Rockfill dam deformation digital twin construction method based on generative AI

Abstract

The application discloses a kind of rock-fill dam deformation digital twin construction methods based on generative AI, core lies in using conditional denoising diffusion probability model, through no classifier guide and continuous condition vector construction conditional sampling mechanism, efficiently fusion finite element simulation data, monitoring data and operation data.Further combined with residual neural network ResNet-18 and K-means clustering method, finite element data is unsupervised classification, introduce combination optimization, identify and monitor data most matching finite element data category, to realize condition guide deformation field generation.The framework has been applied and verified on the world's highest two estuary rock-fill dam (303 meters) at present.The results show that the proposed rock-fill dam deformation digital twin construction method based on generative AI can efficiently reconstruct rock-fill dam deformation, has higher precision and real-time performance, significantly improves the global deformation of rock-fill dam thorough perception ability, provides key technical support for its safe operation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water conservancy and geotechnical engineering, and particularly relates to a rock-fill dam deformation digital twin construction method based on generative AI. BACKGROUND

[0002] Rock-fill dams are an important type of large-scale hydraulic structure, and have become one of the widely used dam types due to their outstanding economy and adaptability. With the continuous development of rock-fill dam construction technology, the dam height has fully entered the 300-meter level, and the structure scale is huge, so the safety operation risk management is crucial. Among them, dam body deformation control is one of the key issues to ensure the safety of the whole life cycle of the rock-fill dam, and is highly concerned in various stages such as design, construction, operation and maintenance. The use of various monitoring technologies for in-situ deformation monitoring is an important basis for deformation control. However, due to factors such as sensor layout density, burial depth, environmental interference and instrument aging, the deformation monitoring data often presents the phenomena of sparse spatial distribution, discontinuity, and even failure, which seriously restricts the comprehensive perception and analysis of the true deformation state of the dam body. In addition, although traditional finite element analysis can analyze the overall deformation of the dam, it relies on parameter inversion, has low calculation efficiency and is sensitive to initial conditions, and it is difficult to meet the engineering needs of dynamic perception in the operation period.

[0003] In recent years, digital twinning has provided more and more intelligent and data support for the safety management of water conservancy engineering, and with the development of artificial intelligence, especially generative artificial intelligence (Generative AI), generative models have been gradually introduced into the field of structural health monitoring and physical field reconstruction to enhance the prediction capability. Although existing researches have adopted deep learning methods for dam body deformation prediction or anomaly identification, most of the models are limited to single-point prediction of existing monitoring points, and it is difficult to realize comprehensive reconstruction of the two-dimensional or even three-dimensional deformation field of the dam body, and the robustness to noise and uncertainty of monitoring data is lacking.

[0004] In summary, how to reconstruct the deformation field of the dam body with high precision and high efficiency under the condition of sparse monitoring data, by integrating the finite element physical prior information and the engineering operation data, to improve the comprehensive perception ability of the deformation of the rock-fill dam, has become a key technical problem that needs to be broken through in the current safety monitoring field of rock-fill dams. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the present application provides a rock-fill dam deformation digital twin construction method based on generative AI, which significantly improves the accuracy, efficiency and physical consistency of deformation reconstruction by using generative diffusion modeling, category guiding mechanism and continuous condition vector joint control generation process.

[0006] According to one aspect of the present application, a rock-fill dam deformation digital twin construction method based on generative AI is provided, comprising:

[0007] constructing a finite element numerical model based on the deformation monitoring data and operation data of the rockfill dam, and obtaining multiple sets of deformation field simulation samples by using a parameter sampling strategy;

[0008] Based on the obtained deformation field simulation samples, two-dimensional deformation field data of key monitoring sections are extracted, and deep features of two-dimensional deformation field images are extracted through a deep neural network, and unsupervised clustering methods are used for classification and label construction of the simulation samples;

[0009] A conditional denoising diffusion probability model is constructed, and the classified two-dimensional deformation field simulation samples, corresponding class labels, and operation data used during simulation are used as conditional inputs for model training, and a controllable generated deformation field reconstruction model is output;

[0010] Combined with actual deformation monitoring data and corresponding section simulation data, a combined optimization framework is constructed to search for the optimal data category, and the searched optimal data category and operation data are jointly input into the controllable generated deformation field reconstruction model to output the reconstructed rockfill dam deformation field.

[0011] As a further technical solution, based on the obtained deformation field simulation samples, after extracting the two-dimensional deformation field data of the key monitoring sections, it further includes:

[0012] The extracted two-dimensional deformation field data is uniformly interpolated and mapped to a fixed resolution image matrix, and the deformation values are converted to single-channel grayscale images through standardization processing.

[0013] As a further technical solution, the deep features of the two-dimensional deformation field images are extracted through a deep neural network, including:

[0014] A pre-trained ResNet-18 network is used as the backbone structure, the end full connection classification layer is removed, and only the convolution layer and global average pooling layer are retained to extract the deep feature vector of each two-dimensional deformation field image.

[0015] As a further technical solution, the unsupervised clustering method is used for classification and label construction of the simulation samples, including:

[0016] The deep feature vectors corresponding to all two-dimensional deformation field images are input into the K-means clustering algorithm, and unsupervised clustering operations are performed to classify the simulation samples and assign unique class labels to each simulation sample.

