High earth and rockfill dam digital twinborn model construction method

Through deep learning and finite element methods, the high-earth and rock dam deformation-seepage digital twin framework is constructed, which solves the shortcomings of the existing monitoring methods in capturing complex feature relationships and generalization capabilities, and achieves high-precision real-time monitoring and security early warning.

CN120145770AActive Publication Date: 2025-06-13SICHUAN UNIV

Patent Information

Application Number
CN202510325363.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-03-19
Publication Date
2025-06-13
Estimated Expiration
2045-03-19

AI Technical Summary

Technical Problem

The existing high-earth and rock dam health monitoring methods are difficult to capture complex multi-character relationships, and there is a problem of insufficient generalization capabilities. Digital twin technology faces challenges in the construction of physical digital models, sensor data integration and data flow feedback.

Method used

Deep learning and finite element methods are used to build a high-earth and rock dam deformation-seepage digital twin framework. Through the LSTM framework with measurement point connection and graph convolution feature extractor, the information transmission relationship between sensor measurement points and finite element nodes is learned, and combined with field adaptive confrontation training and knowledge graph reasoning, a digital twin interaction model for extreme working conditions is constructed.

Benefits of technology

It realizes high-precision real-time perception and prediction of seepage and deformation state of high-altitude and rock dams, improves the matching degree of monitoring data and numerical calculations, enhances the safety warning capabilities of dams, and provides intelligent structural health monitoring solutions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120145770A_ABST
    Figure CN120145770A_ABST
Patent Text Reader

Abstract

The invention provides a high earth and rockfill dam digital twinborn model construction method, and belongs to the technical field of high earth and rockfill dam health monitoring, and the method comprises the steps: constructing a measuring point monitoring model based on sensor monitoring data; constructing a mapping model, and constructing a mapping model dynamic monitoring loss function for adversarial training by adopting a domain adaptive method; inputting predicted values of deformation monitoring effect quantity and osmotic pressure monitoring quantity of the measuring point monitoring model on the seepage process and the deformation process of the high earth and rockfill dam into the mapping model to obtain a digital twinborn model, and constructing an extreme working condition digital twinborn model based on the digital twinborn model; knowledge graph reasoning is combined with a dam slope stability coefficient to realize interaction of an extreme working condition digital twinning model and a solid model, and the high earth and rockfill dam deformation-seepage digital twinning framework is proposed for the first time on the basis of deep learning and finite element methods so as to realize a more economical, efficient and reliable monitoring scheme. And technical support is provided for health monitoring of the high earth and rockfill dam structure in the future.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of health monitoring of high earth-rock dams, and particularly relates to a method for constructing a digital twin model of a high earth-rock dam. Background Art

[0002] Due to advantages such as low cost, strong adaptability, and simple construction, high earth-rock dams are widely used in global water conservancy projects. In recent years, a number of high earth-rock dam projects that have been completed, such as Gushui, Shuangjiangkou, Ruomei, and Lianghekou, etc., their safety is related to the safety of people's lives and property. However, the complex geological environment and high stress conditions make the structural health monitoring of high earth-rock dams challenging. Currently, the monitoring model based on measuring points is the mainstream method for the health monitoring of high earth-rock dams, and statistical models (such as HST) or shallow machine learning methods are often used, but it is difficult for them to capture complex multi-feature relationships and there are problems with insufficient generalization ability. Deep learning models (such as LSTM, GRU, and GCN) have been applied in dam safety monitoring due to their powerful non-linear modeling capabilities. However, the generalization ability of such models is poor under extreme working conditions. Transfer learning can improve the prediction accuracy and generalization of the model through knowledge transfer, providing reliability support for deep learning models under unknown extreme working conditions. In addition, digital twin technology is expected to achieve comprehensive monitoring of high earth-rock dams, but currently faces challenges such as the construction of physical digital models, the integration of sensor data, and the data flow feedback between the digital and the real world. In summary, the existing measuring point monitoring models are difficult to meet the requirements, and the application of digital twin technology is still in its infancy. Summary of the Invention

[0003] Aiming at the above deficiencies in the prior art, the present invention provides a method for constructing a digital twin model of a high earth-rock dam, aiming to solve the technical problems of digital representation of the physical model of a high earth-rock dam and the fusion of monitoring information, and focusing on breaking through the deficiencies of the existing methods in terms of real-time performance, accuracy, intelligent reasoning, and safety warning.

[0004] In order to achieve the above objectives, the technical solution adopted by the present invention is: a method for constructing a digital twin model of a high earth-rock dam, including the following steps:

[0005] S1. Construct a measuring point monitoring model based on sensor monitoring data;

[0006] S2. Use the long short-term memory neural network LSTM framework with measuring point connections to learn the information transfer relationships between the seepage pressure or deformation values of the finite element nodes corresponding to the sensor measuring points, the actual monitoring values of the sensors, and the seepage pressure or deformation values of their surrounding nodes, construct a mapping model, and construct a dynamic monitoring loss function for the mapping model;

[0007] S3. Input the predicted values of the deformation monitoring effect quantities and seepage pressure monitoring quantities of the high earth-rock dam seepage process and deformation process by the measuring point monitoring model into the mapping model, perform adversarial training using domain adaptation, combine with the measuring point monitoring model to obtain a digital twin model, and construct an extreme condition digital twin interaction model based on the digital twin model;

[0008] S4. Construct a high earth-rock dam inference data set based on the knowledge graph, use the high earth-rock dam inference data set to train the extreme condition digital twin interaction model, and use the predicted values of the dam body global deformation monitoring effect quantities and seepage pressure monitoring quantities output by the extreme condition digital twin model as inputs, and use the trained extreme condition digital twin interaction model to output the treatment measures for the cross-section of the physical project to complete the construction of the high earth-rock dam digital twin model.

