Physics-Informed Neural Network water surface predictability for 1D steady-state open channel cases with different flow types and complex bed profile shapes

Advanced Modeling and Simulation in Engineering Sciences

Table 2 Inverse PINN comparison between calibrated roughness values from GLUE experiments and the ones obtained from different AFs and different flow values

Flow (m³ s⁻¹)	n GLUE [19]	Activation function	n	Hidden layers	Neurons per hidden layer	Loss function
0.035	0.555–0.609	Tanh	0.444	3	60	0.024
		ReLU	0.245	9	60	21.275
		Sin	0.445	5	60	0.020
		Sigmoid	0.459	5	60	0.019
0.443	0.105–0.124	Tanh	0.107	7	60	0.025
		ReLU	0.135	7	60	27.487
		Sin	0.130	5	40	0.067
		Sigmoid	0.148	7	60	0.019
0.878	0.092–0.121	Tanh	0.135	3	60	0.024
		ReLU	0.122	9	60	28.317
		Sin	0.129	5	40	0.044
		Sigmoid	0.127	9	60	0.018