ASSESSING THE RELEVANCE OF REDUCED ORDER MODELS FOR BUILDING ENVELOP

The paper focuses on parameter estimation processes for physically meaningful models tuned online and define a process to determine whether a model is relevant or not for GMBA-BEMS tuning purpose. The proposed approach relies on the data coming from the PREDIS/MHI platform. The first step is to calculate realistic parameters with possible intervals because nonlinear optimization, required for physically explicit models, implies initial parameters. The next step is to find the best reduced order model structure using an iterative nonlinear optimization approach using recorded data that leads to parameter estimation. It is based on randomize initial values for parameters to measure the convexity of the search space in studying the convergence. Finally, the last step consists in enhancing the time zones where reduced order model does not fit well with the available data. It points out some non-modeled phenomena. It is based on a weighted iterative estimation method where weights depend on the estimation errors obtain at the previous step.