*Abstract*
Building operations contribute approximately one-third of global CO₂ emissions. Advanced control strategies can reduce these emissions by up to 30%. Such control requires accurate mathematical models that capture the building’s thermal dynamics. Data-driven modeling has emerged as the most scalable approach for this purpose. However, the availability of high-quality building data remains limited. To address this challenge, we propose two methods: (1) a data generation framework that synthesizes realistic building operation data, and (2) a general Transfer Learning model that serves as an effective initialization for modeling new target buildings.
*Bio*
Fabian is a second-year PhD student in the Department of Energy Management Technologies at the Technical University of Munich, supervised by Prof. Dr. Christoph Goebel. His research focuses on using Machine Learning to model building thermal dynamics. Such models are necessary for enabling Model Predictive Control of the building, which can reduce CO₂ emissions by up to 30%.
