Forecasting Green Energy Production in Latin American Countries and Canada via Temporal Fusion Transformer

Abstract

Forecasting green energy is crucial in diminishing dependence on fossil fuels and fostering sustainable development. However, it encounters notable challenges, such as variable demand, restricted data availability, the integration of various datasets, and the necessity for precise long-term projections. This study thoughtfully examines these issues using the temporal fusion transformer (TFT) model to project green energy production across five Latin American nations (Argentina, Brazil, Chile, Colombia, and Mexico) and Canada, drawing on data from 1965 to 2023. The performance of the proposed TFT is more authentic as compared with the gated recurrent unit (GRU), the long short-term memory (LSTM), deep autoregression (DeepAR), and the meta graph-based convolutional recurrent network (MegaCRN). The TFT has a mean square error (MSE) of 0.0003, root mean square error (RMSE) of 0.0173, mean absolute error (MAE) of 0.0112 and mean absolute percentage error (MAPE) of 1.76%. From the preceding results, it is clear that the proposed TFT model can identify dynamic energy patterns that will contribute towards achieving sustainable development goals by the end of 2040.