Table 7. Biot’s equations—inverse modeling: PINN performance as function of hyperparameters.

Relative ${\mathcal{L}}^2$ error of p, u, and v and percentage error of θ₁, θ₂, θ₃, θ₄, and θ₅ for different number of hidden layers, N_hl, and different number of neurons per layer, N_n. The N_tr is fixed at 250. Note that we pick the optimal hyperparameters, i.e., N_hl = 6 and N_n = 20 from the sensitivity analysis of the forward model. Results shown in this table are an average over 27 realizations.

	Nn	2	5	10
	Nhl
p	4	0.00006	0.00004	0.00008
	6	0.00009	0.00004	0.00006
	8	0.00008	0.00006	0.00005
u	4	0.00019	0.00016	0.00047
	6	0.00051	0.00013	0.00036
	8	0.00027	0.00036	0.00034
v	4	0.00042	0.00045	0.00098
	6	0.00114	0.00038	0.00074
	8	0.00060	0.00079	0.00071
θ1	4	0.20180	0.28665	1.06223
	6	0.72512	0.16844	0.97916
	8	0.30534	0.70613	0.99229
θ2	4	0.65507	0.66560	3.56789
	6	2.23747	0.23053	2.80360
	8	1.21152	1.47513	2.45137
θ3	4	0.02968	0.05016	0.23851
	6	0.33292	0.02935	0.10100
	8	0.22593	0.11796	0.07371
θ4	4	0.04834	0.06162	0.19135
	6	0.18104	0.04666	0.13785
	8	0.10711	0.16124	0.12127
θ5	4	0.19710	0.21622	1.10700
	6	1.24043	0.17633	0.60595
	8	0.81199	0.50331	0.39558