Bayesian optimization and ensemble learning algorithm combined method for deformation prediction of concrete dam

Abstract

Concrete dams, as a significant infrastructure project, necessitate a prioritization of operational safety. Deformation is an essential element in the safety monitoring of concrete dams. The development of monitoring models aimed at predicting the trend of dam deformation is the fundamental tool for ensuring dam safety. The accuracy of the monitoring model is paramount in guaranteeing the reliability of subsequent safety assessments. Consequently, this paper proposes a novel combined prediction model for concrete dams, based on Bayesian Optimization (BO) and Random Forests (RF), to achieve high-precision prediction of deformation. Initially, the monitoring data were pre-processed, and the input parameters for the RF model were established. Subsequently, BO based on Gaussian processes optimizes various hyperparameters of the RF models and determines the optimal parameters. Finally, based on the Gini coefficient in this study, the significance of the output characteristic explained the relationship between the dam deformation and its influencing factors. Example analysis and model validation results evinced that the proposed model's prediction accuracy, in the test set, is superior to that of other benchmark models, and the residuals' dispersion is negligible. The proposed combined model has significant engineering implications and provides a novel method for concrete dam safety monitoring.