基于慢特征密度聚类的气路异常检测方法

在结合核技巧、慢特征分析算法与密度聚类方法的基础上，提出了基于混合核慢特征分析和密度聚类的慢特征密度聚类算法，实现了基于民航发动机气路参数原始值的异常检测。核技巧的引入克服了慢特征分析法处理复杂数据时可能存在的维度爆炸问题，充分利用不同核函数的特点和慢特征分析的优势从气路参数原始值中提取出随时间变化最缓慢的特征作为密度聚类算法的输入，最终筛选出异常值。经实验对比发现，该方法针对某些异常拥有最好的聚类效果和最低的虚警率，尤其是检测可调放气活门系统异常时虚警数量不到样本总数量的0.5%，是一种有效的方法。 

Slow feature density clustering-based gas path anomaly detection method

Based on the kernel method，slow feature analysis algorithm and density clustering method，a mixed-kernel slow feature density clustering algorithm was proposed to detect the original gas path parameters of civil aero-engine.The introduction of the kernel method overcame the possible dimensional explosion when the slow feature analysis dealt with complex data，and it took full use of the characteristics of different kernel functions and advantages of slow feature analysis.This algorism can extract the feature with the slowest time-dependent change from the original gas path parameters and use it as the input of the density clustering algorithm.Then the anomalies were found out.Through experimental validation，this method had the best clustering results and the lowest false alarm rate on certain anomalies.Especially，the number of false alarms was less than 0.5% of the total number of samples when detecting the anomaly of the variable bleed valve system，proving it is an efficient method.