I'm not sure from your question if you mean lipid membranes or synthetic filtration membranes. In either case, NpT tends to be representative of a lot of systems, including the biological cell environment. You could then apply the barostat then either independently in either direction or omit some directions if you are dealing with a solid periodic membrane across your simulation box. In the latter case, I would be careful with NpT also because the pressure calculation used in the barostat is meant for homogeneous systems, which is actually not the case if you have a membrane. You could then for example not apply NpT, but instead, use some form of piston (and thermostat) to control the external pressure on your fluid. Your system is then in NVT, but note that although your simulation box might have a constant volume, this doesn't mean that your fluid strictly satisfies the NVT ensemble.

In general, it is also not uncommon to equilibrate in an NpT ensemble and switch to NVE once you know the average system size (specifically when dealing with an equilibrium system). An argument why some people do this is is that you then ensure that the system naturally evolves, without risking that a thermostat or barostat influences the motion of your atoms in a potentially unphysical way. This is mostly not so relevant when you are interested in time-average properties but may affect for example time-correlation functions.

Normally, the solvated systems should be simulated in NPT ensemble. However, for membrane, normally the pressure factor is not very important. If you build the system with a normal/correct workflow, it will be stable and changes little within the normal range of pressure fluctuation. In this case, NVT and NPT would produce very similar results. For test, I recommend you to use NVT, make sure everything is correct, and then proceed to NPT.

Equilibrate with NPT - then move to on NVT.