Aiming at the insufficient adaptability of static concrete dam deformation prediction frameworks under extreme working conditions and to enhance the prediction reliability of deformation monitoring data under various fluctuation scenarios in concrete dam safety monitoring systems, this study proposes a Short-term online prediction method of concrete dam deformation based on concept drift monitoring and transfer learning optimization strategy. In view of the gradual change of data distribution driven by factors such as extreme working conditions or material performance degradation, a concept drift detection method based on sliding window and Kolmogorov-Smirnov test is constructed. According to the test results, the correlation analysis is used to divide the historical deformation data of the same kind of measuring points before drift into two categories: strong correlation and weak correlation. The historical data of the source domain measuring points are used to train the composite model containing the error compensation module, and then the step transfer strategy is used to realize the prediction modeling. Finally, a real-time prediction method of deformation data based on TCN-LSTM and step transfer strategy is formed, which solves the problem of insufficient historical data and difficult feature migration when the performance of the prediction model is improved. Engineering examples show that, the concept drift detection method can monitor the data distribution changes in the deformation sequence in real time. The step transfer strategy proposed in this study can be applied to migration learning together with the error compensation module, and effectively improve the prediction accuracy of short-term deformation data. Compared with direct modeling, the prediction accuracy is significantly improved. Therefore, the proposed method is an effective deformation prediction modeling method for concrete dams under extreme conditions, and also provides a new technical approach for intelligent construction and safety monitoring of dams.

Traditional safety monitoring models can only reflect the isolated deformation patterns at fixed points, typically neglecting the spatial coordination in high concrete dams. As a result, the considerable discrepancies in deformation prediction and interpretation across different points increase the risk of inaccurate prediction and conflicting early-warning. To address this, based on Shapley Additive Explanations Coupled Light Gradient Boosting Machine (SHAP-LightGBM), this paper proposes a joint modeling method for multi-point deformation monitoring to mitigate spatial prediction inconsistency in high concrete dams. Based on analyzing the deformation mechanism and coordination, a multi-point hybrid deformation monitoring model is constructed by introducing spatial coordinates as explanatory factors. An interpretable joint deformation modeling method based on SHAP-LightGBM is then proposed to overcome the high-dimensional non-linear fitting challenges induced by factor multiplication. This approach learns the collaborative multi-point deformation response mechanism, thereby alleviating the spatial prediction inconsistency of high concrete dams. Meanwhile, the spatial coordination is evaluated through both multi-point prediction accuracy and SHAP-based deformation interpretation. Engineering applications demonstrate that the proposed method yields consistently balanced accuracy across all points, and the predictions closely align with the measured deformations. Specifically, at low-quality points, it successfully overcomes prediction distortions inherent in traditional models, reducing the Mean Squared Error on the fitting and prediction sets by approximately 94.5% and 49.0%, respectively. Furthermore, the method achieves reasonable deformation component decoupling, eliminating the risk of conflicting early-warning caused by contradictory interpretations and thereby mitigating the spatial prediction inconsistency of high concrete dams. With further improvement and extension, the proposed model can provide technical support for resolving early-warning conflicts in high dam safety monitoring and facilitating the integrated management of spatial multi-point deformations.

Traditional statistical models for arch dam deformation monitoring consider only functional equivalence between independent variables and deformation components (hydrostatic pressure, temperature, and time effects). However, they neglect the structural characteristic of compatible deformation between horizontal arch rings and cantilever beams under complex loading conditions. This study analyzes deformation mechanisms of arch dams under reservoir hydrostatic pressure and thermal loads based on arch-beam load distribution principles. By examining load sharing and corresponding deformations of arch rings and cantilever beams within the dam system, we derive radial deformation expressions from the cantilever perspective, accounting for beam and foundation deformations. This establishes a mathematical formulation for radial deformation under hydrostatic pressure. Temperature loading is decomposed through the dam thickness into uniform temperature, equivalent linear temperature gradient, and nonlinear temperature gradient. Neglecting the nonlinear gradient (which primarily affects local surface deformation and stress), we develop separate radial deformation expressions for uniform temperature and linear temperature gradient. This yields a functional relationship for the temperature component of arch dam deformation. Building on these foundations, we construct a novel statistical model for analyzing concrete arch dam deformation behavior. Its validity and scientific rigor are demonstrated through engineering case studies.

