Modeling of Brazilian Wind Power Generation Capacity: A Multivariate Analysis with Neural Networks

Over the last twenty years, Brazil has experienced significant changes in its electric energy matrix, notably with the expansion of wind and solar energy sources. In the early 2000s, hydroelectric plants with a capacity of 70 gigawatts (GW) were responsible for generating 90% of the country's electricity, while solar and wind sources contributed merely 1%. Today, the scenario has evolved, with wind and solar energy comprising approximately 25% of the national generation, indicating a departure from the previous hydro-dominated matrix. This research aims to apply specific types of neural networks to multivariate data to forecast Brazil's wind power generation capacity, providing insight into the potential for future energy strategies. This research contributes as a as a decision support tool contemplating relevant information for analysts, researchers, and participants in the renewable energy market. The monetary volume destined to import key components of the wind, considering specific time lags, was utilized to forecast the Installed capacity of the wind farms in Brazil. The Toda-Yamamoto Causality Tests revealed unidirectional Granger causality in the sense of the Import variable for the Brazilian Installed Capacity variable for wind energy production. Another result obtained was the finding that the variable that matters most in the multivariate configuration of the Neural Network is the variation in Installed Capacity in megawatts, followed by the import volume of wind energy components, which exhibited critical lags at 3 and 6 months. Keywords— Renewable Energy; Wind-Solar Growth; Neural Network Forecasting.