Due to the influence of complex loads and adverse environments，the service performance of bridge structures is gradually deteriorating. Structural modal parameters are the characteristic indexes of the overall mechanical behavior of the structure，and the structural service state can be identified by sensitive modal parameters.This paper mainly focused on the problem of false modes ，easy omission of real modes ，and low computational efficiency in traditional modal recognition. The chaotic local search was used to improve the particle swarm algorithm，and the position and size of the window truncation of the acceleration signal were optimized. The modal identification of each sub-signal was carried out by combining covariance-driven stochastic subspace identification. A damage identification model based on proportional flexibility matrix and uniform load curvature was established，and the location and degree of structural damage were determined by the relative change of deflection curvature. Finally, the effectiveness of the proposed method was verified by conducting damage identification tests on the scaled models of cable-stayed bridges. The results show that the proposed chaotic particle swarm stochastic subspace method had a better accuracy of modal identification ，and could realize accurate quantitative and location identification of typical damages such as cables and girders of cable-stayed bridges. The damage at the mid-span position of the structure had a greater impact on structural behavior. In practical engineering，the bearing capacity reserve of the main beam and cable in the mid-span area should be increased to enhance structural safety.