There's more to this than modelling restraints. In fact, since you can see what restraints are currently "active" and "inactive", there are no nonlinearities in the current state of the system.
Like I suggested earlier, a transient may have upset your system resulting in one of your shoes (a +Y support) turning into a line stop.
You must somehow model the spring load (as in springing the system) so that your "operating" state matches the current state of your system.
Here's another guess; probably only one of those shoes is carrying all that spring load. I suggest you model a lateral imposed deflection at what appears to be the one, new, "lateral support" that carries the remaining spring load. If you can replicate the current position of all the nodes, then your model is representing the system.
(I referenced a line stop in describing the "new" boundary condition. But that transient load may have also developed a lateral load component that may also be carried by your new support.)