Paper Conference

Proceedings of BauSim Conference 2020: 8th Conference of IBPSA-Germany and Austria

     

BUILDING SIMULATIONS FOR CONTROL TUNING: ACCOUNTING FOR MODEL USEFULNESS IN CALIBRATION METRICS CHOICE

A. Bres, F. Amblard, S. Hauer

DOI: https://doi.org/10.3217/978-3-85125-786-1-79
Abstract: 1 n Calibration, defined as the adjustment of uncertain parameters to achieve a better agreement between 𝑀𝑆𝐸 = 𝑛 Ξ£(𝑦^i β€” 𝑦i)Β² i=1 (1) simulation and measurements, is a key task in modeling existing systems. Different calibration metrics may be used to quantify this agreement. In the field of building performance simulation, a few The coefficient of variation of the root mean squared error defined in Equation (2) provides a normalized value which can be expressed in percent. metrics are established in practice but there is still little 𝐢𝑉𝑅𝑀𝑆𝐸 = βˆšπ‘€π‘†πΈ (2) evidence on how to assess and select calibration metrics. This paper investigates the assumption that the values of an adequate calibration metric for given simulations models should correlate with the usefulness of these models. Investigations are carried out with synthetic data for an example heat-pump system serving a residential building, looking at model usefulness for the specific task of tuning control parameters. 1 n 𝑛 i=1 i The normalized mean bias error defined in Equation (3) is also expressed in percent, and is known to be only weakly informative because positive and negative biases at different time steps may cancel each other out (Royapoor & Roskilly, 2015). βˆ‘βΏ 𝑦^i β€” βˆ‘βΏ 𝑦i 𝑁𝑀𝐡𝐸 = i=1 n i=1 (3)
Pages: 655 - 662
Paper:
bausim2020_79