A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Using artificial neural networks to predict the impact of daylighting on building final electric energy requirements
Highlights ► 216 integrated daylighting and energy simulation typologies were performed. ► A function approximation using multivariate linear regression was proposed. ► A function approximation using artificial neural networks (ANNs) was proposed. ► The regression models were compared using the coefficient of determination. ► The ANNs performed better besides presents low mean square error.
Abstract Daylighting has a recognized potential for electric energy savings when used as a complement for artificial lighting. Recent studies have indicated that artificial neural networks (ANNs) present high potential for daylighting analysis in buildings. In this study, a statistical performance method based on ANNs is evaluated. The ANN use is assessed in comparison with the multivariate linear regression (MLR) technique in order to represent the potential for energy savings through the daylighting use in office buildings. The data set was based on computer simulation results obtained with 216 models and used to develop a MLR equation. The simulations were performed with Daysim (daylighting) coupled to EnergyPlus (energy). In parallel, an approximation model using ANN was applied. The method adopted for the training and validation of the ANN was a cross-validation procedure, recommended for limited data sets. The results demonstrate that the ANN have shown a better performance in representing this problem, since it presented a lower coefficient of determination and an acceptable mean square error. In addition, the ANNs presented learning skills which allow them to perform the generalization process in a better way. After being validated, the ANN is expected to provide output results for input data which have not yet been submitted to it, providing reasonable estimates of daylighting impact on electric energy need for new environments.
Using artificial neural networks to predict the impact of daylighting on building final electric energy requirements
Highlights ► 216 integrated daylighting and energy simulation typologies were performed. ► A function approximation using multivariate linear regression was proposed. ► A function approximation using artificial neural networks (ANNs) was proposed. ► The regression models were compared using the coefficient of determination. ► The ANNs performed better besides presents low mean square error.
Abstract Daylighting has a recognized potential for electric energy savings when used as a complement for artificial lighting. Recent studies have indicated that artificial neural networks (ANNs) present high potential for daylighting analysis in buildings. In this study, a statistical performance method based on ANNs is evaluated. The ANN use is assessed in comparison with the multivariate linear regression (MLR) technique in order to represent the potential for energy savings through the daylighting use in office buildings. The data set was based on computer simulation results obtained with 216 models and used to develop a MLR equation. The simulations were performed with Daysim (daylighting) coupled to EnergyPlus (energy). In parallel, an approximation model using ANN was applied. The method adopted for the training and validation of the ANN was a cross-validation procedure, recommended for limited data sets. The results demonstrate that the ANN have shown a better performance in representing this problem, since it presented a lower coefficient of determination and an acceptable mean square error. In addition, the ANNs presented learning skills which allow them to perform the generalization process in a better way. After being validated, the ANN is expected to provide output results for input data which have not yet been submitted to it, providing reasonable estimates of daylighting impact on electric energy need for new environments.
Using artificial neural networks to predict the impact of daylighting on building final electric energy requirements
Fonseca, Raphaela Walger da (author) / Didoné, Evelise Leite (author) / Pereira, Fernando Oscar Ruttkay (author)
Energy and Buildings ; 61 ; 31-38
2013-02-01
8 pages
Article (Journal)
Electronic Resource
English