Forecasting of three components of solar irradiation for building applications

Abstract

Solar energy and the concept of passive architecture and Net Zero Energy buildings are being increased. For an optimal management of the building energy, a Model Predictive Control is generally used but requires an accurate building model and weather forecast. For a more reliable modelling, the knowledge of the global solar irradiation is not sufficient; three methods, smart persistence, artificial neural network and random forest, are compared to forecast the three components of solar irradiation measured on the site with a high meteorological variability. Hourly solar irradiations are forecasted for time horizons from h+1 to h+6. The random forest method (RF) is the most efficient and the accuracy of forecasts are in term of nRMSE, from 19.65% for h+1 to 27.78% for h+6 for global horizontal irradiation, from 34.11% for h+1 to 49.08% for h+6 for beam normal irradiation, from 35.08% for h+1 to 49.14% for h+6 for diffuse horizontal irradiation. The improvement brought by the use of RF compared to the two other methods increases with the forecasting horizon. A seasonal study is realized and shows that the forecasting during spring and autumn is less reliable than during winter and summer due to a higher meteorological variability.