Underwater crack detection of concrete dams is commonly hindered by limited generalization, high rates of missed/false detections, which arise from constraints in the scale and quality of training data. To overcome these challenges, a hybrid framework that combines CycleGAN-based data enhancement with an optimized YOLO11 detector is proposed in this study. A cross-domain style-transfer model based on CycleGAN was first developed to synthesize degraded underwater images from clear above-water crack samples. Subsequently, the YOLO11 network was improved by incorporating a cross-attention feature modulation (CAFM) module and adopting the Focal-EIoU loss function, which collectively strengthen texture perception and localization accuracy under complex underwater conditions. Validation on experiments of underwater concrete structures demonstrates that the proposed method achieves 91.8 % precision, 80.3 % recall, and a significantly improved mAP@0.5, effectively mitigating missed/false detections even in low-light and low-contrast scenarios. The proposed approach provides promise for broader applications in small-object and low-quality imaging detection.

To address the challenges of detecting apparent structural defects like cracks in large-scale embankment, an unmanned aerial vehicle (UAV) aerial image-based apparent defect detection method is proposed using UAV aerial photography and computer vision technologies. Firstly, Cycle-Consistent Adversarial Networks (CycleGAN) is employed to generate virtual apparent defect images under different lighting conditions using the limited aerial images via data augmentation, and then an image sample dataset of apparent structural defects is established. Subsequently, Convolutional Block Attention Module (CBAM) is utilized to enhance the feature extraction capability of You Only Look Once version 8 (YOLOv8) network, and Scylla-Intersection over Union (SIoU) is employed as the localization loss function to further improve the training efficiency and detection accuracy. Thereby, an improved YOLOv8 network-based intelligent detection model is established for identifying the apparent defects in embankment. Furthermore, Slice-aided Hyper Inference (SAHI) module is inserted into the improved YOLOv8 network to promote the small target detection capacity from long-distance high-resolution UAV images during inference period, and accordingly, an integrated framework of improved YOLOv8 and SAHI is proposed for apparent defect detection of embankment. Engineering example shows that, compared with YOLOv8 network, the improved YOLOv8 network exhibits higher detection accuracy and lower missed detection rate. Meanwhile, the image sample dataset augmented by virtual images is beneficial for improving the training ability, and the utilization of SAHI significantly promotes the inference performance of the intelligent detection model. A high-precision intelligent apparent defect detection approach is provided for safety management and danger inspection of embankment engineering.

Digital twin-enabled safety monitoring of concrete arch dams requires deformation prediction models that are continuously updateable, physically informed, and uncertainty-aware. Nevertheless, existing approaches often struggle to effectively integrate simulation knowledge with monitoring data, especially under the dual constraints of limited response time and previously unseen real operating conditions. To overcome these challenges, this study proposes a physics-guided multi-fidelity Bayesian neural network (PGMF-BNN) framework for digital twin-oriented probabilistic deformation prediction. Low-fidelity data are generated from a three-dimensional finite element model to represent the dominant deformation behavior under hydrostatic loading initially. A multi-input multi-output (MIMO) architecture is constructed to capture spatial dependencies among monitoring points and enable collaborative prediction. The model is subsequently pre-trained using low-fidelity data to learn a mechanically consistent response skeleton, followed by a two-stage transfer learning process using prototype monitoring data to achieve virtual-physical alignment. Finally, Bayesian inference is incorporated to quantify epistemic and aleatoric uncertainties and to produce calibrated prediction intervals. The effectiveness of the proposed framework is demonstrated using a real concrete arch dam case. The results indicate that the proposed method achieves superior predictive performance, with coefficients of determination exceeding 0.90 at five of the six representative monitoring points (reaching 0.975 at PLA2) and mean absolute percentage errors below 7%. The model successfully reconstructs the spatial correlation structure of multi-point deformation and demonstrates robust performance under limited-data conditions and moderate noise levels (0%-10%). These findings confirm that the proposed framework delivers accurate, physically consistent, and uncertainty-aware deformation predictions under both normal and disturbed operating conditions, providing a reliable and extensible solution for digital twin-based dam safety monitoring and risk early warning.

