基于模型的经典气动参数辨识方法无法解决运载火箭飞行过程中动压、攻角等关键参数难以在线获取和计算效率较低的问题，而基于数据驱动的智能辨识方法又过于依赖离线学习的数据质量，难以保证在线辨识精度。针对以上问题，提出了一种“离线学习+在线修正”的运载火箭气动参数在线辨识方法。在离线学习环节，利用长短期记忆神经网络（LSTM）从速度、位置和视加速度中学习并提取时序依赖关系，并将训练好的网络模型用于在线输出轴向力系数、法向力系数梯度、风向角和风速；在在线修正环节，基于网络模型的输出值，采用递推最小二乘（RLS）在线辨识气动参数误差增量，然后将误差增量与网络输出值叠加以得到气动参数辨识值。在仿真验证中，通过离线引入风场不确定性从而生成足量的训练数据以完成LSTM神经网络的离线学习，然后采用RLS在线修正气动参数，仿真结果表明：相比直接利用神经网络模型辨识气动参数，本文提出的“离线学习+在线修正”的辨识方法能够显著提高辨识精度，同时具备较高的计算效率。

Classical model-based aerodynamic parameter methods cannot solve the key problems in launch vehicle flight， such as the difficulty of obtaining dynamic pressure and angle of attack online， and low computational efficiency. To address the above problems， this paper proposes an online aerodynamic parameter identification method for launch vehicles based on an “offline learning + online correction” framework. In the offline learning process， the Long Short-Term Memory neural network （LSTM） is used to learn and extract the time series features from velocity， position and apparent acceleration， and the trained network model is used to output the axial force coefficient， normal force coefficient gradient， wind direction angle and wind speed. In the online correction part， based on the output value of the network model， the Recursive Least Squares （RLS） is used to identify the error increment of the aerodynamic parameters online， and then the error increment is superimposed with the network output value to obtain the online identification value of the aerodynamic parameters. In the simulation verification， the wind field uncertainty is introduced offline to generate sufficient training data for the LSTM neural network， and then the aerodynamic parameters are corrected online in combination with the RLS. The simulation results show that compared with using the neural network model alone， the identification method of'offline learning + online correction'proposed in this paper can significantly improve the identification accuracy while maintaining high computational efficiency.