NOWCASTING PREDICTION OF SOLAR IRRADIANCE BASED ON FY-4A AND MACHINE LEARNING
In view of the characteristics of abundant radiation resources but lack of observation data in China,this study proposes a short-term solar irradiance forecasting method based on radiation observation data,remote sensing data,McClear,and random forest algorithm,and focuses on analyzing the impact of remote sensing data on radiation forecasting effectiveness.The results show that adding remote sensing data can optimize the forecasting effectiveness at different time horizons and significantly reduce the probability of large prediction errors with a mean absolute percentage error(MAPE)value exceeding 40%.Additionally,the improvement of the forecasting effectiveness with the addition of remote sensing data increases linearly with the time horizon.The difference range of normalized root mean square error(nRMSE)changes from 2.08%to 13.81%,the difference of normalized mean absolute error(nMAE)changes from 1.64%to 14.52%,the difference of R2 shows the most significant change with the time step,changing from-0.03 to-0.43.However,it is worth noting that adding satellite data will significantly increase the time required for model establishment and hyperparameter optimization.
solar irradianceforecastingmachine learningFY-4Aclear sky model
NETL
NSTL
万方数据
