Short-term Forecasting of the Wind Power Generation of Brazilian Power Stations Using an LSTM Model
Resumo
Context: Most of the power generated in Brazil is from hydropower stations. However, the use of other sources is quickly increasing, mainly concerning wind power. In addition, approaches to forecasting wind power generation are important for avoiding problems in the power supply. Problem: Although models for forecasting wind power generation are available in the literature, the use of real-world data from Brazilian power stations is not common. This practical situation is important due to its particularities, such as the availability of complementary data, missing data, and the window of prediction. Solution: Here, we investigate and develop a Long Short-Term Memory (LSTM) model for forecasting the wind power generation of seven Brazillian power stations. IS Theory: This work proposes an intelligent information system and aims to improve the decision-making capacities of information systems. Method: This research is prescriptive, and its evaluation was carried out through computational experiments and quantitative approaches. Summary of Results: The results obtained by the proposed model are compatible with those from other approaches in the literature. Also, we observed that (i) a larger number of input data (from more previous days) and (ii) using an exogenous variable increased the quality of the models generated. Contributions and Impact in the IS area: The proposal generates solutions for improving data-driven decision-making and this kind of approach has become a challenge for researchers in information systems according to the Grand Research Challenges in Information Systems in Brazil 2016-2026.
