Table 3 Training dataset characteristics.

From: Deep recurrent neural networks for water hammer transient prediction and dynamic protection optimization in long distance pipelines

Parameter

Value

Description

Total sequences

18,000

Combined simulation and experimental data

Simulation data

15,847 (88%)

Generated using MOC with validated parameters

Experimental data

2,153 (12%)

Physical testbed measurements

Sensor locations

10

Distributed along 15 km pipeline

Sampling frequency

1000 Hz

Nyquist criterion satisfied

Sequence duration

30 s

Captures complete transient evolution

Pressure range

0.2–2.8 MPa

Covers operational envelope

Initial Flow velocity

0.5–2.5 m/s

Normal operating conditions

Transient types

5 categories

Valve closure, pump trip, emergency stop, flow step, combined events

Training/Validation/Test split

70%/15%/15%

Stratified by transient type