Fig. 6: MFWPN network structure.
From: A machine learning model for hub-height short-term wind speed prediction
MFWPN consists of four parts: spatial encoder-decoder, time units, spatial fusion module, and temporal fusion module. The spatial encoder-decoder is used to extract and reconstruct the spatial features of the wind field. The time unit is used to realize the temporal evolution of the wind field. The spatial and temporal fusion modules are used to fuse the spatial and temporal effects of other meteorological factors on wind field evolution. The last module is a loss function designed according to the wind speed characteristics, which helps the network to complete the fitting more efficiently.
