Wind Speed Prediction with Convolutional Networks

The growth of renewable energy sources is becoming increasingly challenging for the operation of the electrical system. This is because they depend on intermittent resources. For example, wind energy accounts for 11.8\% of the energy supply in Brazil, with wind turbines concentrated in the north and northeast. This paper proposes a methodology based on fully convolutional neural networks for simultaneous prediction of 36 wind speed series, evaluated using a multi-step auto-regressive prediction of 28 steps (two weeks). The results were close to state-of-the-art, showing a smaller error in half of the comparisons from the test set; but with orders of magnitude lower computational cost.