To reduce the servo lag error in adaptive optics to correct the atmosphere turbulence distortion,a kind of neural network prediction algorithm to predict the control voltage of deformable mirror is proposed. The two-layer back propagation neural network prediction method with second-order learning algorithm used to predict the voltage of deformable mirror in advance is studied through numerical simulation,based on the atmospheric turbulence wavefront data influenced by transversal wind. The look-back frame and learning-rate parameter influencing the prediction effect is discussed. The residual error of the adaptive optic system is calculated with neural network prediction algorithm and recursive least-square (RLS) algorithm. The results show that the residual error caused by servo lag in the system is reduced more effectively using the neural network prediction algorithm than using the RLS prediction algorithm.