The nonlinear characteristics of radio telescope servo control system have a negative significant influence on the system dynamics characteristics identification. The measurement and compensation of system nonlinear characteristics will also increase the workload of system identification. In this research, a linear reconstruction method based on nonlinear sampling data is proposed to model dynamic characteristics. By extracting the phase and amplitude of the original sampling data, linear reconstruction of the system sampling data influenced by noise and nonlinear distortion is carried out to reduce the complexity of the model to be identified. A semi-physical experiment platform was built. Based on the actual sampling data of the platform, the linear data were reconstructed, and the dynamics model of the platform was evaluated and identified by singular value method and autoregressive neural network. The experimental results show that the singular value inflection point is reduced from 100 order to 40 order, and model identification is achieved with only 200 trainings of 10 neural network nodes.