Application of data science in the prediction of solar energy for the Amazon basin: a study case

Abstract

The need for renewable energy sources has challenged most countries to comply with environmental protection actions and to handle climate change. Solar energy figures as a natural option, despite its intermittence. Brazil has a green energy matrix with significant expansion of solar form in recent years. To preserve the Amazon basin, the use of solar energy can help communities and cities improve their living standards without new hydroelectric units or even to burn biomass, avoiding harsh environmental consequences. The novelty of this work is using data science with machine-learning tools to predict the solar incidence (W.h/m²) in four cities in Amazonas state (north-west Brazil), using data from NASA satellites within the period of 2013–22. Decision-tree-based models and vector autoregressive (time-series) models were used with three time aggregations: day, week and month. The predictor model can aid in the economic assessment of solar energy in the Amazon basin and the use of satellite data was encouraged by the lack of data from ground stations. The mean absolute error was selected as the output indicator, with the lowest values obtained close to 0.20, from the adaptive boosting and light gradient boosting algorithms, in the same order of magnitude of similar references.