Aggregated independent forecasters of half-hourly global horizontal irradiance

In this study, single and aggregated forecasters of half-hourly global horizontal irradiance are assessed. The models are the standard persistent model and four newly proposed static, dynamic, moving average, and amplified persistent models. These sub-forecasters are aggregated using equal, annual optimal, and monthly optimal weights. A particle swarm optimizer was used to find those weights. Measured data, obtained from two desert sites for the years 2015-2018, was used for fitting and training the different models, while the data of the year 2019 was used to test their prediction capabilities. For the single forecasters, the dynamic model is the most accurate, followed by the static and average models. When the aggregated model of annual optimal weights was tested, the three contributing forecasters were the dynamic, average, and amplified models. The dynamic forecaster held the largest weight due to its prediction superiority during overcast and partially cloudy days. When monthly optimal weights were used, all forecasters contributed, and the dynamic model held the largest weight during winter but not in the summer when the clear sky condition is dominant. The aggregated model was the most precise, with relative mean square errors lower than 15.0% and coefficients of determination higher than 98.8%. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

In this study, different models for forecasting half-hourly global horizontal irradiance were evaluated, with the dynamic model proving to be the most accurate for single forecasting. When aggregate forecasting with annual optimal weights, the dynamic, average, and amplified models contributed the most, with the dynamic model holding the largest weight due to its superior prediction during overcast and partially cloudy days. The aggregated model showed the highest precision with relative mean square errors below 15.0% and coefficients of determination above 98.8%.