Hi all

My problem is as follows: I have measurements (obtained from simulations) of a stream of a fluid (air), which is disturbed by an obstacle (a wind turbine). Of course the flow can be described by solving Navier-Stokes equations, and that is what my simulation environment does, with huge computational cost.

My aim is to use system identification methods to predict the velocity at the wind turbines location, and also, at distance $x$ behind the wind turbine (a so-called wake model). I have measurements in $m$ points in fronts of wind turbine, each includes $N$ samples in time. I can consider these $m$ points as my system inputs, but they are not independent, velocity measured at point $k$ is a function of points ${k-1, k-2, ...}$. One solution I tired was to consider only one of these points, but it has not given me good results. And the other problem is the delay between these $m$ points and my target point is time-varying and depends on wind velocity itself, which adds another level of complexity to the problem. 

Any ideas or references on the subject is highly appreciated.

Thanks.