Workshop on Optimal Control of Thermal Systems in Buildings using Modelica location:University of Freiburg date:23 - 24 March 2015
Building energy systems are complex and non-linear systems. Many different control algorithms exist to control the HVAC system in buildings. A true comparison of different control algorithms is infeasible in practice (identical conditions cannot be guaranteed). Therefore, an emulator should serve as virtual test bench, on which different controllers can be tested in identical conditions. For the comparison to be meaningful, the virtual test bench should represent a real building, with all existing physical dynamic interactions.
The construction of a ‘valid’ detailed emulator model is a cumbersome and iterative process, which is facilitated by the object-oriented, graphical and physical modelling approach in Modelica. The resulting model has a very large amount of parameters of which some values are unknown or uncertain to the modeller. This leads to models, which might represent a building, but are not an appropriate model for the building under study. By tuning the unknown parameter values the model can be calibrated to better fit the measurement data. Considering large simulation times for large building models, the tuning should be handled in a smarter way than single handed adjusting and comparing.
The model calibration can be performed by the following two-step procedure. In a first step, a sensitivity analysis is performed rank the parameters based on their influence on the fit of model output with measurement data. In a second step, a number of these top ranked parameters are entered as optimization variables in Genopt, a general optimization tool. All parameters should be specified with initial guess, min and max values and a stepsize and are optimized to minimize the fit error between model output and measurement data. After these steps, one possible optimal parameter set for the given model highlights the model’s shortcomings and drives model updates and refinements. This procedure is repeated until the model predicts all relevant system phenomena and the model, with a resulting set of parameter values is accepted.This final model will be a good representation of the building. However, it might not contain the optimal parameter values (local optimum in a highly non-linear problem) and should be validated (or recalibrated) for different times in the year. The modeller does improve his confidence in the model and assures himself this is an appropriate model for a building.