基于Transformer模型的四旋翼无人机时空协同航迹预测方法设计
Design of Spatiotemporal Collaborative Trajectory Prediction Method for Quadcopter UAVs Based on Transformer Model
欧洋 1漆雪莲 2胡清月1
作者信息
- 1. 中国电子科技集团公司第10研究所,成都 610036
- 2. 四川省交通运输集团有限责任公司,成都 610000
- 折叠
摘要
无人机在执行任务时面临的飞行环境复杂多变,为了减少事故的风险,并在飞行时对异常情况进行预测和响应,研究一种基于Transformer模型的四旋翼无人机时空协同航迹预测方法;采集四旋翼无人机原始航迹,实施异常点剔除和缺失点插值处理,以优化和清理原始航迹数据,便于后续的航迹预测;使用卷积神经网络实施特征进行数据提取,通过编码和解码过程获取学习数据低维,结合深度学习和表示学习方法完成数据降维;基于Transformer模型实现无人机时空协同航迹的精准预测,通过数据异常点剔除与插值补缺,对采集的四旋翼无人机原始航迹数据实施预处理,提高数据的质量和完整性;实验测试结果表明,设计方法的预测结果虽然相对于真实的坐标点稍有偏差,然而整体结果在可接受范围内,验证集所有数据的均方误差在数据条数为300时仅为0.32 m,拟合优度指标测试结果最接近1,具有良好的航迹预测能力;该方法可以更好地优化无人机的航迹规划,实现多无人机之间的时空协同航迹规划,避免碰撞和冲突,并优化飞行效率.
Abstract
In order to reduce the risk of accidents and predict and respond to abnormal situations in the complex and changeful flight environments faced by unmanned aerial vehicles(UAVs)during mission execution,a spatiotemporal collaborative trajectory prediction method based on Transformer model for quadcopter UAVs is studied.Collect the original trajectory of the quadcopter drone,implement the outlier removal and missing point interpolation processing to optimize and clean up the original trajectory data for subsequent trajectory prediction.Convolutional neural networks are used to conduct the feature extraction,the low dimensional learning data are obtained through encoding and decoding processes,and deep learning and representation learning methods are com-bined to complete the data dimensionality reduction.Based on the Transformer model,the UAV spatiotemporal cooperative trajectory is precisely predicted.By removing the outliers and interpolation,the collected raw trajectory data of quadcopter UAVs is prepro-cessed to improve the quality and completeness of the data.The experimental test results show that although the prediction results of the design method have a slight deviation from the actual coordinate points,the overall results are within an acceptable range.The mean square error of all data in the validation set is only 0.32 m when the number of data is 300,and the goodness of fit index is clo-sest to 1,indicating its good trajectory prediction ability.This method can better optimize the trajectory planning of drones,achieve spatiotemporal collaborative trajectory planning among multiple drones,avoid collisions and conflicts,and optimize flight efficiency.
关键词
Transformer模型/四旋翼无人机/表示学习/时空协同航迹预测Key words
Transformer model/quadcopter UAV/represent learning/spatiotemporal collaborative trajectory prediction引用本文复制引用
出版年
2024
NETL
NSTL
万方数据