Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Nowcasting and Short Term GHI Forecasting Using GOES-16 Shortwave Radiance Data: A Machine Learning Case Study of Petrolina-PE, Brazil
Submission Author:
Paulo Alexandre Costa Rocha , CE , Brazil
Co-Authors:
Paulo Alexandre Costa Rocha, Hugo Pedro, Carlos Coimbra
Presenter: Paulo Alexandre Costa Rocha
doi://10.26678/ABCM.COBEM2021.COB2021-0247
Abstract
In this work it was used the visible and near-IR channels of the GOES-16 satellite to model (nowcasting) and forecast GHI intra-hourly. The signals were used as predictors, both in their raw and normalized forms. Five different machine learning models were tested, namely Artificial Neural Networks, k-Nearest Neighbors, LASSO, Support Vector Machines and XGBoost. Their performance was compared by the RMSE and Forecast Skill (FS) relative to the persistence model. The first four methods presented similar behavior, with the RMSE values ranging from 180.42 to 202.05 W.m-2. The XGBoost outperformed all the other methods, obtaining RMSE values between 144.37 and 160.34 W.m-2. The FS improved over the time horizon for all methods, being negative for nowcasting and 15-min forecasting, with the exception of XGBoost, which performed positively for all forecasting time horizons, reaching 40.78 % for 60-min horizons.
Keywords
renewable energy, solar irradiance forecasting, full disk GOES-16 satellite, machine learning, caret R package
