Energy management of heating, ventilating and air-conditioning (HVAC) systems is a main concern in the design and project of buildings. Artificial neural networks (ANNs) are very useful in representing highly non-linear problems, such as the HVAC system. Neural networks' correct sizing is important to have a trade-off between model accuracy and computational cost. Therefore, the main idea of this work was to identify the optimal size of the neural network used to model the relationship between temperature and energy demand in HVAC system building. To this, a scholastic building with low and high energy performance levels located in Bolzano, Perugia and Catania was modelled in EnergyPlus environment to obtain thermo-electric profile databases to be employed for the training of the feedforward neural network. To find out the optimal size of hidden layers, different trainings of the ANN were carried out by varying the neurons' number of the first and second hidden layer and, as a pilot study, the optimal sized ANN was used to predict the thermo-electric model of the HVAC scholastic building. The validation error and the standard deviation were calculated for each combination of neurons' number of the first and second hidden layer. Results demonstrated that the validation error was always lower than 0.014 and its minimum value was obtained by increasing the number of neurons of both hidden layer and, therefore, the complexity of the ANN.

Palermo, M., Forconi, F., Belloni, E., Quercio, M., Lozito, G.M., Riganti Fulginei, F. (2023). Optimization of a feedforward neural network's architecture for an HVAC system problem. In International Conference on Electrical, Computer, Communications and Mechatronics Engineering, ICECCME 2023 (pp.1-6). Institute of Electrical and Electronics Engineers Inc. [10.1109/ICECCME57830.2023.10252568].

Optimization of a feedforward neural network's architecture for an HVAC system problem

Palermo M.;Forconi F.;Belloni E.;Quercio M.;Riganti Fulginei F.
2023-01-01

Abstract

Energy management of heating, ventilating and air-conditioning (HVAC) systems is a main concern in the design and project of buildings. Artificial neural networks (ANNs) are very useful in representing highly non-linear problems, such as the HVAC system. Neural networks' correct sizing is important to have a trade-off between model accuracy and computational cost. Therefore, the main idea of this work was to identify the optimal size of the neural network used to model the relationship between temperature and energy demand in HVAC system building. To this, a scholastic building with low and high energy performance levels located in Bolzano, Perugia and Catania was modelled in EnergyPlus environment to obtain thermo-electric profile databases to be employed for the training of the feedforward neural network. To find out the optimal size of hidden layers, different trainings of the ANN were carried out by varying the neurons' number of the first and second hidden layer and, as a pilot study, the optimal sized ANN was used to predict the thermo-electric model of the HVAC scholastic building. The validation error and the standard deviation were calculated for each combination of neurons' number of the first and second hidden layer. Results demonstrated that the validation error was always lower than 0.014 and its minimum value was obtained by increasing the number of neurons of both hidden layer and, therefore, the complexity of the ANN.
2023
979-8-3503-2297-2
Palermo, M., Forconi, F., Belloni, E., Quercio, M., Lozito, G.M., Riganti Fulginei, F. (2023). Optimization of a feedforward neural network's architecture for an HVAC system problem. In International Conference on Electrical, Computer, Communications and Mechatronics Engineering, ICECCME 2023 (pp.1-6). Institute of Electrical and Electronics Engineers Inc. [10.1109/ICECCME57830.2023.10252568].
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11590/456247
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? ND
social impact