An AutoML framework for transfer learning applied to solar radiation prediction: A case study em Juiz de Fora, Minas Gerais, Brazil.

Resumo

Accurate solar radiation prediction is fundamental for efficiently planning and operating solar energy systems, assessing regional climate, and modeling crop growth. This study proposes a framework based on Automated Machine Learning (AutoML) combined with transfer learning techniques to forecast solar radiation. The objective is to streamline the selection and tuning of predictive models, minimizing the time and technical expertise needed for deployment. The framework enables knowledge transfer from regions with abundant data to those lacking measurement histories by leveraging historical data on solar radiation, meteorological conditions, and geographic features. This approach makes it possible to generate reliable forecasts even in cities without extensive historical records. A case study was conducted in Juiz de Fora, Minas Gerais, Brazil. The results show that the use of transfer learning alongside AutoML offers a promising strategy to allow cities that do not have a significant history of measurements to have good predictions.