Annie Axt

Monday, September 07, 2026, 15:00 - 16:00

Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany

Abstract: Many input signals relevant to building simulation show periodic behavior, which allows them to be decomposed into sinusoidal frequency components. We therefore investigate which properties of such signals determine how reliably their relevant frequency components can be identified, and how accurately the output can be approximated from them. To answer this question, we identify dominant frequencies in an input signal via
spectral analysis and estimate their amplitude and phase using the least-squares method. We determine the required number of frequencies sequentially, based on the normalized error between the approximated and simulated output, and obtain the total output by superposing the individual frequency contributions. We apply this methodology, which is generally applicable to LTI systems, to an RC network model of a building with multiple inputs. The results show that approximation quality depends on the concentration of signal power in the spectrum, the sensitivity of the output to the respective input, and nonlinearities in signal generation. How strongly these properties are present and how they combine differs across the four inputs studied, with the overall error largely determined by the least approximable signals. The RMSE of the approximated total output is about 3.8C, well above the VDI 6020 reference value of 1.5C.