Accurate prediction of seawall deformation is essential for further structural health assessment and early warning of potential instability. However, inclinometer monitored deformation data usually exhibit strong depth-dependent heterogeneity, nonlinear temporal evolution, and spatiotemporal coupling, which make conventional single-sequence prediction methods insufficient. To address this issue, this study proposes a spatiotemporal prediction method for seawall inclinometer-monitored deformation data based on Dynamic Time Warping (DTW) clustering and an improved seahorse optimization-based Transformer (ISHO-Transformer). First, considering that deformation sequences at different depths may present similar deformation trends with nonlinear temporal misalignment, DTW is employed to measure the similarity between depth-wise monitoring sequences, and hierarchical clustering is introduced to classify depths with similar deformation patterns. Subsequently, for each clustered depth group, a Transformer-based prediction model is constructed to characterize the coupled evolution of monitoring location, depth, and time. To further enhance model performance and reduce the uncertainty of manual parameter tuning, the improved seahorse optimization (ISHO) algorithm is used to adaptively optimize key Transformer hyperparameters. Ultimately, the proposed method is validated using measured seawall inclinometer data and compared with several benchmark models. The results demonstrate that the proposed framework can effectively improve prediction accuracy, providing a useful tool for seawall deformation analysis and safety monitoring.
Ding et al. (Thu,) studied this question.