Abstract: The theory about stochastic parallel gradient descent(SPGD) algorithm and the use of SPGD algorithm for coherent beam combination are introduced. For solving the difficulty for adjusting the key parameters in SPGD algorithm, a new method of combining the hardware and software is proposed, which can online collect and analyze the experimental data and automatically adjust the key parameters in SPGD in real time. Experiments of 4 fiber-laser coherent beam combination are developed, in which different adjusting methods are compared. The results show that using new methods can efficiently achieve the value of the gain coefficient and the stochastic perturbation amplitude in SPGD algorithm. The experiments are profitable. The new method proposed in the paper is novel and effective, which can also guide the coherent beam combining experiments in the future.