Master's Thesis Thomas Rosen

 

A hybrid data- and design-driven grey-box modeling approach for characterizing building thermal performance

Prediction of internal air temperature based on grey-box modeling Copyright: EBC Prediction of internal air temperature based on grey-box modeling

The consideration of embedding building models into a district heating system can be an important factor as the building sector is huge energy consumers and provides high optimization potential. This work describes design-driven and data-driven building grey-box models, which use lumped capacity models to represent the thermal behavior and way to combine these models to a hybrid model. While the design-driven model parameters are calculated mainly on basis of wall materials and sizes providing a better physical representation, the data-driven model parameters are computed by non-linear parameter estimation and provides more robustness, as not modeled influences can also be merged into the structure. Within the scope of this work, the data-driven approach was implemented into the tool PyMPC. Several estimation methods were implemented, analyzed on their behaviour and the influence of data, including CTSM-R, which estimates via an extended Kalman filter. On the basis of TEASER, a tool providing methods for design-driven building grey-box modeling, a hybrid model was formed, allowing to produce more physical parameters for unknown buildings, than in the pure data-driven approach.