A deep learning method for predicting the displacement of concrete arch dams considering the effect of cracks

Displacement represents one of the most palpable effects of dam's structural response to both internal and external influences and the monitoring of concrete arch dam displacements traditionally relied upon statistical methods, characterized by limited precision, which paid scant attention to the effect of cracks. In an endeavor to holistically consider the effect of cracks on displacements, a monitoring model, denoted as HSCT, has been devised, which integrates crack opening displacements as factors influencing displacement, establishing an initial set of influential factors. Subsequently, by employing the methods of the Max-Relevance and Min-Redundancy (mRMR) algorithm in conjunction with the Least absolute shrinkage and selection operator (Lasso), factors with the utmost predictive capacity to displacements in concrete arch dams with cracks can be identified, approaching excise detrimental features while preserving the accuracy of the model and the results are compared with the outcomes of Stepwise regression and Pearson correlation criteria for factor selection. Ultimately, a deep learning model Convolutional Neural Network with Long Short-Term Memory (CNN-LSTM) for displacement prediction is formulated. This model is applied to the filtered dataset for predictive analysis, which contrasted with Multivariate Linear Regression (MLR) as well as machine learning models (PSO-RVM and SSA-ELM). As such, it forms a deep learning method for the prediction of concrete arch dam displacements, accounting for the effect of cracks. Engineering instance substantiates the necessity of considering the influence of cracks when establishing a displacement monitoring model for concrete arch dams with cracks. Methods of mRMR and Lasso for feature selection and factor screening prove to be efficient, as they are capable of identifying a majority of crucial variables in a single computation, ensuring both accuracy and computational efficiency. CNN-LSTM model for dam displacement exhibits commendable predictive performance, adequately addressing the longterm dependencies of monitoring values while capturing short-term local characteristics. In this study, the deep learning prediction method contributes as a scientific foundation and technical support for the secure monitoring and operational management of concrete dams with cracks, introducing novel perspectives into the realm of safety monitoring and health service diagnostics for hydraulic structures with cracks.