[0009] The beneficial effects of the present invention are as follows: Based on deep learning and finite element methods, the present invention first proposes a deformation-seepage digital twin framework for high earth-rock dams, aiming to achieve a more economical, efficient, and reliable monitoring scheme, and provide technical support for the future structural health monitoring of high earth-rock dams.

[0010] Further, the S1 includes the following steps:

[0011] Construct a graph convolutional feature extractor: Extract environmental impact factors as input variables, and use GCN to adopt a feature extractor-decoder architecture to construct a graph convolutional feature extractor;

[0012] Seepage-deformation feature integration: Use the graph convolutional feature extractor to respectively construct a seepage feature extractor and a deformation feature extractor, extract environmental-seepage impact features and environmental-deformation impact features, and splice them to obtain seepage-deformation coupling features Among them, the seepage-deformation coupling features Adopt a shared trainable feature decoder to simultaneously output seepage monitoring predicted values And deformation monitoring predicted values

[0013]

[0014] Among them, σ represents a non-linear activation function, W represents a trainable weight matrix, and b represents a bias;

[0015] Extreme condition physical feature constraint: Perform supervised learning on the hydraulic load feature and soil-rock creep feature of the earth-rock dam deformation feature extractor. Among them, in the process of the earth-rock to the deformation feature extractor, the expression of the loss function of multi-teacher distillation is as follows:

[0016] L MYD =αL task (l S ,t)+(1-α)LKD (p(f s ,T),E(p(f T H ,T),p(f T θ,T)))

[0017] L KD = DF(p(f S ,T),p(f T ,T))

[0018] L = αL task +(1 - α)L KD

[0019] p(f,T) = solfmax(f / T)

[0020] where L MYD represents the loss function of the soil-rock deformation feature extractor, α represents the loss balance parameter, L task represents the loss of the student network regression prediction of the soil-rock dam deformation feature extractor, l S represents the deformation prediction value of the student network, t represents the measured deformation value, L KD represents the gap between the teacher network and the student network, p represents the distillation loss calculation method, f s represents the student network feature, T represents the temperature parameter, E() represents the integrated output function of the teacher network, f T H represents the hydraulic load feature of the teacher network, f T θ represents the soil-rock creep feature of the teacher network, DF() represents the distance metric function, f T represents the teacher network feature, f represents the extracted deformation feature, L represents the knowledge distillation loss function calculation method;

[0021] Time-varying effect model update: The construction of the monitoring point model is completed by sharing parameters - fine-tuning the feature decoder trained on the long-term monitoring data source task on the new sensor monitoring data set. Among them, the monitoring point model outputs the predicted values of the deformation monitoring effect quantity and the seepage pressure monitoring quantity.

[0022] The technical effects brought by the above further solutions are: realizing more accurate prediction of the dam structure condition under extreme water level conditions and realizing continuous iterative update to improve the prediction performance of the time-varying problem of parameters caused by particle breakage.

[0023] Furthermore, the expression of the graph convolution feature extractor is as follows:

[0024]

[0025] where H0 Represents the environmental features input to the graph convolutional feature extractor, and represents the set of hydraulic load influence factors, temperature influence factors, and aging influence factors at time t respectively, W (l) and W 0 both represent trainable weight matrices, H (l+1) represents the environmental influence factor feature matrix of the nodes in the l-th layer, and σ represents the non-linear activation function, represents the normalized degree matrix, represents the influence factor adjacency matrix with self-connections added, H (l+1) and H (l) represent the environmental influence factor feature matrices of the nodes in the (l + 1)-th layer and the l-th layer respectively, A represents the influence factor adjacency matrix, and I represents the identity matrix.

[0026] The technical effect brought by the above further solution is: using the graph convolutional feature extractor to extract the complex correlation relationships of numerous environmental quantity influence factors.

[0027] Furthermore, the expression of the seepage-deformation coupling feature is as follows:

[0028]

[0029] Among them, represents the seepage-deformation coupling feature, represents the environmental-seepage influence feature, represents the environmental-deformation influence feature.

[0030] The technical effect brought by the above further solution is: through the seepage-deformation coupling feature, realizing the mutual coupling modeling of the seepage process and the deformation process.

[0031] Furthermore, the expression of the mapping model is as follows:

[0032]

[0033] θ δ =(W, b)

[0034] Among them, represents the seepage pressure value of the nodes around the finite element node, δ Monitoring point represents the measured value of the monitoring point, Model LSTM () represents the model constructed based on the long short-term memory neural network LSTM, represents the calculated value of the finite element node corresponding to the sensor measurement point, ⊙ represents element-wise multiplication, W, b, and θ δ all represent trainable parameters, and [] represents feature fusion.

[0035] The technical effect brought about by the above further solution is: the present invention utilizes a mapping model with measuring point connections to achieve a breakthrough transition from point monitoring to full-domain engineering monitoring and improve prediction accuracy.

[0036] Furthermore, the expression of the dynamic monitoring loss function of the mapping model is as follows:

[0037]

[0038] L Digital twin =λ 1 L accuracy +λ 2 L space

[0039]

[0040] Among them, θ Digital twin represents the parameters obtained by domain adaptive adversarial training of the mapping model, L Digital twin represents the loss function of the mapping model for domain adaptive adversarial training, θ represents the trainable parameters of the mapping model, and λ 1 and λ 2 Both represent weight parameters, which are used to balance the impact of accuracy loss and spatial variability loss. accuracy represents the accuracy loss function, L space represents the spatial variability loss function, N represents the number of finite element nodes, l represents the loss calculation method, i represents the cyclic calculation, and represents the spatial gradient between the predicted value of the mapping model and the numerical simulation result, α represents the loss balance parameter, Indicates the loss when the input is the sensor measurement point, It represents the loss when the input is the sensor point corresponding to the finite element node. represents the losses of all finite element nodes.

