基于集成孤立森林的无人机异常检测算法

UAV Anomaly Detection Algorithm Based on Integrated Isolated Forest

无人机(Unmaned Aerial Vehicle,UAV)运行过程中,通过嵌入式传感器感知飞行环境,会受到通信链路及恶劣操作环境等不确定因素的影响,导致其运行风险,从而威胁公共安全,这是亟待解决的问题。UAV异常状态检测是控制运行风险的重要手段。针对UAV异常检测的特点,利用UAV在安全状态运行过程中监管平台搜集到的相关飞行数据建立模型,提出了一种集成孤立森林(Integrated Isolated Forest,IIF)无监督异常检测算法,以便检测出异于UAV正常运行的情况。使用不同类型的UAV数据集验证了该算法的检测性能,实验结果表明,算法检测无矫正措施的UAV飞行异常的F1分数(F1-score)在0.94以上,对无矫正措施的UAV飞行异常的F1分数在0.75以上,相较于传统异常检测算法,该算法具有更好的检测性能。

During the operation of unmanned aerial vehicles(UAVs),the flight environment perceived by embedded sensors will be affected by uncertain factors such as communication links and disadvantaged operating environment,resulting in operation risks and threats to public safety,which are urgent problems to be solved.Abnormal state detection of UAV is an important means to control operation risk.According to the characteristics of UAV anomaly detection,an Integrated Isolated Forest(IIF)unsupervised anomaly detection algorithm is proposed by using the relevant flight data collected by the monitoring platform during the operation of UAV in safe state to establish a model,so as to detect the situation different from the normal operation of UAV.Different types of UAV data sets are used to verify the detection performance of the algorithm.Experimental results show that the F1-score of the algorithm for detecting UAV flight anomalies without correction measures is above 0.94,and the F1-score of the algorithm for detecting UAV flight anomalies without correction measures is above 0.75.Compared with the traditional anomaly detection algorithm,this algorithm has better detection performance.