ANN model for predicting the direct normal irradiance and the global radiation for a solar application to a residential building

An accurate solar potential estimation of a specific location is basic for the solar systems evaluation. Generally, the global solar radiation is determined without considering its different contributes, but systems as those concentrating solar require an accurate direct normal irradiance (DNI) evaluation. Solar radiation variability and measurement stations non-availability for each location require accurate prediction models. In this paper two Artificial Neural Network (ANN) models are developed to predict daily global radiation (GR) and hourly direct normal irradiance (DNI). Two heterogeneous set of parameters as climatic, astronomic and radiometric variables are introduced and the data are obtained by databases and experimental measurements. For each ANN model a multi layer perceptron (MLP) is trained and validated investigating nine topological network configurations. The best ANN configurations for predicting GR and DNI are tested on different new dataset. MAPE, RMSE and R2 for the GR model are respectively equal to 4.57%, 160.3 Wh/m2 and 0.9918, while for the DNI they are equal to 5.57%, 17.7 W/m2 and 0.994. Hence, the proposed models show a good correlation both between measured and predicted data and with the literature. The main results obtained are the DNI and the GR models predicting which have allowed the evaluation of the electric energy production by means of two different photovoltaic systems used for a residential building. Hence, the developed ANN models represent a good tool to support the assessment of the green energy production evaluation.