[0041] The technical effect brought about by the above further scheme is: the present invention realizes dynamic deep fusion of numerical simulation calculation results and sensor measurement results through the loss function of the mapping model.

[0042] Furthermore, the expression of the extreme working condition digital twin model is as follows:

[0043]

[0044] in, Indicates the parameters after fine-tuning the extreme working condition data parameters, θ δThe trainable parameters for fine-tuning parameters under extreme working conditions are denoted as, and L() represents the loss calculation method. Represents the predicted value of the digital twin model under extreme working conditions. Represents the calculated value of the finite element model under extreme working conditions. Represents the seepage prediction value of the measuring point monitoring model. Represents the deformation prediction value of the measuring point monitoring model, θ Digital twin Represents the parameters obtained through domain adaptation adversarial training under the time-series monitoring data of the mapping model, f Digital twin Represents the digital twin model.

[0045] The beneficial effect of the above further solution is that the present invention realizes the accurate extrapolation of the deformation and seepage conditions of the digital twin model under extreme working conditions by optimizing the parameter fine-tuning strategy under extreme working conditions.

[0046] Furthermore, the expression of the digital twin interaction model is as follows:

[0047]

[0048] J(θ) = |F s - f|

[0049] Where, P() represents the failure mode of the digital twin model, S represents the disposal method inferred from the high earth-rockfill failure knowledge graph, Represents the predicted value of the digital twin model under extreme working conditions, σ represents the activation function, W s and b s Represents the trainable parameters, θ t+1 Represents the optimized parameters, θ t Represents the parameter W of the digital twin interaction model s and b s Set, η represents the learning rate, J(θ t ) represents the loss function, F s Represents the real-time calculated dam slope stability safety factor, and f represents the recommended value of the dam slope stability safety factor related specifications.

[0050] The beneficial effect of the above further solution is that the present invention realizes the guidance of the digital twin model for the entity engineering treatment measures in the case of structural damage through the knowledge graph enhanced digital twin interaction model. Brief Description of the Drawings

[0051] Figure 1 Is the method flow chart of the present invention.

[0052] Figure 2 Is the layout diagram of the deformation and seepage measuring points of the 210 monitoring section of the PB Dam.

[0053] Figure 3 It is a schematic diagram of the process line of the measured values at the monitoring measurement points.

[0054] Figure 4 It is a table of the finite element model mesh division and zoning conditions.

[0055] Figure 5 It is a schematic diagram for comparing the deformation prediction accuracies of the proposed model, the finite element model, and the ablation model.

[0056] Figure 6 It is a schematic diagram for comparing the seepage prediction accuracies of the proposed model and the finite element model.

[0057] Figure 7 It is a schematic diagram for comparing the prediction accuracies of the monitoring model at the measurement points and the baseline model. Detailed implementation manners

[0058] The following describes the detailed implementation manners of the present invention to facilitate those skilled in the art of this technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed implementation manners. For those of ordinary skill in the art of this technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

[0059] Embodiment

[0060] The present invention aims to solve the technical problems of digital representation of physical models of high earth-rock dams and integration of monitoring information, and focuses on breaking through the deficiencies of existing methods in terms of real-time performance, accuracy, intelligent reasoning, and safety warning. First of all, the existing digital models have poor adaptability to environmental factors (such as water level changes, temperature effects, and aging effects), and it is difficult to accurately describe the seepage and deformation states of the dam body under complex working conditions. To address this issue, the present invention proposes a digital twin framework for high earth-rock dam structures based on deep learning, transfer learning, and physical constraints, enabling the model to have the ability to adapt to environmental changes and improving the real-time reflection and prediction accuracy of the dam body state. Secondly, there are defects in the fusion method of monitoring data and finite element simulation data, resulting in deviations in the mapping of the model to the true state of the dam body. In traditional methods, it is difficult to accurately correspond the node relationships calculated by finite element with the monitoring data, affecting the accuracy of dam safety assessment. The present invention uses a long short-term memory neural network (LSTM) with measuring point connections and a dynamic monitoring loss function, and combines the finite element method to establish a mapping model, thereby optimizing the fusion of monitoring data and numerical calculation and improving the data matching accuracy and calculation efficiency. In addition, there are deficiencies in the intelligent reasoning feedback between the digital twin model and the physical dam body in the existing technology, and it is difficult to effectively use the monitoring data to guide the dam maintenance and optimal operation. The present invention combines slope stability analysis and knowledge graph reasoning technology to establish an intelligent feedback mechanism, enabling the digital twin model to infer the structural safety state of the dam body based on real-time monitoring data and providing a scientific basis for engineering management and decision-making. Finally, the prediction and safety warning capabilities of the dam body stability under extreme working conditions (such as heavy rain) are limited, and traditional monitoring systems are difficult to provide reliable warning information. The present invention adopts a transfer learning method to enable the model to have the ability to update time-varying effects and extrapolate and predict extreme working conditions, thereby enhancing the dam safety warning ability and providing technical support for disaster prevention and mitigation. The main purpose of the present invention is to construct a digital twin framework for high earth-rock dams based on deep learning, finite element method, and knowledge graph reasoning, realizing high-precision real-time perception and prediction of the seepage and deformation states of the dam body. At the same time, the deep learning method integrating physical constraints is used to improve the matching degree of monitoring data and numerical calculation, ensuring the accurate mapping of the digital model to the real dam body. By integrating slope stability analysis and intelligent reasoning mechanisms, reliable assessment and warning for the dam structure safety are provided, and the disaster prevention and mitigation ability is enhanced. In addition, the time-varying update ability of the model is enhanced, enabling it to adapt to long-term working condition changes and having the ability to extrapolate and predict under extreme conditions, providing intelligent support for dam operation and maintenance.

