Measured displacement data-driven efficient interpretation and real-time risk assessment method for the service performance of arch dams with cracks

Current methods for predicting arch dam displacement rarely consider the impact of cracks on displacement and the interpretability of factors, nor do they reasonably assess the risk probability of arch dam. To address these issues, first, clustering partitions are conducted based on the Ward criterion, and the comprehensive displacement is obtained through the Criteria Importance Through Intercriteria Correlation (CRITIC) method. Secondly, considering the impact of cracks, a displacement monitoring model HSCT is constructed, and feature selection for the HSCT model factors is performed using the Max-Relevance and Min-Redundancy (mRMR), while Kernel Principal Component Analysis (KPCA) is utilized for feature extraction of crack factors. Furthermore, to enhance interpretability, an attention mechanism is incorporated into the Convolutional Neural Network with Long Short-Term Memory (CNN-LSTM) model, establishing a CNN-LSTM-Attention model to predict comprehensive displacement and visualize the importance of influencing factors. Finally, Kernel Density Estimation (KDE) is applied to the residuals of the comprehensive displacement, and a multivariate Copula function is used to construct the joint distribution to calculate the overall risk rate. The results indicate that the proposed methods and models are reasonable and feasible, providing scientific basis and technical support for the health diagnosis of hydraulic structures.