model is not able to minimize loss for PointSetBC

[ryabs07]

Hi all,
I am dealing with a problem where an Initial Condition cannot be defined by a function of spatial co-ordinates x. But I have values of initial conditions at discretized points as an array. I used PointSetBC in order to solve the initial condition from the given array. However, after a few hundred iterations of optimization, the model is not able to minimize the loss for initial condition further more. A snippet of the code implementation is below:

x = np.linspace(x_start, x_end, total_points)
X, T_initial = np.meshgrid(x,0)  #creating a mesh of discretized points and initial time 0
IC_input = np.hstack((X.flatten()[:,None], T_initial.flatten()[:,None]))

#Define the PDEs
def pde(a, b):
    #a is a 2D input array where a[:,0:1] represents spatial domain in x direction and a[:,1:2] represents the temporal domain
    #a is the input of the neural net
    #b is a the output array where y[:,0:1] represents water depth , y[:,1:2] represents velocity, y[:,2:3] represents bed elevation (topography)
    #b is the output of the neural net
    x, t = a[:, 0:1], a[:, 1:2]
    h, u, z = b[:, 0:1], b[:, 1:2], b[:, 2:3]
    ...defining rest of the equations...


ic_h = dde.icbc.PointSetBC(IC_input, h_initial, component=0) #initial water depth as an array
ic_u = dde.icbc.PointSetBC(IC_input, u_initial, component=1) #initial water velocity as an array
ic_z = dde.icbc.PointSetBC(IC_input, z_initial, component=2) #initial topography as an array

data = dde.data.TimePDE(geomtime, pde, [ic_h, ic_u, ic_z, bc_l_h, bc_r_h, bc_l_u], num_domain=20000, num_boundary=1000, anchors = input)

I have tried multiple network width and depth, weights, activation functions. But nothing seems to work. Please let me know if there is any solution to this or if my implementation is not correct.
Thank you in advance :)