Adaptive optics (AO) based on stochastic parallel gradient descent (SPGD) algorithm can be used to correct the phase aberration without the wavefront sensor,but the low speed of convergence confines its application in real-time system. Considering the relationship between the stochastic perturbation and convergence speed of SPGD algorithm,a technique is proposed to optimize the proportion of stochastic perturbation with Zernike mode. Aiming at atmospheric turbulence,a 61-element AO model based on SPGD algorithm is set up,and a group of phase aberration with the Kolmogorov spectrum is simulated numerically to research the effect on convergence characteristic resulting from this method. Results show that,comparing with the AO before optimization,the convergence speed can be improved efficiently although it gives up a little correction precision.