Aiming at the problems of weak interpretation ability, low precision and low resolution of inversion results of traditional geophysical prospecting methods for hidden safety hazards of earth-rock dams, a method for identifying internal defects of earth-rock dams based on multi-source data information interpretation of geophysical prospecting is proposed. In order to explore the distribution attributes of internal defects in earth-rock dams, the implementation process of internal defect identification of earth-rock dams based on the fusion of ground penetrating radar (GPR) and electrical resistivity tomography (ERT) geophysical information is constructed. Through the preprocessing, spatial registration and information fusion of the original geophysical data from the same source and different sources, and combined with the prior knowledge of the formation medium of the earth-rock dam, the imaging interpretation method of the internal defects of the dam based on different information fusion strategies is established. The method realizes the accurate identification of the full-chain structural defects of the hidden danger detection information of the earth-rock dam from data processing fusion to inversion imaging. Combined with the actual engineering exploration research, it is shown that the proposed fusion theory and method can effectively identify the hidden defects inside the dam. Compared with the traditional single geophysical prospecting method, the inversion results of this method have stronger interpretation ability, which can effectively solve the problems of misidentification and omission of traditional methods, and the degree of identification and visualization are significantly enhanced. It can provide method support for accurate diagnosis of hidden defects in earth-rock dams. In addition, after some improvement and expansion, the method can also be applied to the detection and analysis of hidden dangers in other earth-rock structure buildings.

To address the high-dimensional multi-objective optimization challenge arising from the dependence of concrete fracture phase-field model parameters on empirical estimation or trial-and-error calibration,this study con-ducted a quantitative analysis of four solution parameters,including time-step,convergence tolerance,mesh size,and the ratio of mesh size to length scale.Furthermore,a piecewise multi-objective optimization framework was proposed toward the full-curve mechanical response of concrete,enabling the systematic optimization of solution parameters while simultaneously balancing model accuracy and computational efficiency.The results show that these four param-eters all have statistically significant effects on solution efficiency,with mesh size exhibiting more pronounced gradi-ent variations in goodness-of-fit metrics across the elastic,plastic,and softening stages,thereby making it the key computational parameter governing both the accuracy and efficiency of full-curve fitting.Further verification demon-strates that the framework maintains the consistent convergence direction of the optimal solution across different opti-mization algorithms,indicating good applicability and robustness.Compared with schemes targeting only fitting accu-racy,the overall computational efficiency of the model can be improved by more than 60%after an auxiliary optimiza-tion objective for solution efficiency is introduced within an allowable 5%reduction in accuracy.

