Day-Ahead Hourly Solar Irradiance Forecasting Based on Multi-Attributed Spatio-Temporal Graph Convolutional Network.

Area of Science:

• Renewable Energy Systems
• Atmospheric Science
• Data Science

Background:

• Solar energy commercialization requires accurate solar irradiance forecasting to manage output variability.
• Existing forecasting models often analyze limited meteorological variables and lack comprehensive spatiotemporal context.
• Understanding inter-station weather dynamics is crucial for improving solar power predictions.

Purpose of the Study:

• To propose a novel solar irradiance forecasting model using attributed dynamic networks.
• To analyze temporal changes in atmospheric parameters across multiple stations.
• To evaluate the impact of spatial adjacency, temporal dynamics, and variable diversity on forecasting accuracy.

Main Methods:

• Representing multi-station atmospheric parameters as an attributed dynamic network.
• Extending spatio-temporal graph convolutional network (ST-GCN) models to analyze network temporal changes.
• Comparing the proposed model against existing methods for hourly solar irradiance prediction.

Main Results:

• The proposed model demonstrated superior performance in solar irradiance forecasting.
• Analysis confirmed the synergistic contributions of spatial station adjacency, temporal variable changes, and diverse meteorological data.
• Single-aspect analyses were insufficient to capture the complex interdependencies observed.

Conclusions:

• The novel dynamic network approach significantly enhances solar irradiance forecasting accuracy.
• Integrating diverse meteorological variables and considering spatiotemporal correlations is key for robust solar energy prediction.
• This study highlights the importance of holistic atmospheric context in renewable energy forecasting.