Considering that greedy characteristics lead to slow convergence and low accuracy problems of surrogate-based optimization solutions in the later stage, a collaborative improvement aggregation strategy is proposed to address this. This strategy is extended to an efficient surrogate-based optimization method for expensive black-box problems. The proposed approach integrates probability improvement and mean improvement criteria using Chebyshev decomposition and achieves a balance between global exploration and local search capabilities through random weight coefficients. In addition, the differences between surrogate-based optimization and surrogate-assisted optimization methods were analyzed from the perspective of candidate point sets, further exploring the role of randomness in optimization design. The experimental results show that this method can effectively improve the convergence accuracy of the optimized solution in black-box problems. Compared with similar techniques, this method has higher convergence accuracy and stronger robustness of optimized solutions.