ObjectiveDetermining the physical and mechanical parameters of dike soils accurately is the prerequisite for objective safety assessment. However, in practical engineering, dikes often face the challenge of limited soil strength tests and sparse on-site monitoring data. This scarcity makes it difficult to accurately characterize the probabilistic distribution of sensitive soil parameters, leading to deviations in safety evaluations. To address the problem where traditional deterministic methods ignore parameter variability and probabilistic methods rely heavily on sufficient data, a hybrid parameter inversion framework is proposed. This study aims to accurately invert the permeability coefficients and shear strength parameters under limited data conditions by characterizing them as interval variables and random variables, respectively, thereby improving the reliability of dike safety assessment.MethodsA data-model driven inversion method for sensitive soil parameters was constructed based on the distinct data characteristics of permeability and shear strength. Firstly, regarding the permeability coefficient, due to the spatial sparsity and discontinuity of monitoring data (such as piezometric heads), it was characterized as an interval variable. The elevation of the seepage exit point on the backwater side, which reflects the macro-seepage behavior and is easily observable, was selected as the mapping index. A Gaussian Process Regression (GPR) surrogate model was established to map the non-linear relationship between the seepage exit point elevation and the permeability coefficient. The Latin Hypercube Sampling (LHS) method was employed to generate uniform samples within the prior interval. These samples were input into a Finite Element Method (FEM) model to calculate the corresponding seepage exit elevations, forming the training and verification datasets. The squared exponential kernel function was selected for the GPR model to predict the equivalent interval of permeability coefficients based on the measured exit point elevation range. Secondly, regarding the shear strength parameters (cohesion c and internal friction angle φ), they were characterized as random variables since they originate from standardized discrete sampling tests. A stochastic inversion model was built based on Bayesian Updating with Subset Simulation (BUS). The likelihood function was constructed based on the error between the simulated and measured structural responses. The parameter inversion problem was transformed into a structural reliability problem, where the posterior distribution samples were generated using the Markov Chain Monte Carlo (MCMC) simulation within the subset simulation framework. This approach effectively solves the Bayesian updating problem for small probability events under small sample conditions. Finally, taking a dike in the lower reaches of the Yellow River as a case study, a 2D finite element model (5844 elements, 6040 nodes) was constructed. The proposed method was applied to invert the parameters of three soil layers (upper body, lower body, and foundation). The inverted parameters were then used to calculate the Safety Factor (FS) and Failure Probability (Pf) under continuous high water levels, and the results were compared with those obtained from deterministic methods using field detection values.Results and Discussions The proposed method was validated through the engineering application, yielding specific inversion results and safety indicators. In the interval inversion of permeability coefficients, the GPR surrogate model achieved a composite correlation coefficient (R2) of 0.99, demonstrating high prediction accuracy. Based on the field measured seepage exit point elevation interval of [1.9,2.3] m, the equivalent permeability coefficient intervals for the three soil layers were inverted. The results showed that the permeability coefficient for the upper dike body ranged from 3.54×10-5cm/s to 6.35×10-5cm/s; the lower dike body ranged from 2.45×10-5cm/s to 5.73×10-5cm/s; and the dike foundation ranged from 2.74×10-6cm/s to 4.60×10-6cm/s. The deterministic calculation values fell within these intervals, verifying the rationality of the interval inversion. In the stochastic inversion of shear strength parameters, the posterior distributions obtained via BUS showed reduced uncertainty compared to the prior distributions. Specifically, the mean cohesion of the upper dike soil decreased from 8 kPa (prior) to 5.64 kPa (posterior), a reduction of approximately 30%, reflecting the impact of environmental factors and human activities. Conversely, the mean cohesion of the lower dike soil increased to 10.07 kPa, likely due to consolidation under the weight of the upper layer. The safety assessment results under different parameter combinations were analyzed. Using the field detection values (deterministic method), the critical Safety Factor (FSa) was 1.305, which satisfied the stability requirement (FS>1.30). However, when using the parameters inverted by the proposed hybrid method, the safety indicators changed significantly. Combining the BUS-updated shear strength parameters (τBUS) with the maximum value of the inverted permeability interval (kmax), the average critical Safety Factor (FSf) dropped to 1.192, and the Failure Probability (Pf) increased to 3.89%. Compared to the conservative result based on detection values, the stability safety decreased by 8.66%. This assessment aligns with the historical concentrated leakage events observed in this dike section, indicating that the traditional method might overestimate the safety margin. The proposed method revealed potential risks that were masked by the uncertainties in the parameters.ConclusionsA sensitivity soil parameter inversion method combining interval theory and Bayesian statistics is established to address the difficulty of parameter characterization in dike safety assessment. The GPR-based interval inversion effectively handles the spatially sparse data of permeability coefficients by utilizing the seepage exit point elevation, while the BUS-based stochastic inversion updates the probability distribution of shear strength parameters under small sample conditions. The engineering application demonstrates that the proposed method can accurately capture the spatial variability and time-varying characteristics of soil parameters. The safety assessment results derived from the inverted parameters are more consistent with the actual operation status and historical hidden dangers of the dike compared to traditional deterministic analysis. This hybrid framework provides a scientific basis for dike stability evaluation and risk management under data-scarce conditions.

Since the updating of deformation monitoring models for arch dams often lags behind the evolution of structural performance, it relies on the back-analysis of dam performance parameters. However, the large number of parameters influencing dam structural performance reduces the efficiency of parameter back-analysis. Therefore, this study employs the Morris method to assess the sensitivity of structural performance parameters to arch dam deformation, thereby identifying the key parameters for back-analysis. Additionally, to fully leverage the respective advantages of the RSM, Kriging, and RBF surrogate models, they are combined to equivalently back-analyze the physical and mechanical parameters of concrete arch dams.Given that arch dam deformation is influenced by the coupled effects of multiple factors such as water pressure, temperature, and time-dependent effects, and based on the obtained equivalent back-analyzed parameter values, a hybrid monitoring model for concrete arch dam deformation is established using an SVM-Transformer model. By leveraging the Transformer model’s strong capability to capture complex patterns, it can automatically extract deep and global temporal features. Meanwhile, the SVM model flexibly and efficiently handles complex nonlinear relationships among features to improve prediction accuracy. In this way, dynamic updates to the tracking and monitoring model of arch dam deformation behavior have been achieved.The verification results of the engineering example demonstrate that the proposed method exhibits good scientific soundness and effectiveness. Taking the PLA1 measurement point as an example, the prediction MAE of the SVM-Transformer model is 0.678, which is significantly lower than those of PSO-SVM (1.016) and SVM (1.219), confirming the superiority of this model in arch dam deformation prediction.