[0017] As a further technical solution, the training of the conditional denoising diffusion probability model includes:

[0018] Gaussian noise is gradually added to the simulation deformation field data of each section of the rockfill dam through a forward diffusion process to obtain noisy images;

[0019] inputting the noisy image into the U-Net network, training the U-Net network to predict a noise component added to an original two-dimensional deformation field image;

[0020] reconstructing the original deformation field image by gradually removing the noise in the back diffusion process by using the trained U-Net network.

[0021] As a further technical solution, the method further comprises:

[0022] encoding the discrete category label into a high-dimensional vector through a fully connected layer, and using a category guiding mechanism to randomly mask the category label with a set probability in the training stage;

[0023] transforming the running data into a feature representation through a fully connected layer, and embedding and fusing the category label under the category guiding mechanism, to jointly guide the decoding path of the U-Net network.

[0024] As a further technical solution, the actual deformation monitoring data and the simulation data corresponding to the section are combined to construct a combination optimization framework, to search for an optimal data category, and the searched optimal data category and the running data are jointly input into a controllable generated deformation field reconstruction model, including:

[0025] The multi-category data extracted from the finite element simulation sample is constructed into a candidate set, and a combination optimization model is introduced to select several categories closest to the measured deformation value as the conditional generation guide.

[0026] A genetic algorithm is used to optimize and solve the combination optimization model to obtain the optimal data category.

[0027] The label of the optimal data category and the running data are jointly constructed into a conditional vector, which is jointly input into the trained conditional denoising diffusion probability model for conditional sampling to obtain a generative deformation field result consistent with the measured data.

[0028] According to an aspect of the present application, a rock-fill dam deformation digital twin construction system based on generative AI is provided for implementing the method, and the system comprises:

[0029] A first main module is configured to construct a finite element numerical model based on rock-fill dam deformation monitoring data and running data, and to obtain a plurality of deformation field simulation samples by using a parameter sampling strategy;

[0030] A second main module is configured to extract two-dimensional deformation field data of a key monitoring section based on the obtained deformation field simulation samples, to extract deep features of the two-dimensional deformation field image by using a deep neural network, and to classify and label the simulation samples by using an unsupervised clustering method;

[0031] The third main module is used for constructing a conditional denoising diffusion probability model, and taking the classified two-dimensional deformation field simulation sample, the corresponding category label and the running data used in the simulation as a conditional input to train the model, and outputting a controllable generated deformation field reconstruction model;

[0032] The fourth main module is used for combining the actual deformation monitoring data and the simulation data of the corresponding section to construct a combined optimization framework, searching for an optimal data category, and inputting the searched optimal data category and the running data into the controllable generated deformation field reconstruction model to output a reconstructed rock-fill dam deformation field.

[0033] According to an aspect of the present application, a rock-fill dam deformation digital twin construction device based on generative AI is provided, comprising a memory and a processor, the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the rock-fill dam deformation digital twin construction method based on generative AI.

[0034] According to an aspect of the present application, a non-transitory computer readable storage medium is provided, which stores computer instructions, and the computer instructions make the computer execute the rock-fill dam deformation digital twin construction method based on generative AI.

[0035] Compared with the prior art, the present application has the following advantages:

[0036] (1) The present application supports the full-area deformation thorough perception of the rock-fill dam under the condition of sparse monitoring data. The present application fuses finite element simulation data, in-situ monitoring records and running data, breaks through the limitation of the traditional model which is strongly dependent on monitoring points and weak in spatial reconstruction capability, and can still realize the high-precision reconstruction of the two-dimensional deformation field under the condition of limited monitoring data. The reconstruction result of the dam deformation field not only has a high degree of match with the measured data in the amplitude, but also has good physical consistency in the deformation space-time, thereby supporting the safety evaluation of the dam body.

[0037] (2) The present application supports the dynamic real-time reconstruction of the rock-fill dam deformation field. Compared with the traditional parameter inversion finite element method which needs to consume several hours for single simulation, the model of the present application can complete the generation of the deformation field in several seconds after being trained, and the reconstruction efficiency is improved by more than four orders of magnitude, thereby having the ability of real-time monitoring and rapid response.

[0038] (3) The present application has high robustness and strong generalization capability. The generative diffusion model training mechanism is adopted, which has stable noise modeling and prediction capability, and through the combined optimization strategy, the deformation field reconstruction under different monitoring data inputs is realized, so that reliable rock-fill dam deformation field can still be generated under the condition of reduced number of monitoring points, thereby enhancing the generalizability and practical value of the method in actual rock-fill dam engineering. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 A flowchart of a rockfill dam deformation digital twin construction method based on generative AI provided by the embodiments of the present application;

[0041] Figure 2 An application process diagram of the DDPM used by the embodiments of the present application in the rockfill dam deformation field prediction: (a) rockfill dam model example and core wall cross-section deformation field; (b) deformation field prediction using DDPM;

[0042] Figure 3 A CDDPM model structure diagram for the reconstruction of the rockfill dam deformation field of the embodiments of the present application;

