摘要
文中提出一种融合时序注意力的CNN-BiGRU模型用于四轴无人机的轨迹预测.通过卷积层捕捉无人机轨迹切片在窗口内的空间特征,输入双向门控循环单元获取前向与后向时间特征的影响关系,对解码阶段输出向量的时间步维度引入注意力机制,以增强提取关键特征的能力.结果表明:基于所采集的多种四轴无人机轨迹数据集,所提方法在预测时长5 s时,相比于门控循环单元网络的平均MAE、RMSE、MAPE分别降低12.9%、8.0%、3.9%.
A CNN-BiGRU model integrating temporal attention was proposed for trajectory prediction of four-axis UAV.The convolution layer captured the spatial characteristics of UAV trajectory slices in the window.The influence relationship between forward and backward time features was obtained by inputting the bidirectional gated cyclic unit,and attention mechanism was introduced into the time step dimension of the output vector in the decoding stage to enhance the ability of extracting key features.The results show that the average MAE,RMSE and MAPE of the proposed method are reduced by 12.9%,8.0% and 3.9%,respectively,compared with the gated cyclic cell network,when the prediction duration is 5 s based on the collected trajectory data sets of various four-axis UAVs.
作者
李娜
羊钊
王业萍
高翔宇
LI Na;YANG Zhao;WANG Yeping;GAO Xiangyu(College of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;College of General Aviation and Flight,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)
出处
《武汉理工大学学报(交通科学与工程版)》
2025年第4期749-755,共7页
Journal of Wuhan University of Technology(Transportation Science & Engineering)
基金
国家重点研发计划资助(2022YFB3104502)
国家自然科学基金项目(52172328)
工信部专项(TC220A04A-79)。
关键词
四轴无人机
时间序列
注意力机制
循环神经网络
轨迹预测
four-axis drone
time series
attention mechanism
recurrent neural network
trajectory prediction
作者简介
第一作者:李娜(1999-),女,硕士生,主要研究领域为低空空域无人机运行规划与管控。
