Table 2 The hyper-parameter setting of KE-ANN

From: Towards parameter identification in pipeline hydraulics: integrating data-driven discovery and knowledge embedding

Network section

Hyper-parameters

Range

Time interval

30 s

5 min

Hydraulic feature extraction network

Number of layers

1–10

3

5

Neural units

8–128

[200, 100, 20]

[2000, 500, 10]

Batch size

16–512

256

512

Activation function

[Relu, Tanh, Sigmoid]

Relu

Relu

Dropout

0–0.5

0.1

0.1

Parameter identification network

Number of layers

1–10

2

2

Neural units

8–128

[20,80]

[700, 100]

Batch size

16–512

256

512

Activation function

[Relu, Tanh, Sigmoid]

Relu

Relu

Dropout

0–0.5

0.1

0.1