[0043] Figure 4 A related situation diagram of the Lihe estuary rockfill dam applied by the embodiments of the present application: (a) aerial view of the dam; (b) BIM model; (c) construction-operation process and material partition;

[0044] Figure 5 A process diagram for generating a cross-section deformation field by a 200-step gradual denoising generation using a U-Net network when the CDDPM is used by the embodiments of the present application;

[0045] Figure 6 A deformation field reconstruction situation diagram of the core wall and the 3-3 cross-section at a typical time by the embodiments of the present application: (a) iteration process of the objective function in the combination optimization; (b) deformation field reconstructed by the CDDPM;

[0046] Figure 7 A part time consumption statistical diagram when the embodiments of the present application are applied;

[0047] Figure 8 A statistical analysis diagram of the cross-section deformation field reconstruction results without using the CFG guide condition by the embodiments of the present application: (a) average value of the 3-3 cross-section; (b) average value of the core wall cross-section; (c) median of the 3-3 cross-section; (d) median of the core wall cross-section; (e) maximum value of the 3-3 cross-section; (f) maximum value of the core wall cross-section;

[0048] Figure 9 An SSIM index heat map under the condition of gradually reducing the measuring points by the embodiments of the present application. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0050] This invention provides a method for constructing a deformable digital twin of a rockfill dam based on generative AI, which improves the accuracy of rockfill dam simulation. It mainly consists of four parts: dataset preparation, finite element simulation data preprocessing, CDDPM model training, and CDDPM conditional sampling. (See reference...) Figure 1 The specific steps include:

[0051] Step S1: Collect finite element deformation simulation data, operational data, and deformation monitoring data of the rockfill dam profile. The finite element simulation data is obtained through multiple finite element analyses. The operational data includes the construction and operation duration T and the reservoir water level. H R Deformation monitoring data is obtained from various monitoring equipment and technologies used in rockfill dam projects.

[0052] In step S1, the acquisition of deformation monitoring data, operational data, and finite element simulation data includes:

[0053] Step S1.1: Various monitoring equipment and technologies, including internal deformation monitoring equipment such as water pipe settling meters, electromagnetic settling rings, and flexible inclinometers, and external monitoring technologies such as measuring robots, GNSS monitoring, and InSAR, are used to obtain in-situ deformation monitoring data.

[0054] Step S1.2: Water level change data and construction and operation duration are typically provided by the rockfill dam project management unit. Both types of data are used not only in finite element numerical simulation analysis but also in subsequent model training and condition sampling processes.

[0055] Step S1.3, the acquisition of finite element simulation data is obtained by constructing a finite element model and calculating by using a suitable constitutive model and analysis software. This step constructs a finite element model of the rockfill dam, and applies a numerical sampling method to generate a plurality of sets of constitutive model parameter combinations, which are substituted into the finite element model of the rockfill dam for numerical simulation to obtain a large amount of finite element simulation data.

[0056] Step S1.3 further comprises the following steps:

[0057] (1) According to the geometric structure, material partition, construction condition and boundary condition of the rockfill dam, a three-dimensional finite element model is constructed. The element type used can be a three-dimensional solid element (such as C3D8), and the finite element analysis software used includes but is not limited to ABAQUS, ANSYS, etc.;

[0058] (2) Select a constitutive model suitable for describing the deformation behavior of rockfill materials, such as Duncan-Chang E-B model for simulating nonlinear elastic response, creep model for characterizing long-term effect, and wetting deformation model for describing volume change behavior under water action, etc.;

[0059] (3) A plurality of representative constitutive parameter combinations are generated by using a parameter design method such as Latin Hypercube Sampling (LHS), and are substituted into the finite element model for simulation calculation one by one to reflect the influence of material parameter uncertainty on the deformation field of the dam body;

[0060] (4) According to the actual filling and water storage process, a plurality of loading steps are set to simulate the whole process deformation response of the rockfill dam from filling to operation. The simulation process can include the dam body response at a plurality of typical time points (such as construction completion, initial water storage, maximum water storage, etc.), and the two-dimensional deformation results of the key sections are extracted;

[0061] (5) Each set of simulation results is mapped to the two-dimensional deformation field data under the regular grid, and global normalization processing is performed to construct a standardized finite element simulation sample set required for training and generating a model.

[0062] Step S2, the simulation data is preprocessed. The two-dimensional deformation field data of the key monitoring section is extracted first, then the deep feature vector thereof is extracted by using a ResNet-18 residual neural network, and the simulation data is classified by using a K-means clustering method to obtain a two-dimensional deformation field simulation sample with a class label.

[0063] In the step S2, the simulation data is preprocessed, mainly including unsupervised deep clustering and data classification, and the main steps are as follows:

[0064] Step S2.1, image format standardization. The extracted two-dimensional deformation field data includes the horizontal and vertical coordinates of each nodex , y ) and the corresponding deformation value (D δ ), to form a two-dimensional deformation field data (D x , y , δ ). The extracted two-dimensional deformation field data is uniformly interpolated and mapped to a fixed resolution image matrix (such as 128x128), and a standardization process (such as min-max normalization or [-1, 1] interval standardization) is used to convert the deformation value into a single-channel grayscale image, facilitating subsequent neural network input processing.

