基于时空关联的飞控传感器数据异常检测

A spatio-temporal correlation method for flight control sensor data anomaly detection

飞控传感器数据的异常检测对保证飞行器安全稳定运行有重要意义。然而,现有的异常检测方法大多仅从传感器数据之间的关联性或不同传感器间数据的相关性变化角度出发,当飞行器运行工况动态变化时,可能会由于对传感器数据的特征提取不充分而导致异常检测结果的准确率偏低且虚警率偏高。对此,本文提出一种基于时空关联的飞控传感器数据异常检测方法,实现对传感器数据时间与空间2个维度变化规律的融合建模。首先,同时构建时序演化与空间相关性特征提取模块来对传感器数据进行时间与空间两个维度特征的并行提取。其次,对2个模块的预测输出进行时空关联融合,得到时空关联的预测飞控传感器数据。最后,基于预测数据与实际数据残差的统计量进行阈值选取并对传感器数据进行异常检测。在无人机惯性测量单元仿真与实测数据集上对本文方法进行验证,结果表明相较于相关向量机等典型的异常检测方法,本文方法的异常检测准确率至少提高0.4%且虚警率至少降低1.8%。

The normal operation of flight control sensors is the key of the safety and stability of the aircraft.However,most of the anomaly detection methods only consider the correlation between sensors data or the correlation change of data between different sensors.When the operating conditions of aircraft change dynamically,the accuracy of anomaly detection results may be low and the false alarm rate is high due to insufficient feature extraction.In this article,a flight control sensor data anomaly detection method based on spatiotemporal correlation is proposed to achieve the fusion modeling of sensor data changes in time and space.Firstly,the feature extraction module of temporal evolution and spatial correlation is formulated to extract the features in time and space in parallel.Secondly,the spatio-temporal correlation fusion is carried out to obtain spatio-temporal correlation predictive data.Finally,based on the statistic of the residual between the predicted data and the actual data,the threshold is selected and the sensor data is detected.Through the verification of simulation and measurement data,compared with typical anomaly detection methods such as RVM,the anomaly detection accuracy rate of the proposed method is at least 0.4%higher and the false alarm rate is at least 1.8%lower.