Dynamic calibration and long-term fidelity method for simulation model of concrete dam operation

Abstract

The application discloses a kind of simulation model dynamic calibration and long-acting fidelity method for long-term operation of concrete dam, the method is monitored by real-time acquisition of data by multi-source sensor array, after data calibration by edge computing node, establish concrete dam finite element numerical model and determine to be identified parameter and its change mode;Through the construction training parameter group carries out finite element positive analysis, establishes parameter-response sample set and trains nonlinear mapping model;Parameter is identified in current period in combination with optimization search algorithm and is rewritten to finite element model and carries out mutual feedback simulation;The updated response result is compared with new monitoring data, when deviation exceeds set threshold, trigger sample set augmentation and model updating, repeat parameter identification and simulation calculation, form the continuous dynamic calibration closed loop across monitoring period, realize simulation model long-acting fidelity.The application can effectively solve the problem of simulation model long-term operation precision attenuation, enhance online monitoring early warning and forecasting ability.

Description

Technical Field

[0001] This invention belongs to the field of hydraulic structure safety monitoring technology, specifically involving a method for dynamic calibration and long-term fidelity preservation of simulation models for the long-term operation of concrete dams. Background Technology

[0002] During long-term operation, the properties of concrete materials in the dam body (such as elastic modulus and creep characteristics) evolve with age and environmental factors. The mechanical parameters of the dam foundation rock mass may also change due to water level variations, seepage, and geological tectonic activity. Furthermore, the stability conditions of the high slope rock mass exhibit time-varying characteristics. These factors collectively lead to significant time-varying and uncertainties in the structural response of concrete dams. Therefore, using monitoring data to invert the mechanical parameters of the dam body, foundation, and high slope rock mass, and calibrating numerical models, is a crucial means of accurately evaluating the long-term operational performance of dams.

[0003] Existing parameter inversion methods typically identify dam body or foundation parameters by selecting a subset of monitored parameters as the target response and combining finite element analysis, response surface methodology, neural networks, or optimization algorithms. While these methods improve parameter analysis efficiency to some extent, their technical approach still suffers from the following shortcomings: (1) Monitoring data often contains outliers, missing values ​​and invalid data. Existing data preprocessing and parameter inversion are usually performed offline in steps, lacking a complete real-time online link from data collection and cleaning to parameter inversion, which affects the accuracy and timeliness of the identification results; (2) Existing methods generally regard inversion as a one-time parameter calibration task, failing to build a continuous dynamic closed loop between real-time monitoring data and simulation models, making it difficult to realize the continuous adaptive evolution of simulation model parameters as the actual performance of the dam body evolves; (3) Although the existing technology has proposed the idea of ​​mutual feedback correction between monitoring and simulation, its essence is still the quasi-static parameter calibration based on the current monitoring data. The accuracy of the corrected model will gradually decrease with the long-term evolution of concrete material properties, changes in dam foundation boundary conditions and the influence of extreme working conditions. It cannot guarantee continuous fidelity within a service cycle of several to several decades. That is, the existing technology lacks the understanding of the independent technical problem of "the long-term operation accuracy decay of simulation model", and has not proposed a complete technical path to achieve long-term fidelity through cross-time dynamic calibration and continuous expansion of sample set. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a dynamic calibration and long-term fidelity method for simulation models for long-term operation of concrete dams, so as to solve the problems in the prior art where there is no continuous dynamic closed loop between monitoring data and simulation models, and the accuracy of early warning and forecasting is affected by the decline in the accuracy of simulation models during long-term operation.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams, comprising the following steps: S1. Real-time acquisition of multi-source monitoring data through a multi-source sensor array deployed in the concrete dam body, dam foundation and high slope rock mass. The monitoring data includes one or more of the following: deformation monitoring data, stress-strain monitoring data, temperature monitoring data and seepage pressure monitoring data. S2. Receive raw monitoring data reported by the multi-source sensor array through the deployed edge computing nodes, perform calibration processing on the monitoring data, and form a calibration monitoring dataset. The calibration processing includes: Select monitoring data according to either the full-time period mode or the manually set time period mode; Organize monitoring data according to single-point, multi-point, single-project, or multi-project models; The effectiveness of monitoring data is identified and invalid data is removed using a professional dam safety monitoring model, and missing data is supplemented. S3. On the computing server, establish a finite element numerical model of the concrete dam body, dam foundation, and high slope rock mass. The finite element numerical model includes dam body structural information, dam foundation and high slope geological information, as well as load and boundary condition information. S4. Determine the parameters to be identified and their variation patterns. The parameters to be identified include the thermal and mechanical parameters of the dam concrete, as well as the mechanical or seepage parameters of the dam foundation rock mass, high slope rock mass, and main structural surfaces. The variation patterns include one or more of the following: unknown constant pattern, piecewise constant pattern, and polynomial time-varying pattern. S5. Combine the load and time-sensitivity factors that affect the finite element analysis results, and set different values ​​for the parameters to be identified to construct a training parameter set. S6. Input the training parameter set into the finite element numerical model for forward analysis, obtain the calculation results of the monitoring quantity of the concrete dam under the corresponding working condition, and establish a sample set between the parameters to be identified, the load and the time-effect elements and the monitoring quantity response. S7. Establish a nonlinear mapping model between the parameters to be identified, load and time-effect elements and the monitoring quantity response based on the sample set; S8. Input the calibration monitoring data into the nonlinear mapping model and combine it with the optimization search algorithm to identify the parameters to be identified for the current time period; S9. Write back the identified parameters to be identified to the finite element numerical model for mutual feedback simulation to obtain the updated concrete dam response results. S10. Compare the updated concrete dam response results with the newly added monitoring data. When the deviation exceeds the set threshold, trigger the calibration update process: Using the working conditions corresponding to the newly added monitoring period as the augmented working conditions, the finite element positive analysis was carried out again, the sample set was expanded and the nonlinear mapping model was retrained, and steps S8 to S10 were repeated using the updated nonlinear mapping model. If the deviation does not exceed the set threshold, the current parameter identification result and simulation model are retained, and subsequent new monitoring data are received to perform the next round of deviation discrimination. By using bias discrimination and condition-triggered sample augmentation and model retraining, a continuous dynamic calibration closed loop is formed across monitoring periods. This ensures that the deviation between the updated concrete dam response results and the newly added monitoring data does not exceed a set threshold during the long-term operation of the concrete dam, thus achieving long-term fidelity of the simulation model.