During long-term service, concrete arch dams exhibit significant spatially heterogeneous variations in their mechanical parameters due to complex environmental scenarios. These variations pose serious uncertainty challenges for structural safety assessment. To address the limitations of conventional inversion approaches, which fail to provide reliable uncertainty quantification under varying load conditions, this study proposes a probabilistic multi-zone parameter inversion framework based on a monitoring-simulation feedback mechanism. The core concept of the framework lies in the integration of physical constraints and high-fidelity surrogate modeling. First, global sensitivity analysis was conducted to identify the key inversion parameters, thereby reducing the problem dimensionality. An adaptive intelligent sampling strategy was subsequently introduced to efficiently construct a high-information-value sample set for surrogate model training. Based on this sample set, a Hierarchical Bayesian Neural Network (Hier-BNN) model was developed. By incorporating shared hyperpriors, the model imposes physical constraints, while the embedded Transformer-based Cross-Attention mechanism effectively captures complex spatial correlations and inter-zone coupling effects. Finally, the Markov Chain Monte Carlo (MCMC) algorithm was employed to perform posterior inference, yielding a complete characterization of the parameter distributions. The proposed framework was applied to an actual arch dam project. The results demonstrate that the framework achieves rapid and accurate inversion, and the feedback-based state identification is in close agreement with engineering observations. Compared with multiple benchmark models, the proposed method exhibits significant advantages in both prediction accuracy and the reliability of uncertainty quantification. This study provides a robust theoretical foundation and an efficient computational pathway for the structural performance identification of dam structures.

Ensuring the safe operation of concrete dams requires effective evaluation of their structural safety state. To address the issue, methods that can provide real-time assessments while ensuring interpretability of the results is required. In this work, a novel comprehensive evaluation framework is developed, integrating multiple monitoring data and numerical simulation technique. Specifically, a hierarchical indicator system and a precise grading classification method based on multi-dimensional monitoring data series reflecting the structural response are established initially. Then, the dynamic weighting strategy is proposed considering the relationship between the values of monitoring quantity and its importance in the process of the evaluation using grey relational analysis. Ultimately, with the combination of the enhanced set-pair analysis model and cloud theory, the inherent randomness and fuzziness in the procedure of evaluation is comprehensively handled. The systematically improved evaluation method is validated through a case study of a typical concrete gravity dam section in Mianhuatan hydropower station, demonstrating desirable evaluation performance and superior stability (LCR of 99.2%, 98.7% and 96.4% with 5%, 10% and 20% added noisy data) compared to the conventional methods. This research presents a practically reliable method for the real-time structural safety assessment of concrete dams, significantly advancing the robustness and precision of structural health evaluation.

In the intelligent monitoring system of concrete dam, there is a drift phenomenon of data uncertainty in the monitoring sequence, which still needs to be identified manually. The construction process of dam monitoring model depends on the stage of static intelligent model established manually. The spatial correlation of multi-point monitoring data is rarely considered, and a large amount of spatio-temporal information contained in prototype observation data needs to be further explored. Therefore, This study proposes a Dam multi-source heterogeneous monitoring data fusion and synchronization method based on time series analysis. Aiming at the massive monitoring data of concrete dam operation and maintenance management and control platform, the CEEMDAN-wavelet threshold joint denoising algorithm is used to filter the original monitoring data, and the mutation threshold is set and the absolute variation range of the adjacent difference of the time series is calculated to identify the different drift intervals of the time series. By averaging the relative intervals of the monitoring values of different drift intervals, a combined model based on SSA-CNN-GRU is established to obtain the difference between adjacent drift intervals. Finally, the same monitoring benchmark is automatically synchronized for each drift interval, the historical samples are reused, and the SSA-CNN-GRU model that fuses multi-source monitoring information is dynamically updated to obtain the fitting value of the corresponding drift interval. The results confirm that the proposed method can accurately simulate the dynamic evolution process of vertical displacement of concrete dam, and compared with the conventional monitoring model, its fitting accuracy is higher, which provides a new method and means for dam safety monitoring.

