Accurate prediction of the atmospheric turbulence evolution in the next few sampling cycles can compensate for the time delay in the control systems of adaptive optics (AO) systems. In this paper, a predictive correction method in AO systems is proposed, and its robustness is analyzed. Under the frozen-flow assumption, the translational motion can be identified using the slope measurements of a Shack-Hartmann wavefront sensor (SHWS) in AO systems. Using the transverse wind information, prediction of the future slope can be achieved by Fourier translation. The shape of the deformable mirror (DM) can be calculated by the direct-gradient wavefront reconstruction algorithm. The aberrated wavefront is corrected by the DM. With a known transverse wind, the proposed predictive correction can provide a perfect compensation for the decline of the dynamic performance caused by delays in the control system. With estimated wind parameters, improvement of the correction efficiency can be obtained as long as the wind-velocity estimation error is less than the velocity itself, while the wind direction is estimated accurately, or the direction error is less than 60° while the wind speed is estimated accurately. With a simultaneous wind-velocity and direction error, the correction efficiency can still be improved within a large error range.