The fundamental theory for coherent beam combing (CBC) via using stochastic parallel gradient descent (SPGD) algorithm is introduced and an adaptive SPGD algorithm is proposed. The convergence rate is increased by adaptively controlling the perturbations and stages of the SPGD algorithm. The results show that for CBC of laser arrays with 25 channels, 49 channels and 100 channels by using adaptive SPGD algorithm, the convergence rates are increased by 36.6%, 59.8% and 80.2%, respectively. This method has an advantage for CBC of laser arrays with large number of lasers.