异常形变检测方法探讨

Method for Detecting Anomalous Deformation

在讨论以往方法的基础上 ,将拟准检定法拓展用于发现异常形变。该方法直接以真误差为研究对象 ,避免了其他方法以最小二乘残差及其函数为研究对象的缺点。最后通过一个算例 。

Anomalous deformation detection is an important step for the whole block movement analysis because the moving trend of the whole block not have been got if the anomalous deformation could not be eliminated.Anomalous deformation detection has some similarities to the gross error detection,so it was previously implemented by the method of detecting gross errors,e.g.hypothesis test method or robust estimation method.Quasi_accurate method is a new approach to de_tect the gross errors,so this paper extends it to detect anomalous deformations. The new method takes the real errors as the study object so it can avoid the defects of taking least square residuals and their function as the study object.The method is that the real errors are calculated directly by adding the condition that the norm of quasi_accurate observations is minimum.The key of it is how to select the quasi_accurate observations (QAOs) correctly.A successful scheme is that the QAOs are chosen through preliminary and fine choices because we do not know which observations have the anomalous deformations in advance.First,the observations are classified into four classes,'0 Sort','1 Sort','2 Sort','3 Sort',according to a certain criterion.The QAOs are selected finely according to the magnitudes of the estimators of real errors.The fine selection ends till the QAOs do not vary.The observations corresponding to the greater magnitudes of the estimators of real errors are the anomalous observations.