A Practical Method to Hourly Forecast the Solar Irradiance

Based on the fact that the weather bureau currently does not provide solar irradiance forecast data, which is a key parameter to predict energy where photovoltaic (PV) is generated, a practical and indirect solar irradiance prediction model that is based on the RBF neural network and input by several of hourly sequences is proposed. In this paper, the model is built by using UV index sequence, theoretical extraterrestrial irradiance sequence, sequence of air temperature, sequence of weather types and historical solar irradiance sequence. The network is trained by GP-RBF algorithm to forecast the solar irradiance in a period of time to the future by dividing conditions into four types. The experiment results show that, when compared with the other forecasts and the real curves, the new model, which is based on UV index and theoretical extraterrestrial irradiance hourly sequences, is practical and highly accurate.