基于SVM-MCD的大坝变形监测数据异常值判定

Abnormal values determination of concrete dam deformation monitoring data based on SVM-MCD

变形是最能直观反映混凝土坝运行性能的宏观效应量。考虑到监测数据异常值对监控模型精度与大坝安全性态诊断的不利影响,提出了一种基于支持向量机的混凝土坝变形监测数据异常值判定方法。在分析混凝土坝原型变形监测数据显性异常点的基础上,通过构建基于支持向量机理论(SVM)的高精度计算体系,充分利用结果效应量与驱动环境量之间的映射交互关系,结合最小协方差矩阵(MCD)稳健估计理论对残差序列进行异常值判定,而后将其异常确定值进行有机性替代处理,解决了内蕴复杂环境干预的监测信息挖掘难点。工程实例分析表明:所建判定系统的精准性及泛化能力均得以提升,相比于传统方法具有较好的实用性和鲁棒性,能有效地避免变形监测数据预处理中的误判漏判等困扰。此外,所提出的判定方法经一定的优化和拓展,亦可推广应用于其他水工建筑物的数据异常值判定分析。

Deformation is the macroscopic effect that can most directly reflect the operation performance of concrete dams.Considering the adverse effects of abnormal monitoring data on the accuracy of monitoring model and diagnosis of dam safety state,this paper proposed a method for determining abnormal values of concrete dam deformation monitoring data based on support vector machine and minimum covariance matrix theory.Firstly based on the analysis of the dominant abnormal points in the deformation monitoring data,a high-precision calculation system based on the support vector machine theory(SVM)was constructed,and the mapping interaction between the result effect and the driving environment was fully utilized.Then combined with the minimum covariance matrix(MCD)robust estimation theory,the abnormal values of the residual sequence were determined,and then the abnormal determined values were organically substituted,so as to solve the difficulty of monitoring information mining in the intervention of complex environment.The analysis of engineering examples showed that the accuracy and generalization ability of the judgment system constructed in this paper were improved.Compared with the traditional method,it has better practicability and robustness,and can effectively avoid the misjudgment and omission in the preprocessing of deformation monitoring data.In addition,the proposed method can also be applied to the determination and analysis of abnormal data of other hydraulic structures after certain optimization and expansion.