Solar radiation forecast using neural networks for the prediction of grid connected PV plants energy production (DSP project)

The work presented in this paper is part of a project aimed to develop a prototype device (DSP) able to forecast with a day in advance the energy produced by PV plants. The energy forecast is required by the National Authority for the electricity in order to control the high instabilities of the electric grid induced by unpredictable energy sources such as photovoltaic. In the paper several models to forecast the hourly solar irradiance with a day in advance using Artificial Neural Network (ANN) techniques are described. Statistical (ST) models that use only local measured data and Hybrid model (HY) that also use Numerical Weather Prediction (NWP) data are tested for the University of Rome “Tor Vergata” site. The performance of ST, NWP and HY models, together with the Persistence model (PM), are compared. The ST models and the NWP model exhibit similar results improving the performance of the PM of around 20%. Nevertheless different sources of forecast errors between ST and NWP models are identified. The Hybrid models give the better performance, improving the forecast of approximately 39% with respect to the Persistence model.