[0061] As Figure 1 shown, the present invention provides a method for constructing a digital twin model of a high earth-rock dam, and the implementation method is as follows:

[0062] S1. Construct a measuring point monitoring model based on sensor monitoring data, and the implementation method is as follows:

[0063] Construct a graph convolutional feature extractor: Extract environmental impact factors and use them as input variables. Utilize GCN to adopt a feature extractor-decoder architecture to construct a graph convolutional feature extractor;

[0064] Seepage-deformation feature integration: Use the graph convolutional feature extractor to construct a seepage feature extractor and a deformation feature extractor respectively, extract environmental-seepage impact features and environmental-deformation impact features, and splice them to obtain seepage-deformation coupling features Among them, the seepage-deformation coupling features Adopt a shared trainable feature decoder to simultaneously output seepage monitoring prediction values and deformation monitoring prediction values

[0065] Physical feature constraint under extreme working conditions: Conduct supervised learning on the hydraulic load characteristics and soil creep characteristics of the earth-rock dam deformation feature extractor;

[0066] Time-varying effect model update: Complete the construction of the monitoring point model by sharing parameters and fine-tuning the feature decoder trained on the long-term monitoring data source task on the new sensor monitoring data set. Among them, the monitoring point model outputs the predicted values of the deformation monitoring effect quantity and the seepage pressure monitoring quantity.

[0067] In this embodiment, a graph convolutional feature extractor is constructed. First, extract environmental impact factors as input variables based on the HST model, and adopt a feature extractor-decoder architecture based on GCN.

[0068]

[0069] Among them, H 0 represents the environmental features input to the graph convolutional feature extractor, and represent the sets of hydraulic load impact factors, temperature impact factors, and aging impact factors at time t respectively. W (l) and W 0 both represent trainable weight matrices, H (l+1) represents the environmental impact factor feature matrix of the l-th layer nodes, σ represents the non-linear activation function, represents the normalized degree matrix, represents the impact factor adjacency matrix with self-connections added, H (l+1) and H (l)They respectively represent the environmental impact factor feature matrices of the nodes in the (l + 1)-th layer and the l-th layer. A represents the impact factor adjacency matrix, and I represents the identity matrix. The essence of this formula is to normalize and aggregate the information of the adjacent nodes of the impact factors, so that the impact factor node features of each layer not only depend on themselves but also on the information of their neighboring impact factor nodes. Among them, the output of Equation (1) is the input of Equation (2).

[0070] In this embodiment, for the integration of seepage-deformation characteristics: a graph convolution feature extractor is used to construct a seepage feature extractor and a deformation feature extractor respectively, and the environment-seepage impact features and the environment-deformation impact features are extracted. Then, feature fusion is achieved by means of splicing:

[0071]

[0072] Among them, represents the seepage-deformation coupling feature, represents the environment-seepage impact feature, represents the environment-deformation impact feature.

[0073] The seepage-deformation coupling feature uses a shared trainable feature decoder to simultaneously output the seepage monitoring prediction value and the deformation monitoring prediction value

[0074]

[0075] Among them, σ represents the non-linear activation function, W represents the trainable weight matrix, and b represents the bias.

[0076] In this embodiment, supervised learning is simultaneously performed on the hydraulic load characteristics and the soil creep characteristics of the earth-rock dam deformation feature extractor. The loss function of multi-teacher distillation in the process can be expressed as:

[0077] L MYD = αL task (l S , t) + (1 - α)L KD (p(f s , T), E(p(f T H , T), p(f T θ , T))) (5)

[0078] L KD = DF(p(f S , T), p(f T , T)) (6)

[0079] L = αL task +(1 - α)L KD (7)

[0080] p(f, T) = solfmax(f / T) (8)

[0081] Wherein, L MYD represents the loss function of the soil-rock deformation feature extractor, α represents the loss balance parameter, and L task represents the loss of the student network regression prediction of the soil-rock dam deformation feature extractor, l S represents the deformation predicted value of the student network, t represents the measured deformation value, and L KD represents the gap between the teacher network and the student network, p represents the distillation loss calculation method, and f s represents the student network feature, T represents the temperature parameter, E() represents the integrated output function of the teacher network, and f T H represents the hydraulic load feature of the teacher network, and f T θ represents the soil-rock creep feature of the teacher network, DF() represents the distance metric function, and f T represents the teacher network feature, f represents the extracted deformation feature, and L represents the knowledge distillation loss function calculation method.

[0082] In this embodiment, the time-varying effect model is updated: by sharing - fine-tuning the parameters of the feature decoder trained on the long-term monitoring data source task on the new recent monitoring data set, the prediction performance of the model is improved.

[0083] θ shared (W decoder , b decoder ) = θ pretrain (W decoder , b decoder ) (9)

[0084]

[0085] Wherein, θ pretrain represents the training parameter of the feature decoder, θ shared and θ update respectively represent the shared and updated parameters, f Cell boby (·) represents the measuring point monitoring model, L Recent monitoring data represents the loss function for updating the recent monitoring data task, W decoder and b decoder represent the trainable parameters, θ pretrain represents the obtained parameter, and θ represents the set of trainable parameters, Sets representing the hydraulic load influence factor, temperature influence factor, and aging influence factor at time t, respectively.

