Statistical Analysis of Novel Ensemble Recursive Radial Basis Function Neural Network Performance on Global Solar Irradiance Forecasting

Abstract

Reliable operation of energy management systems, grid stability, and managing energy demand responses are becoming challenging because of the flickering nature of solar irradiance. Accurate forecasting of global solar irradiance, i.e., global horizontal irradiance (GHI), plays a significant role in energy policy-making and the energy market. This paper proposes a novel global solar irradiance forecasting model based on the ensemble recursive radial basis function neural networks (ERRBFNNs). The various atmospheric inputs based on the built ensemble recursive radial basis function neural networks make the network more stable and robust to climatic uncertainty. This paper statistically investigates the performance of novel feed-forward neural networks based on forecasting models with various hidden nodes for global solar irradiance forecasting applications. We validated the proposed ERRBFNN global solar irradiance forecasting model using real-time data sets. The simulation results confirm that the proposed ensemble recursive radial basis function neural network based on global solar irradiance forecasting improves the accuracy, generalization, and network stability. Furthermore, the proposed ERRBFNN lowers the forecasting error to the least compared to other state-of-the-art forecasting models.