Different Optimizers-Based Gated Recurrent Unit Network to Forecast One Step Ahead Solar Irradiance

Abstract

AbstractWith the variable and stochastic behavior of climate and weather, the solar interconnected grids are prone to failure due to imbalanced operation, planning, and management. Therefore, precise and accurate solar irradiance forecasting becomes one of the important tasks in grid interconnected solar systems. This study is dedicated to the development of a solar forecasting model using the gated recurrent unit (GRU) deep learning network based on different optimizers. Three optimizers: adaptive moment estimation with momentum (ADAM), stochastic gradient descent with momentum (SGDM), and root mean square propagation (RMSprop) are used in the study with the GRU forecaster to forecast the one step ahead solar global horizontal irradiance (GHI). The developed models are trained with a one-year dataset of Indian location: Ahmadabad, Gujarat. Whereas, the monthly forecasting is performed using the trained model for discussed optimizers. To observe the performance of all models, root mean square error (RMSE) and mean absolute percentage error (MAPE) is used. In addition to this, the developed models are also compared with the naïve model considered as a baseline model. The results of the study show that the ADAM-GRU model outperforms SGDM-GRU, RMSprop-GRU as well baseline model.KeywordsGated recurrent unit (GRU)Solar forecastingAdaptive moment estimation with momentumStochastic gradient descent with momentumAnd root mean square propagationClear sky index (CSI)