Multiple-Points Dam Deformation Modeling and Prediction Based on Extreme Learning Machine

Abstract

The multiple-points dam deformation model establishes vital connections among monitoring points on the dam, which are crucial for assessing the overall behavior of the dam. This study examines the statistical process of the multiple-point dam deformation model and introduces three methods: extreme learning machine, backpropagation neural networks, and stepwise regression for modeling multiple-point dam deformation. In the case study of Wuqiangxi Dam, the performance of the three multiple point models is analyzed and evaluated. Multiple-points dam deformation models are shown to be effective in modeling and predicting the displacements of all monitoring points on one tension wire at the same time, though the performance of the models would present different results at different monitoring points. All three models yielded satisfactory results, with average fitting residual RMS values of 1.12 mm, 0.91 mm, and 1.15 mm, and average prediction residual RMS values of 1.68 mm, 1.76 mm, and 1.61 mm, respectively. The fitting and prediction results indicate the feasibility of multiple-points dam deformation model. In comparison to the other two methods, the extreme learning machine-based multiple-point model, serving as a single hidden-layer feedforward neural network, demonstrates the advantages of simplicity, flexibility, and efficiency.