An adaptive stochastic parallel gradient descent (SPGD) control algorithm with a variable gain coefficient is applied in coherent beam combining (CBC) of a large scale fiber laser array. The influence of different gain coefficients on convergence speed of the control algorithm is computed. The relationship between control bandwidth, iteration rates, beam quality of combination, laser numbers and the feasibility of a large scale coherent beam combining based on this adaptive algorithm is analyzed. The results show that in CBC of 7 fiber lasers, this adaptive SPGD algorithm using a variable gain coefficient control strategy holds advantage of high iteration rates, high control bandwidth and good applicability to coherent beam combining of fiber laser array. The fast convergence speed can also be obtained and the convergence speed is increased by 37.8%, 63.8% and 75.0% respectively, when this algorithm is applied in CBC of 37, 91 and 100 fiber lasers. We believe the proposed adaptive SPGD technique has the potential to be scaled to a large-scale array with high output power.