Generalized global solar radiation forecasting model via cyber-secure deep federated learning.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Climate Modeling

Background:

• Rising global solar energy adoption necessitates precise solar irradiance prediction.
• Data scarcity and privacy concerns hinder traditional solar forecasting methods.
• Distributed data collection for centralized analysis presents significant challenges.

Purpose of the Study:

• To develop a privacy-preserving global solar radiation forecasting model.
• To leverage federated learning (FL) and convolutional neural networks (CNN) for enhanced prediction accuracy.
• To establish a global supermodel capable of forecasting in data-scarce regions.

Main Methods:

• Utilized federated learning (FL) for decentralized model training across multiple clients.
• Employed convolutional neural networks (CNN) for analyzing diverse climatic data from eight Iranian regions.
• Tested the global supermodel's efficacy on three new Iranian regions (Abadeh, Jarqavieh, Arak) with no prior training data.

Main Results:

• The FL-CNN global supermodel achieved high accuracy in new regions: 95% (Abadeh), 92% (Jarqavieh), and 90% (Arak).
• Conventional machine learning and deep learning models failed to forecast solar radiation in these new regions due to lack of training data.
• The proposed FL approach demonstrated superior performance compared to conventional methods, especially in data-scarce environments.

Conclusions:

• Federated learning combined with CNNs offers a robust solution for global solar irradiance forecasting.
• The developed model effectively addresses data privacy concerns and overcomes limitations of data availability.
• This privacy-preserving, data-efficient approach significantly advances the field of renewable energy prediction.