To address the shortcomings of blind search and the tendency to get trapped in local optima caused by high-dimensional variables and strong nonlinear mechanical constraints in the design of complex engineering structures, a multi-strategy improved starfish optimization algorithm based on the mixed dimension selection is proposed. The algorithm adopts a hybrid dimension selection strategy based on maximum deviation, prioritizes updating dimensions with large deviations from the global optimal solution, and randomly updates some dimensions to balance directed search and random exploration. Then the algorithm introduces a a nonlinear adaptive probability factor to dynamically balance global exploration and local exploitation. Simultaneously the algorithm builds a dual elite boundary mechanism to effectively enhance the diversity of the population and uses Cauchy variation to perturb the optimal solution to strengthen the algorithm’s ability to escape from local extreme values. The calculation results of the standard test functions and the application examples in optimal design of engineering structures with high-dimensional constraints show that the proposed algorithm has prominent advantages in convergence speed, optimization accuracy and solution robustness.