This thesis explores the application of graph theoretical and topological concepts in architecture and investigates the use of graph machine learning methods in the context of architectural analysis, with a particular focus on energy efficiency as a key performance metric. To this end, a synthetic architectural dataset containing geometric, categorical, dimensional, energetic, topological and relational information is generated by integrating various space partitioning algorithms combined with architectural control functions into an automated generation pipeline. Subsequently, a classification model and a regression model are trained on the generated knowledge graph dataset to evaluate the prediction and classification accuracy in terms of energy efficiency.

The resulting dataset and the code for generating and training the model will be made publicly available to further research in the field of graph machine learning in architectural applications. This research demonstrates the potential of a closer integration of various mathematical concepts and computer science methods into the architectural design and verification process, and shows the potential of applying knowledge graphs for the abstraction, representation and analysis of architectural objects.

TECHNIQUES : PYTORCH | DGL | TOPOLOGICPY | PYTHON