基于KICA-RVM的大坝缺失监测数据插值方法

An interpolation method based on KICA-RVM for missing monitoring data of dam

监测数据的缺失会给大坝的健康监控带来困难,因此,需要采用科学合理的插补方法对缺失数据进行插补,从而获取完整可靠的监测数据。根据测点的空间相关性,本文提出一种基于核独立分量分析(KICA)和关联向量机(RVM)相结合的大坝缺失监测数据插值方法。KICA-RVM采用KICA对原始自变量进行非线性变换并提取独立分量,形成特征自变量,通过特征值谱分析确定最佳特征变量个数,消除冗余信息干扰;采用RVM对KICA变换后的样本数据进行回归建模,从而获得强非线性表达能力且性能良好的模型。通过对某工程大坝监测资料进行分析,结果表明该插值方法具有良好的插值精度及稳定性。

The lack of monitoring data can bring difficulties to the dam＇s healthy monitoring , therefore, we need to present a scientific and rational interpolation method for interpolating missing data in order to gain complete and reliable monitoring data. A data interpolation method integrating Kernel Independent Component Analysis （KICA） with Relevance Vector Machine （RVM） for interpolating missing data of dam was proposed according to the spatial correlations of the measured datapoints. The KICA - RVM transformed the original independent variables nonlinearly by KICA and extracted the independent components to form the characteristic independent variable. The number of characteristic independent variables was determined by the Eigenvalue spectrum analysis and the disturbance of redundant information was eliminated. RVM was applied for regression modelling of the sampling data transofermed by using KICA, thus to obtain a model with strong nonlinear expression ability and good performance. Through analysis of the dam monitoring data of a project, the results showed that the interpolation method had a good preci- sion and statability.