Deep learning approach for one-hour ahead forecasting of solar radiation in different climate regions

Abstract

The accurate prediction of hourly global solar radiation is critical to solar energy conversion systems selecting appropriate provinces, and even future investment policies. With this viewpoint, the aim of this study is to predict one hour ahead the global solar radiation of six provinces that have different climate regions in Turkey. A deep learning technique based on the LSTM neural network is used for this purpose. To see the success of this proposed model, the deep learning technique GRU, a machine-learning approach ANFIS with FCM, and standard statistical models ARMA, ARIMA, and, SARIMA are also applied. Forecasting models developed to estimate the future values of hourly global solar radiation are based on past time series data. The study discusses four different statistical metrics (MAE, RMSE, NSE, and, R) to determine the success of these algorithms. The results indicate that the R, MAE, NSE, and, RMSE values of the two approaches range from 0.9352 to 0.9806, from 29.0737 to 59.4840 Wh/m2, from 0.6445 to 0.9686 and from 57.4492 to 94.3287 Wh/m2, respectively. The results revealed that the methods used in the study performed satisfactorily in terms of hourly solar radiation estimation, but the LSTM method performed better.