基于误差序列的离线数据无监督快速异常检测方法研究

Research on Unsupervised Fast Anomaly Detection of Off-line Data Base on Error Sequences

摘要:

针对利用多维度、多模态的遥测参量实现快速定位检测故障需求的问题,在分析和归纳飞行器遥测参量特点及故障检测方法的基础上,研究了基于误差序列的离线数据无监督快速异常检测方法。该方法改变了传统的主要依靠人工分析进行故障判定模式,结合历史数据的使用,通过无监督自适应阈值的设定,实现自动适配并检测遥测参量异常记录状态的功能,降低了需要大量人员分析的时间成本,减少了遂行处理工作中快速定位故障的压力,解决了故障定位通用性不强的问题。经过实际应用,效果良好,满足数据处理要求,具有较好的自适应性,易于实现。

Abstract:

To solve the problem of using multi-dimensional and multi-modal telemetry parameters to realize fast location fault detection,based on the analysis and induction of the characteristics of telemetry parameters and fault detection methods of aircraft,an unsupervised fast anomaly detection method for offline data based on error sequences is studied.This method changes the traditional fault judgment mode mainly relying on manual analysis, and combines the use of historical data with the setting of unsupervised adaptive wide value,realizing the function of automatically adapting and detecting the abnormal recording state of the telemetry parameter,reducing the time cost requiring a large number of personnel to analyze, and problem that the fault location is not universal is solved.After practical application,the results are good ,meet the requirements of data processing,have better self-adaptability,easy to implement.