To address the complex and unknown aerodynamic characteristics of the hypersonic glide vehicle (HGV) during high-speed, large-domain flight, this paper presents an intelligent online identification approach that combines offline learning with real-time adjustment. In the offline phase, an unsupervised pretraining and BP network framework is established to learn from wind tunnel and flight test data, constructing a neural network proxy model that maps flight states to aerodynamic coefficients. In the online phase, an aerodynamic sample library is built based on flight mechanics principles and measurement data. To ensure online identification accuracy, an adaptive moment estimation (Adam) algorithm is employed to rapidly update the aerodynamic network, making it more consistent with the actual aerodynamic characteristics of the HGV. Simulation results demonstrate that the offline aerodynamic model achieves a fitting accuracy of over 95%, while the online aerodynamic identification error remains below 15%. The proposed approach features low computational cost and high efficiency, making it well-suited for applications in HGV flight control.