@inproceedings{simbuild2024_2144,
	doi = {},
	url = {https://publications.ibpsa.org/conference/paper/?id=simbuild2024_2144},
	year = {2024},
	month = {May},
	publisher = {IBPSA-USA},
	author = {Chul-Hong Park and  Seongkwon Cho and  Tae Yong Song and  Seon-Young Heo and  Cheol-Soo Park},
	title  = {Physics-Informed Hybrid Modeling Approach for Room Temperature Prediction Using an RC Model and Siamese Neural Network},
	booktitle = {Proceedings of SimBuild Conference 2024},
	volume  = {11},
	isbn = {},
	address  = {Denver, Colorado},
	series  = {IBPSA-USA Building Simulation Conference},
	pages = {698--704},
	abstract = {In this paper, a novel hybrid modeling approach is proposed to to combine the advantages of physics-based and data-driven approaches for predicting the thermal behavior of the building. It incorporates an RC model and neural networks by designing custom layers and utilizing a neural network modeling technique known as “Siamese neural network”. The neural network is used to predict various time-invariant and time-varying parameters in the RC model. The modeling technique allows flexibility in the model design, simultaneous training with both time-invariant and varying parameters present, warmup period and multiple-timestep forecasting per input during the training phase, and training the model with a limited number of measured states. To validate the proposed approach, it was applied to an existing building located in South Korea, using the measured data from a single air handling unit (AHU) serving an office area located on the 5th floor. The trained model was used to predict the room air temperature for the test period. It was found that a simple RC model combined with the Siamese neural network was good enough to predict room air temperature.},
	issn = {},
	Organisation = {IBPSA-USA},
	Editors = {}
}