Autoregressiv analys på tidsseriedata från en kontorsbyggnad – Smarta byggnader i teori och praktik
InformationFörfattare: Astrid Gustafsson, Clara Grönlund
Beräknat färdigt: 2020-05
Handledare: Maria Alexandersson
Handledares företag/institution: Sweco
Ämnesgranskare: Kristiaan Pelckmans
PresentationerPresentation av Astrid Gustafsson
Presentationstid: 2020-06-05 14:15
Presentation av Clara Grönlund
Presentationstid: 2020-06-05 15:15
Opponenter: Julia Stålenheim, Johanna Holmqvist Larsson
The building sector is responsible for around 40% of the energy consumption in Europe, and one way to work towards sustainable societies could be to make the buildings more energy efficient. One approach to make a building more energy efficient is to use knowledge gained from digitalization of the building and to make the building smart. This thesis aims to study the area of smart buildings, and the ongoing work with smart solutions within the real estate sector.
Two parallel investigations are used to study the area. One is an interview study in order to map the ongoing work with smart buildings. The situation on the market, the matureness of technical solutions as well as ongoing trends and challenges are amongst other things studied. The second investigation consists of a pilot project which aims to exemplify how time series data analysis could be used in order to make a building smarter. Time-series prediction provides a way to discover and quantify regularities in such data, and methods of time series prediction point to how to make building management more efficient.
The result of the survey shows that the smart building market is not yet stabilized, but that the interest in working with smart buildings is big amongst real estate managers. There are many smaller solutions which are being tested and implemented, but there is no consensus of what the definition of a smart building really is. The results of the data analysis indicate two results: (1) it provides insight in the data, and reports how one should prepare the data for subsequent analysis, and (2) we report results for different autoregressive (AR)-based time-series models. For (2), we indicate how methods of K-means improve over linear AR-based modelling, pointing to the possible use of nonlinear modelling. We however question whether performance improvements are sufficiently large for this application to justify the additional (computational) demands.