[0006] In the preferred embodiment, in step S2, the edge computing node and the multi-source sensor array are connected through a real-time data link. The calibration process is completed at the edge of the dam area. The calibrated monitoring dataset is transmitted to the computing server through a dedicated network. The computing server and the edge computing node constitute a distributed computing architecture that decouples data preprocessing and parameter inversion.

[0007] In the preferred embodiment, the dynamic calibration and long-term fidelity of the simulation model is a continuous process spanning multiple time periods; When new monitoring data arrives in subsequent periods, a calibration update process triggered by the deviation judgment results is executed, forming a continuous calibration closed loop for the simulation model of the dam's long-term operation.

[0008] In a preferred embodiment, the calibration process in step S2 further includes outlier identification, dimensional unification, time-series alignment, and preprocessing of the monitoring data to form a calibration monitoring dataset that can be used for parameter identification and analysis.

[0009] In a preferred embodiment, the thermal and mechanical parameters of the dam concrete include one or more of the following: adiabatic temperature rise, elastic modulus, thermal conductivity, linear expansion coefficient, and creep parameters.

[0010] In a preferred embodiment, the mechanical parameters or seepage parameters of the dam foundation rock mass, high slope rock mass, and main structural surfaces include one or more of deformation modulus and permeability coefficient.

[0011] In a preferred embodiment, the load and time-effect elements in step S5 include one or more of the following: water level load, temperature load, self-weight load, uplift pressure, seepage boundary conditions, and time effect.

[0012] In a preferred embodiment, the monitoring data calibration in step S2 includes one or more of the following modes: The full-time mode is used to identify parameters based on the entire historical monitoring process. The manual time period setting mode is used to identify parameters based on a specified construction period, water storage period, operation period, or abnormal period. Single-point mode is used to identify local parameters based on data from a single monitoring point. Multi-point mode is used to perform overall parameter identification based on data from multiple monitoring points; Single-project mode is used for parameter identification based on data from a single type of monitoring project. The multi-project mode is used to jointly identify parameters based on multiple types of monitoring projects.

[0013] In a preferred embodiment, the change pattern of the parameter to be identified in step S4 is as follows: Unknown constant mode, that is, the parameter to be identified takes a single constant value during the target analysis period; Piecewise constant mode, which means that the parameter to be identified takes different constant values ​​in different time periods; A polynomial time-varying model is a model in which the parameters to be identified are represented as polynomial functions of time to reflect the evolution of the parameters with age or running time.

[0014] In a preferred embodiment, the nonlinear mapping model in step S7 is a neural network model. The neural network model is trained using the sample set established in step S6 to characterize the nonlinear correspondence between the parameters to be identified, the load and the time-effect elements, and the monitoring response of the concrete dam.

[0015] In a preferred embodiment, the optimization search algorithm in step S8 is a genetic algorithm, which performs a global search on the nonlinear mapping model to find the parameter value or parameter combination that minimizes the deviation between the calibration monitoring data and the model response results.

[0016] In the preferred embodiment, the deviation threshold in step S10 is determined comprehensively based on the concrete dam design specifications, monitoring noise level, and statistical characteristics of historical monitoring residuals.

