Decomposition-Based Multistep Sea Wind Speed Forecasting Using Stacked Gated Recurrent Unit Improved by Residual Connections

Table 5

Parameters of the proposed model.

Stacked GRU predictor	Parameters
GRU layer 1	Number of units: 500
	Input shape: (24, 1)
GRU layer 2	Number of units: 50
GRU layer 3 ()	Number of units: 50
GRU layer 4 ()	Number of units: 50
Flatten layer	None
Dense layer	Number of units: 3
	Output shape: (3)
General setting	Batch size: 24
	Learning rate: 0.001