Feature selection helps improve the accuracy and computational time of solar forecasting. However, FS is often passed by or conducted with methods that do not suit the solar forecasting issue, such as filter or linear methods. In this study, we propose a wrapper method termed Sequential Forward Selection (SFS), with a Kernel Conditional Density Estimator (KCDE) named SFS-KCDE, as FS to forecast day-ahead regional PV power production in French Guiana. This method was compared to three other FS methods used in earlier studies: the Pearson correlation method, the RReliefF (RRF) method, and SFS using a linear regression. It has been shown that SFS-KCDE outperforms other FS methods, particularly for overcast sky conditions. Moreover, Wrapper methods show better forecasting performance than filter methods and should be used.

Increasing the utilization of renewable energy is at the center of most sustainability policies. Solar energy is the most abundant resource of this type on Earth, and optimizing its use requires the optimal estimation of surface solar irradiation. Heliosat-2 is one of the most popular methods of global horizontal irradiation (GHI) estimation. Originally developed for the Meteosat satellite, Heliosat-2 has been modified in previous work to deal with GOES-13 data and named here GOES_H2. This model has been validated through the computation of indicators and irradiation maps for the Guiana Shield. This article proposes an improved version of GOES_H2, which has been combined with a radiative transfer parameterization (RTP) and the McClear clear-sky model (MC). This new version, hereafter designated RTP_MC_GOES_H2, was tested on eight stations from the Baseline Surface Radiation Network, located in North and South America, and covered by GOES-13. RTP_MC_GOES_H2 improves the hourly GHI estimates independently of the type of sky. This improvement is independent of the climate, no matter the station, the RTP_MC_GOES_H2 gives better results of MBE and RMSE than the original GOES_H2 method. Indeed, the MBE and RMSE values, respectively, change from $-11.93$% to $-2.42$% and $23.24$% to $18.24$% for North America and from $-4.35$% to $1.79$% and $19.97$% to $17.37$ for South America. Moreover, the flexibility of the method may allow to improve results in the presence of snow cover and rainy/variable weather. Furthermore, RTP_MC_GOES_H2 results outperform or equalize those of other operational models.