[0065] Step S2.2, deep feature extraction. The above two-dimensional deformation field image is extracted by using ResNet-18 residual neural network. The specific operation is as follows: taking the pre-trained ResNet-18 network as the backbone structure, removing the end full connection classification layer, and only retaining the convolution layer and global average pooling layer (Global Average Pooling) to extract the deep feature vector of each image, representing its deformation structure feature.

[0066] Step S2.3, unsupervised clustering classification. The deep feature vector corresponding to all two-dimensional deformation field samples is input into the K-means clustering algorithm to perform unsupervised clustering operation and classify the simulation samples. The number of cluster centers K can be set according to the sample diversity and training requirements (such as K = 50). The clustering result will assign a unique class label to each simulation sample for subsequent generative model training.

[0067] Step S2.4, data consistency check. Ensure that the samples of each class have certain representativeness and diversity in spatial form and deformation amplitude, and balance the number of class samples if necessary to improve the stability and generalization ability of model training.

[0068] Step S3, the classified two-dimensional deformation field simulation data, corresponding class label and running data used in simulation are collectively used as input to train the conditional denoising diffusion probability model CDDPM, so as to obtain the controllable generated deformation field reconstruction model.

[0069] In this step, in order to make the generated deformation field closer to the real deformation, CFG (Classifier-Free Guidance, Classifier-Free Guidance mechanism) and continuous condition vector are introduced, and multi-source monitoring data and running data are used as the generation conditions of the model to guide the model to generate a deformation field more consistent with the actual measurement. The finite element data class label obtained by deep clustering and the running data of the corresponding time step are used as the condition input for training.

[0070] In step S3, CDDPM (Conditional Denoising Diffusion Probabilistic Model) training is performed, and the main steps include:

[0071] Step S3.1, Model Input Construction: The following three types of data will be used as model input:

[0072] (1) Two-dimensional deformation field simulation image: derived from the deformation field simulation sample after standardization in step S2, and represented as a single-channel grayscale image with a fixed resolution (e.g., 128×128);

[0073] (2) Category label: The cluster category number assigned to each simulation sample, using one-hot or multi-hot encoding as discrete condition input;

[0074] (3) Operational data: including total construction and operation time T Current water level H R These are combined to form a continuous conditional vector, which serves as a representation of the actual running data.

[0075] Step S3.2, Model Structure Design. This method uses CDDPM as the core model for generating the deformation field of the rockfill dam. Its core idea is to gradually add noise to the original data through a forward diffusion process and train a neural network model to gradually remove noise during the back diffusion process, thereby reconstructing the original data. The model consists of two main parts: the forward diffusion process and the parameter back diffusion process.

[0076] (1) Forward diffusion process: In this process, Gaussian noise is gradually added to the simulated deformation field data of each section of the rockfill dam. This process constitutes a Markov chain, that is, the deformation field data is gradually diffused into pure Gaussian noise. Specifically, from a uniformly distributed... U (1, 2,…, T Sample one time step in ) t and from the standard normal distribution N ( 0 , I Sampling in) t Sub-Gaussian noise ε 1, ε 2,…, ε t Add them sequentially to the initial image. x 0~ q ( x 0), of which x 0 represents the original two-dimensional deformation field image of a certain cross-section of the rockfill dam, from which the noisy image is finally obtained. x tThe image is then input into the U-Net network structure, which is trained to predict the noise component added to x 0 ε . The denoising objective function is:

[0077]

[0078] where x 0 represents the clean deformation field image at the initial time (t = 0), t x t represents the diffusion t step after adding the noise version, β t is a fixed scalar parameter of the diffusion intensity, with a value range of (0, 1), I represents the covariance matrix in the standard normal distribution N(0, I ).

[0079] (2) Parameterized reverse diffusion process: The reverse diffusion process is a process of gradually restoring the deformation field image from the Gaussian noise image. Specifically, the process starts with a pure noise image N 0 , I ) sampled from the standard normal distribution x T . At each diffusion step t , the current noisy image x t is input into the U-Net network to predict its corresponding Gaussian noise component , while a noise term N 0 , I ) is resampled and the next time image z x t-1 is calculated using the reparameterization formula. Through multiple iterations, the target deformation field image x 0 is finally generated, and the process can be represented as:

[0080]

[0081] where ε t represents the noise component predicted by the neural network, and the entire process is equivalent to learning the mean vector μ x t , t ) and the covariance matrix Σ( x t , t ), parameter α ​​​​​t =1- β indicates the first t step reserved original image ratio, which reflects the overall original information retention degree in the cumulative product of the previous t step.

[0082] Step S3.3, conditional information embedding. Discrete category labels are encoded into high-dimensional vectors through a fully connected layer, and CFG mechanism is used to randomly mask category labels with a certain probability in the training stage c cat ∈{0, 1} C , that is:

[0083]

[0084] Further, the model realizes flexible learning and generalization control of conditional information, guiding the generative model to learn different deformation modes.

