To address the low convergence speed and the challenge of selecting an optimal iterative step size in the stochastic parallel gradient descent (SPGD) algorithm, an adaptive SPGD algorithm based on energy feedback adaptation is proposed. This improved algorithm adjusts the step size adaptively based on the energy feedback value of the far-field spot, thereby accelerating convergence. Instead of requiring a specific initial step size, it only needs a defined convergence interval, allowing the initial step size to be chosen freely within this range. The proposed algorithm resolves the difficulty of initial step size selection. Simulations of 3-way, 19-way, and 37-way beam combinations demonstrate improved convergence speeds by 7.14%, 33.33%, and 42.78%, respectively, and enhanced convergence accuracy by 0.0018, 0.0016, and 0.0075, respectively, compared to the traditional algorithm. A 3-way beam combination experiment further confirms these findings, showing increases of 57.69% and 0.0297 in the convergence speed and convergence accuracy, respectively, with a final voltage increase of 0.052 V. The feasibility of the improved algorithm is validated through simulations and experiments comparing it to the traditional approach. The proposed algorithm effectively reduces system convergence time and increases beam energy, contributing valuable insights to multibeam synthesis technology.