基于GRU和自注意力机制的无人机故障检测方法  

UAV fault detection method based on GRU and self-attention mechanism

作　　者：梁本豪 黄庆南[1,2] 张恩泽 LIANG Benhao;HUANG Qingnan;ZHANG Enze(School of Automation,Guangxi University of Science and Technology,Liuzhou 545616,China;Institute of Intelligent Systems and Control(Guangxi University of Science and Technology),Liuzhou 545616,China)

机构地区：[1]广西科技大学自动化学院,广西柳州545616 [2]智能系统与控制研究所(广西科技大学),广西柳州545616

出　　处：《广西科技大学学报》2025年第2期69-76,共8页Journal of Guangxi University of Science and Technology

基　　金：国家自然科学基金项目(62161003);柳州市科技计划项目(2021AAF0103)资助。

摘　　要：针对无人机(unmanned aerial vehicle,UAV)在实际飞行中偶发的故障事件难以收集足够的故障数据来训练有效的故障检测模型,且现有的故障检测方法存在准确率低的问题,本文提出一种基于门控循环单元(gated recurrent unit,GRU)和自注意力机制结合的无人机故障检测方法。首先,在模型训练阶段使用无人机正常飞行的历史数据作为GRU的输入,对无人机的飞行趋势进行预测;为关注不同时刻对当前无人机飞行状态预测的影响,在GRU层后引入自注意力机制,对不同时刻的状态赋予不同的权值,提高对当前时刻无人机状态的预测准确度。其次,使用正常飞行状态下训练得到的残差来计算故障检测阈值。最后,使用无人机故障飞行序列在训练好的无人机故障检测模型中进行预测,并使用阈值进行检测。实验结果表明:本文提出的无人机故障检测方法的准确率达到了97.37%,与其他的故障检测方法相比,本方法在提高故障检测准确率、精确率和召回率的同时减少了故障误报的发生。In order to address the problems of difficulty in collecting enough fault data to train the unmanned aerial vehicle(UAV)fault detection model due to the faults of UAV as episodic events in real flight and the low accuracy of existing fault detection methods,this paper proposes an UAV fault detection method based on the combination of gated recurrent unit(GRU)and self-attention mechanism.Firstly,historical data of normal UAV flights were used as inputs to the GRU during the model training phase to predict UAV flight trends.To focus on the influence of different moments on the prediction of the current UAV flight state,a self-attention mechanism was added after the GRU layer to give different weights to the states at different moments to improve the prediction accuracy of the UAV state at the current moment.Subsequently,the residuals obtained from training in normal flight conditions were used to compute the fault detection thresholds.Finally,the UAV fault flight sequences were used in the trained UAV fault detection model for prediction and detection using thresholds.The experimental results show that the accuracy of the UAV fault detection method proposed in this paper reaches 97.37%,and compared with other fault detection methods,this method reduces the occurrence of fault false alarms while improving the accuracy,precision and recall.

关 键 词：无人机(UAV) 故障检测 门控循环单元 自注意力机制 

分 类 号：V279.2[航空宇航科学与技术—飞行器设计]