[0086] S2. Use a long short-term memory neural network (LSTM) framework with measurement point connections to learn the information transfer relationships between the seepage pressure or deformation values of the finite element nodes corresponding to the sensor measurement points, the actual sensor measurement values, and the seepage pressure or deformation values of their surrounding nodes, construct a mapping model, and construct a dynamic monitoring loss function for the mapping model;

[0087] In this embodiment, an LSTM framework with measurement point connections is used to learn the seepage pressure or deformation values of the finite element nodes corresponding to the sensor measurement points The actual sensor measurement value δ Monitoring point And the seepage pressure or deformation values of its surrounding nodes To obtain the mapping model of the information transfer relationship between them:

[0088]

[0089] θ δ =(W, b) (12)

[0090] Where Represents the seepage pressure value of the surrounding nodes of the finite element node, δ Monitoring point Represents the measurement value of the monitoring point, Model LSTM () Represents the mapping model constructed based on the long short-term memory neural network LSTM, Represents the calculated value of the finite element node corresponding to the sensor measurement point, ⊙ represents element-wise multiplication, W, b, and θ δ All represent trainable parameters, and [] represents feature fusion.

[0091] In this embodiment, a dynamic monitoring loss function is constructed. During the model training process, a dynamic monitoring loss function is used to adjust the learning of the mapping model for the association between the finite element mesh nodes and the actual sensor monitoring. The loss function is divided into an accuracy loss function L accuracy And a spatial variability loss function L space Two parts:

[0092]

[0093] L Digital twin =λ 1 L accuracy +λ 2 L space (14)

[0094]

[0095] Where θDigital twin denote the parameters obtained from the domain adaptation adversarial training of the mapping model, L Digital twin denote the loss function of the domain adaptation adversarial training of the mapping model, θ denote the trainable parameters of the mapping model, λ 1 and λ 2 both denote the weight parameters, which are used to balance the influence of the accuracy loss and the spatial variability loss, L accuracy denote the accuracy loss function, L space denote the spatial variability loss function, N denote the number of finite element nodes, l denote the loss calculation method, i denote the loop calculation, and denote the spatial gradient of the predicted value of the mapping model and the numerical simulation result, α denote the loss balance parameter, denote the loss when the input is the sensor measurement point, denote the loss when the input is the finite element node corresponding to the sensor measurement point, denote the loss of all finite element nodes.

[0096] S3. Input the predicted values of the deformation monitoring effect quantity and the seepage pressure monitoring quantity of the high earth-rock dam seepage process and deformation process obtained by the measurement point monitoring model into the mapping model, perform adversarial training using domain adaptation, combine the measurement point monitoring model to obtain the digital twin model, and construct the extreme condition digital twin interaction model based on the digital twin model;

[0097] In this embodiment, the digital twin model is constructed. The digital twin model can be obtained by inputting the predicted values of the high earth-rock dam seepage process and deformation process obtained by the measurement point monitoring model into the measurement point mapping model:

[0098]

[0099] In this embodiment, the extreme condition digital twin model is constructed. The generalization of the digital twin model is improved by simulating extreme conditions such as high water levels:

[0100]

[0101] Among them, denote the parameters after fine-tuning the data parameters of the extreme conditions, θ δ denote the trainable parameters for fine-tuning the parameters under extreme conditions, L() denote the loss calculation method, denote the predicted value of the digital twin model under extreme conditions, denote the calculated value of the finite element model under extreme conditions, denote the seepage prediction value of the measurement point monitoring model, denote the deformation prediction value of the measurement point monitoring model, θ Digitaltwin represents the parameters obtained by domain - adaptive adversarial training under the time - series monitoring data of the mapping model, f Digital twin represents the digital twin model.

[0102] S4. Construct a high - earth - rock dam inference dataset based on the knowledge graph, use the high - earth - rock dam inference dataset to train the digital twin interaction model for extreme conditions, and take the predicted values of the dam body's overall deformation monitoring effect quantity and seepage pressure monitoring quantity output by the digital twin model for extreme conditions as inputs. Then, use the trained digital twin interaction model for extreme conditions to output the treatment measures for the physical project of the cross - section, and complete the construction of the digital twin model for the high - earth - rock dam.

[0103] In this embodiment, the digital twin model interacts with the physical model by combining knowledge graph reasoning and the dam slope stability coefficient:

[0104]

[0105] J(θ)=|F s - f| (21)

[0106] where P() represents the failure mode of the digital twin model, S represents the disposal method obtained by reasoning in the high - earth - rock dam failure knowledge graph, represents the predicted value of the digital twin model under extreme conditions, σ represents the activation function, W s and b s represent the trainable parameters, θ t+1 represents the optimized parameters, θ t represents the parameters W of the digital twin interaction model s and b s set, η represents the learning rate, J(θ t ) represents the loss function, F s represents the real - time calculated dam slope stability safety factor, and f represents the recommended value of the dam slope stability safety factor in the relevant specifications.

[0107] The following further explains the present invention.

