•  
  •  
 

Abstract

Dam displacement is a crucial indicator for assessing the safety of a concrete dam through structural health monitoring. Since the displacement data exhibits a non-linear and complex relationship with influencing factors like waterhead, time and temperature, machine learning models are deployed to accurately predict dam displacement. Furthermore, the limited availability of monitored data in the majority of the dams renders the studies conducted with a large number of observations valueless. In order to address the aforementioned issues, this study proposes a feature selection approach to predict dam displacement by examining the ability of four ensemble machine learning models on different input combinations. The results reveal that the extreme gradient boost performs the best with a coefficient of determination (R2) and RMSE value of 0.960 and 0.275 mm respectively. Random forests and decision tree models exhibited better performance on using single predictor variables waterhead and age respectively. AdaBoost exhibited moderate performance but was unaffected by the negative influence of extra predictor variables. The comparison results indicated that models developed with only ambient air temperature as input data are insufficient to predict dam deformation. The outcomes of this study are resourceful in prioritizing models based on the data availability for accurate prediction of dam displacement.

Download

Included in

Engineering Commons

Share

COinS