To enhance the accuracy of dam displacement prediction, this paper proposes a hybrid model combining Random Forest (RF), a Convolutional Neural Network (CNN), and a Residual Attention Informer (RA-Informer). Firstly, RF is utilized to assess the importance of input features, selecting key factors that significantly influence dam displacement. Then, CNN is employed to perform deep feature extraction on the input data, mining effective information. Subsequently, the Informer model integrated with a residual attention mechanism establishes the mapping relationship between the extracted features and dam displacement, enhancing the focus on critical features. Finally, the Northern Goshawk Optimization (NGO) algorithm is adopted to optimize the model’s hyperparameters. Experimental results on actual engineering data demonstrate that the proposed model exhibits superior prediction accuracy and stability compared to other typical models, offering higher precision and reliability.

In view of the problem that traditional dam outlier identification methods mostly rely on single-monitoring-point models and do not fully consider the spatio-temporal correlation characteristics of deformation between monitoring points, which can easily lead to the misdiagnosis of outliers, this paper proposes a novel Ward-VMD-BiLSTM-Iforest method for identifying gross errors in dam deformation monitoring. By integrating spatio-temporal clustering, variational mode decomposition (VMD), and BiLSTM neural networks, the method effectively identifies outliers while avoiding the misclassification of data mutations caused by environmental changes. Compared to traditional models (GRU, LSTM, and BiLSTM), the HHO-BiLSTM model demonstrates superior performance, achieving an R² of 0.97775 at TCN08, with a reduced MAE and better accuracy. In comparison with the Raida and Romanovsky criteria, the proposed method achieves 100% precision and 100% recall, significantly improving detection accuracy and reducing misjudgment. This method provides an effective and reliable solution for dam deformation outlier detection.

Monitoring data outliers comprises isolated mode outliers, cluster mode outliers, and normal points. To identify and distinguish the data hopping problems caused by outliers and environmental mutations in the displacement monitoring data of concrete AMS, this paper proposes a method based on wavelet transform, DBSCAN clustering algorithm combined with isolated forest and reinforcement learning algorithm to identify outliers in concrete dam monitoring data. In this paper, the trend line of measuring point data are extracted by the wavelet transform algorithm, and the residual data are obtained by subtracting it from the original process line. Subsequently, the DBSCAN clustering algorithm divides the residual data according to density. Therewith, the outlier scores of different data clusters are calculated, the iterative Q values are updated, and the threshold values are set. The data exceeding the threshold are finally marked as outliers. Finally, the water level and displacement data were compared by drawing the trend to ensure that the water level change did not cause the final identified concrete dam displacement data outliers. The results of the example analysis show that compared with the other two outlier detection methods (“Wavelet transform combined with DBSCAN clustering” or “W-D method”, “Wavelet transform combined with isolated forest method” or “W-IF method”). The method has the lowest error rate and the highest precision rate, recall rate, and F1 score. The error rate, precision rate, recall rate, and F1 score were 0.0036, 0.870, 1.000, and 0.931, respectively. This method can effectively identify data jumps caused by an environmental mutation in deformation monitoring data, significantly improve the accuracy of outlier identification, reduce the misjudgement rate of outliers, and have the highest detection accuracy.

To improve the accuracy of deformation prediction in concrete buildings and large-scale infrastructures such as dams, this study proposes an Attention-BiLSTM model integrated with a parameter-optimized Variational Mode Decomposition (VMD). Specifically, the Pelican Optimization Algorithm (POA) is employed to optimize VMD parameters, enhancing signal decomposition efficiency for structural deformation time series. The optimized VMD is then coupled with a BiLSTM network embedded with an attention mechanism, forming a hybrid prediction framework that captures both temporal dependencies and key feature weights in monitoring data. Using three sets of engineering-measured deformation datasets, the proposed model is validated through comparative analyses with conventional single models (e.g., standalone BiLSTM and VMD-BiLSTM without attention). Results demonstrate that the developed model achieves superior accuracy and stability, significantly outperforming all comparative methods, with the highest R² reaching 0.996, while reducing MAE and RMSE by over 60% and 30%, respectively. Quantitative evaluation indicators (e.g., RMSE, MAE, and R²) confirm that the approach effectively captures both short-term fluctuations and long-term trends of structural deformation. These findings verify its reliability and applicability for intelligent safety monitoring of concrete buildings and infrastructures.