[0017] In the preferred embodiment, the deviation threshold is a dynamically adjusted threshold that is adaptively corrected as monitoring data accumulates and the model is updated.

[0018] In a preferred embodiment, the sample set augmentation in step S10 includes adding the load combination corresponding to the newly added monitoring period and the range of values ​​of the parameters to be identified to the training parameter group, performing finite element positive analysis again, and adding the new parameter-response samples to the sample set.

[0019] In a preferred embodiment, in step S10, when the deviation exceeds a set threshold, the calibration update process is triggered based on the magnitude of the deviation between the newly added monitoring data and the feedback simulation results. For engineering scenarios where the required update frequency is higher than the threshold trigger frequency, periodic updates are performed according to a preset time interval.

[0020] In the preferred embodiment, in step S10, when the deviation exceeds the set threshold, the monitoring item category that dominates the deviation is identified from the monitoring data according to the monitoring physical quantity type corresponding to the deviation, and the output weights of the parameter subset to be identified or the nonlinear mapping model corresponding to the monitoring item category are adjusted when expanding the sample set.

[0021] The present invention provides a method for dynamic calibration and long-term fidelity preservation of simulation models for the long-term operation of concrete dams. By adopting the above method, the following beneficial effects can be achieved: (1) A complete cross-time period continuous dynamic calibration closed loop is formed by “data calibration → sample construction → proxy mapping → parameter identification → write-back simulation → deviation discrimination → sample augmentation”. This closed loop mechanism makes the simulation accuracy of the finite element numerical model independent of the accuracy of the initial modeling conditions. Instead, it gradually approaches the real dam response through continuous calibration, which effectively solves the problem of simulation model accuracy decay caused by material performance evolution, dam foundation boundary condition change and extreme working conditions during the long-term operation of concrete dams. It realizes the long-term fidelity of the simulation model in the service cycle of several years to several decades. (2) The edge computing nodes deployed on the edge side of the dam area receive the raw monitoring data reported by the multi-source sensor array in real time. The edge side completes the preprocessing operations such as data calibration, validity identification, missing value completion and outlier removal. The calibrated monitoring dataset is transmitted to the computing server through a dedicated network, so that the computing server and the edge computing nodes form a distributed computing architecture that decouples data preprocessing and parameter inversion. This effectively alleviates the bandwidth pressure of massive monitoring data transmission and the processing burden of the computing server, and improves the real-time response capability of the online monitoring system. (3) Through the sample augmentation mechanism of deviation discrimination and condition triggering, the calibration update process is triggered only when the deviation between the simulation model response result and the newly added monitoring data exceeds the set threshold, rather than blindly updating at a fixed time period, thus avoiding unnecessary consumption of computing resources. At the same time, by expanding the sample set and retraining the nonlinear mapping model, the prediction accuracy of the proxy model continues to evolve with the accumulation of monitoring data during the operation period, thus achieving the synergistic optimization of computing efficiency and model accuracy. (4) By constructing unknown constant patterns, piecewise constant patterns and polynomial time-varying patterns of the parameters to be identified in the target analysis period, it can meet the requirements for characterizing the evolution of parameters of concrete dams in different stages such as construction period, water storage period, operation period and abnormal period. It solves the problem that the parameter identification in the construction period and operation period lacks connection and is difficult to reflect the continuity and transfer of dam parameters in different stages in the existing technology, and enhances the engineering adaptability of parameter identification method. (5) Through the full-link automated closed loop of “real-time acquisition by multi-source sensors → rapid calibration of edge computing nodes → parameter inversion of computing servers and dynamic calibration of models → engineering evaluation based on the calibrated model”, a complete data flow closed loop from data acquisition to model update is formed, which does not require manual intervention and significantly improves the automation level and early warning and forecasting capabilities of the online monitoring system. (6) It fully integrates the thermal and mechanical parameters of dam concrete, the mechanical parameters of dam foundation rock mass and the seepage parameters of high slope rock mass, and the joint identification of multiple types of parameters to be identified. Through single-parameter identification or multi-parameter joint identification, it can simultaneously identify dam concrete parameters, dam foundation rock mass parameters, high slope rock mass parameters and main structural surface parameters. The parameter identification results have clear physical meaning and engineering interpretability, overcoming the problems of non-unique parameter inversion results and lack of engineering interpretability in the existing technology. Attached Figure Description

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart illustrating the dynamic calibration and long-term fidelity maintenance of the simulation model in the method of this invention across time periods.

[0023] Figure 2 This is a flowchart of sample augmentation and dynamic calibration based on newly added monitoring data.

[0024] Figure 3 This is a finite element model diagram of an embodiment of the present invention.

[0025] Figure 4 The figure shows the finite element positive analysis temperature field calculation results of an embodiment of the present invention.

[0026] Figure 5 The figure shows the calculation results of the deformation field along the river direction by the finite element forward analysis in an embodiment of the present invention.

