Il a été développé 3 approches pour l'estimation d'occupation (nombre d'occupants et potentiellement leur activité) à base de techniques de machine learning :- une technique d'apprentissage supervisé exploitant différentes typologies de capteurs,- une technique à base de connaissance sans capteurs- et une technique interactive sollicitant lorsque c'est opportun les occupants via des questions. / Building energy management and monitoring systems (EMMS) should not only consider building physics and HVAC systems but also human behavior. These systems may provide information and advice to occupants about the relevance of their behavior regarding the current state of a dwelling and its connected grids. Therefore, advanced EMMS need to estimate the relevance of occupant activities. Additionally, innovative end-user services such as replay past situations, anticipate the future or mirror the current state are under development and require models together with building state estimations including the human part of the state. However, to define the state of a zone, non-measured values should be known in both physical (i.e. heat flows) and human part (i.e. occupancy and activities).The problem is to identify and calculate data processed from sensors, calendars, etc… that could be used in a classification model to estimate the number of occupants and various activities happening in offices/homes. The sensor data must provide a rich context for a classifier to have a broad separation plane and represent the office situation closely. Since the use of video cameras is a problem in many areas, the solution must respect privacy issues and relies largely on non-intrusive sensors.The thesis identifies the most relevant calculation from the sensor data in order to classify the number of people in a zone and their activities in offices/homes at a given time period. The proposed approach is inspired from machine learning and interactive learning to avoid using the camera and build a general estimation method.Three approaches are proposed for occupancy and activities estimation:- supervised learning approach. It starts to determine the common sensors that shall be used to estimate and classify the approximate number of people (within a range) in a room and their activities. Means to estimate occupancy include motion detection, power consumption, CO2 concentration sensors, microphone or door/window positions. It starts by determining the most useful measurements in calculating the information gains. Then, estimation algorithms are proposed: they rely on decision tree learning algorithms because it yields decision rules readable by humans, which correspond to nested if-then-else rules, where thresholds can be adjusted depending on the considered living areas. An office has been used for testing.- knowledge base approach using sensor data and knowledge coming respectively from observation and questionnaire. It relies on hidden Markov model and Bayesian network algorithms to model a human behavior with probabilistic cause-effect relations and states based on knowledge and questionnaire. Different applications have been studied for validation: an office, an apartment and a house.- an interactive learning approach is proposed. It estimates the number of occupants in a room by questioning occupants when relevant, meaning limiting the number of interactions and maximizing the information gains, about the actual occupancy. Occupancy and activities estimation algorithms use information collected from occupants together with common sensors. A real-time application has been done in an office case study.
Identifer | oai:union.ndltd.org:theses.fr/2017GREAT112 |
Date | 03 October 2017 |
Creators | Amayri, Manar |
Contributors | Grenoble Alpes, Ploix, Stéphane |
Source Sets | Dépôt national des thèses électroniques françaises |
Language | English |
Detected Language | English |
Type | Electronic Thesis or Dissertation, Text |
Page generated in 0.0022 seconds