[0085] Convert the continuous type condition vector T , H R into a feature representation, embed it into the neural network through an independent fully connected layer, and use it together with the category label embedding. Conditional CDDPM learns to predict noise under conditional and unconditional settings. When CFG conditions are available during inference, the U-Net will predict noise ε θ ( x t , c CFG ), while when it is shielded, it will predict ε θ ( x t ). During the sampling process, the two predictions are combined in a linear combination as follows:

[0086]

[0087] In the formula, ω represents the condition guiding strength, which can flexibly control the influence of CFG conditions in the generation process, and balance the diversity and fidelity of the reconstructed deformation field.

[0088] In addition, the continuous condition vector c cont = [ T , H R ] is extracted from the running data. The vector is embedded into the network through a fully connected layer and directly guides the CDDPM together with the CFG condition label as follows:

[0089]

[0090] clock, f cat (·) and f cont (·) denotes an independent fully connected layer, λ is a hyper-parameter controlling the influence of the continuous condition vector. These fused embeddings jointly guide the decoding path of the U-Net, shaping the spatial patterns and amplitudes of the reconstructed deformation field. Compared with label-driven conditioning, this dual-condition mechanism enhances the adaptability of the model to complex and variable dam operation scenarios, while achieving high-fidelity and controllable deformation reconstruction under sparse monitoring data.

[0091] Step S3.4, model training mechanism: model training follows the basic process of DDPM, adding Gaussian noise to real images through a forward diffusion process, and then training the network to predict the added noise; the loss function uses mean square error (MSE) or its improved form to measure the error between the model's predicted noise and the real noise; to enhance the performance of the model, training techniques such as automatic mixed precision (AMP), learning rate scheduler, gradient clipping, etc. can be introduced.

[0092] Step S3.5, model training output. After training is completed, the CDDPM model is obtained, which can be used for subsequent conditional sampling generation; the model has the ability to generate deformation field images with reasonable spatial distribution and good structural consistency, conditioned on specified category labels and operation data.

[0093] Step S4, combined with actual monitoring data and corresponding simulation data of the section, a combined optimization framework is constructed to search for the finite element simulation data category that is closest to the measured deformation. The optimal category label and actual operation data are jointly input into the CDDPM for conditional sampling, and finally the reconstructed generative deformation field is obtained.

[0094] In the step S4, according to the monitoring data and the actual operation data, the conditional sampling of the CDDPM model is performed to realize the reconstruction of the deformation field.

[0095] This step, when reconstructing the deformation field, collects the monitoring data and operation data at the corresponding time as input, uses a combined optimization method to select the finite element data category that best matches the monitoring data, and uses it as a CFG guiding condition. The selected optimal category label and actual operation data are jointly input into the CDDPM model to realize the reconstruction of the deformation field during the operation of the rockfill dam.

[0096] The step S4 further comprises:

[0097] Step S4.1, Modeling the Combinatorial Optimization Problem. A candidate set is constructed from the multi-category data extracted from the finite element simulation samples. This candidate set is then introduced into the combinatorial optimization model, and several categories that most closely resemble the measured deformation values ​​are selected as guiding conditions for generation. This problem is formalized as a discrete combinatorial optimization problem, with the objective function being:

[0098]

[0099] In the formula, x i This is a selection scalar for multi-hot encoding, with a value of 0 or 1, indicating whether to select the first hot code. i Class data; δ i,j For the first i Simulation data in the first j Deformation values ​​at each measuring point; The measured deformation value is used. By minimizing this objective function, the data combination that best matches the actual monitoring results is found.

[0100] Step S4.2, Solving with Intelligent Algorithm. The classic Genetic Algorithm is used to solve the above combinatorial optimization model. During the optimization process, the individual encoding adopts a multi-hot encoding format, which is suitable for parallel searching of the global optimal solution for multiple class combinations, and has good search capabilities and convergence characteristics.

[0101] Step S4.3, Condition Input Construction and Sampling Generation. The optimal simulation data category labels obtained through combined optimization are then combined with the actual construction and operation data condition vectors […]. T , H R The conditional vectors are jointly constructed and input into the trained CDDPM for conditional sampling, ultimately obtaining generative deformation field results that match the measured data.

[0102] As one implementation method, this invention uses the aforementioned generative AI-based method for constructing a digital twin of a rockfill dam deformation to reconstruct the deformation field of the Lianghekou core rockfill dam during its operation.

[0103] Figure 2 (a) and (b) with Figure 3 The working principle of the proposed CDDPM is further clarified. Figure 4 (a) through (c) illustrate the relevant circumstances of the rockfill dams used in the embodiments of the present invention.

[0104] Three types of data sources are first obtained from the rockfill dam project: in-situ deformation monitoring data, construction and operation data, and finite element simulation data. The in-situ deformation monitoring data comes from actual equipment such as water pipe type settlement meter, flexible inclinometer, electromagnetic settlement ring, etc. The operation data includes water level change process and construction and operation time length. The finite element simulation data is generated by a three-dimensional finite element model established under the conditions of real geometry and material zoning of the rockfill dam. The model uses C3D8 elements, sets 87 loading steps, and covers the filling and impoundment process. Each set of simulation samples corresponds to a set of constitutive parameters and a two-dimensional deformation field of multiple time steps. The two-dimensional deformation field data of the core wall and the 3-3 typical section in the operation period stage (59-87 steps) are extracted as the focus, which are used for subsequent model training.

