Solar Photovoltaic Power Forecasting

Table 5

The references of the statistical methods used in the forecasting approach.

References	Approaches
[14]	Grid-tie PV power-forecasting model for 0–6 h ahead, also called the 2D-interval forecasts based on SVR-2D, that computes directly the 2D-interval forecasts from the previous historical solar power and meteorological data by using the SVR method. The parameters of the forecasting model were the solar and the weather data that included the solar irradiance, temperature, humidity, and wind speed provided from the “Australian photovoltaic data” for two years sampled for every 1, 5, and 30 min along with the past data of PV power. At the same time, the mean absolute interval deviation (MAID), MRE, and interval coverage probability (ICP) were used to perform the forecasting model accuracy.
[77]	AR model that had comparable performances with the ARMA model to produce the short-term PV power forecasting, and the forecasting parameters include the climate state of previous time samples. Therefore, the forecasting model used for false data injection attacks (FDIAs) detection showed performance results in the security and the control of power grid. To that end, the phase-phase correlation (PPC) was used for evaluating the accuracy of forecasts.
[78]	Cloud and irradiance forecasting of 15 min to 5 hours ahead based on the satellite images and SVM. The 4 years of historical satellite images, meanwhile, were used to learn the model. Consequently, this application showed an improvement for the EMS in terms of RMSE, MRE, and the coefficient of determination R^2.
[79]	The parametric approach that relied on the mathematical models with several parameters that describe the PV system, whereas the nonparametric approach was based on quantile regression forests with training and forecast stages. In the meantime, the forecasting parameters are the meteorological variables from the NWP models. In this case, this forecasting model showed better results in terms of mean-based error (MBE), RMSE, MAE, and skill scores (SS). Therefore, the forecasting engine has been used for calculating the hourly power delivered to the grid.
[80]	Multilinear adaptive regression splines and persistence method used for the short-term PV power forecasting model. The forecasting parameters, meanwhile, include the weather forecasts from the “US Global Forecasting Service (GFS)” and PV power output data (estimated to 1.3 MW) of a PV power plant located in the Borkum city of Germany country. Therefore, the application of this forecasting model showed better results in terms of R^2, RMSE, MAE, and MBE, and in this situation, the forecasting process was advantageous for calculating the day-ahead production from a PV power plant.
[81]	A classical statistical method based on neural network modelling. The forecasting model parameters, meanwhile, are the number of sunny hours, length of the day, air pressure, maximum temperature, insolation of the day, and cloudiness. The forecasting model showed better results in terms of Pearson’s linear correlation coefficients, kurtosis, skewness, and RMS, and it was developed to perform the short-term PV power forecasting model.
[82]	A multistep method used for forecasting the PV power in different ranges of time, respectively, 10 s, 1 min, 5 min, 30 min, and 2 hours. The forecasting model, meanwhile, was based on the persistence method and the auto regressive exogenous (ARX) model, which presented better results in terms of RMSE and MAE once trained by the forecasting parameters, which included the data from the NREL radiometer grid, Hawaii (USA), and the Microgen database, East Midlands (UK).