation model. The residential energy demand model is then used to project decarbonisation scenarios to 2050. Under the assumption of consistent energy demand fuel share allocation, modelling results suggest that emissions from the residential sector can be reduced from 125 MtCO2 to 44 MtCO2 after all major energy efficiency measures have been applied, the power sector is decarbonised and all newly constructed dwellings are zero carbon. Meeting future climate change targets will thus not only require extensive energy efficiency upgrades to all existing dwellings but also the complete decarbonisation of end use energy demand. Such a challenge can only be met through the transformation of existing building regulations, models that properly allow for the effects of human behaviour, and flexible policies capable of maximising impact from a heterogeneous residential building stock.

In diesem Buch wurde eine Übersicht und kritische Zusammenfassung des derzeitigen Forschungsstandes zu Human Activity Recognition (HAR), zu Deutsch Aktivitätserkennung bei Menschen, durchgeführt. Dabei ergab sich eine Forschungslücke im Rahmen der nicht überwachten Lernalgorithmen für HAR-Systeme. Für überwachte Lernalgorithmen muss je Anwendung ein annotierter Datensatz über mehrere Wochen mühselig erstellt werden, bevor das HAR-System zum Einsatz kommen kann. Dies entfällt mit dem neuen HAR-System. Des Weiteren ist das neue System in der Lage auch parallel laufende Aktivitäten des täglichen Lebens (ADL, aus dem Englischen Activity of Daily Living) zu erkennen. Viele HAR-Systeme aus dem aktuellen Stand der Forschung sind dazu nicht in der Lage, da sie z.B. sequenziell arbeiten. Beide Probleme wurden mit dem neuen HAR-System erfolgreich gelöst. Das in diesem Buch vorgestellte HAR-System ist eine neuartige Kombination aus einem stochastischen Modell und einem kognitiven Ansatz. Das HAR-System wird in drei Phasen angewandt. Die erste Phase ist die sogenannte Initialphase. In dieser ersten Phase wird a priori Wissen gesammelt. Das neue HAR-System benötigt im Gegensatz zu den Systemen der aktuellen Forschung nur sehr wenig a priori Wissen. Es wird die Art und Anzahl der Sensoren und der ADL benötigt, welche in eine sinnfällige initiale Verbindung miteinander gebracht werden. Diese Verbindung ist eine vorläufige und gleichverteilte Initialbelegung der Sensor-ADL-Beziehung, die in der Lernphase individuell an die jeweilige Person und Anwendungsfall angepasst wird. Es wird ein neuartiges Markov Modell (MM) und ein neu entwickeltes Impulsmodell (IM) erlernt. Das genutzte MM unterscheidet sich von den aktuellen MM durch dessen Zustandsdefinitionen, die die Sensorereigniskombinationen abbilden, wodurch das Segmentierungsproblem wegfällt. Dadurch können auch wichtige Strukturen aus dem MM extrahiert werden, die das menschliche Verhalten darstellen. Diese Strukturen werden durch neuartige Modellvergleiche bewertet. Das Resultat dieser Bewertung wird wiederum in Kombination mit dem neuen kognitiven IM in einem speziell dafür entwickelten iterativen Ansatz verwendet, um die initiale Sensor-ADL-Beziehung zu individualisieren. Diese neue Sensor-ADL-Beziehung ist Grundlage für die dritte und letzte Phase: der Anwendungsphase. Im IM wird die Sensor-ADL-Beziehung in Kombination mit neu entwickelten Regeln angewandt, um eine finale ADL-Wahrscheinlichkeitsverteilung der erkannten ADL zu berechnen. Diese besagt, welches ADL derzeit am wahrscheinlichsten ausgeführt wird und welche ADL gerade parallel zu anderen ADL ausgeführt werden. Das neue HAR-System wurde mit drei Datensätzen unterschiedlichen Anspruchs und einem Benchmark getestet. Dieser Benchmark beinhaltete vier verschiedene stochastische Modelle des aktuellen Stands der Forschung. Das neue HAR-System ist in der Lage eine höhere Erkennungsrate als der Benchmark zu leisten