In order to obtain the optimal control voltages of adaptive optics system, an closed-loop iterative control algorithm based on improved singular value decomposition is proposed. By adjusting parameters g1, g2 and w, the convergent velocity of model can be optimized, and the control signal can converge a reliable local optimal solution rapidly. An adaptive optical system based on micromachined membrane deformable mirror(MMDM) which optical influence functions of every actuator are measured is built. The quasi-square relationship between voltage and displacement of an actuator and the quasi-linear superposing among the actuators are experimentally verified. Wavefront from a model eye and a human eye are used respectively as the incident wavefront of adaptive optics system. Experimental results show that the improved algorithm can correct static and dynamic distorted wavefront aberration rapidly and effectively, is very suitable for adaptive optical system based on MMDM.