Experimental and numerical studies were conducted to examine an idea of selecting a roll-cell pattern in 2D flow among many possible stable states. Neural networks (NN) were applied in order to recognize a certain flow pattern. And Pyragas' delayed feedback control (DFB) theory is used to overcome the instability caused by process time delay of feedback loop. Notable approach of this study toward active control of fluid flow is that suitable control rule and control parameters can be chosen by way of judging the flow pattern or flow state. To selecting a certain mode of flow, particle image velocimetry (PIV) is utilized for estimating flow field, then NN switches to the suitable control rule for stabilizing the mode. 3D numerical simulations were also carried out in order to find optimal parameters of Pyrags control.