und war im Durchschnitt 3,2% akkurater. Es erzielte eine 95-97%-ige Wiedererkennung der ADL. Durch die erstellten Konfusionsmatrizen ergab sich eine durchschnittliche Verbesserung von 42% in den Metriken für Sensitivität, Wirksamkeit und F-Maß. Ein weiterer großer Unterschied zum Benchmark ist, dass das neue HAR-System unüberwacht lernt. Dadurch fällt die Datenakquise im Vergleich zum Benchmark sehr gering aus und das neue HAR-System wirkt attraktiver für den Markt in dessen Anwendbarkeit.:1 Einführung in die Domäne 1 1.1 Motivation 1 1.2 Ambient Assistive Living 2 1.2.1 Menschliches Verhalten und technische Assistenz 4 1.2.2 Aktivitäten des täglichen Lebens 5 1.3 Zielstellung und Abgrenzung 6 2 AAL-Systeme 9 2.1 Human Activity Recognition 9 2.1.1 Sensorik 11 2.1.2 Lernansatz 13 2.1.3 Ereignisstromanalyse 16 2.1.4 Abbildungsgrad menschlichen Verhaltens 18 2.1.5 Fazit und Anforderungen 19 3 Human Activity Recognition - Modelle 21 3.1 Datenbasierte Modelle 23 3.1.1 Deterministische Modelle 23 3.1.2 Stochastische Modelle 24 3.1.2.1 Bayes'sche Netze 25 3.1.2.2 Hidden Markov Modelle 26 3.1.2.3 Conditional Random Field 29 3.1.2.4 Neuronale Netze 30 3.1.2.5 Support Vector Machines 32 3.1.3 Vergleich und Fazit 34 3.2 Wissensbasierte Modelle 40 3.2.1 Datenbanken 40 3.2.2 Ontologien 41 3.2.3 Wahrscheinlichkeitsbasierte Ontologien 44 3.2.4 Fazit 45 3.3 Anforderungen und Forschungsfragen 46 3.4 Forschungsnaher Stand der Technik 49 3.4.1 Unüberwachter Klassifizierungsalgorithmus 49 3.4.2 Unüberwachtes Lernen mittels einer Ontologie 50 3.4.3 Wahrscheinlichkeitsbasierte Ontologie 51 3.4.4 Fazit 53 4 Lösungskonzept 57 4.1 Sensordatenschnittstelle und Definitionen 60 4.2 Initialphase 62 4.3 Lernphase 63 4.3.1 Markov Modell und menschliche Angewohnheiten 63 4.3.1.1 Das Markov Modell 63 4.3.1.2 Erlernen des Markov Modells 64 4.3.1.3 Lernen menschlicher Angewohnheiten 69 4.3.1.4 Schlussfolgerung 73 4.3.2 Abbildung des menschlichen Erinnerungsvermögen 74 4.3.2.1 Menschliches Lernen und Vergessen 74 4.3.2.2 Impulsmodell 75 4.3.2.3 Bestimmung der ADL für MM-Strukturen 78 4.3.2.4 Erlernen der Relevanzfaktoren 79 4.3.2.5 Schlussfolgerung 85 4.4 Anwendungsphase 85 4.4.1 Impulsmodell in der Anwendungsphase 85 4.4.2 ADL Erkennung 86 4.4.3 Wahrscheinlichkeitsverteilung der ADL 88 4.4.4 Schlussfolgerung und Zusammenfassung 90 5 Bewertung der Lösung 91 5.1 Testszenarien und Datensets 91 5.2 Datensätze des Benchmarks 93 5.3 Evaluierung 95 5.4 Komplexität der Lösung 107 5.5 Einschätzung der Vor- und Nachteile 108 6 Zusammenfassung und Ausblick 115 6.1 Zusammenfassung 115 6.2 Ausblick und Weiterentwicklung 116 A Fallbeispiel 119 B Evaluierung 123 B.1 Ergebnisse des Benchmarks 123 B.2 Datenschnittstelle 126 B.3 Markov Modell 129 B.4 Strukturen und deren Signifikanz 135 B.5 Ergebnisse der Relevanzfaktorenberechnung 138 B.6 ADL-Wahrscheinlichkeitsverteilung der Lösung 138 Tabellenverzeichnis 141 Abbildungsverzeichnis 143 Literaturverzeichnis 145

No / This paper outlines the current challenges of driver behaviour modelling for real-world applications and presents the novel method to identify the pattern of usage to predict upcoming journeys in probability sense. The primary aim is to establish similarity between observed behaviour of drivers resulting in the ability to cluster them and deploy control strategies based on contextual intelligence and datadriven approach. The proposed approach uses the probability density function (PDF) driven by kernel density estimation (KDE) as a probabilistic approach to predict the type of the upcoming journey, expressed as duration and distance. Using the proposed method, the mathematical formulation and programming algorithm procedure have been indicated in detail, while the case study examples with the data visualisation are given for algorithm validation in simulation.

