随机并行梯度下降（SPGD）算法已被证明是一种较为有效的像清晰化系统控制算法，具有不依赖波前传感器直接对系统性能指标进行优化的特点。其控制参数增益系数和扰动幅度决定了算法的收敛速度以及收敛稳定性。参数取值范围较窄，超出范围将导致收敛后期的震荡，或者较慢的算法收敛速度。研究了算法增益系数和扰动幅度对校正效果和收敛速度的影响，提出了一种参数自适应优化的方法。基于52单元变形镜、位置敏感传感器等器件建立了SPGD控制算法的像清晰化实验平台，验证该方法的有效性。实验结果表明，该方法扩展了参数取值范围，提高算法收敛速度的同时具有较好的收敛稳定性。

The stochastic parallel gradient descent (SPGD) algorithm had been proved to be an effective system control method of image clarity in experiments and applications, which is independent of wavefront sensor and can optimize the system performance directly. The convergence speed and stability are determined by the value of the control parameters. The ranges of the control parameters are narrow, and out of range will lead to the vibration of the convergence or reduce the convergence speed. Based on SPGD, an image clarity test bed is built with a 52-element deformable mirror and a position sensitive detector. A method of automatic adjustment of parameters is proposed. The principle of SPGD control algorithm was demonstrated through examining the effects of gain and perturbation amplitude on correction capability and convergence rate, and the new method is proved to be effective. Experimental results show that by using the method of automatic adjustment of parameters, the ranges of parameters are extended. The practicality and convergence speed of the algorithm are improved with better convergence stability.