A Graph Neural Network Based Deep Learning Predictor for Spatio-Temporal Group Solar Irradiance Forecasting

The fast growth of photovoltaic (PV) power generation raises the concern of grid instability due to its intermittent nature. Solar irradiance forecasting is becoming an effective way to support the power ramp rate control, mitigate power intermittency, and improve grid resilience. However, most previous research focuses on the predictor applied to the central PV farm but fails to fit the distributed PV system well. This article proposes a novel deep learning architecture, namely, group solar irradiance neural network, used for solar irradiance forecasting. The strategy adopts a convolutional graph neural network to mine distributed PVs' graphs and identify the critical features. Meanwhile, the scheme includes a long-short-term memory recurrent neural network to capture the temporal correlations. The proposed solution shows an improved universality for model training and results in accurate and reliable predictions compared to other solutions. Furthermore, the learning architecture can be applied to the master-slave deployment scheme to reduce overall system cost. The proposed algorithm is tested with the data sourced from the U.S. National Renewable Energy Laboratory. The evaluation demonstrates the highest accuracy and best fitting for solar irradiance forecasting in comparison with other prediction methods.

Automated summary

This article proposes a novel deep learning architecture for solar irradiance forecasting. It improves the accuracy and reliability of predictions and is applicable to distributed PV systems.