Based on the data-driven stochastic subspace identification (SSI), an improved ITD method, an improved STD method and an improved complex index method are put forward. SSI has high identification accuracy, and Hankel matrix orthogonal projection calculation of data can retain all the useful information in the original data and simultaneously remove noise. The data obtained in P-matrix is taken as input for the ITD , STD and complex index methods, and then the three methods would no longer need random decrement technique or NExt method for pre-processing, thereby eliminating the error caused by the two pretreatment methods.The improved methods are applied to the modal parameter identification for structure shaking table test model by using the simulated earthquake response data from Tongji University. The recognition results further prove the correctness and feasibility of the improved methods.