[0105] The extracted two-dimensional deformation field data is uniformly interpolated into fixed size images (128x128) and converted into grayscale images. The pre-trained ResNet-18 residual neural network is used to extract the deep features of the images, and then the K-means algorithm is used for cluster analysis of the simulation samples. Each simulation sample is assigned a class label according to its deformation pattern, a total of 50 classes, providing structural information guidance for the subsequent conditional generation model.

[0106] The model takes three types of condition vectors as input: two-dimensional deformation field images, data clustering class labels, and information such as operation period water level and construction and operation total time length. The CDDPM architecture is a deep U-Net network that combines time embedding, condition embedding, and spatial coding to support flexible control and fusion of conditions. Noise is added through forward diffusion, and the model is retrained to learn the reverse denoising process, thereby realizing conditional generation.

[0107] Figure 5 The process of generating deformation field from pure Gaussian noise using the CDDPM proposed in the present application is shown. Figure 6 (a) shows the convergence of the objective function in the combinatorial optimization process, which converges to below 0.01 after 20 iterations, indicating that the optimization method proposed in the present application can effectively identify representative simulation data categories based on monitoring data for CFG condition guidance. Figure 6 (b) shows the deformation field results of the typical section of Lianghekou obtained using the reconstruction framework. Since the conditional DDPM model is trained based on finite element data, the generated results are highly consistent with the finite element deformation field in terms of spatial distribution pattern.

[0108] Table 1 shows the relative error between the reconstruction results and the monitoring data. The reconstruction error of all measuring points on the two sections is basically controlled within 10%, and the overall reconstruction accuracy is 92.8%, verifying the effectiveness and accuracy of the method in the high-precision deformation perception task of rockfill dams.

[0109] Table 1 Relative error between reconstructed deformation field and monitoring data

[0110] .

[0111] Figure 7 The calculation time consumption of each stage of the method of the present application under the conditions of the foregoing equipment is shown. Similar to the conventional parameter inversion method, the calculation time is mainly concentrated in the finite element sample preparation stage. After the CDDPM model training is completed, real-time reconstruction of the deformation field can be realized, and it only takes about 3 seconds to generate a single deformation field image, which can meet the demand of real-time monitoring. In contrast, a single finite element analysis based on inverted parameters takes an average of about 6 hours, and a structural health monitoring model based on time series monitoring data and deep learning also needs to be trained before it can be used for prediction. Both of these methods are difficult to achieve real-time deformation field generation. Therefore, the deformation field reconstruction framework proposed in the present application combines high accuracy and real-time performance, providing effective support for engineering applications.

[0112] To further verify the effectiveness of the present application, the input construction and operation data were adjusted for conditional generation during the generation process without the CFG guide mechanism. Figure 8 The reconstruction results of 10 times under different construction and operation data are counted. Overall, with the advancement of dam construction and operation, the reconstruction deformation field of each section gradually increases, which conforms to the physical law that the deformation of the dam increases with time. However, the generated deformation value under this condition is high, with most of the maximum settlement values exceeding 5 meters, which is more than 1 meter higher than the results of finite element analysis and deformation field reconstruction under the complete guide condition, making it difficult to directly meet the accuracy requirements of rockfill dam deformation analysis. Therefore, the DDPM model in the present application mainly relies on deformation monitoring data for guided sampling, and the operation data are introduced as auxiliary conditions to enhance the physical reasonableness of the generated deformation field, making the size and evolution trend of the reconstructed deformation more close to the actual working conditions.

[0113] As the service time of the dam extends, the monitoring instruments gradually age and even fail, further reducing the number of available monitoring data. Therefore, to evaluate the robustness of the framework under the condition of gradually reducing monitoring data, the present application conducts deformation field reconstruction experiments with step-by-step reduction of the number of monitoring points on the 3-3 section and the core wall section. Specifically, in the initial stage, 10 and 5 monitoring points are selected on the 3-3 section and the core wall section, respectively. In each experiment, 2 and 1 monitoring points are sequentially removed from the two sections for reconstruction. To evaluate the consistency of the generated deformation field and the original deformation field in terms of structural information, the structural similarity index (Structural Similarity Index, SSIM) is introduced. The value of SSIM ranges from 0 to 1, and the higher the value, the higher the similarity between the two images. Figure 9The SSIM indexes of the reconstruction results of two sections under different numbers of measuring points are shown. Overall, the SSIM of all the reconstruction results is greater than 0.75, indicating that they are in good consistency with the reference deformation field in the structural form. The result verifies that the conditional DDPM model proposed in the application has good robustness under different numbers of measuring points. This is mainly due to the fact that the model effectively learns the spatial distribution pattern of the finite element deformation field during the training phase, enabling it to generate deformation field images with physical consistency and high quality during the sampling process. In summary, the reconstruction framework proposed in the application shows good adaptability in practical engineering and can provide reliable support for the whole-process deformation perception and safety monitoring of rockfill dams during operation.

[0114] The implementation basis of each embodiment of the application is achieved by programmed processing of a device with processor function. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the application are packaged into various modules. Based on this actual situation, on the basis of the above-mentioned embodiments, an embodiment of the application provides a rockfill dam deformation digital twin construction system based on generative AI, which is used to execute the generative AI-based rockfill dam deformation digital twin construction method in the above-mentioned method embodiment.

