Componentnet: Processing U- and V-components for spatio-Temporal wind speed forecasting

The increasing presence of intermittent renewables in modern power systems motivates the development methods for forecasting. More accurate forecasts may implicate less operational costs systems. In this context, paper proposes a family architectures based on fully convolutional neural networks wind speed prediction, ComPonentNet (CPNet) family. CPNet produces multi-site spatio-temporal forecasting phenomena which be decomposed into multiple components (e.g., wind, u- and v-wind). includes three - core CPNet, fully-fused bottom-fused CPNet. Each architecture processes phenomenon different manner separate branches operations, same branch, or mixing joint branches. This investigates performance each speed. Moreover, framework is compared against U-Net architecture. results indicate that proposed promising, splitting processing beneficial to forecasting, with outperform U-Net.