Bachelor's thesis Robert Plümpe
Intelligent modelling of connected data structures for energy systemsCopyright: EBC
In order to achieve future climate goals, more and more machine learning algorithms are being used in building energy systems to reduce energy consumption. These algorithms require a large amount of structured data and semantic information from building energy systems in their development and application. In order to collect these data and for these data and to make them available for further applications, so-called context information systems are being developed. There are already databases that store context information and offer semantic interoperability through the use of ontologies, as well as applications that facilitate the retrieval of information from different sources and thus increase data exchange. A combination of these applications and databases has already been proposed in an approach by Li et al. .
The aim of this thesis is to apply the methodology of Li et al. . Therefore, a context information system based on a property graphmodel was createdwith the help of Python and the graph database Neo4j. Parts of the ontologies Brick and SAREF4BLDG were used as well as the BUDO-key, which proved to be a useful tool for initializing the context information system. The methods presented in this thesis for initializing a context information system are tested on the data points of the E.ON ERC main building. In addition, the TICC algorithm is used to show how other applications can access the context information system to add or retrieve data. In addition, external context suppliers are integrated into the context information system in form of a time series database.
Overall, it has proven useful to create a context information system. However, there is a lack of further context information on themore accurate structure of the building energy system of the E.ON ERC main building, so that no major benefit can be derived fromthis context information model. An implementation with more context information and on a larger scale, however, will be of future relevance. Since it is particularly important for the acceptance and use of the systems that little manual effort is required to set them up, further methods for automatic information retrieval should be explored. In addition, it is important to adapt ontologies to property graph models in the future and to extend additional devices, relationships and attributes so that a comprehensive use of context information systems becomes possible.