Abstract: The numerical simulation method of ground motion can provide ground motion input for seismic hazard analysis and seismic design of structures. Due to the uncertainties associated with ground motion simulation parameters and the limitations of simulation technology, the simulated ground motions may be quite different from the actual ground motions of the target area. With the increased number of observed regional strong ground motions, it is meaningful to integrate the measured ground motion characteristics into the simulated ground motions to reflect the local characteristics of ground motion. But at present, there is no synthetic ground motion method considering the characteristics of ground motion in the target area. To solve this problem, the ground motion time history is synthesized by machine learning method combined with the actual observed ground motion in the target region. The principal component analysis algorithm is applied to extract the mother waves which represent the characteristics of the ground motion from the huge target ground motion database; the reasonable prediction equation is selected to predict the response spectrum; then the improved particle swarm optimization algorithm is applied to quickly solve the combination coefficient of the mother waves of the ground motion. Thus, the weighted new response spectrum matches that from the prediction equation, and the synthetic time history of ground motion not only meets the results of seismic hazard analysis, but also conforms to the characteristics of the local ground motion. Based on the method proposed in this paper, the feasibility and effectiveness of the method are verified by using the strong motions in Sichuan of China, which provides a new idea and method for the synthesis of ground motions considering the characteristics of regional ground motions.