A multi-target interval prediction method for arch dam displacement based on SE-MPRN and corrected ensemble of regressor chains

For concrete arch dams, the displacement at different monitoring points exhibits a high degree of correlation and uncertainty. In order to improve the accuracy of displacement prediction by using displacement correlation, this study first employs adaptive weighted derivative dynamic time warping and clustering algorithms to determine the regional segmentation of the arch dam. Furthermore, the residual network (ResNet) was enhanced by incorporating convolution kernels of varying sizes and introducing squeeze and excitation (SE) blocks, resulting in the development of multi-path ResNet with SE block (SE-MPRN). Next, the maximal information coefficient was employed to assess the displacement correlation of monitoring points within the region, followed by adjustments to the regressor chain (RC), the corrected ensemble of RCs (CERC) was proposed. By integrating CERC with SE-MPRN, the multi-target (MT) displacement prediction method for arch dams, SE-MPRN CERC , was established. Finally, to quantify the uncertainty in arch dam displacement prediction, SE-MPRN CERC was combined with upper and lower bound estimation (LUBE) to establish a MT displacement interval prediction method. The results of the engineering case study demonstrate that SE-MPRN exhibits exceptional ability in learning the complex relationship between influencing factors and displacement. Compared to other deep learning-based MT prediction methods, the proposed MT method, SE-MPRN CERC , significantly enhances the accuracy of arch dam point value predictions. The integration of SE-MPRN CERC with LUBE for MT displacement interval prediction greatly improves the quality of the prediction intervals, providing a new reliable technological support for arch dam safety monitoring.