A dual-objective displacement prediction model for concrete arch dams with cracks based on signal decomposition and feature selection

Arch dam displacement is an important indicator for assessing the safety of these structures. However, existing displacement prediction models often overlook the spatial correlation of displacement and the influence of cracks. Therefore, this study proposes a dual-objective displacement prediction method for arch dams with cracks, integrating signal decomposition and feature selection. An initial displacement factor set is established considering temperature hysteresis and cracks, with the primary feature set identified utilizing the sequential backward selection-random forest (SBS-RF) algorithm. The primary feature set is then decomposed with the application of whale optimization algorithm-variational mode decomposition, and intrinsic mode functions (IMFs) are selected via SBS-RF to form the IMF component set. These are combined with displacement components obtained through seasonal-trend decomposition using loess to construct a prediction dataset. Using mean squared error and displacement shape similarity index as optimization objectives, a dual-objective displacement prediction model is developed based on temporal convolutional networks and multi-kernel relevance vector machine. Finally, residual correction is applied to refine the model, yielding the corrected model. Engineering case studies demonstrate that the dataset, processed through decomposition and feature selection, significantly enhances the predictive accuracy of the model. The corrected model strengthens the spatial correlation at monitoring points and improves prediction accuracy. This study introduces a novel approach to displacement prediction for arch dams with cracks, offering substantial practical value for dam structural health monitoring and safety management. It also serves as an important reference for research and applications in related fields.