Early damage detection and reinforcement of civil engineering structures are crucial. To ensure timely maintenance in the later stage, the civil structure is subjected to modal parameter identification. In vibration signal recognition, traditional stochastic subspace vibration signal recognition has modal omissions. Previous studies have shown that using singular value decomposition can suppress signals with lower energy. Therefore, zero phase filtering technology is used to improve the random subspace identification method. Singular value decomposition is used for structural parameter analysis and noise suppression. The zero phase filtering technology is combined to solve the low energy signal loss. Therefore, an improved stochastic subspace identification method for civil engineering structural parameter identification is proposed. This study validates the proposed method through experiments, analyzes the corresponding design parameters and experimental data results, and verifies the advantages and feasibility of this method. According to the research results, there were multiple peaks within 0-10Hz under environmental effects, with 3.15Hz and 4.84Hz corresponding to the first two modes of the system. Under the action of vehicle load, two peaks were displayed around 3.15Hz and 4.84Hz, with two clearly stable axes. The system order was determined to be N=4 through the stability diagram. Under relevant amplitude conditions, the estimated damping ratio of the two test points in the second mode was always equal to 0.55%. It can effectively analyze the natural frequency and damping ratio in civil engineering structures, laying the foundation for damage detection in civil engineering structures.