[0115] The system comprises: a first main module for constructing a finite element numerical model based on rockfill dam deformation monitoring data and operation data, and obtaining a plurality of deformation field simulation samples by using a parameter sampling strategy; a second main module for extracting two-dimensional deformation field data of a key monitoring section based on the obtained deformation field simulation samples, extracting deep features of two-dimensional deformation field images by using a deep neural network, and classifying and labeling the simulation samples by using an unsupervised clustering method; a third main module for constructing a conditional denoising diffusion probability model, inputting the classified two-dimensional deformation field simulation samples, corresponding category labels and operation data used during simulation as conditional inputs for model training, and outputting a controllable generated deformation field reconstruction model; and a fourth main module for constructing a combined optimization framework by combining actual deformation monitoring data and simulation data of the corresponding section, searching for an optimal data category, and inputting the searched optimal data category and operation data into the controllable generated deformation field reconstruction model to output a reconstructed rockfill dam deformation field.

[0116] The rockfill dam deformation digital twin construction system based on generative AI provided by the embodiment of the application faces the current situation of how to fuse finite element physical prior information and engineering operation data under the condition of sparse monitoring data, realize high-precision and high-efficiency dam deformation field reconstruction, and improve the overall thorough perception ability of rockfill dam deformation. By using the aforementioned modules, the generation of diffusion modeling, the category guiding mechanism and the joint control generation process of continuous conditional vectors are adopted, which significantly improves the accuracy, efficiency and physical consistency of deformation reconstruction.

[0117] It should be noted that the system embodiments provided by the present application are used to implement the methods in the above method embodiments, and are also used to implement the methods in other method embodiments provided by the present application, the difference is only that the corresponding function modules are set, and the principle is basically the same as that of the above system embodiments provided by the present application. As long as the person skilled in the art improves the modules in the above system embodiments on the basis of the above system embodiments, refers to the specific technical solutions in other method embodiments, obtains the corresponding technical means by combining technical features, and the technical solutions composed of these technical means, on the premise of ensuring the practicability of the technical solutions, the corresponding system class embodiments are obtained, which are used to implement the methods in other method class embodiments.

[0118] Based on the same inventive concept as the foregoing embodiments, the present application embodiment also provides a rockfill dam deformation digital twin construction device based on generative AI, comprising a memory and a processor, the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the rockfill dam deformation digital twin construction method based on generative AI.

[0119] Based on the same inventive concept as the foregoing embodiments, the present application embodiment also provides a non-transitory computer readable storage medium, which stores computer instructions, and the computer instructions make the computer execute the rockfill dam deformation digital twin construction method based on generative AI.

[0120] In summary, the core of the present application is to propose a rockfill dam deformation digital twin technology framework based on generative artificial intelligence and multi-source data fusion, to efficiently fuse finite element simulation data, monitoring data and operation data by using a conditional denoising diffusion probability model (CDDPM), through a classifier-free guidance (CFG) and a continuous condition vector to construct a conditional sampling mechanism. Further combining residual neural network ResNet-18 and Kmeans clustering method, unsupervised classification of finite element data is performed, combined optimization is introduced, and the most matched finite element data category of identification and monitoring data is identified, so as to realize the generation of conditionally guided deformation field. The framework has been applied and verified on the currently built world's highest Shuanghekou rockfill dam (303 meters). The results show that the rockfill dam deformation digital twin construction method based on generative AI can efficiently reconstruct the rockfill dam deformation, has high precision and real-time performance, and significantly improves the comprehensive deformation perception ability of the rockfill dam, providing key technical support for its safe operation.

[0121] The application significantly improves the accuracy, efficiency and physical consistency of deformation reconstruction by generating process control through generative diffusion modeling, category guide mechanism (CFG) and continuous conditional vector. Compared with the traditional finite element inversion method, the single reconstruction takes only about 3 seconds after training, which improves the efficiency by about four orders of magnitude and has real-time response capability; the structural consistency index (SSIM) is higher than 0.75 in the scene of gradually reducing the number of actual measuring points, showing strong robustness and good generalization ability; the method has been verified in the two river mouth rockfill dam project, and a reconstruction accuracy of 92.8% has been achieved, showing its adaptability and engineering practical value in complex working conditions. The examples of the application improve the level of thorough perception of rockfill dam deformation, and can provide important support for the deformation safety evaluation of rockfill dam.

[0122] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary general hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute the method described in each embodiment or some parts of the embodiment.

[0123] In this patent, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or device. Without more limitations, the elements defined by the statement "include" do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0124] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the application.

