Monitoring model group of seepage behavior of earth-rock dam based on the mutual information and support vector machine algorithms

Abstract

The establishment of a high-precision piezometric water level monitoring model ensures the safe operation of earth-rock dams. The hysteresis effect of the upstream water level and rainfall should be considered during modeling. In the traditional method, the average factors are used to express this effect, and linear regression modeling is adopted. These factors reduce the accuracy of the model. In this paper, the mutual information (MI) and support vector machine (SVM) algorithms are proposed. MI has a powerful correlation analysis capability, and it is innovatively used to address hysteresis effects. SVM has a strong nonlinear modeling ability, and it is used as a modeling algorithm. During this study, it was found that the lag time of rainfall varied. In view of this characteristic, the concept of an innovative model group, which is an important extension of the traditional single model, is proposed. In the example, the mean square error (MSE) is used as the precision index. Compared with the traditional single model established by linear regression, the MSE of the MI–SVM model group can be reduced by approximately 60.98%–68.75%. Compared with the model group established by linear regression, the MSE of the MI–SVM model group can be reduced by approximately 41.28%–45.45%. The new method effectively improves the accuracy of the model and can precisely monitor the seepage state of the dam. Moreover, it is beneficial for improving the level of dam safety management and can be extended to other fields involving hysteresis effects and nonlinear modeling.