Table 1 Hyperparameter optimization of the baseline and H-LSTM-GLE models.

From: Sequence to sequence architecture based on hybrid LSTM global and local encoders approach for meteorological factors forecasting

Hyperparameter

Value

Grid

Epochs

25

25

Batches per epoch

64

[32, 64, 128]

Loss

RMSE

RMSE

Dropout

0.2

[0.1, 0.2, 0.3]

Parameter optimization

SGD

SGD

Momentum

0.9

0.9

Initial learning rate

0.001

0.001

Learning rate decay

0.5

0.5

Overfitting termination

10

[6, 8, 10, 12]

Input sequence length

96

96

Output sequence length

4

4

Sliding window step size

6

[1, 2, 4, 6, 8, 10, 12]

Hidden states size

128

[16, 32, 64, 128, 256]

Back to article page