Claims

1. A method for constructing a digital twin of rockfill dam deformation based on generative AI, characterized in that, The method comprises the steps of: Based on the deformation monitoring data and operation data of the rockfill dam, a finite element numerical model is constructed, and a parameter sampling strategy is used to obtain multiple sets of deformation field simulation samples; Based on the obtained deformation field simulation samples, the two-dimensional deformation field data of the key monitoring section is extracted, and the deep features of the two-dimensional deformation field image are extracted through a deep neural network, and an unsupervised clustering method is used for classification and label construction of the simulation samples; A conditional denoising diffusion probability model is constructed, and the classified two-dimensional deformation field simulation samples, the corresponding category labels, and the operation data used in the simulation are input as conditional inputs for model training, and a controllable generated deformation field reconstruction model is output; the conditional denoising diffusion probability model comprises a forward diffusion process and a parameter back diffusion process, wherein in the forward diffusion process, Gaussian noise is gradually added to the simulation deformation field data of each section of the rockfill dam, and in the parameter back diffusion process, the deformation field image is gradually restored from the Gaussian noise image; Combined with the actual deformation monitoring data and the simulation data of the corresponding section, a combined optimization framework is constructed, the optimal data category is searched, and the searched optimal data category and the operation data are jointly input into the controllable generated deformation field reconstruction model to output the reconstructed deformation field of the rockfill dam.

2. The method of claim 1, wherein, After extracting the two-dimensional deformation field data of the key monitoring section based on the obtained deformation field simulation samples, the method further comprises the steps of: The extracted two-dimensional deformation field data is uniformly interpolated and mapped to a fixed resolution image matrix, and the deformation values are converted to single-channel grayscale images through standardization processing.

3. The method of claim 2, wherein, The deep features of the two-dimensional deformation field image are extracted through a deep neural network, which comprises the steps of: A pre-trained ResNet-18 network is used as the backbone structure, the end full connection classification layer is removed, and only the convolution layer and the global average pooling layer are retained to extract the deep feature vector of each two-dimensional deformation field image.

4. The method of claim 3, wherein, The classification and label construction of the simulation samples are performed through an unsupervised clustering method, which comprises the steps of: The deep feature vectors corresponding to all two-dimensional deformation field images are input into the K-means clustering algorithm, an unsupervised clustering operation is performed, the simulation samples are classified, and each simulation sample is assigned a unique category label.

5. The method of claim 1, wherein, The training of the conditional denoising diffusion probability model comprises the steps of: Gaussian noise is gradually added to the simulation deformation field data of each section of the rockfill dam through the forward diffusion process to obtain a noisy image; The noisy image is input into the U-Net network to train it to predict the noise component added to the original two-dimensional deformation field image; The trained U-Net network is used to gradually remove the noise in the back diffusion process to reconstruct the original deformation field image.

6. The method of claim 5, wherein, The method further comprises the steps of: Discrete category labels are encoded into high-dimensional vectors through a full connection layer, and a category guiding mechanism is used to randomly mask the category labels with a set probability during the training stage; The operation data is converted into feature representation through a full connection layer and embedded and fused with the category labels under the category guiding mechanism to jointly guide the decoding path of the U-Net network.

7. The method of claim 1, wherein, Combined with the actual deformation monitoring data and the simulation data of the corresponding section, a combined optimization framework is constructed, the optimal data category is searched, and the searched optimal data category and the operation data are jointly input into the controllable generated deformation field reconstruction model, which comprises the steps of: Multi-class data extracted from finite element simulation samples is constructed as a candidate set, a combinatorial optimization model is introduced, and several categories closest to the measured deformation values are selected as the condition generation guide; A genetic algorithm is used to optimize and solve the combinatorial optimization model to obtain the optimal data category; The label of the optimal data category is combined with the running data to form a condition vector, which is input into the trained conditional denoising diffusion probability model for conditional sampling to obtain a generative deformation field result consistent with the measured data.

8. A system for building a digital twin of rockfill dam deformation based on generative AI for implementing the method of any one of claims 1 to 7, characterized in that, The system comprises: A first main module for constructing a finite element numerical model based on rockfill dam deformation monitoring data and running data, and obtaining multiple sets of deformation field simulation samples using a parameter sampling strategy; A second main module for extracting two-dimensional deformation field data of key monitoring sections based on the obtained deformation field simulation samples, extracting deep features of two-dimensional deformation field images through a deep neural network, and classifying and labeling simulation samples using an unsupervised clustering method; A third main module for constructing a conditional denoising diffusion probability model, inputting the classified two-dimensional deformation field simulation samples, corresponding category labels, and running data used during simulation as conditions for model training, and outputting a controllable deformation field reconstruction model; the conditional denoising diffusion probability model comprises a forward diffusion process and a parameter reverse diffusion process, wherein in the forward diffusion process, Gaussian noise is gradually added to the simulation deformation field data of each section of the rockfill dam, and in the parameter reverse diffusion process, the deformation field image is gradually restored from the Gaussian noise image; A fourth main module for constructing a combinatorial optimization framework by combining actual deformation monitoring data and corresponding simulation data of sections, searching for the optimal data category, and inputting the searched optimal data category and running data into the controllable deformation field reconstruction model to output a reconstructed rockfill dam deformation field.

9. A device for construction of a rockfill dam deformation digital twin based on generative AI, characterized by, A memory and a processor, the memory storing program instructions executed by the processor, the processor invoking the program instructions to execute the method for constructing a generative AI-based rockfill dam deformation digital twin according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, comprising: The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the method for constructing a generative AI-based rockfill dam deformation digital twin according to any one of claims 1 to 7.

Images