To address the challenges of mesh dependency, high computational cost, and low accuracy in Finite Element Method (FEM)-based internal damage identification for concrete dams, a multi-fidelity surrogate model is proposed synergistically calibrated by monitoring data and numerical simulations, aiming to enhance the accuracy and efficiency of internal damage identification in concrete dams. Firstly, the extended finite element method (XFEM) is employed to construct the simulation analysis model for concrete dams. The methodologies of displacement field reconstruction, boundary tracking of structural damage, and sub-element integral are provided. Secondly, the elastic modulus is taken as the key parameter to evaluate the structural damaged degree. Based on the inversion results of the comprehensive elastic modulus of dam concrete, latin hypercube sampling (LHS) is utilized to obtain samples of damaged parameters, such as coordinates, size, and degrees. Then, the sample set of low-fidelity surrogate model is constructed using damaged parameters and simulated structural responses, and the multi-fidelity surrogate model for the structural damage identification (SDI) of concrete dam is established by incorporating the high-fidelity deformation responses under actual loads via space mapping method. Finally, genetic algorithm (GA) is employed to search for the optimal solution of objective function for damage identification. Numerical and engineering examples show that, the multi-fidelity surrogate model can accurately identify the damaged region of concrete dam. Meanwhile, compared with FEM, XFEM-based numerical simulation model avoids the problem of repeated mesh division, and the computational efficiency is significantly improved. A data and physics-driven method is provided for efficiently and accurately identifying the damaged regions of concrete dams.

The deformation prediction model of concrete dams provides a scientific basis for long-term safety evaluations. However, traditional models primarily focus on characterizing the weight differences of deformation explanatory factors at a specific moment while ignoring the variations in causality induced by the evolution of structural performance. Additionally, the complex nonlinear mapping relationships significantly increase the difficulty of rapidly identifying the structural performance evolution states across different regions of dams. To address this issue, a multi-point hybrid prediction model for concrete dam deformation based on the rapid identification of structural performance evolution state is proposed in this paper. During the identification process, standard particle swarm optimization is employed to optimize the objective function, while hierarchical surrogate-model-assisted (HSMA) mechanism is proposed to improve the multi-dimensional adaptability by deeply coupling the iteration. With the identification results, a multi-point hybrid model for deformation prediction based on Bayesian optimized support vector regression is constructed. Taking an aging concrete dam as example, the application results indicate that the HSMA mechanism significantly improves the efficiency and accuracy of multi-dimensional identification of structural performance, while the fitting and prediction results of the proposed model exhibit a favorable agreement with the prototype monitoring data. After certain improvements and extensions, this method can be applied to construct the digital twin platforms for water conservancy engineering, providing technical support for monitoring spatiotemporal deformation.

Accurately predicting dam deformation is crucial for understanding its operational status. However, existing models struggle to effectively capture the spatiotemporal correlations in monitoring data and quantify uncertainty within dam systems. This paper presents an innovative uncertainty quantification model for evaluating regional deformation in arch dams. First, a method to extract the spatiotemporal correlation features is proposed. Considering the multidimensional deformation at measurement points, correlations among various points are analyzed through improved self-organizing map clustering and federated Kalman filtering. Second, a temporal convolutional network is employed for improved lower and upper bound estimation, and a quality-driven loss function is adopted to optimize model parameters. Finally, engineering case studies demonstrate that this model can generate reliable prediction intervals for regional deformation, thereby aiding in risk analysis and diagnostics.

Evaluation of the effectiveness of reinforcement measures for the diseased concrete gravity dam is a critical issue in the field of hydraulic structural health control. The traditional probabilistic reliability analysis method is constrained by factors such as the randomness of the uncertain parameters, the high nonlinearity of the performance function, etc., when it is applied to the evaluation of the effectiveness evolution of reinforcement measures for the diseased concrete gravity dam. Therefore, a time-varying non-probabilistic reliability (NR) analysis method is proposed in the research. To characterize the uncertainty of the parameters, we adopt the form of interval numbers to present parameters and construct the interval parameter-based reliability analysis framework. Specifically, combining interval mathematics and non-probabilistic reliability theory, the NR calculation model for the gravity dam is proposed. In addition, considering of the aging of the structure, the time-varying NR (TV-NR) calculation model utilized to evaluate the long-term reinforcement effectiveness is established. To tackle the issue of expensive computation existed in the TV-NR model, response surface method (RSM) is coupled in the TV-NR calculation model to replace the original high nonlinearity performance function. Ultimately, the proposed method is verified in engineering examples that it can detect the most possible initial locations and failure processes of the gravity dam varying with time and can efficiently and accurately evaluate the reinforcement effectiveness of the gravity dams. The results indicate that the use of reinforcement measures can provide an improvement for the safety service of gravity dams.

