A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Time-Resolved Neural Network Surrogate Models as Digital Twins
Digital Twins (DTs) are a relatively new concept in the construction industry. They usually rely on calibrated physics-based simulation models for prediction purposes. Such models are resource-intensive and therefore expensive to build and deploy. To address this, surrogate models provide a suitable alternative for DTs as these are much faster to evaluate (and hence calibrate) and potentially more flexible to deploy. This research devises a neural network-based surrogate model implementation using easily accessible building parameters and weather data. The methodology constitutes parameter mapping, time-series surrogate fitting, and output validation. The resulting output provides the time-series building energy consumption patterns for any use case. To demonstrate the applicability of this methodology, this study uses the US Department of Energy Medium Office archetype for time-series surrogate formulation. When considering electricity and natural gas consumption on a 10-min resolution, the CNN surrogate provides a better fit for electricity (R2 = 0.977) against natural gas (R2 = 0.824). Furthermore, the predictions were more accurate for natural gas predictions (RMSE = 0.094 kWh) against electricity predictions (RMSE = 2.473 kWh).
Time-Resolved Neural Network Surrogate Models as Digital Twins
Digital Twins (DTs) are a relatively new concept in the construction industry. They usually rely on calibrated physics-based simulation models for prediction purposes. Such models are resource-intensive and therefore expensive to build and deploy. To address this, surrogate models provide a suitable alternative for DTs as these are much faster to evaluate (and hence calibrate) and potentially more flexible to deploy. This research devises a neural network-based surrogate model implementation using easily accessible building parameters and weather data. The methodology constitutes parameter mapping, time-series surrogate fitting, and output validation. The resulting output provides the time-series building energy consumption patterns for any use case. To demonstrate the applicability of this methodology, this study uses the US Department of Energy Medium Office archetype for time-series surrogate formulation. When considering electricity and natural gas consumption on a 10-min resolution, the CNN surrogate provides a better fit for electricity (R2 = 0.977) against natural gas (R2 = 0.824). Furthermore, the predictions were more accurate for natural gas predictions (RMSE = 0.094 kWh) against electricity predictions (RMSE = 2.473 kWh).
Time-Resolved Neural Network Surrogate Models as Digital Twins
Environ Sci Eng
Wang, Liangzhu Leon (editor) / Ge, Hua (editor) / Zhai, Zhiqiang John (editor) / Qi, Dahai (editor) / Ouf, Mohamed (editor) / Sun, Chanjuan (editor) / Wang, Dengjia (editor) / Kotha, Rajeev (author) / Lédée, François (author) / Shamsi, Mohammad Haris (author)
International Conference on Building Energy and Environment ; 2022
Proceedings of the 5th International Conference on Building Energy and Environment ; Chapter: 157 ; 1519-1528
2023-09-05
10 pages
Article/Chapter (Book)
Electronic Resource
English
Digital Twins of Cities vs. Digital Twins for Cities
| Springer Verlag | 2024
Development of surrogate models using artificial neural network for building shell energy labelling
| BASE | 2014
| Wiley | 2021
Analysis of Moving Loads on Beams Using Surrogate Models Inspired from Artificial Neural Network
| Springer Verlag | 2024