[0027] Figure 6 This is a comparison chart of the finite element positive analysis temperature calculation results and the measured temperature data in an embodiment of the present invention.

[0028] Figure 7 This is a comparison chart of the process lines between the finite element forward analysis results of deformation along the river and the measured deformation data in an embodiment of the present invention. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] Example 1: This example uses a concrete gravity dam as an example. The dam section selected is the No. 2 spillway section. Temperature measuring point T1 and deformation measuring point P1 are selected. The process curves of the measuring points are as follows: Figure 6 , Figure 7 As shown.

[0031] S1. Real-time acquisition of multi-source monitoring data Multi-source monitoring data is collected in real time by deploying a multi-source sensor array on the concrete dam body, dam foundation and high slope rock mass.

[0032] In this embodiment, historical monitoring data and newly added monitoring data on temperature and deformation of the gravity dam are selected. Temperature monitoring data is acquired through temperature sensors embedded in the dam body, while deformation monitoring data is acquired through sensors such as displacement gauges and joint gauges deployed in the dam body and foundation. The sampling frequency of the sensor array is set according to the type of monitoring item, with a temperature monitoring sampling interval of 1 day and a deformation monitoring sampling interval of 1 hour. All monitoring data are aggregated by the on-site data acquisition unit and then reported to the edge computing node.

[0033] S2, Edge Computing Node Data Calibration Processing The deployed edge computing nodes receive the raw monitoring data reported by the multi-source sensor array, perform calibration processing on the monitoring data, and form a calibration monitoring dataset.

[0034] In this embodiment, the calibration period is set from October 2001 to December 2021, and temperature and deformation monitoring data are organized according to a multi-point and multi-project mode. The calibration process specifically includes: selecting all historical monitoring data within the time period in a full-time mode; selecting monitoring data from temperature measuring point T1 and deformation measuring point P1 in a multi-point mode; simultaneously organizing monitoring data for temperature monitoring projects and deformation monitoring projects in a multi-project mode; using a dam safety monitoring professional model to identify the validity of the monitoring data and remove outliers and invalid data caused by sensor failure, signal transmission interruption, etc.; completing missing data using linear interpolation or time-series compensation; unifying the dimensions and aligning the time series of the completed monitoring data to ensure that the collection time points of different monitoring projects are consistent, and finally forming a calibration monitoring dataset that can be used for parameter identification and analysis.

[0035] S3. Establishing a finite element numerical model On the computing server, a finite element numerical model of the concrete dam body, dam foundation, and high slope rock mass is established. The finite element numerical model includes dam body structural information, dam foundation and high slope geological information, and load and boundary condition information.

[0036] In this embodiment, a finite element model is established based on the geometric structure, material zoning, and geological data of the No. 2 spillway section, such as... Figure 3As shown, both the dam body and bedrock were meshed using hexahedral solid elements, while the dam bedrock was simulated using tetrahedral elements. The model consisted of 365,578 elements and 411,831 nodes. In the temperature field simulation, the transverse joints of the dam section, the four sides of the foundation, and the bottom surface were set as adiabatic boundaries, while the upstream, downstream, and top surfaces of the dam were set as heat exchange boundaries. The ambient temperature was applied based on the measured air temperature time history curve. In the deformation field simulation, water level load, temperature load, and self-weight load were applied. The water level load was applied based on the actual reservoir water level process line, the temperature load was imported from the temperature field calculation results, and the self-weight load was applied based on the unit weight of the dam material.

[0037] S4. Determine the parameters to be identified and their variation patterns. The parameters to be identified and their variation patterns are determined. The parameters to be identified include the thermal and mechanical parameters of the dam concrete, as well as the mechanical or seepage parameters of the dam foundation rock mass, high slope rock mass, and main structural surfaces.

[0038] In this embodiment, the parameters to be identified are the thermal conductivity coefficient and elastic modulus of the dam concrete. The thermal conductivity coefficient affects the accuracy of the calculated temperature field distribution of the dam body, while the elastic modulus is a key parameter for calculating the deformation field. Its variation pattern is an unknown constant pattern, meaning that each parameter takes a single constant value during the target analysis period (October 2001 to December 2021). The initial range of the thermal conductivity coefficient is determined to be 2.0–3.5 W / (m·K) based on concrete mix proportions and experience from similar projects, while the initial range of the elastic modulus is determined to be 20–45 GPa based on indoor test results and design data.

[0039] S5. Construct the training parameter set The load and time-sensitivity factors that affect the finite element analysis results are combined, and different values ​​are set for the parameters to be identified to construct a training parameter set.

[0040] In this embodiment, the load and time-dependent factors are water level load, temperature load, and self-weight load. Specifically, the water level load is calculated from the dead water level to the flood control limit water level, the temperature load is applied according to the annual variation curve of the monthly average air temperature, and the self-weight load is constant. A total of 200 training parameter sets are generated by Latin hypercube sampling within their respective value ranges for the parameters to be identified (thermal conductivity coefficient and elastic modulus).

