A Federated Learning Approach for Distributed Solar Irradiance Forecasting

Título: A Federated Learning Approach for Distributed Solar Irradiance Forecasting

Autores: Cristiano Gregory Monfrim Camandaroba, Sophia Simião Guimarães Werneck, Julia Zibetti Castilhos de Mellos, Caian Dutra de Jesus, Matheus Malta, Frederico Gadelha Guimarães & Eduardo Pestana de Aguiar

Resumo: This paper focuses on the development of a federated learning framework for predicting solar irradiance in photovoltaic power plants while preserving data privacy. The integration of decentralized machine learning techniques with two neural network architectures, Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP), is discussed and applied across multiple geographically distributed clients. Model evaluation was conducted to assess the predictive performance and robustness of the approaches. A comparative analysis of the effectiveness of LSTM and MLP models is presented, highlighting the superior capability of LSTM in capturing temporal dependencies inherent in solar irradiance data. Finally, the proposed federated learning approach demonstrates enhanced privacy protection and operational scalability for real-world solar energy systems. The reported results show that federated learning combined with LSTM provides a promising, privacy-preserving solution for accurate solar irradiance forecasting in distributed photovoltaic plants.

Palavras-chave: Federated Learning; Solar Irradiance Forecasting; Time Series; Distributed Machine Learning.

Páginas: 8

Código DOI: 10.21528/CBIC2025-1175590

Artigo em PDF: CBIC_2025_paper1175590.pdf

Arquivo BibTeX:
CBIC_2025_1175590.bib