[0108] PB Hydropower Station is a large - scale hydropower project mainly for power generation, with additional benefits in flood control, sediment retention, and other comprehensive utilization aspects. The dam type is a gravel - filled soil - core rock - fill dam, with a dam crest elevation of 856m, a maximum dam height of 186m, a normal storage level of 850m, and a dead water level of 790m. As Figure 2As shown, the typical section of PB Dam at 0+210 was studied emphatically. A set of horizontal displacement gauges using tension wires were arranged at the downstream filter layer, transition layer and rockfill body at elevations of 808m, 758m and 731m. These gauges were numbered from CH1 to CH16, with a total of 16 measurement points, including deformation measurement points and seepage measurement points. Currently, there are three seepage pressure monitoring points on the downstream dam slope of this section, numbered P22, P29 and P32. The process line of the measured reservoir water level and the values of some deformation and seepage monitoring points are as Figure 3 shown Figure 3 In Figure 3 , Date represents time. A dataset was created using the deformation and seepage monitoring data of the 0+210 section of PB Dam from May 21, 2012 to May 23, 2024 and the finite element simulation data. The division ratios of training, validation and test data are shown in Table 1, which is a table of the dataset situation. To more accurately evaluate the performance of the proposed model, the following nine baseline models were selected for comparison: the classical statistical HST (water pressure - temperature - aging model), FEM (finite element model), five baseline deep learning models (long short - term memory neural network LSTM, GRU gated recurrent unit model, GCN graph convolutional neural network model, Transformer model, TGCN spatio - temporal graph convolutional model), a shallow machine learning model (SVM support vector machine model) and an ablation model. The ablation model only uses the long short - term memory neural network LSTM framework as the mapping model.

[0109] Table 1

[0110]

[0111] The digital twin model of the typical section of PB Dam at 0+210 was established using the present invention, and the specific steps are as follows:

[0112] (1) Establishment of the finite element model. In the finite element model, actual situations such as the dam zoning layout and the thick foundation overburden layer were considered. In addition, according to the results of on - site drilling tests, the finite element model divides the dam into upper and lower regions, with the dead water level of 790m as the boundary. The simulation range is 2 times the dam height upstream, downstream and below the dam foundation. The finite element model consists of 6572 elements and 13348 nodes. The finite element mesh and material division are as Figure 4As shown (Main rockfill area: the main body and main load-bearing structure of the dam, Core wall: the central impervious body of the earth-rock dam composed of materials such as clay, Secondary rockfill area: to protect the stability of the main rockfill area and the downstream dam slope, Overburden layer: loose accumulations and sediments of various origins covering the bedrock, Bedrock: the intact new mineral rock under the weathered layer on the surface of the continental crust). The PB dam is divided into the upstream dike area, transition layer, filter layer, gravelly soil core wall, downstream secondary dike area, and downstream main dike area. These design parameter values are shown in Tables 2 and 3, while the designed permeability coefficients of each partition of the PB dam are shown in Table 4. Table 2 is the parameter table of the seven-parameter creep model of the PB dam, Table 3 is the parameter table of the Duncan-Chang E-B constitutive model of the PB dam, and Table 4 is the permeability coefficient table of the PB dam.

[0113] Table 2

[0114]

[0115] Table 3

[0116]

[0117] Table 4

[0118]

[0119]

[0120] (2) Establishment of the monitoring model for measuring points. Build a monitoring model for measuring points. The inputs are the hydraulic load factors (H, H 2 , H 3 , H 4 ) constructed based on the HST (Hydraulic Pressure-Temperature-Aging Model), the temperature factor and the aging factor (θ, ln(θ), e -θ ). The outputs are the predicted values of the deformation monitoring effect quantities numbered CH1-CH16 and the seepage pressure monitoring quantities numbered P10, P22, P29, and P32. Training is carried out using the Pycharm IDE (Python integrated development environment) and Pytorch 2.5.1 (open-source deep learning framework) based on Python 3.10 under the Windows 10 system environment. The server configuration is as follows: the CPU is an Intel Core i7-13700, and the GPU is an Nvidia RTX 4070 SUPER. The learning rate is set to 0.001, the batch size is 64, the Adam optimizer is selected to update the network parameters, the number of iterations is set to 500, the model hidden layer is 64, the graph convolutional layer is 2, and the multi-layer perceptron layer is 2. The comparison with the baseline model is as Figure 7 shown, Figure 7Among them, MAE represents the mean absolute error, RMSE represents the root mean square error, and MAPE represents the mean absolute percentage error.

[0121] (3) Mapping model establishment. The inputs for constructing the mapping model are the deformation monitoring effect quantities numbered CH1-CH16, the seepage pressure monitoring quantities numbered P10, P22, P29, and P32, and their corresponding finite element node calculated values, and the output is the predicted values of all finite element nodes. During the domain adaptation adversarial training process, the loss function is divided into two parts: spatial variability loss and accuracy loss (the loss of predicting the actual monitoring quantities, the loss of predicting the calculated values of the nodes corresponding to the layout measurement points, and the loss of predicting the calculated values of all finite element nodes), to dynamically adjust the comprehensive learning of the actual monitoring situation and the finite element calculation situation. Training is carried out using Pycharm IDE (Python integrated development environment) and Pytorch 2.5.1 (open-source deep learning framework) based on Python 3.10 under the Windows 10 system environment. The server configuration is as follows: the CPU is Intel Core i7-13700, and the GPU is Nvidia RTX 4070 SUPER. The learning rate is set to 0.001, the batch size is 64, the Adam optimizer is selected to update the network parameters, the number of iterations is set to 500, and the model hidden layer is 64.

[0122] (4) Construct the digital twin model for extreme working conditions. Input the predicted values of the deformation monitoring effect quantities CH1-CH16 and the seepage pressure monitoring quantities P10, P22, P29, and P32 output by the measurement point monitoring model into the measurement point mapping model, and output the deformation and seepage processes of the cross-section. The comparison with the finite element model and the ablation model is Figure 5 and Figure 6 as shown.

[0123] (5) Construct the digital twin interaction model. Based on the knowledge graph, construct an inference dataset for high earth-rock dams (digital twin state - high earth-rock dam failure - treatment measures). Use the predicted values of the dam body's global deformation monitoring effect quantities and seepage pressure monitoring quantities output by the digital twin model as inputs, and output the treatment measures for the physical project of the cross-section (such as adding drainage facilities, correcting the dam slope angle, water level regulation measures, etc.).