[0041] S6. Conduct finite element analysis to establish a sample set. The training parameter set is input into the finite element numerical model for forward analysis to obtain the calculation results of the monitoring quantity of the concrete dam under the corresponding working condition, and to establish a sample set between the parameters to be identified, the load and the time-effect elements and the monitoring quantity response.

[0042] In this embodiment, the monitored parameters are temperature and deformation. 200 sets of training parameters are input one by one into the finite element model for multi-field coupled numerical simulation. Each set calculates the temperature and deformation field distributions of the dam body under the corresponding working condition. The calculated temperature at measuring point T1 and the calculated deformation value at measuring point P1 are extracted. Each set of training parameters (thermal conductivity coefficient, elastic modulus, water level condition) and its corresponding monitored parameter calculation results (T1 temperature, P1 deformation) are paired to form 200 sets of parameter-response samples. The finite element forward analysis temperature and deformation fields of this dam section at a typical moment (February 16, 2016) are as follows: Figure 4 , Figure 5 As shown.

[0043] S7. Establish a nonlinear mapping model Based on the sample set, a nonlinear mapping model is established between the parameters to be identified, load and time-sensitivity elements, and the response of the monitored quantities.

[0044] In this embodiment, the nonlinear mapping model is a neural network model, specifically a three-layer BP neural network (input layer with 3 nodes for thermal conductivity, elastic modulus, and water level load; hidden layer with 10 nodes; output layer with 2 nodes for T1 temperature and P1 deformation). It is trained using the 200 sets of samples established in step S6, with the training set and validation set divided in a 4:1 ratio. The hidden layer activation function is the ReLU function, and the output layer uses a linear activation function. The optimizer is the Adam algorithm, with a learning rate of 0.001 and 5000 training epochs. An early stopping mechanism terminates training when the validation set loss function does not decrease for 100 consecutive epochs. The trained neural network model is used to characterize the nonlinear correspondence between the parameters to be identified, the load and time-sensitivity factors, and the monitoring responses of the concrete dam.

[0045] S8, Parameter Recognition The calibration monitoring data is input into the nonlinear mapping model, and the optimized search algorithm is used to identify the parameters to be identified for the current time period.

[0046] In this embodiment, the optimization search algorithm is a genetic algorithm. The measured values ​​of temperature T1 and deformation P1 in the calibration monitoring dataset are used as target values. A neural network model is used as the fitness evaluation function, and a genetic algorithm is employed for global search. The genetic algorithm parameters are set as follows: population size 100, maximum number of generations 200, crossover probability 0.8, and mutation probability 0.05. By minimizing the root mean square error between the monitored values ​​and the model predicted values, the parameter values ​​to be identified that minimize the deviation between the calibration monitoring data and the model response results are obtained. In this embodiment, the thermal conductivity coefficient obtained by inversion is 2.3 W / (m·K), and the elastic modulus is 38 GPa.

[0047] S9, Write-back Simulation and Feedback Simulation The identified parameters are written back to the finite element numerical model for mutual feedback simulation to obtain the updated concrete dam response results.

[0048] In this embodiment, the thermal conductivity coefficient (2.3 W / (m·K)) and elastic modulus (38 GPa) obtained in step S8 are substituted into the corresponding material properties of the finite element numerical model, and the finite element calculation of the temperature field and deformation field is performed again to obtain the updated temperature distribution and deformation distribution of the dam body.

[0049] S10, Deviation Judgment and Dynamic Calibration The updated concrete dam response results are compared with the newly added monitoring data. When the deviation does not exceed the set threshold, the updated parameters and simulation results for the current time period are output; when the deviation exceeds the set threshold, the finite element positive analysis is carried out again with the working condition corresponding to the newly added monitoring time period as the augmented working condition, the sample set is expanded and the nonlinear mapping model is updated, and steps S8 to S10 are repeated to achieve cross-time dynamic calibration and long-term fidelity of the simulation model.

[0050] In this embodiment, monitoring data from October 2001 to December 2016 was used as the training period to complete the first round of parameter identification, and monitoring data from January 2017 to December 2021 was used as new monitoring data for deviation discrimination. The temperature deviation threshold was set to ±0.5℃, and the deformation deviation threshold was set to ±1.0mm. The T1 temperature time history and P1 deformation time history calculated by the updated finite element model were compared with the new measured data from January 2017 to December 2021 on a time-by-time basis. The finite element forward analysis temperature and deformation data corresponding to the inverted parameters were compared with the measured temperature and deformation data. Figure 6 , Figure 7 As shown, the average temperature deviation for each time period was 0.3℃ and the average deformation deviation was 0.6mm, both of which did not exceed the set threshold, indicating that the current parameter identification results and simulation model remained valid from January 2017 to December 2021.

