Integrator modeling using Matlab and system identification toolbox

Hi!

I'm modeling a tank and its dynamics by black box experiments. I do know that the tank level is linear, thus it should be no problem to identify a model. A PI controller is used for level control. The controller is mounted after the tank, hence the integrator gain is negative. I change the manual output of the level controller and get a response. See attached pdf of identification and validation data set. Up to this point everything has worked. I have a somewhat rich identification and validation set.

Here's the problem. When using matlab for identification, nothing makes sense. Matlab develops models that are way of. I've tried with different routines(procest, n4sid, tfest, etc.) and nothing returns a decent model that fits my validation set. By looking at the data set provided, the model is definitively an integrator with negative gain as assumed. This should not be difficult, but somehow I don't manage to see what I do wrong when using Matlab for identification.

Anybody who could help here?

Best!