Photovoltaic power forecasting through temperature and solar radiation estimation

Abstract

The Utilization of the photovoltaic power as a source of electricity has been strongly growing. The unpredictability of the PV power energy induces frequency fluctuations and power system instabilities. Thus, short term PV power prediction, from one hour to several hours, becomes very important to ensure grid stability. The photovoltaic power depends on different weather conditions mostly temperature and solar radiation. Therefore, weather data forecasting becomes highly recommended. This paper presents a comparison study between the adaptive neuro-fuzzy inference system and the feed forward neural network for one hour ahead temperature and solar radiation estimation using different input data. Two and four hours ahead forecasting of the metrological data are done using the feed forward neural network model. Using the forecasted weather data, the photovoltaic power is deduced. The accuracy of the topologies is based on the normalized root mean square error (NRMSE), and the mean absolute percentage error (MAPE) The simulation results show that the FFNN outperforms the ANFIS model.