[0051] If the temperature deviation exceeds 0.5℃ or the deformation deviation exceeds 1.0mm during a certain period, the water level and temperature conditions corresponding to that period are used as augmented conditions. The augmented conditions are added to the training parameter group in step S5, and the finite element positive analysis is carried out again. The newly generated parameters—response samples—are added to the sample set, the neural network model is retrained, and steps S8 to S10 are repeated until the deviation falls back to the threshold range.

[0052] Cycle termination condition: The repeated steps S8 to S10 can be executed cyclically until the deviation between the updated concrete dam response result and the newly added monitoring data does not exceed the set threshold, or the preset maximum number of iterations (set to 10 times in this embodiment) is reached. Then the current cycle calibration is terminated, the optimal parameter identification result and simulation result of the current cycle are output, and a calibration report is generated for engineering evaluation reference.

[0053] Relationship between periodic updates and threshold triggering: In this embodiment, the deviation threshold-triggered calibration update is the core calibration mechanism of this invention. The calibration update process is triggered only when the deviation between the simulation model response result and the newly added monitoring data exceeds a set threshold, avoiding the waste of computational resources caused by blind updates at fixed time periods. For engineering scenarios where the update frequency requirement is higher than the threshold triggering frequency (such as extreme working conditions or monitoring periods with special requirements for model real-time performance), periodic updates can be performed according to a preset time interval as a supplement to the threshold triggering mechanism. The periodic update is still driven by newly added monitoring data; its essence is a proactive preventative calibration performed in scenarios where the threshold triggering condition has not yet been met but higher requirements for model timeliness are needed, rather than a blind update detached from monitoring data. In normal operating scenarios, deviation threshold triggering remains the core calibration mechanism, with periodic updates serving only as an optional auxiliary means.

[0054] The updated parameters can be pushed to the online monitoring platform for further use in predicting and forecasting the behavior of concrete dams under different conditions.

[0055] Example 2: Based on the above embodiment 1, this embodiment further illustrates the joint parameter identification in the multi-item mode.

[0056] In this embodiment, in addition to temperature and deformation monitoring data, seepage pressure monitoring data of the dam foundation is further acquired. The parameter to be identified is the permeability coefficient of the dam foundation rock mass, which is added to the parameter to be identified in Embodiment 1. The objective function for monitoring during operation is composed of the weighted deviation between the measured and calculated values ​​of the three monitoring items: temperature, deformation, and seepage pressure. The weights of each monitoring item are set according to data quality, representativeness of the monitoring location, and sensitivity of the parameter to be identified (weight of temperature item 0.3, weight of deformation item 0.4, weight of seepage pressure item 0.3). The range of permeability coefficient values ​​in the parameter prior constraints is determined to be 1×10 based on the results of the field pressure water test. -7 ~1×10 -5 cm / s. By combining temperature, deformation, and seepage pressure monitoring information for parameter inversion, three parameters—the thermal conductivity coefficient of the dam concrete, the elastic modulus, and the permeability coefficient of the dam foundation rock—can be identified simultaneously, further improving the multi-physics consistency and engineering interpretability of the parameter identification results. The remaining steps are the same as in Example 1.

[0057] Example 3: Based on the above embodiment 1, this embodiment further illustrates the long-term parameter evolution identification under the piecewise constant mode.

[0058] In this embodiment, the operation period from October 2001 to December 2021 is divided into three periods: October 2001 to December 2006 (initial operation), January 2007 to December 2014 (mid-operation), and January 2015 to December 2021 (late operation). The parameters to be identified (thermal conductivity and elastic modulus) take a single constant value in each period, which may be different between different periods. A piecewise constant model is used for parameter inversion. Steps S3 to S9 are executed for each period to obtain the parameter identification results for each period: thermal conductivity 2.6 W / (m·K) and elastic modulus 38.6 GPa in the initial operation; thermal conductivity 2.4 W / (m·K) and elastic modulus 38.2 GPa in the mid-operation; and thermal conductivity 2.3 W / (m·K) and elastic modulus 38 GPa in the late operation. The piecewise constant model can reflect the gradual evolution trend of the thermal and mechanical properties of dam concrete with age and under the influence of long-term operating environment. The parameter identification results between different time periods have clear physical continuity and transferability. The remaining steps are the same as in Example 1.

[0059] Example 4: Based on the above embodiment 1, this embodiment further illustrates the specific implementation method of distributed collaborative processing between edge computing nodes and computing servers.

