Segmented sequence decomposition-Informer model for deformation of arch dams

Abstract

Deformation serves as a key index to characterize the operational condition of dams. However, the prediction accuracy of deformation in dams remains limited due to the influence of multiple factors. Accordingly, this study innovatively combines the Informer with the segmented sequence decomposition and proposes a segmented sequence decomposition-Informer model (SD-Informer) for the deformation prediction of arch dams, which significantly improves the prediction accuracy and stability. The segmented sequence decomposition divides the predicted time series into annual segments and decomposes them in a segment-by-segment manner, thereby minimizing the reduction of prediction accuracy over long sequences and the boundary effects in decomposition. In addition, the Informer extracts macro- and micro-level information from deformation sequences using a multi-head attention mechanism, which significantly improves the prediction accuracy. LYX arch dam and XW arch dam, which have been in operation for more than 20 years, are taken as case studies. The results show that the performance of the SD-Informer surpasses that of wavelet neural networks, long short-term networks, and Informer, demonstrating that the SD-Informer is an accurate, robust, practical deformation prediction of arch dams for engineering applications.