You can try distance restraints. gromacs manual has information on how to implement it.

I would say that if ligand moved than either interaction is not enough to hold the ligand or something wrong with force-field. Also, in a dynamical system, I would expect that ligand will change interactions during the simulations. At the end, you have to look for probability that how many times it remain in different types of interaction.

What do you mean by "receptor-ligand" interaction change ? And why would you want to constrain interactions ? I think that the goal of protein-ligand complex MD simulation is to find out interactions.

I want to constraint the interaction because before the simulation , ligand interact with earlier reported residues. But after the MD, the orientation of the ligand changed and it misses the important interaction which were important for protein catalytic activity.

You can position restrain the particular residues of importance. But I am not sure if the resulting simulation will be physically significant or not. You can read the GROMACS manual or post in GROMACS forum. The forum is very useful for these questions.

okay thanks