[0060] In this embodiment, the edge computing node is deployed in the on-site monitoring room of the dam area and is connected to the data acquisition unit of the multi-source sensor array via industrial Ethernet to receive the raw monitoring data stream in real time. The edge computing node is pre-installed with a Linux operating system and a data calibration processing software module. After the raw monitoring data arrives at the edge computing node, calibration processing operations such as data decoding, validity identification, outlier removal, missing value imputation, unit unification, and time sequence alignment are completed in real time at the edge. The size of the processed calibration monitoring dataset is about 30% of the original data and is transmitted to the computing server deployed in the remote cloud computing center through the dedicated fiber optic network of the dam area. The computing server is equipped with a high-performance GPU for accelerating neural network model training and parameter inversion. After receiving the calibration monitoring dataset, it executes steps S3 to S10. This distributed architecture physically decouples data preprocessing and parameter inversion. The edge computing node is dedicated to high-frequency data acquisition and real-time cleaning, while the computing server focuses on computationally intensive finite element forward analysis and neural network training, effectively reducing the data transmission bandwidth requirements, avoiding the occupation of network resources by directly transmitting massive amounts of raw monitoring data, and shortening the overall response time from data acquisition to parameter inversion. The remaining steps are the same as in Embodiment 1.

[0061] Example 5: Based on the above embodiment 1, this embodiment further illustrates the specific implementation method of dynamic adjustment of deviation threshold.

[0062] In this embodiment, the deviation threshold is not a fixed value, but rather adaptively adjusted as monitoring data accumulates and the model is updated. Specifically, after each round of parameter identification and feedback simulation, the residual sequence between the updated model response and the corresponding measured data is calculated. Using the residual sequences of the current round and previous rounds as samples, the mean μ and standard deviation σ of the residuals are calculated, and the deviation threshold for the next round is dynamically set to μ ± 3σ (i.e., a 99.7% confidence interval). Simultaneously, this dynamic threshold is not lower than the physical lower limit of sensor accuracy (0.1℃ for temperature sensors, 0.1mm for deformation sensors) nor higher than the engineering warning values ​​(1.0℃ for temperature, 2.0mm for deformation). When new monitoring data is added, μ and σ are recalculated using the updated residual sequence, thereby updating the deviation threshold for the next round. In this embodiment, when the first round of calibration was completed in January 2017, the mean temperature residual was 0.25℃ and the standard deviation was 0.12℃. Based on this, the temperature deviation threshold for the next round was set at 0.61℃. By the second round of calibration in January 2019, the mean temperature residual had decreased to 0.20℃ and the standard deviation had decreased to 0.09℃, and the deviation threshold was adaptively adjusted to 0.47℃. This dynamic adjustment mechanism ensures that the deviation threshold always matches the current accuracy level of the model, avoiding the problems of model inaccuracy due to an overly loose fixed threshold or frequent invalid triggers due to an overly strict threshold. The remaining steps are the same as in Embodiment 1.

Claims

1. A method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams, characterized in that... Includes the following steps: S1. Real-time acquisition of multi-source monitoring data through a multi-source sensor array deployed in the concrete dam body, dam foundation and high slope rock mass. The monitoring data includes one or more of the following: deformation monitoring data, stress-strain monitoring data, temperature monitoring data and seepage pressure monitoring data. S2. Receive raw monitoring data reported by the multi-source sensor array through the deployed edge computing nodes, perform calibration processing on the monitoring data, and form a calibration monitoring dataset. The calibration processing includes: Select monitoring data according to either the full-time period mode or the manually set time period mode; Organize monitoring data according to single-point, multi-point, single-project, or multi-project models; The effectiveness of monitoring data is identified and invalid data is removed using a professional dam safety monitoring model, and missing data is supplemented. S3. On the computing server, establish a finite element numerical model of the concrete dam body, dam foundation, and high slope rock mass. The finite element numerical model includes dam body structural information, dam foundation and high slope geological information, as well as load and boundary condition information. S4. Determine the parameters to be identified and their variation patterns. The parameters to be identified include the thermal and mechanical parameters of the dam concrete, as well as the mechanical or seepage parameters of the dam foundation rock mass, high slope rock mass, and main structural surfaces. The variation patterns include one or more of the following: unknown constant pattern, piecewise constant pattern, and polynomial time-varying pattern. S5. Combine the load and time-sensitivity factors that affect the finite element analysis results, and set different values ​​for the parameters to be identified to construct a training parameter set. S6. Input the training parameter set into the finite element numerical model for forward analysis, obtain the calculation results of the monitoring quantity of the concrete dam under the corresponding working condition, and establish a sample set between the parameters to be identified, the load and the time-effect elements and the monitoring quantity response. S7. Establish a nonlinear mapping model between the parameters to be identified, load and time-effect elements and the monitoring quantity response based on the sample set; S8. Input the calibration monitoring data into the nonlinear mapping model and combine it with the optimization search algorithm to identify the parameters to be identified for the current time period; S9. Write back the identified parameters to be identified to the finite element numerical model for mutual feedback simulation to obtain the updated concrete dam response results. S10. Compare the updated concrete dam response results with the newly added monitoring data. When the deviation exceeds the set threshold, trigger the calibration update process: Using the working conditions corresponding to the newly added monitoring period as the augmented working conditions, the finite element positive analysis was carried out again, the sample set was expanded and the nonlinear mapping model was retrained, and steps S8 to S10 were repeated using the updated nonlinear mapping model. If the deviation does not exceed the set threshold, the current parameter identification result and simulation model are retained, and subsequent new monitoring data are received to perform the next round of deviation discrimination. By using bias discrimination and condition-triggered sample augmentation and model retraining, a continuous dynamic calibration closed loop is formed across monitoring periods. This ensures that the deviation between the updated concrete dam response results and the newly added monitoring data does not exceed a set threshold during the long-term operation of the concrete dam, thus achieving long-term fidelity of the simulation model.

2. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: In step S2, the edge computing node and the multi-source sensor array are connected through a real-time data link. The calibration process is completed at the edge of the dam area. The calibrated monitoring dataset is transmitted to the computing server through a dedicated network. The computing server and the edge computing node constitute a distributed computing architecture that decouples data preprocessing and parameter inversion.

3. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The dynamic calibration and long-term fidelity of the simulation model is a continuous process that spans multiple time periods. When new monitoring data arrives in subsequent periods, a calibration update process triggered by the deviation judgment results is executed, forming a continuous calibration closed loop for the simulation model of the dam's long-term operation.

4. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The calibration process in step S2 also includes outlier identification, dimension unification, time series alignment, and preprocessing of the monitoring data to form a calibration monitoring dataset that can be used for parameter identification and analysis.

5. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The thermal and mechanical parameters of the dam concrete include one or more of the following: adiabatic temperature rise, elastic modulus, thermal conductivity, linear expansion coefficient, and creep parameters.

6. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The mechanical or seepage parameters of the dam foundation rock mass, high slope rock mass, and main structural surfaces include one or more of deformation modulus and permeability coefficient.

7. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The load and time-effect elements in step S5 include one or more of the following: water level load, temperature load, self-weight load, uplift pressure, seepage boundary conditions, and time effect.

8. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The monitoring data calibration in step S2 includes one or more of the following modes: The full-time mode is used to identify parameters based on the entire historical monitoring process. The manual time period setting mode is used to identify parameters based on a specified construction period, water storage period, operation period, or abnormal period. Single-point mode is used to identify local parameters based on data from a single monitoring point. Multi-point mode is used to perform overall parameter identification based on data from multiple monitoring points; Single-project mode is used for parameter identification based on data from a single type of monitoring project. The multi-project mode is used to jointly identify parameters based on multiple types of monitoring projects.

9. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The change pattern of the parameter to be identified in step S4 is as follows: Unknown constant mode, that is, the parameter to be identified takes a single constant value during the target analysis period; Piecewise constant mode, which means that the parameter to be identified takes different constant values ​​in different time periods; A polynomial time-varying model is a model in which the parameters to be identified are represented as polynomial functions of time to reflect the evolution of the parameters with age or running time.

10. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The nonlinear mapping model in step S7 is a neural network model. The neural network model is trained using the sample set established in step S6 and is used to characterize the nonlinear correspondence between the parameters to be identified, the load and the time-effect elements and the monitoring response of the concrete dam.

11. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The optimization search algorithm in step S8 is a genetic algorithm, which performs a global search on the nonlinear mapping model to find the parameter value or parameter combination that minimizes the deviation between the calibration monitoring data and the model response results.

12. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The deviation threshold in step S10 is determined comprehensively based on the concrete dam design specifications, monitoring noise levels, and historical monitoring residual statistical characteristics.

13. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 12, characterized in that: The deviation threshold is a dynamically adjusted threshold that is adaptively corrected as monitoring data accumulates and the model is updated.

14. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: The sample set augmentation in step S10 includes adding the load combination corresponding to the newly added monitoring period and the range of values ​​of the parameters to be identified to the training parameter group, re-performing the finite element positive analysis, and adding the new parameter-response samples to the sample set.

15. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: In step S10, when the deviation exceeds the set threshold, the calibration update process is triggered to execute based on the magnitude of the deviation between the newly added monitoring data and the mutual feedback simulation results. For engineering scenarios where the required update frequency is higher than the threshold trigger frequency, periodic updates are performed according to a preset time interval.

16. The method for dynamic calibration and long-term fidelity preservation of simulation models for long-term operation of concrete dams according to claim 1, characterized in that: In step S10, when the deviation exceeds the set threshold, the monitoring item category that dominates the deviation is identified from the monitoring data according to the type of monitoring physical quantity corresponding to the deviation, and the output weight of the parameter subset to be identified or the nonlinear mapping model corresponding to the monitoring item category is adjusted when expanding the sample set.

Images