Naslov
Proposed RC Model Structure for Estimation of Building Performance Using Limited Data

Autori
Badun, Nikola ; Zakula, Tea

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Digital proceedings of the 17th SDEWES Conference on Sustainable Development of Energy, Water and Environment Systems / - , 2022, 1-13

Skup
17th Conference on Sustainable Development of Energy, Water and Environment Systems (SDEWES 2022)

Mjesto i datum
Pafos, Cipar, 06.11.2022. - 10.11.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
predictive control, building simulation, RC model, parameter estimation, root-mean-square-error

Sažetak
The greatest obstacle to large-scale utilization of predictive control in buildings is acquiring a suitable building model. Due to its interpretability and smaller training periods, the resistance-capacitance (RC) model is one of the most widely used models for buildings. However, in most cases, a modeler needs to develop a model structure specific to the respective building and approximate its parameters based on the building’s physics, which is error-prone and time-consuming. This paper proposes an RC model structure to predict temperatures for various buildings based solely on four easily measured inputs. The goal of the paper is to show that this simple structure trained on a limited dataset can effectively predict internal temperatures for various inputs. Model parameters were estimated using training data for different parts of the year, namely, heating, transitional and cooling periods, to generate three models. The least squares regression and global start method were used to reduce non-convexity issues. The developed models were validated in heating, transitional and cooling validation periods, each lasting 48 hours. The root-mean-square errors (RMSE) ranged from 0.2°C to 0.8°C, indicating low-error models. The conclusion is that the proposed model structure can be used for temperature prediction without considering the specific characteristics of the building in question, as long as the limited data set is available. To some extent, the model parameters capture the building’s physics. The models trained over a heating period can be used for predictions over the cooling period and vice versa, which is demonstrated by comparing prediction errors for three different validation periods.

Izvorni jezik
Engleski

Znanstvena područja
Strojarstvo

POVEZANOST RADA