Machine Learning (ML)-driven approaches have been employed to replace computationally intensive seismic simulations of hydraulic engineering structures. For the complex seismic responses of arch dams, constructing a metamodel that captures the nonlinear relationship between ground motion inputs and structural response outputs using a limited set of numerical simulations can significantly reduce the computational cost. However, conventional deterministic predictions and fragility analyses fail to account for the high aleatory and epistemic uncertainties inherent in the seismic response of arch dams. To this end, this paper proposes an efficient fragility analysis method for arch dams that integrates probabilistic ML algorithms with the traditional Incremental Dynamic Analysis. By constructing a Natural Gradient Boosting (NGBoost) metamodel for the arch dam dynamic response, not only can the predicted mean value of each response sample be obtained, but also its conditional probability distribution. Superimpose the simulation data with the response distribution predicted by NGBoost, and the binary parameters of the fragility function are estimated, thereby generating both the fragility curve and the uncertain fragility interval of arch dam. Additionally, representative Ground Motion Records (GMRs) for the arch dam are selected using the Partitioning Around Medoids (PAM) unsupervised clustering technique, determining the minimum subset proportion that effectively represents the whole GMR dataset. The effectiveness of the proposed method is validated in a super-high arch dam. The 40% GMR proportion is found to adequately reproduce the fragility curves of the whole dataset, with the reference curve falling within the derived uncertainty interval, achieving a 56.8% reduction in computational cost. The 60% GMR proportion ensured fragility curves with balanced accuracy and effectiveness, exhibiting maximum mean differences of 0.058 and maximum standard deviation differences of 0.031 from reference curves across all damage levels, while reducing computational cost by 39.7%. Comparative results demonstrate the superiority of NGBoost and PAM over existing deterministic metamodels and GMRs selection techniques, respectively. The efficient fragility analysis method proposed in this study ultimately enables the direct characterization of uncertainties in arch dam seismic responses.

Accurate dam displacement prediction is vital for optimizing maintenance and ensuring structural safety. Nevertheless, current models often struggle to effectively capture the complex relationships between structural responses and environmental variables, alongside the interactions between temporal dynamics and multivariate data, resulting in suboptimal predictive accuracy. Therefore, we propose a dual-branch interactive fusion network (DBIFN) for dam displacement prediction using parallel temporal representation and gated cross-attention. The dual-branch architecture, which parallelly integrates the enhanced Transformer (eTransformer) and long short-term memory (LSTM), is designed to optimize feature extraction and interaction modeling across multiple dimensions. Specifically, eTransformer is dedicated to extracting features from targeted displacement sequences, while LSTM effectively processes auxiliary environmental dynamics, enabling a comprehensive analysis of underlying patterns within monitoring data. To fully fuse the interpreted temporal features from dual-branch outputs, we introduce a new cross-attention module to utilize the multi-dimensional gated attention unit to efficiently encode them into semantic representations, followed by a Kolmogorov-Arnold network mapping for further representation enhancement. The effectiveness of the proposed model is validated using real-world monitoring datasets collected from a concrete dam project, with experiments conducted across multiple monitoring points. Results demonstrate that DBIFN achieves superior prediction accuracy compared to both single-branch and conventional baseline models. Across all monitoring points, the proposed model can effectively capture temporal variations, attaining an average coefficient of determination of over 0.95 on the test set and outperforming comparative models in most metrics. Furthermore, statistical significance testing confirms the reliability and reproducibility of the results, while computational efficiency is maintained within inference time constraints. These findings offer valuable insights into the practical application of DBIFN-based monitoring models and support informed decision-making.