[0124] In this example, the digital twin framework of high earth-rock dams is applied to high earth-rock dam projects. This method can effectively realize the digital twin of the deformation-seepage coupling process of high earth-rock dams, and the average error between the predicted deformation and the actual monitoring data is only 3.17%. Compared with the traditional numerical simulation method, it can significantly improve the simulation accuracy, and the deformation simulation accuracy is increased by 64.42%. In addition, the proposed dynamic monitoring loss method can effectively improve the fusion prediction accuracy of the finite element model and the actual monitoring data, and the deformation prediction accuracy is increased by 32.55%. Compared with the baseline model, the proposed monitoring point prediction model achieves the best prediction accuracy and better generalization ability. The results show that compared with the ablation model, the proposed deformation-seepage coupling prediction method can effectively improve the prediction accuracy of the monitoring point model by 19.44%. This method can more accurately and real-timely reflect the deformation and seepage processes of high earth-rock dams.

[0125] In summary, the current deformation monitoring of high earth-rock dams mainly relies on measurement point models based on statistical methods or shallow machine learning. These models have limitations in capturing the relationship between complex environmental variables and dam deformation characteristics and time characteristic modeling, and it is difficult to comprehensively reflect the operation state of high earth-rock dams. In addition, although deep learning models have powerful non-linear modeling capabilities, their generalization ability for unknown extreme working conditions is poor, and it is difficult to meet the requirements of real-time monitoring of high earth-rock dams. The present invention first proposes a digital twin framework applicable to high earth-rock dams to solve the limitations of existing monitoring methods. The framework combines graph convolutional neural networks, transfer learning, LSTM, and finite element methods to establish a measurement point model, a measurement point mapping model, an interaction model, and an extrapolation prediction model for extreme working conditions that integrate seepage and deformation monitoring, so as to realize the real-time perception, prediction, and early warning of the deformation and seepage states of high earth-rock dams. The framework can dynamically update the monitoring model, enhance its extrapolation ability for unknown working conditions, and further evaluate the structural safety and intelligently generate treatment measures through slope stability analysis. Compared with the traditional numerical simulation method, the proposed method improves the simulation accuracy by integrating the actual monitoring information of sensors, avoids a large amount of numerical calculations, has real-time performance, and improves the timeliness. Compared with the model based on sensor measurement points, its prediction accuracy and generalization ability are excellent. This method realizes a breakthrough transformation from the traditional "point monitoring" of high earth-rock dam structures to "global monitoring", can more accurately and real-timely reflect the surface deformation and seepage processes of high earth-rock dams, and can use mechanism methods such as slope stability analysis to judge the structural safety in real time, providing an economical and efficient intelligent monitoring solution for actual projects and having broad application prospects.

Claims

1. A method for constructing a digital twin model of a high earth-rock dam, characterized in that: The following steps are involved: S1. Construct a monitoring model of measuring points based on sensor monitoring data; S2. Using the LSTM framework of the long short-term memory neural network with measuring point connections, learn the information transmission relationship between the finite element node seepage pressure or deformation value corresponding to the sensor measuring point, the actual monitoring value of the sensor and the seepage pressure or deformation value of the surrounding nodes, respectively, build a mapping model, and build a dynamic monitoring loss function of the mapping model; S3. Input the predicted values ​​of the deformation monitoring effect and seepage pressure monitoring quantity of the high earth-rock dam seepage process and deformation process of the measuring point monitoring model into the mapping model, use domain adaptation for adversarial training, combine the measuring point monitoring model to obtain the digital twin model, and build the digital twin interaction model of extreme working conditions based on the digital twin model; S4. Construct a high earth-rock dam reasoning dataset based on the knowledge graph, use the high earth-rock dam reasoning dataset to train the extreme working condition digital twin interaction model, and use the dam body global deformation monitoring effect and seepage pressure monitoring prediction value output by the extreme working condition digital twin interaction model as input. Use the trained extreme working condition digital twin interaction model output to obtain the cross-section treatment measures for the physical project, and complete the construction of the high earth-rock dam digital twin model.

2. The method for constructing a digital twin model of a high earth-rock dam according to claim 1, characterized in that: The S1 comprises the following steps: Construct a graph convolutional feature extractor: extract environmental influencing factors and use them as input variables, use GCN to adopt a feature extractor-decoder architecture, and construct a graph convolutional feature extractor; Infiltration-deformation feature integration: Use graph convolution feature extractor to construct infiltration feature extractor and deformation feature extractor respectively, extract environment-infiltration influence features and environment-deformation influence features, and then perform splicing to obtain infiltration-deformation coupling features. Among them, the seepage-deformation coupling characteristics Use a shared trainable feature decoder to simultaneously output seepage monitoring predictions and deformation monitoring prediction value Among them, σ represents a nonlinear activation function, W represents a trainable weight matrix, and b represents a bias; Physical feature constraints of extreme working conditions: supervised learning of hydraulic load characteristics and soil-rock creep characteristics of the earth-rock dam deformation feature extractor. The loss function of multi-teacher distillation in the process of soil-rock deformation feature extractor is expressed as follows: L MYD =αL task (l S ,t)+(1-α)L KD (p(f s ,T),E(p(f T H ,T),p(f T θ ,T))) L KD =DF(p(f S ,T),p(f T ,T)) L=αL task +(1-α)L KD p(f,T)=solfmax(f / T) Among them, L MYD represents the loss function of soil-rock deformation feature extractor, α represents the loss balance parameter, and L task represents the loss of the student network regression prediction of the earth-rock dam deformation feature extractor, l S represents the predicted value of the student network deformation, t represents the measured value of the deformation, and L KD represents the gap between the teacher network and the student network, p represents the distillation loss calculation method, and f s represents the characteristics of the student network, T represents the temperature parameter, E() represents the integrated output function of the teacher network, and f T H represents the hydraulic load characteristics of the teacher network, f T θ represents the soil-rock creep characteristics of the teacher network, DF() represents the distance metric function, and f T represents the teacher network features, f represents the extracted deformation features, and L represents the calculation method of the knowledge distillation loss function; Time-varying effect model update: The construction of the measurement point monitoring model is completed by parameter sharing-fine-tuning the feature decoder trained on the long-term monitoring data source task on the new sensor monitoring dataset. The measurement point monitoring model outputs the predicted values ​​of the deformation monitoring effect quantity and the seepage pressure monitoring quantity.

