基于密度聚类的监测数据漂移动态校正算法

Dynamic correction method for monitoring data drift based on density clustering

由于系统故障或外界环境扰动影响,工程监测数据常会出现漂移的现象。对于水工建筑物,由于结构的相似性和测点布置的相关性,其监测量往往具有显著的空间关联性,使得监测数据存在相似的变化规律,以此可为测点的漂移校正提供判别依据。依据上述原理,提出基于相似测点的密度聚类分析,并运用DBSCAN算法判定测点漂移区间和漂移量;同时为克服校正突变和类簇粘连等问题,引入滑动窗口模式建立监测量漂移的动态校正模型,其校正过程分为窗口内数据校正与窗口滑动校正两个部分。工程实例表明:该方法具有较强的适用性及较高精度,为结构中存在相似测点的数据漂移问题提供了新的自动校正思路。

Due to the system fault or external disturbance,the drift phenomenon of monitoring data often occurs.As for hydraulic structures,because of the structure similarity and the correlation of measuring points arrangement,the monitoring data often have significant spatial correlation making them possess similar variation rules,so as to provide a discriminant basis for the drift correction of the measuring points.According to the above principle,the density clustering analysis based on similar measuring points was proposed,and the DBSCAN algorithm was used to determine the drift interval and drift amount of measuring points.At the same time,in order to overcome the problems of correction mutation and cluster adhesion,the sliding window mode was introduced to establish a dynamic correction model for monitoring data drift.The correction process was divided into two parts,data correction within window and window sliding correction.The engineering example showed that this method has strong applicability and high precision,which provides a new automatic correction idea for the data drift problem of similar measuring points in hydraulic structures.