This thesis deals with the energy saving in smart building with focus on the impact of the user behaviour on the energy consumption. The problem of human behaviour modelling has been widely studied in the state of the art, but it is still an open problem in the field of smart building since the stochastic nature of the behaviour is difficult to be accurately represented by numerical tools. An interdisciplinary approach is proposed in order to identify the suitable user features from the psychological and social point of view and to integrate such a representation into a DSS for appliance scheduling and energy cost reduction. The proposed method has exploited location-based features of the users in order to represent their habits and needs and to compute the schedules that maximize the user acceptance toward an “energy-aware” behaviour. The obtained results point out a reduction of the peak-to-average ratio higher than 40% also considering the user constraints imposed by their presence into the building.

In partnership with iMonSys - an emerging company in the passive care field - a new system, 'Verity', is being developed to fulfil the role of a passive behaviour monitoring and alert detection device, providing an unobtrusive level of care and assessing an individual's changing behaviour and health status whilst still allowing for independence of its elderly user. In this research, a Hidden Markov Model incorporating Fuzzy Logic-based sensor fusion is created for the behaviour detection within Verity, with a method of Fuzzy-Rule induction designed for the system's adaptation to a user during operation. A dimension reduction and classification scheme utilising Curvilinear Distance Analysis is further developed to deal with the recognition task presented by increasingly nonlinear and high dimension sensor readings, and anomaly detection methods situated within the Hidden Markov Model provide possible solutions to identification of health concerns arising from independent living. Real-time implementation is proposed through development of an Instance Based Learning approach in combination with a Bloom Filter, speeding up the classification operation and reducing the storage requirements for the considerable amount of observation data obtained during operation. Finally, evaluation of all algorithms is completed using a simulation of the Verity system with which the behaviour monitoring task is to be achieved.

Depuis de nombreuses années, nous nous sommes efforcés de comprendre le comportement humain et nos interactions avec l'environnement sociotechnique. Grâce à l'avancée de nos connaissances dans ce domaine, nous avons contribué à la conception de technologies et de processus de travail nouveaux ou améliorés. Historiquement, une part importante du travail d'analyse des interactions sociales fut entreprise au sein des sciences sociales. Cependant, la simulation informatique a apporté un nouvel outil pour tenter de comprendre et de modéliser les comportements humains. En utilisant une approche à base d'agents, cette présentation décrit mon travail sur la construction de modèles informatiques du comportement humain pour guider la conception par la simulation. A l'aide d'exemples issus de projets des deux domaines d'application que sont la gestion des crises et de l'urgence et la gestion de l'énergie, je décris comment mon travail aborde certains problèmes centraux à la simulation sociale à base d'agents. Le premier concerne le processus par lequel nous développons ces modèles. Le second problème provient de la nature des systèmes sociotechniques. Les sociétés humaines constituent un exemple parfait de système complexe possédant des caractéristiques d'auto-organisation et d'adaptabilité, et affichant des phénomènes émergents tels que la coopération et la robustesse. Je décris comment la théorie des systèmes complexes peut être appliquée pour améliorer notre compréhension des systèmes sociotechniques, et comment nos interactions au niveau microscopique mènent à l'émergence d'une conscience mutuelle pour la résolution de problèmes. A partir de systèmes de simulation à base d'agents, je montre comment la conscience du contexte peut être modélisée. En terme de perspectives, j'expliquerai comment la hausse de la prévalence des agents artificiels dans notre société nous forcera à considérer de nouveaux types d'interactions et de comportements coopératifs.