Various missing values inevitably exist in the settlement monitoring data of prolonged operation of earth-rockfill dams, which delays or even interrupts the dam structure analytical procedure. This study aims to develop a reliable method for addressing the missing data problem and monitoring earth-rockfill dam settlement behavior. First, an imputation model, support vector regression based on the finite element method and time input (FEMTSVR), is proposed and offers superior performance when capturing complex nonlinear mappings from environmental variables to settlement on small sample data. Herein, an improved particle swarm optimization algorithm (IPSO) is developed, realizing the nonlinear adjustment of weights and parameter reduction during hyperparameter optimization. Then, this study establishes a sequential prediction model based on gate recurrent unit (GRU) networks to monitor dam behavior on the imputed and complete dataset. Eventually, the proposed method is evaluated using real-world earth-rockfill dam monitoring data with the help of statistical indicators, demonstrating its efficiency in monitoring settlement data subjected to large-scale missingness. This study provides a robust database for dam structural health monitoring, while also providing a promising framework for the safety assessment of other civil or hydraulic engineering based on raw monitoring data.

When using a differential evolution algorithm to solve the joint flood optimization scheduling problem of cascade reservoirs, a greedy random optimization strategy is prone to premature convergence. Therefore, a new, improved Elite Conservative Differential Evolution Algorithm (ECDE) was proposed in this study. This algorithm divides a population into elite and general populations. The elite population does not undergo differential mutation, whereas the general population uses an adaptive differential mutation strategy based on successful historical information to participate in differential mutation. This elite conservative strategy effectively improves the diversity of the population evolution process and enhances convergence accuracy and stability. In a numerical experiment involving 10 test functions, the proposed ECDE performed the best overall (seven functions had the best stable convergence solution, while the remaining three performed the best), while in the single-objective flood control optimization scheduling problem of cascade reservoirs in the middle reaches of the Gan River, some algorithms could not even stably converge to feasible solutions (taking the 1973 inflow as an example, the peak shaving rate of the ECDE calculation results was 3.4%, 13.72%, and 11.73% higher than those of SHADE, SaDE, and GA, respectively). The proposed ECDE algorithm outperformed the SHADE, SaDE, GA, PSO, and ABC algorithms in terms of both convergence accuracy and stability. Finally, ECDE was used to analyze the multi-objective flood control scheduling problem of cascade reservoirs in the middle reaches of the Gan River, and it was found that the weight setting in multi-objective optimization should follow an upstream priority program or equilibrium programs. Adopting a downstream priority program results in poor upstream flood control performance. The above analysis fully verifies the superiority of the proposed algorithm, which can be used to solve and analyze the joint optimization scheduling problem of cascade reservoirs.

Multicomplex constraints often need to be considered in the optimal operation of cascade reservoirs with high-dimensional decision variables, making it difficult for traditional optimization methods and modern intelligent algorithms to solve such problems. Therefore, this study proposes a constraint handling method combining a penalty function nested DPSA-POA and an intelligent algorithm, to determine the optimal flood control operation of cascade reservoirs in the middle reaches of the Ganjiang River, a problem with decision variables of up to 2196 dimensions. The results indicate that the constraint handling method proposed in this paper can solve high-dimensional optimization problems in three modes: continuous nesting (Mode 1), optimization with intelligence after obtaining a feasible solution (Mode 2), and optimization with the DPSA-POA after obtaining a feasible solution (Mode 3). Of the three modes, Mode 2 has the highest accuracy, but its calculation time is approximately 10 h. Although the accuracy of Mode 3 is slightly worse, can only achieve 98%–99% of Mode 2, its calculation time is only approximately 1–3 h. The comprehensive performance of Mode 1 is poor, the convergence accuracy can only reach 97% of Mode 2, which corresponds to a calculation time of approximately 4–6 h. The existing superiority of feasibility (SF), stochastic ranking (SR), penalty function (PF), ε-constraint (EC) and adaptive ε-constraint (Adaptive EC) methods cannot converge to a feasible solution stably, and the accuracy of the results of these methods under the condition of obtaining a feasible solution is significantly lower than that of Mode 2 and Mode 3, reaching only 95%.