As a highly dynamic, strongly coupled, and parameter time-varying uncertain nonlinear dynamic system, Hypersonic Flight Vehicle (HFV) have the problem of difficulty in accurately characterizing their full envelope dynamic characteristics in uncertain environments using traditional analytical models. In response to this issue, a hybrid data-driven digital twin modeling method based on analytical models and dynamic regression neural network is proposed in this paper. Firstly, the residuals between the actual state variables of the HFV and the state variables of the analytical model are calculated. Introducing a dynamic regression neural network based on Gate Recurrent Unit (GRU) encoder-decoder to predict model residuals and actuator input instructions. Secondly, based on the predicted values of the actuator input instructions, the state variables of the analytical model are calculated as the reference values, and the residuals predicted values are used to dynamically compensate the reference values online to ensure the high fidelity of the HFV model. Finally, the parameters of the digital twin model are further updated online according to the actual data measured by the sensor to realize the real-time dynamic matching calibration of the hybrid digital twin model and the HFV dynamics. The digital twin model not only integrates the efficiency and interpretability of the traditional analytical model but also improves the approximation accuracy and rate of convergence and reduces the amount of online training computation by using the analytical model as a priori knowledge compared with the traditional deep learning model. This paper verifies the effectiveness of the proposed hybrid data-driven digital twin model through simulation.