3. The method for constructing a digital twin model of a high earth-rock dam according to claim 2, characterized in that: The expression of the graph convolutional feature extractor is as follows: Among them, H 0 represents the environmental features input to the graph convolutional feature extractor, and represents the set of hydraulic load influence factors, temperature influence factors and time-effect influence factors at time t, respectively, W (l) and W 0 Both represent trainable weight matrices, H (l+1) represents the characteristic matrix of environmental impact factors of the l-th layer nodes, σ represents the nonlinear activation function, represents the normalized degree matrix, represents the adjacency matrix with the impact factor of self-connection, H (l+1) and H (l) They represent the environmental impact factor feature matrices of the l+1th layer nodes and the lth layer nodes respectively, A represents the impact factor adjacency matrix, and I represents the unit matrix.

4. The method for constructing a digital twin model of a high earth-rock dam according to claim 2, characterized in that: The expression of the seepage-deformation coupling characteristic is as follows: in, represents the seepage-deformation coupling characteristics, Represents the environmental-seepage influence characteristics, Represents the environment-deformation influence characteristics.

5. The method for constructing a digital twin model of a high earth-rock dam according to claim 1, characterized in that: The expression of the mapping model is as follows: i δ =(W,b) in, represents the seepage pressure value of the nodes around the finite element node, δ Monitoringpoint Indicates the measurement value of the monitoring point, Model LSTM () represents a model built based on the long short-term memory neural network LSTM, represents the calculated value of the finite element node corresponding to the sensor measurement point, ⊙ represents the element multiplication, W, b and θ δ All represent trainable parameters, and [] represents feature fusion.

6. The method for constructing a digital twin model of a high earth-rock dam according to claim 5, characterized in that: The expression of the dynamic monitoring loss function of the mapping model is as follows: Among them, θ Digitaltwin represents the parameters obtained by domain adaptive adversarial training of the mapping model, L Digitaltwin represents the loss function of the mapping model for domain adaptive adversarial training, θ represents the trainable parameters of the mapping model, λ1 and λ2 represent weight parameters, which are used to balance the impact of accuracy loss and spatial variability loss, and L accuracy represents the accuracy loss function, L space represents the spatial variability loss function, N represents the number of finite element nodes, l represents the loss calculation method, i represents the cyclic calculation, and represents the spatial gradient between the predicted value of the mapping model and the numerical simulation result, α represents the loss balance parameter, Indicates the loss when the input is the sensor measurement point, It represents the loss when the input is the sensor point corresponding to the finite element node. represents the losses of all finite element nodes.

7. The method for constructing a digital twin model of a high earth-rock dam according to claim 1, characterized in that: The expression of the extreme working condition digital twin model is as follows: in, Indicates the parameters after fine-tuning the extreme working condition data parameters, θ δ represents the trainable parameters for fine-tuning parameters under extreme working conditions, L() represents the loss calculation method, represents the predicted value of the digital twin model under extreme conditions, represents the calculated value of the finite element model under extreme conditions, represents the seepage prediction value of the monitoring model at the measuring point, represents the deformation prediction value of the monitoring model of the measuring point, θ Digitaltwin represents the parameters obtained by domain adaptive adversarial training under the time series monitoring data of the mapping model, f Digitaltwin Represents a digital twin model.

8. The method for constructing a digital twin model of a high earth-rock dam according to claim 1, characterized in that: The expression of the digital twin interaction model is as follows: J(θ)=|F s -f| Among them, P() represents the failure mode of the digital twin model, S represents the treatment method obtained by reasoning in the knowledge graph of high soil and rock failure, represents the predicted value of the digital twin model under extreme conditions, σ represents the activation function, and W s and b s represents the trainable parameters, θ t+1 represents the optimized parameters, θ t Parameter W representing the digital twin interaction model s and b s set, η represents the learning rate, J(θ t ) represents the loss function, F s It represents the real-time calculated dam slope stability safety factor, and f represents the recommended value of the dam slope stability safety factor in relevant specifications.

Citation Information

Patent Citations

  • Pixel-level PM2.5 (Particulate Matter 2.5) prediction method based on land change prediction and space lag model

    CN116402199A

  • Dam deformation monitoring method, device and equipment based on knowledge graph-multiple monitoring points and medium

    CN116429055A

  • Spatial variation slope reliability analysis method and system based on machine learning algorithm

    CN117521482A

  • Dam safety monitoring system and method based on digital twinning

    CN119624146A

  • Method for simulating temperature field of distributed underground facility in mountain mass

    US20160364509A1

Cited By

  • Pick body gradient heat treatment method and system based on stress nephogram

    CN120874475A

  • A Gradient Heat Treatment Method and System for Cutting Tooth Body Based on Stress Cloud Diagram

    CN120874475B

  • Rock-fill dam deformation digital twinborn body construction method based on generative AI

    CN121031235A

  • Rockfill dam deformation digital twin construction method based on generative AI

    CN121031235B

  • Dam bending monitoring section layout method and device

    CN121475131A