In partnership with iMonSys - an emerging company in the passive care field - a new system, 'Verity', is being developed to fulfil the role of a passive behaviour monitoring and alert detection device, providing an unobtrusive level of care and assessing an individual's changing behaviour and health status whilst still allowing for independence of its elderly user. In this research, a Hidden Markov Model incorporating Fuzzy Logic-based sensor fusion is created for the behaviour detection within Verity, with a method of Fuzzy-Rule induction designed for the system's adaptation to a user during operation. A dimension reduction and classification scheme utilising Curvilinear Distance Analysis is further developed to deal with the recognition task presented by increasingly nonlinear and high dimension sensor readings, and anomaly detection methods situated within the Hidden Markov Model provide possible solutions to identification of health concerns arising from independent living. Real-time implementation is proposed through development of an Instance Based Learning approach in combination with a Bloom Filter, speeding up the classification operation and reducing the storage requirements for the considerable amount of observation data obtained during operation. Finally, evaluation of all algorithms is completed using a simulation of the Verity system with which the behaviour monitoring task is to be achieved.

La consommation énergétique dans le secteur bâtiment dépend de diverses facteurs parmi lesquels ses caractéristiques physique, ses équipements, l’environnement extérieur, etc… mais il ne faut pas oublier le comportement des habitants qui est déterminant pour la consommation énergétique globale. Or, la plupart des travaux et outils représentent les occupants par des profils d’occupation. Cette thèse s’intéresse à la représentation plus détaillée du comportement des occupants, en particulier les mécanismes cognitifs, réactifs et délibératifs. Le comportement dynamique des occupants est modélisé et co-simulé avec les aspects physiques et des éventuels systèmes de gestion énergétique. L’analyse de la consommation de différents équipements électroménagers met en évidence que le consommation énergétique est très dépendante des comportements des occupants. L’analyse des consommations et des actions des habitants permet d’élaborer un modèle du comportement des occupants impactant la consommation énergétique. Le modèle représente des mécanismes cognitifs, qui représente les causes qui motivent les actions, incluant des échange avec d’autres acteurs humains. Une approche à base d’agents logiciels a été développée. Outre les aspects techniques, une méthodologie de réglage des paramètres des modèles de comportement est proposée. Ces outils sont utilisés pour réaliser une co-simulation représentant la physique du bâtiment, le comportement réactif, c’est-à-dire sensible aux données physiques, et délibératif des habitants mais aussi un système de gestion énergétique qui peut ajuster directement la configuration du logement ou simplement conseiller ces occupants. L’impact de différents types de comportements, avec et sans gestionnaire énergétique est analysé. Ces travaux ouvrent de nouvelles perspectives dans la simulation bâtiment, dans la validation de gestionnaires énergétiques mais aussi dans la représentation des bâtiments dans les réseaux d’énergie dits intelligents, dans lesquels des signaux peuvent être envoyés aux utilisateurs finaux pour les inviter à moduler leur consommation. / Energy consumption in buildings is affected by various factors including its physical characteristics, the appliances inside, and the outdoor environment, etc. However, inhabitants’ behaviour that determines the global energy consumption must not be forgotten. In most of the previous works and simulation tools, human behaviour is modelled as occupancy profiles. In this thesis the focus is more on detailed behaviour representation, particularly the cognitive, reactive, and deliberative mechanisms. The inhabitants’ dynamic behaviour is modelled and co-simulated together with the physical aspects of a building and an energy management system. The analysis of different household appliances has revealed that energy consumption patterns are highly associated with inhabitants’ behaviours. Data analysis of inhabitants’ actions and appliances’ consumptions is used to derive a model of inhabitants’ behaviour that impacts the energy consumption. This model represents the cognitive mechanisms that provide causes that motivate the actions, including the communication with other inhabitants. An approach based on multi-agent systems is developed along with a methodology for parameter tuning in the proposed behaviour model. These tools are used to co-simulate, not only the physical characteristics of the building, the reactive behaviour that is sensitive to physical data, and deliberative behaviour of the inhabitants, but also the building energy management system. The energy management system allows the direct adjustment of the building parameters or simply giving advice to the inhabitants. The impact of different types of inhabitants’ behaviours, with and without the inclusion of an energy management system is analyzed. This work opens new perspectives not only in the building simulation and in the validation of energy management systems but also in the representation of buildings in the smart grid where signals can be sent to end users advising them to modulate their consumption.