基于SSA-DBSCAN的边坡安全监测数据粗差探测方法

Gross Error Detection of Slope Safety Monitoring Data Based on SSA-DBSCAN

考虑到边坡安全监测数据中存在粗差这一问题,提出了一种基于奇异谱分析(SSA)和密度聚类算法(DBSCAN)的粗差探测法,该方法结合SSA在提取信号和DBSCAN算法在区分粗差和异常值上的优势:首先使用SSA对监测序列进行分解重构,准确提取主信号并获取残余分量;然后使用DBSCAN聚类算法对残余分量进行分析;最后联合2种方法确定粗差点并剔除。通过引入多因素影响的边坡监测序列实例进行验证,并且将SSA-DBSCAN粗差探测法与中位数绝对偏差法(MAD)和格拉布斯准则法(Grubbs)进行比较分析。结果表明,本文提出的SSA-DBSCAN粗差探测法与上述方法相比性能优异、误判率低,可为后续监测数据分析处理乃至于预测预警奠定基础。

A method of detecting the gross error of slope monitoring data is presented based on singular spectrum analysis(SSA)and density-based spatial clustering of applications with noise(DBSCAN).The method integrates the advantages of SSA in signal extraction and DBSCAN in distinguishing gross errors and outliers.Firstly,SSA is used to decompose and reconstruct the monitoring series to accurately extract the main signal and obtain the residual components.Secondly,DBSCAN is employed to analyze the residual components.The two methods are combined to determine and eliminate the gross errors.Examples of slope monitoring series affected by multiple factors are introduced for verification.Moreover,the present method is compared with the median absolute deviation method(MAD)and Grubbs criterion method(Grubbs),and results suggest that the present SSA-DBSCAN method is of excellent performance and low misjudgment rate compared with the abovementioned methods.