ETH Zurich :
Computer Science :
Pervasive Computing :
Distributed Systems :
Student Projects :
Opportunistic Sensing for Smart Heating Control Systems (M)
Global warming, the lack of fossil fuels and the resulting increase in energy prices induce an urgent need for sustainable buildings. Heating accounts for 70% of the domestic energy consumption in Switzerland. Some of this energy may be saved using thermostats which allow the temperature to fall to a deep setback while the occupants are not present. However, as manual thermostats are often cumbersome to use (e.g. re-programming is often forgone due to their complexity), automatic solutions - so-called "smart thermostats" - should be preferred.
In order for these to work automatically we need to sense occupancy e.g. find out whether someone is home or not. There are several approaches for sensing occupancy using different types of sensors. A desirable way of doing this is “opportunistic sensing”, in which already present sensors are used, avoiding the need to purchase and install additional sensors.
Examples for this would be reading the electric load curve recorded by smart meters or using data from existing motion sensors used by security systems.
The goal is then to build a system which combines the data from several heterogeneous sensors and applies machine learning algorithms to learn and be able to predict the occupancy state of the household. In this thesis the aim is to build such a system on a single low-cost single-board computer (i.e. a RaspberryPi). This not only reduces cost, making it more feasible for residential households than alternative solutions, but also reduces the privacy concerns, since all data is recorded and processed locally.
These predictions, along with a model of the building, are then used to control the thermostats and the heating system of the building in order to heat in an optimized way (e.g. save as much energy as possible, while having pleasant temperature whenever somebody is home). For this purpose RF-controlled heating valves (Honeywell HR-20) will be used. The machine predicting and sensing the occupancy states of a household then uses state-of-the-art infrastructure (CoAP) and control algorithms (MP) to execute the necessary actions.
The thesis can be split into multiple parts, in the first one the system will be defined. In order to achieve this, temperature measurements will be taken and different models will be analyzed and compared against each other to find the one best fitting. For the measurements a first system will be installed in the targeted house which can later be altered and run algorithms necessary for later parts of the thesis. Nodes will be installed in different rooms of the building and the single-board computer can read out the current temperature by issuing CoAP requests. Additionally the sensors and algorithms for occupancy sensing have to be identified. The focus will be on using opportunistic sensing . Included in this part is also a literature review in the corresponding fields.
In a second part the control infrastructure will be built up. Temperatures of the individual rooms will have to be recorded in order to regulate them according to the desired specification. Furthermore a best control algorithm (MPC, ) to regulate the heating system has to be identified and optimized to run on a single-board computer.
The control will be done on a per-house level, this means while different rooms can be defined to have different setback and comfort temperatures, the occupancy will be modeled as a binary variable (e.g. somebody is home or not). The idea here is not to find a perfect solution but rather find limitations of these approaches.
Single-room control (optional)
This part will be optional, the control infrastructure would here be expanded to a single-room version. Here occupancy will not only be binary, instead the number of people that are currently home and the identity of these people will also be taken to account. These expansions will make the system a lot more complex, since we also have to look at thermal dependencies between different rooms (e.g. how much more will we have to heat a room if the one next to it is at the setback temperature compared to if it is at the comfort temperature.
 L. Yang and M. B. Srivastava, “Inferring Occupancy from Opportunistically Available Sensor Data.”
 D. Rogers, M. Foster, and C. Bingham, “,” Build. Environ., vol. 72, pp. 356–367, 2014.
 S. Prívara, J. Široký, L. Ferkl, and J. Cigler, “Model predictive control of a building heating system: The first experience,” Energy Build., vol. 43, pp. 564–572, 2011.
Student/Bearbeitet von: Marc Hueppin
Contact/Ansprechpartner: Wilhelm Kleiminger