摘要
针对无人机飞行模式切换导致飞行数据在线异常检测准确率低的问题,提出基于过采样投影近似基追踪(OSPABP)的在线异常检测框架。首先利用滑窗和Z-score变换消除飞行数据流量纲,并抽取相关的飞行数据子集;然后过采样当前时刻子集的输入数据,放大异常数据对数据子空间的影响,并通过在线估计和追踪匹配过采样后数据子空间的投影近似基方向变化,从而判断子集实时输入数据的异常程度。同时该方法还可抑制飞行模式切换对异常检测效果的影响。采用Flight Gear模拟飞行数据和明尼苏达大学真实无人机飞行数据的实验结果表明,所提出方法对飞行模式切换敏感度低,可明显降低异常检测的误检率,并有效提高检测准确率。此外,算法的计算和存储复杂度均可满足机载处理要求。
Aiming at the low accuracy problem of flight data online anomaly detection caused by flight mode switching of UAV,an online anomaly detection method is proposed based on Over Sampling Projection Approximation Basis Pursuit（ OSPABP）. Firstly, the dimension of flight data stream is eliminated with sliding window and Z-score transformation,and correlated flight data subset is extracted from sliding window. Secondly,the multivariate data of the subset in current time is oversampled to amplify the influence of abnormal data on data subspace. Through online estimation and pursuing match of the direction change of the projection approximation basis of the data subspace after oversampling,the anomaly of the real time input data of the subset is determined. Meanwhile,the method can also suppress the influence of flight mode switching on anomaly detection result. The experiments on the simulated flight data from Flight Gear and the real UAV flight data from University of Minnesota were conducted; the experiment results show that the proposed OSPABP method can reduce the false positive rate of anomaly detection significantly and improve the accuracy of anomaly detection effectively as its sensitivity to flight mode switching is low. In addition,the computational and storage complexity of the OSPABP method meets the requirement of flight data real-time processing.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2016年第7期1468-1476,共9页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61571160)
部委重点基金课题(9140A17050114HT01054)项目资助
关键词
无人机
在线异常检测
数据流
过采样
投影近似基
unmanned aerial vehicle（UAV）
online anomaly detection
data stream
oversampling
projection approximation basis
