Comparison between GRU and BP Neural Networks for Short-Term Prediction of Solar Irradiance

In this study, we present a data-driven approach for predicting solar irradiance by applying dimensionality reduction techniques to a dataset comprising of ground meteorological station data and FY-4A remote sensing data collected from January 1st to December 31st, 2018. Specifically, we use Principal Component Analysis (PCA) to reduce the dimensionality of the dataset. Next, we employ a Gated Recurrent Unit (GRU) neural network-based short-term irradiance prediction model to realize the short-term prediction of solar irradiance. We then evaluate the performance of the GRU model by comparing its prediction results with those obtained using a traditional Backpropagation (BP) neural network model with measured solar irradiance. The results indicate that the root mean square error (RMSE) of the GRU neural network model is 41% lower than that of the BP neural network model, indicating the improved performance of the proposed model.