Table 2 Scope and results of CNN-BiLSTM-attention hyperparameter search based on Bayesian algorithm.

From: Photovoltaic power interval prediction with conditional error dependency using Bayesian optimized deep learning

Optimized component

Hyperparameters

Range

Result

CNN

n_channels 1

[8,512]

150

n_channels 2

[8,512]

382

n_channels 3

[8,512]

210

pool_size1

[1,5]

1

pool_size1

[1,5]

2

pool_size1

[1,5]

1

BiLSTM

hidden_size1

[16,512]

226

hidden_size1

[16,512]

278

hidden_size1

[16,512]

272

dropout_rate1

[0.1,0.5]

0.2

dropout_rate1

[0.1,0.5]

0.2

dropout_rate1

[0.1,0.5]

0.3

Learning rate

[1e−5, 0.5]

2e−3

Back to article page