Use the ordinary least squares estimator applied to the measurement equation. Otherwise, use the unbiased FIR (UFIR) filter. In the latter case, set the power of the system matrix to zero and obtain an estimate related to the initial point of the averaging horizon.

Thank you 

You can use Linear regression method if the mapping is linear or else use Nonlinear regression method when the mapping is non-linear. Or else just use a time-series approximation on the input and output.

Both the OLS and UFIR estimators are regression estimators. The Kalman filter requires the system state model, whereas the problem is stated to ignore system dynamics.

If you don't want to use system parameters or system dynamics, does it mean that the model of the system is not known ?

yes I want a system for which model is not known, but input and output are measurable

I found 4SID (Subspace state space system identification) methods extremely useful if you also planned to identify the state-space model of your system. You can search n4sid and ssest commands in MATLAB.

Thank you all for your advices. I found adaptive observers are suitable to my problem. I request if somebody can give suggestions on adaptive observers using neural networks.