Traditional coherent beam combination (CBC) system architecture has revealed inadequacies in meeting the concurrent demands of large-scale deployment and high-bandwidth requirements. Addressing this challenge, we propose a distributed CBC system architecture based on the optimized stochastic parallel gradient descent (SPGD) algorithm. Our strategy segments the large-scale laser array into multiple independent smaller-scale subarrays, ensuring their efficient phase convergence through the introduction of corresponding reference lasers while avoiding interference when integrating different subarrays. Moreover, the piecewise SPGD algorithm is proposed and the intensity of the reference laser is modulated to further improve the convergence speed and accuracy within subarrays, enhancing the algorithm's compatibility across laser arrays of varying scales. We have validated the feasibility of the distributed CBC architecture through numerical analysis and assessed the strategy's performance in both static and dynamic environments using simulation software. The simulation findings indicate that, compared to traditional CBC systems, distributed architecture with 3, 7, and 19 subarrays and utilizing the piecewise SPGD algorithm, has experienced phase control bandwidth enhancements by factors of approximately 3.6, 10.4, and 32.5 respectively, maintaining superior average power output in dynamic noise environments. The proposed architecture and strategy also accommodate subarrays of variable scales and obviates the necessity for large-aperture optical components on the emitted plane, demonstrating exceptional scalability and adaptability.