In the process of damage identification for high-order nonlinear structure such as cable-stayed bridges by the standard genetic algorithm, premature convergence would appear. In order to avoid this, an improved hierarchic genetic algorithm was proposed. The cable force change was used to establish the optimization function and threetypes of standard genetic algorithm were combined with variable fine-tuning and hierarchic strategy.To establish a hierarchical genetic algorithm with catastrophe characteristics A single-tower cable-stayed bridge model was used in the numerical simulation and the result showed that the probability of premature convergence was reduced in the improved hierarchic genetic algorithm and and the cable-stayed bridge damage was identified effectively. The anti-noise performance was better.