Master's Thesis David Wackerbauer

 

Identification of building systems and operation characteristics from building energy data

Illustration of the data-mining process Copyright: EBC Illustration of the data-mining process for the extraction of information from building energy data

Dynamic simulation by means of physical models depends on information about the buildings' equipment and operation. These information are often not available and acquisition requires manual efford. Building energy data becomes increasingly available through the digitalization of building technology, esecially Smart-Meters. Data-Mining methods can potentially extract required information from this data. The methodology introduced in this study enables the extraction of certain characteristics of buildings' technical equipment and operation, using measured data on electric and thermal load as well as climatic data. A feature-based approach is implemented to gain information by means of classification and regression algorithms from building energy data. These algorithms depend on a dataset with information on the seeked characteristics. An evaluation of the datasets from the Jülich Research Centre showed, that methodology can not be developed using this dataset due to a lack of the required information about these buildings. Therefore the methodology is developed and tested on simulation datasets. Testing shows, that a selection of different algorithms is applicable for the analysis of the datasets. A detailed optimization and evaluation is performed using Support Vector Machines for calssification and K-Nearest Neighbour for regression. The results yield high identification rates for the seeked characteristics using simulation data. These models based on simulation data are then used to extract information from real datasets. Comparing the results with verifiable characteristics from real buildings shows high accordance. Causes are discussed for characteristics that could not reliably be identified. Furthermore the features calculated from the datasets within the process can reveal defferences between the simulated and real datasets. An analysis indicates areas in which simulation models might be improved to represent real buildings.