Scholarlink

<p>To address the trajectory deviation issue occurring in individual members of unmanned aerial vehicle (UAV) formations from predefined flight paths, a trajectory prediction and correction framework utilizing joint neural networks (NNs) is proposed. The methodology consists of four principal components: First, flight trajectory data of the UAV is generated through simulations. Second, a hybrid Convolutional Neural Network with Long Short Term Memory (CNN-LSTM) network is built up for trajectory prediction, combining advantages of spatial feature extraction of the CNN with the temporal feature extraction of the LSTM. Third, a prediction fusion network is designed to mitigate prediction errors arising from coordinate system transformations between multiple observing UAVs. The integrated framework enables accurate trajectory estimation for deviating UAVs through multi-source predictive fusion. Finally, experimental validation demonstrates the superior prediction accuracy of both the CNN-LSTM model and the fusion network. Compared with baseline network, the prediction errors of the CNN-LSTM are reduced by 34.5%, and compared with single network, the prediction errors of the fusion network are reduced by 96.1%. The results show that the proposed joint-network algorithm is feasible and effective for formation flight maintenance in complex aerial environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Trajectory deviation prediction of UAV formation by joint neural networks

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Jingzhou Dai Ruan,
  • Shiqian Liu,
  • Han Chen,
  • Weizhi Lyu,
  • Qian Zhang

摘要

To address the trajectory deviation issue occurring in individual members of unmanned aerial vehicle (UAV) formations from predefined flight paths, a trajectory prediction and correction framework utilizing joint neural networks (NNs) is proposed. The methodology consists of four principal components: First, flight trajectory data of the UAV is generated through simulations. Second, a hybrid Convolutional Neural Network with Long Short Term Memory (CNN-LSTM) network is built up for trajectory prediction, combining advantages of spatial feature extraction of the CNN with the temporal feature extraction of the LSTM. Third, a prediction fusion network is designed to mitigate prediction errors arising from coordinate system transformations between multiple observing UAVs. The integrated framework enables accurate trajectory estimation for deviating UAVs through multi-source predictive fusion. Finally, experimental validation demonstrates the superior prediction accuracy of both the CNN-LSTM model and the fusion network. Compared with baseline network, the prediction errors of the CNN-LSTM are reduced by 34.5%, and compared with single network, the prediction errors of the fusion network are reduced by 96.1%. The results show that the proposed joint-network algorithm is feasible and effective for formation flight maintenance in complex aerial environments.