干旱区生态环境知识资源中心

Illuminance measurements do not make up a part of routine measurements in meteorological stations in Brazil, therefore, they are very rare. This information is important for evaluating the potential contribution of natural illumination in commercial buildings, which would significantly reduce the consumption of electric energy that is used for artificial illumination and refrigeration systems.

To face this lack of information, different models known as luminous efficacy, were created that made possible the estimation of illuminance in regions where there only exists information on solar irradiation. In general, they are statistical models that empirically correlate the relationship between illuminance and solar irradiation with other meteorological variables and / or sky conditions.

In this work, an estimation of hourly luminous efficacy was made by means of the MLP (Multilayer perceptron) artificial neural networks (ANN). The hourly global luminous efficacy was estimated by considering a group of physical variables from the same locality that were collected in a simultaneous way. The data input of the ANN was the following: dew temperature, precipitable water, sky brilliance index, clearness index of Perez and clearness index.

The results were compared with the statistical models of Perez et al., (1990) and Alados (1996), adjusted with local coefficients. The artificial neural network model shows a statistical performance that is similar to these models: for the city of Recife the RMSE was 5.8% and for the city of Pesqueira 3.6%.