< Back to previous page

Project

User profiling methods for personalized products

Environmental issues such as growing waste streams, Earth Overshoot Day that occurs earlier each year and the increasingly visible impact of climate change on the world have, since the last two decades or so, led to a growing environmental awareness, spurring a vast body of research into sustainability (e.g. circular economy, renewable energy, etc.) and incentivising humanity to adopt more environmentally conscious behaviour. On top of this, ever cheaper, more powerful electronics and the associated rise of the internet-of-things are facilitating the development of intelligent systems that automatically optimise the efficiency-comfort trade-off. Such systems possess an inherent potential to substantially reduce resource consumption, while maintaining a high degree of comfort, as they exploit historical usage information to tailor system functionality to user requirements. Whereas interactive or manually scheduled control is generally sub-optimal, resulting in more discomfort, elevated consumption or both.

A wide variety of applications can be addressed by these intelligent, personalized products, such as smart heating, dynamic power management, fault detection and load shifting. Yet, techniques reported in literature are typically directed at a single application. Moreover, they are generally limited to modelling a single source of usage information and lack a mechanism to deal with evolving data, yielding the first research goal.

Research Objective 1: The development of a generic user profiling strategy for intelligent control systems aimed at reducing resource consumption, that is:
a) able to tune the savings-comfort trade-off,
b) adapt to changing behaviour,
c) and deal with multi-user environments.


To comply with the generic nature of the envisaged system, a framework for resource minimising intelligent control systems was iteratively developed and refined on academic and industrial case studies. However, evaluation by means of a proof-of-concept is limited, due to the complexity and cost of real-life test setups, to smart (zone-)heating systems. In literature, performance analysis of smart heating systems mainly relies on simulations, which illustrate their energy saving potential. However, the feasibility of such approaches remains to be corroborated by real-life implementations. The second research objective is directed at this lack of real-world experiments.

Research Objective 2: An extensive evaluation of deploying the developed smart heating system in the field, which can be used to assess its energy and financial savings, and environmental impact reduction potential.

Evaluation of the presented framework has demonstrated the modelling algorithm’s competence to infer qualitative usage profiles for several applications of varying complexity and exploit them to forecast future usage. Furthermore, its energy saving potential, whilst retaining comfort, as a smart heating system has been proven by two experiments; first in a single-user office environment and second in student rooms of a university residence hall. Finally, to generalise the findings of the experiments, the required minimal energy savings of a smart heating system, on a yearly basis, to obtain environmental and economic benefits for residential heating in Flanders, Belgium were estimated by life cycle analysis and life cycle costing methods.

Overall, such a smart heating system is easy to install and can be retrofitted in virtually any building. It constitutes a cost efficient approach to achieve substantial energy savings, compared to the extensive renovation associated with upgrading the building’s insulation or heating system. Although, a smart heating system will (almost) always yield environmental benefits, their usefulness is limited in modern, energy efficient buildings that, for example, rely on heat pumps making use of clean/green energy. Nevertheless, even then energy efficiency remains important. The presented system is thus particularly suited for offices, student rooms, rental suites, etc. Evidently, the savings potential is greatest in case of old or protected heritage, poorly insulated buildings in cold regions, heated with costly, strongly polluting fuel. And especially in those where residents pay a flat fee for heating costs, as in this case residents’ heating behaviour might be less energy efficient.

Date:9 Dec 2013 →  21 Jan 2022
Keywords:Intelligent systems
Disciplines:Adaptive agents and intelligent robotics, Data mining, Machine learning and decision making
Project type:PhD project