Estefania Alvarez del Castillo Cardoso

Abstract: 
This thesis presents a numerical system for the energy management in the residential sector. The objective is to include the inhabitants in the process of energy management so that they can learn to have more energy sober behaviour. We propose a tool of self-experimentation of energy and comfort impacts. This tool can allow to associate sensor measurements (the effects) to its causes (human behaviour). Given that the residences include a wide diversity of contexts and multiple life events may happen, even simultaneously, it is proposed that the inhabitants can decide the life events they wish to analyse. We made the hypothesis that the inhabitants wonder about the energy consumption or comfort impacts that activities or contexts might have. The self-exploration tool, here proposed can be configured by the inhabitants so that they run their experiments. A survey was carried out to know the kind of questions that the inhabitants might have as well as the interest in the proposed tool. Some prototypes of the interface of the system are presented and analysed. Two processes to associate causes and effects, are provided : annotations-free recognition and annotations-based recognition. Annotations-free recognition can be carried out when the sensors information is enough to answer to the inhabitants question. The annotations based recognition is applicable when certain information cannot be recovered using only sensors data. Specifically, information often linked to intentions and preferences of the inhabitants, as this is what corresponds to their psychological experience. A couple of methodologies to support the occupants in the annotation process are presented : the interactive and cooperative learning (Silva et al., 2022) and the a posteriori annotations system. The interactive learning processes, is an supervised learning method that define the moments in which information is requested to the inhabitants and ensures the quality of the database. At the same time, the sensors information (effects) is associated to the annotations provided by the inhabitants (human behaviour). The cooperative learning process is also a supervised learning method, that spots incoherences between the recorded associated information, the system notifies this to the occupants and asks for a verification. There is a symmetrical interaction between the system and the inhabitant to improve the learned knowledge. The a posteriori annotations system, allows the occupants to provide their annotations later in time, rather than answering when annotations are requested. It is based on a Change Point Detection method. It spots the moments of changes in the sensors measurements , which can indicate a change of activities. The idea is to show the moments in which one activity could have ended to start a new one. This should work as a reference to help the inhabitants remember past life events so that they can annotate, even in a posterior moment. In the end of the experiment carried out by the inhabitants they receive the information related to their study. It is proposed that the inhabitants select the graphical visualizations and the information that help them to better understand the results. Annotations were recovered from two different sites and the annotation process was evaluated on how it helps inhabitants to better understand the impacts of different situations in the residential 1 Abstract context. Some tests of the interactive and cooperative learning were carried out using the annotations recovered as well as the data from the installed sensors in the sites.