Abstract: The systematic signal and random signal in the monitoring sequence are difficult to distinguish in the conventional monitoring models of the dam, thus the forecasting accuracy of the conventional model can be promoted. In this paper, we separate the systematic signal from random signal by their frequency features based on wavelet decomposition. According to the advantages of managing signals of stepwise regression and Support Vector Machine(SVM), in association with grid search and cross validation methods for determining the sensitive parameters of SVM, we present a monitoring model of dam displacement based on multivariate statistical combined with wavelet decomposition and support vector machine. Then the calculating procedures are compiled. The engineering examples indicate that both the systematic signal and random signal can be separated effectively in the composite model, with high forecasting accuracy and good optimization ability. Finally, the composite model is effective and the method can be applied to high slope monitoring and other warning indicators of dam projects.

Key words: dam displacement, wavelet decomposition, parameter optimization, support vector machine, monitoring model