Daylight harvesting is a well-known strategy to address building energy efficiency. However, few simplified tools can evaluate its dual impact on lighting and air conditioning energy consumption. Artificial neural networks (ANNs) have been used as metamodels to predict energy consumption with high precision, few input parameters and instant response. However, this approach still lacks the potential to estimate consumption when there is daylight harvesting, at the ambient level, where the effect of orientation can be noted. This study investigates this potential, in order to evaluate the applicability of ANNs as a tool to aid the architectonic design. The ANNs were approached as metamodels trained based on EnergyPlus thermo-energetic simulations. The network configuration focused on determining its simplest feasible form. The input parameters adopted as the main variables of the building envelope were as follows: orientation, window-to-wall ratio and visible transmission. The effects of the encoding of orientation as a network input parameter, the number of examples of each variable for network training and changing the parameters used for the training were evaluated. The networks predicted the individualized consumption according to the end use with errors below 5%, indicating their potential to be applied as a simplified tool to support the design process, considering the elementary variables of the building envelope. The discussion of results focused on guidelines and challenges to achieve this purpose when contemplating the broadening of the metamodel scope.
An approach for determining relative input parameter importance and significance in artificial neural networks
