航空学报 > 2025, Vol. 46 Issue (23): 131483-131483   doi: 10.7527/S1000-6893.2025.31483

基于循环神经网络的非失速和近失速飞行数据气动参数辨识

惠哲1,2,3, 都东岳4, 刘雨亭1,2, 昌敏1,2,3(), 白俊强1,2,3,4   

  1. 1.西北工业大学 无人系统技术研究院,西安 710072
    2.西北工业大学 无人飞行器技术全国重点实验室,西安 710072
    3.西北工业大学 无人机技术集成攻关大平台,西安 710072
    4.西北工业大学 航空学院,西安 710072
  • 收稿日期:2024-11-01 修回日期:2024-12-16 接受日期:2025-04-02 出版日期:2025-04-11 发布日期:2025-04-10
  • 通讯作者: 昌敏 E-mail:changmin@nwpu.edu.cn
  • 基金资助:
    国家自然科学基金(12402272);陕西省自然科学基础研究计划项目(2024JC-YBQN-0024);中央高校基本科研业务费专项资金(D5000240030)

Aerodynamic parameter identification of non-stall and near-stall flight data using recurrent neural networks

Zhe HUI1,2,3, Dongyue DU4, Yuting LIU1,2, Min CHANG1,2,3(), Junqiang BAI1,2,3,4   

  1. 1.Unmanned System Research Institute,Northwestern Polytechnical University,Xi’an 710072,China
    2.National Key Laboratory of Unmanned Aerial Vehicle Technology,Northwestern Polytechnical University,Xi’an 710072,China
    3.Integrated Research and Development Platform of Unmanned Aerial Vehicle Technology,Northwestern Polytechnical University,Xi’an 710072,China
    4.School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China
  • Received:2024-11-01 Revised:2024-12-16 Accepted:2025-04-02 Online:2025-04-11 Published:2025-04-10
  • Contact: Min CHANG E-mail:changmin@nwpu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(12402272);Natural Science Basic Research Program of Shaanxi Province(2024JC-YBQN-0024);Fundamental Research Funds for the Central Universities(D5000240030)

摘要:

提出了一种结合门控循环单元(GRU)神经网络模型和高斯-牛顿(GN)优化算法的气动参数辨识方法,旨在分别从自主研发的小型无人机和ATTAS飞机生成的纵向飞行数据中准确辨识未知的气动参数。经过设计和训练的GRU网络模型用以表征所选飞机系统的动力学,同时避免对其假设的动力学模型进行数值积分。GN优化算法结合训练完毕的GRU网络模型,通过迭代最小化与未知气动参数相关的代价函数来获得高置信度的参数辨识结果。本文用到的实测飞行数据包括:自研无人机的非失速飞行数据和ATTAS飞机的近失速飞行数据。研究结果表明:GRU网络模型可以通过调整相应的网络参数(诸如,隐藏层的数量、单个隐藏层中的单元数量、丢弃率、时间步长和学习率)来保持对非失速和近失速飞行数据的可靠预测效果。此外,通过将所选飞机气动参数的辨识值与相对应的风洞测量值或参考值进行比较,证实了所提参数辨识方法的有效性。

关键词: 气动参数辨识方法, GRU网络模型, GN优化算法, 实测飞行数据, 网络参数

Abstract:

This paper proposes a new aerodynamic parameter identification method that combines a Gated Recurrent Unit (GRU) neural network model and the Gauss-Newton (GN) optimization algorithm, aiming to accurately identify the unknown aerodynamic parameters from longitudinal flight data generated by a self-developed small-sized Unmanned Aerial Vehicle (UAV) and ATTAS aircraft. The GRU network model was designed and then trained to characterize the chosen aircraft systems’ dynamics while avoiding numerical integrals of the postulated dynamic models. The GN optimization algorithm incorporates the trained GRU network model to obtain high-confidence identification results by iteratively minimizing a cost function concerning the unknown aerodynamic parameters. The measured flight data used in this study includes non-stall flight data from the self-developed UAV and near-stall flight data from the ATTAS aircraft. The study results demonstrate that the GRU network model can maintain reliable prediction effects for the non-stall and near-stall flight data by adjusting the corresponding network parameters such as the number of hidden layers, number of units in each hidden layer, dropout rate, time-step, and learning rate. Furthermore, the effectiveness of the proposed parameter identification method is confirmed by comparing the identified aerodynamic parameters with the corresponding wind tunnel or reference values.

Key words: aerodynamic parameter identification method, GRU network model, GN optimization algorithm, measured flight data, network parameters

中图分类号: