A soil and rock dam stress and deformation field time sequence prediction method and system based on finite element prior and monitoring data assimilation

Abstract

The application discloses a kind of earth-rock dam stress deformation field time series prediction method and system based on finite element prior and monitoring data assimilation, belong to the technical field of water conservancy and hydropower engineering, method includes: multiple types of monitoring data and complete finite element field establish unified physical space registration relationship, and the prior stress deformation field of earth-rock dam is encoded into prior latent variable;Prior stress deformation field is mapped to sparse field monitoring space, and compared with field monitoring data matrix to construct monitoring residual matrix and feature extraction, correct prior latent variable in low-dimensional latent space based on sparse monitoring correction feature, and reconstruct as finite element complete stress deformation field after monitoring assimilation;Through autoregressive propagation model, online assimilation and rolling prediction are carried out.The application realizes dynamic correction and future state prediction of finite element full field prior under sparse monitoring condition, improves the spatial integrity, real-time performance and accuracy of safety state evaluation of earth-rock dam construction period and initial operation period, and has significant practicability and wide application prospect.

Description

Technical Field

[0001] This invention belongs to the field of time series prediction technology for water conservancy and hydropower projects, specifically involving a time series prediction method and system for stress and deformation fields of earth-rock dams based on finite element priors and assimilation of monitoring data. Background Technology

[0002] Earth-rock dams undergo complex processes of settlement, horizontal displacement, stress, and pore pressure evolution during construction, impoundment, and long-term service. The dam body exhibits distinct material zoning, with materials such as rockfill and impermeable layers displaying prominent nonlinear characteristics. Construction path, water level changes, boundary conditions, and time effects all influence the dam's stress-deformation response. Therefore, timely understanding of the complete stress-deformation field within the dam body and its evolution is crucial for construction control, operation management, and safety assessment.

[0003] Traditional finite element methods can provide relatively complete spatial field distributions and have a clear mechanical basis. However, finite element calculation results are easily affected by material parameter values, constitutive model applicability, construction process simulation, boundary conditions, mesh scale, and scale effects, leading to systematic deviations from the actual engineering response. In engineering practice, finite element calculations often show deviations between the calculated and measured engineering response values ​​under different dam heights or working conditions, making it difficult to directly use finite element results for real-time operational performance evaluation.

[0004] While on-site monitoring data can reflect the true response of the dam body, the number of monitoring points is limited and their spatial distribution is sparse. Typically, only a few settlement points, horizontal displacement points, inclinometers, plumb lines, piezometers, or fiber optic monitoring lines can be provided, making it difficult to independently reconstruct the complete spatial field of the dam body. If the spatial distribution of the entire field is inferred solely from monitoring point interpolation or empirical statistical models, it is easy to lose spatial correlations caused by local stress concentrations, deformation extrema, and material zonal coupling.

[0005] Existing machine learning or physical information neural network methods often use field monitoring data as the fitting object or part of the training loss, focusing on single-point sequence prediction or local response correction, making it difficult to achieve dynamic correction of finite element priors at the whole field scale. At the same time, some time series prediction models only perform static point prediction and lack the ability to continuously update the state for construction or water level stages, making it difficult to meet the needs of overall field state perception and future trend prediction during the construction and early operation of earth-rock dams.

[0006] Therefore, there is an urgent need for a technical solution that can simultaneously utilize the prior information of the complete finite element field and the local authenticity of the field monitoring data, so that the finite element prior field can be continuously corrected with the field monitoring data, and rolling predictions can be made for the future construction or operation stages based on the corrected state, so as to improve the accuracy, real-time performance and spatial integrity of the safety status assessment of earth-rock dams. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for time-series prediction of stress and deformation fields in earth-rock dams based on finite element priors and assimilation of monitoring data, which can effectively solve the aforementioned problems.

[0008] The technical solution adopted in this invention is as follows:

[0009] In a first aspect, the present invention provides a time-series prediction method for the stress-deformation field of earth-rock dams based on finite element prior and monitoring data assimilation, comprising:

[0010] Step S1: Determine the current engineering conditions and current engineering stage of the earth-rock dam, and input them into the constructed earth-rock dam finite element model to perform a complete stress-deformation response simulation, and obtain a high-dimensional finite element complete stress-deformation field as the prior stress-deformation field of the current engineering stage.

[0011] Step S2: Based on the observation operator, the prior stress deformation field is mapped to the sparse field monitoring space to obtain a prior monitoring prediction matrix that is comparable to the field monitoring data matrix. A monitoring residual matrix is ​​constructed based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix. The monitoring residual matrix is ​​then used to extract features from the monitoring residual matrix through a pre-trained monitoring encoder to obtain sparse monitoring correction features.

[0012] Step S3: The prior stress deformation field is encoded into prior latent variables in a low-dimensional latent space using a pre-trained shared VAE encoder; the prior latent variables are corrected in the low-dimensional latent space based on the sparse monitoring and correction features using a pre-trained latent variable updater to obtain posterior latent variables.

[0013] Step S4: Input the posterior latent variable into the pre-trained shared VAE decoder, and the shared VAE decoder reconstructs the posterior latent variable into the monitored and assimilated complete finite element stress-deformation field;

[0014] Step S5: The monitored and assimilated finite element complete stress-deformation field is used as the posterior stress-deformation field of the current engineering stage. Combined with the engineering conditions of the next engineering stage, the finite element complete stress-deformation field of the next engineering stage is predicted by the autoregressive propagation model. Then, return to step S2 to realize online assimilation and rolling prediction.

[0015] Furthermore, the engineering conditions include at least one of the following: geometric parameters of the earth-rock dam body, material parameters, construction loading path, water level change path, and boundary conditions; the engineering stages include the construction period, the impoundment period, and the operation period; the complete finite element stress-deformation field includes at least one of the following: horizontal displacement field, vertical settlement field, stress field, and pore pressure field; and the observation operator includes at least one of the following: interpolation operator, projection operator, and integration operator.

[0016] Furthermore, the method of mapping the prior stress-deformation field to a sparse field monitoring space based on the observation operator to obtain a prior monitoring prediction matrix comparable to the field monitoring data matrix includes:

[0017] The sparse field monitoring space has multiple monitoring points with corresponding field monitoring data. Based on the observation operator, the prior stress-deformation field is mapped to the sparse field monitoring space to obtain the prior monitoring prediction value for each monitoring point, thereby obtaining the prior monitoring prediction value matrix of the sparse field monitoring space, specifically including:

[0018] For displacement or settlement monitoring points, the prior stress deformation field is interpolated based on the element shape function of the finite element unit where the monitoring point is located to obtain the prior monitoring prediction value at the corresponding monitoring point.

[0019] For inclinometer, plumb line or stratified settlement monitoring data, the prior stress-deformation field is projected or extracted in layers according to the monitoring path or monitoring depth to obtain the prior monitoring prediction value at the corresponding depth position.

[0020] For fiber optic monitoring data, the prior monitoring prediction values ​​are linearly sampled or path integrated according to the fiber optic deployment path to obtain the prior monitoring prediction values ​​for each monitoring point along the line.

[0021] Furthermore, when constructing the monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix, considering the missing field monitoring data, a missing data mask matrix is ​​introduced, and the monitoring residual matrix is ​​obtained using formula (1):

[0022] (1)

[0023] in: The missing measurement mask matrix indicates whether the field monitoring data of each monitoring point in the sparse field monitoring space is valid. If it is valid, the corresponding element in the missing measurement mask matrix is ​​1; otherwise, it is 0. This represents element-wise multiplication; Indicates the current stage of the project. The on-site monitoring data matrix; Indicates the current stage of the project. The prior monitoring prediction matrix; This represents the monitoring residual matrix.

[0024] Furthermore, sparse monitoring correction features are obtained by extracting features from the monitoring residual matrix using a pre-trained monitoring encoder, including:

[0025] Monitor residual matrix Monitoring point location matrix Monitoring point type matrix Missing test mask matrix Prior monitoring prediction matrix and on-site monitoring data matrix A common input is used to pre-train a monitoring encoder, which obtains sparse monitoring correction features. The expression is:

[0026] (2)

[0027] in: Indicates monitoring encoder;

[0028] As online assimilation and rolling prediction continue to advance, the number of monitoring points with on-site monitoring data in the sparse field monitoring space is constantly increasing. The monitoring encoder adopts an ensemble coding structure, firstly extracting features from each monitoring point, and then performing pooling aggregation on the features extracted from each monitoring point. The expression is as follows:

[0029] (3)

[0030] in: The feature extraction function represents a single monitoring point. Indicates the current stage of the project. The number of monitoring points; Indicates the current stage of the project. Monitoring points , Indicates the current stage of the project. The number of monitoring points; , , , , and , respectively representing the monitoring points The on-site monitoring data includes prior monitoring predictions, monitoring residuals, monitoring point locations, monitoring point monitoring data, and missing measurement masks.

[0031] Furthermore, the shared VAE encoder and the shared VAE decoder constitute a shared variational autoencoder; the monitoring encoder and the latent variable updater constitute a latent variable correction model; the training method for the shared variational autoencoder and the latent variable correction model during the offline training phase is as follows:

[0032] Step A1: Based on the finite element model of the earth-rock dam, construct a complete stress-deformation field sample library for multiple working conditions using finite element methods.

[0033] Step A2 involves training the shared variational autoencoder using the multi-condition finite element complete stress-deformation field sample library. Through training, the shared VAE encoder is used to map the finite element complete stress-deformation field into latent variables in a low-dimensional latent space. This includes: mapping the finite element complete stress-deformation field into latent variable distribution parameters in the low-dimensional latent space, including the mean and standard deviation; obtaining the latent variables based on the mean and standard deviation using a reparameterization method; and through training, the shared VAE decoder is used to reconstruct the latent variables in the low-dimensional latent space to obtain the reconstructed finite element complete stress-deformation field.

[0034] Step A3: Take the complete finite element stress-deformation field of each sample in the multi-condition finite element complete stress-deformation field sample library as the target field, construct a biased prior field paired with each target field through a perturbation algorithm, and then obtain a sample library of paired biased prior fields and target fields; based on the trained shared variational autoencoder, use the paired sample library to train the latent variable correction model offline.

[0035] Furthermore, the multi-condition finite element complete stress-deformation field sample library is characterized as follows: ;

[0036] Wherein: the sample library Finite element complete stress-deformation field sample Characterization Engineering Conditions and engineering phase Finite element complete stress-deformation field ; The number of samples representing the complete stress-deformation field of a finite element method; engineering conditions. , Indicates the geometric parameters of the dam body. Indicates material parameters, Indicate boundary conditions, Indicates the construction loading path. Indicates the path of water level change; dam geometric parameters Including dam type, dam height, and dam body zoning;

[0037] During training, the shared VAE encoder will generate a complete finite element stress-deformation field. Mapped to latent variables in a low-dimensional latent space The latent variable distribution parameters, including the mean and standard deviation The expression is , Indicates a shared VAE encoder; based on the mean. and standard deviation The latent variables are obtained by reparameterization. The expression is: ;in, This represents a random perturbation vector sampled from a standard multivariate normal distribution. This represents a standard multivariate normal distribution with a mean of 0 and a covariance matrix of identity matrix I, where I represents the identity matrix. This represents element-wise multiplication;

[0038] During training, the shared VAE decoder will store latent variables in the low-dimensional latent space. Reconstruction yields the complete finite element stress-deformation field. The expression is , Indicates a shared VAE decoder;

[0039] The training loss of the shared variational autoencoder Represented as:

[0040] (4)

[0041] (5)

[0042] in: This represents the reconstruction loss of the complete stress-deformation field in the finite element method. Latent space regularization constraint; Indicates the regularization weight; Indicates KL divergence; This represents the complete stress-deformation field of the input finite element model as determined by the shared VAE encoder. latent variables obtained The approximate posterior distribution, This represents the pre-defined prior distribution of the latent variables.

[0043] Furthermore, step A3 includes:

[0044] Step A31: Construct paired samples of the biased prior field and the target field;

[0045] Each finite element complete stress-deformation field sample in the multi-condition finite element complete stress-deformation field sample library Finite element complete stress-deformation field As the target field, it is re-represented as ;in, Indicates engineering conditions , This represents the engineering phase; the target field is constructed using a perturbation algorithm. Paired biased prior fields ;

[0046] Step A32, using the same observation operator , respectively target field and biased a priori fields Mapping to the sparse field monitoring space yields the target monitoring prediction matrix for the sparse field monitoring space. And biased monitoring prediction matrix The expression is and ;

[0047] Step A33, generate the target monitoring prediction value matrix. As a pseudo-site monitoring data matrix, the biased monitoring prediction matrix will be used. As the prior monitoring and prediction value matrix, the formula is used. Obtain the pseudo-monitoring residual matrix The data is then input into the monitoring encoder to obtain pseudo-sparse monitoring correction features. ;

[0048] Step A34, respectively target fields and biased a priori fields The input is sent to the shared VAE encoder, which will then input the target field. Encoding target latent variables in a low-dimensional latent space There will be a biased prior field Encoded as biased prior latent variables in a low-dimensional latent space ;

[0049] Step A35, the pseudo-sparse monitoring correction feature and the biased prior latent variables The latent variable updater is input, and the latent variable updater learns the pseudo-sparse monitoring and correction features. For the biased prior latent variables The latent variable correction amount is used to update the biased prior latent variable. Obtain biased posterior latent variables Through training, the biased posterior latent variables are made... Continuously approaching the target latent variable The monitoring encoder and the latent variable updater are trained to obtain the trained latent variable correction model.

[0050] Furthermore, the target field is constructed using a perturbation algorithm. Paired biased prior fields ,include:

[0051] Engineering condition perturbation algorithm: for engineering conditions According to the set deviation strength parameters To perform the perturbation, use the deviation generation function. Generate system deviation The expression is ; to system bias Acting on the target field , obtain the target field Paired biased prior fields ;

[0052] Latent space perturbation algorithm:

[0053] target field The input is sent to the shared VAE encoder, which will then input the target field. Encoding target latent variables in a low-dimensional latent space ;

[0054] target latent variables Perturbation is performed to obtain the perturbed target latent variables. The expression is: ;in, Indicates the intensity of the disturbance; Let represent the latent space perturbation term, and let represent the random perturbation vector sampled from the multivariate normal distribution. This indicates that the mean is 0 and the covariance matrix is... The multivariate normal distribution, It is the identity matrix;

[0055] Using the shared VAE decoder, the perturbed target latent variables are... Reconstructed into a complete finite element stress-deformation field, which serves as the target field. Paired biased prior fields The expression is ;, This indicates a shared VAE decoder.

[0056] Secondly, this invention provides a time-series prediction system for the stress-deformation field of earth-rock dams based on finite element prior and monitoring data assimilation, comprising:

[0057] The finite element model construction and simulation module for earth-rock dams is used to determine the current engineering conditions and current engineering stage of earth-rock dams, and input them into the constructed finite element model of earth-rock dams to perform complete stress-deformation response simulation, and obtain a high-dimensional complete finite element stress-deformation field as the a priori stress-deformation field for the current engineering stage.

[0058] An observation operator is used to map the prior stress-deformation field to a sparse field monitoring space to obtain a prior monitoring prediction matrix that is comparable to the field monitoring data matrix.

[0059] A monitoring residual matrix construction module is used to construct a monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction value matrix.

[0060] A monitoring encoder is used to extract features from the monitoring residual matrix to obtain sparse monitoring correction features;

[0061] A shared VAE encoder is used to encode the prior stress deformation field into prior latent variables in a low-dimensional latent space;

[0062] A latent variable updater is used to correct the prior latent variable based on the sparse monitoring correction feature in a low-dimensional latent space to obtain the posterior latent variable;

[0063] A shared VAE decoder is used to reconstruct the posterior latent variables into a complete finite element stress-deformation field after monitoring assimilation;

[0064] The autoregressive propagation model is used to take the monitored and assimilated finite element complete stress-deformation field as the posterior stress-deformation field of the current engineering stage, combine it with the engineering conditions of the next engineering stage, predict the finite element complete stress-deformation field of the next engineering stage, and take the finite element complete stress-deformation field of the next engineering stage as the prior stress-deformation field of the next engineering stage, and transmit it to the observation operator and the shared VAE encoder. This process is repeated to achieve online assimilation and rolling prediction.

[0065] The beneficial effects of this invention are as follows:

[0066] The present invention provides a method and system for time-series prediction of stress and deformation field of earth-rock dam based on finite element prior and monitoring data assimilation, which has the following advantages: The present invention establishes a unified physical space registration relationship between multi-type monitoring data and complete finite element field through observation operators. Under sparse monitoring conditions, it can realize dynamic correction of the full-field prior of finite element of earth-rock dam and prediction of future state, improve the spatial integrity, real-time performance and accuracy of safety state evaluation during the construction period and early operation of earth-rock dam, and has significant practicality and broad application prospects. Attached Figure Description

[0067] Figure 1 The overall flowchart of the time-series prediction method for stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation provided in the embodiments of the present invention is shown below.

[0068] Figure 2 A schematic diagram of the observation operator provided in an embodiment of the present invention;

[0069] Figure 3 The flowchart for online rolling prediction provided in the embodiments of the present invention. Detailed Implementation

[0070] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the invention.

[0071] This invention provides a method and system for time-series prediction of stress and deformation field of earth-rock dam based on finite element prior and monitoring data assimilation. It relates to safety monitoring of water conservancy and hydropower projects, numerical analysis of earth-rock dams, artificial intelligence prediction and data assimilation technology. Specifically, it relates to a method for state assimilation, complete field reconstruction and future stage prediction of stress and deformation field of earth-rock dam by integrating finite element prior field, sparse field monitoring data and autoregressive time-series prediction.

[0072] This invention discloses a method and system for time-series prediction of stress-deformation field of earth-rock dams based on the assimilation of finite element priors and monitoring data. This method utilizes finite element calculation results to provide a complete spatial field prior, avoiding the ill-posed problem when sparse monitoring data is directly used to reconstruct the complete field. It establishes a unified physical space registration relationship between various types of monitoring data and the complete finite element field through observation operators, improving the ability of monitoring data to correct the prior field. It applies monitoring residuals to the latent space posterior update, rather than directly to the high-dimensional complete field, thereby improving model stability, generalization ability, and physical interpretability. Furthermore, it enables dynamic correction of the current stress-deformation field through decoding and reconstruction, and achieves online rolling prediction of future construction or operational stages through an autoregressive propagation model, thus supporting dynamic safety evaluation of earth-rock dams during construction and early operation.

[0073] See Figure 1 The above is a flowchart of the time-series prediction method for stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation provided in the embodiments of the present invention. It mainly includes an offline training stage and an online assimilation and prediction stage.

[0074] The time-series prediction method for stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation provided in this invention is used in a system that mainly includes an earth-rock dam finite element model, observation operator, monitoring encoder, latent variable updater, shared variational autoencoder and autoregressive propagation model; wherein, the shared variational autoencoder includes a shared VAE encoder and a shared VAE decoder.

[0075] The offline training phase is primarily used to train the monitoring encoder, latent variable updater, and shared variational autoencoder to achieve the corresponding performance in the online assimilation and prediction phase. The online assimilation and prediction phase enables time-series prediction of the stress-deformation field of an earth-rock dam based on its initial state. Specifically, it generates a posterior stress-deformation field from the prior stress-deformation field of the current engineering stage and field monitoring data, and further predicts the posterior stress-deformation field for future stages.

[0076] To facilitate understanding of this invention, the time-series prediction process of the stress-deformation field of earth-rock dams in the online assimilation and prediction stage is first introduced, see reference. Figure 1 and Figure 3 The time-series prediction method for stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation includes steps S1 to S5:

[0077] Step S1: Determine the current engineering conditions and current engineering stage of the earth-rock dam, and input them into the constructed earth-rock dam finite element model to perform a complete stress-deformation response simulation, and obtain a high-dimensional finite element complete stress-deformation field as the prior stress-deformation field of the current engineering stage.

[0078] Specifically, in the online assimilation and prediction phase, it is used to achieve online rolling prediction according to engineering stages, representing the current engineering stage as... The corresponding engineering conditions are expressed as follows: Therefore, by performing a complete stress-deformation response simulation using the finite element model of the earth-rock dam, a high-dimensional complete finite element stress-deformation field is obtained, which serves as the basis for the current engineering stage. a priori stress-deformation field .

[0079] As one embodiment, the engineering conditions It includes at least one of the following: geometric parameters of the earth-rock dam body, material parameters, construction loading path, water level change path, and boundary conditions; the project stages include the construction period, the impoundment period, and the operation period; the complete finite element stress-deformation field includes at least one of the following: horizontal displacement field, vertical settlement field, stress field, and pore pressure field.

[0080] Step S2: Based on the observation operator, the prior stress deformation field is mapped to the sparse field monitoring space to obtain a prior monitoring prediction matrix that is comparable to the field monitoring data matrix. A monitoring residual matrix is ​​constructed based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix. The monitoring residual matrix is ​​then used to extract features from the monitoring residual matrix through a pre-trained monitoring encoder to obtain sparse monitoring correction features.

[0081] In this embodiment of the invention, an observation operator H is defined. During the online phase, it is determined through the current engineering phase. Observation operator , the prior stress deformation field Mapping to a sparse field monitoring space to obtain a matrix of field monitoring data. Comparable prior monitoring prediction matrix The characteristics are as follows: .

[0082] On-site monitoring data matrix It can be represented as: ,in, This represents the complete field corresponding to the actual engineering response. This indicates monitoring noise or measurement error.

[0083] As one embodiment, the observation operator includes at least one of an interpolation operator, a projection operator, and an integration operator. Based on the observation operator... The prior stress deformation field Mapping to a sparse field monitoring space to obtain a matrix of field monitoring data. Comparable prior monitoring prediction matrix ,include:

[0084] The sparse field monitoring space has multiple monitoring points with corresponding field monitoring data. Based on the observation operator, the prior stress-deformation field is mapped to the sparse field monitoring space to obtain the prior monitoring prediction value for each monitoring point, thereby obtaining the prior monitoring prediction value matrix of the sparse field monitoring space. (See reference...) Figure 2 This is a schematic diagram illustrating the mapping of a priori stress-deformation fields to settlement points, horizontal displacement points, inclinometer tubes, or fiber optic monitoring points through interpolation, projection, or integration. It includes:

[0085] For displacement or settlement monitoring points, corresponding to point monitoring data such as horizontal displacement and settlement, the prior stress deformation field is interpolated based on the element shape function of the finite element unit where the monitoring point is located to obtain the prior monitoring prediction value at the corresponding monitoring point.

[0086] Specifically, for a monitoring point located inside a finite element element, the prior monitoring prediction value is obtained by interpolation based on the element shape function: ;in, Indicates the current stage of the project. At monitoring points Prior monitoring and prediction values ​​at the location, Indicates monitoring points The first finite element element The shape function corresponding to each node; Indicates monitoring points Coordinates in the local coordinate system of this finite element; Indicates the current stage of the project. The finite element element is the first The prior stress-deformation field values ​​of each node; This indicates the number of nodes contained in the finite element element.

[0087] For monitoring data such as inclinometer tubes, inverted plumb lines, or stratified settlement data deployed along the depth, the prior stress-deformation field is projected or extracted in layers according to the monitoring path or monitoring depth to obtain the prior monitoring prediction value at the corresponding depth location.

[0088] For fiber optic monitoring data, the prior monitoring prediction values ​​are linearly sampled or path integrated according to the fiber optic deployment path to obtain the prior monitoring prediction values ​​for each monitoring point along the line.

[0089] For example, for the monitoring data of optical fibers deployed along the line, its prior monitoring prediction value It can be represented as:

[0090]

[0091] in: This represents the monitoring path of the j-th optical fiber. Indicates the path length. Represents path coordinates. It represents a function that extracts optical fiber observables from a priori stress-deformation fields.

[0092] By using observation operators, a unified comparison relationship can be established between the finite element full-field results and different types of field monitoring data.

[0093] As one embodiment, when constructing the monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix, the missing field monitoring data situation is considered, and a missing measurement mask matrix is ​​introduced. Using formula (1), the monitoring residual matrix is ​​obtained:

[0094] (1)

[0095] in: The missing measurement mask matrix indicates whether the field monitoring data of each monitoring point in the sparse field monitoring space is valid. If it is valid, the corresponding element in the missing measurement mask matrix is ​​1; otherwise, it is 0. This represents element-wise multiplication; Indicates the current stage of the project. The on-site monitoring data matrix; Indicates the current stage of the project. The prior monitoring prediction matrix; This represents the monitoring residual matrix.

[0096] As one embodiment, sparse monitoring correction features are obtained by extracting features from the monitoring residual matrix using a pre-trained monitoring encoder, including:

[0097] Monitor residual matrix Monitoring point location matrix Monitoring point type matrix Missing test mask matrix Prior monitoring prediction matrix and on-site monitoring data matrix A common input is used to pre-train a monitoring encoder, which obtains sparse monitoring correction features. The expression is:

[0098] (2)

[0099] in: Indicates monitoring encoder;

[0100] As online assimilation and rolling prediction continue to advance, the number of monitoring points with on-site monitoring data in the sparse field monitoring space is constantly increasing. The monitoring encoder adopts an ensemble coding structure, firstly extracting features from each monitoring point, and then performing pooling aggregation on the features extracted from each monitoring point. The expression is as follows:

[0101] (3)

[0102] in: The feature extraction function represents a single monitoring point. Indicates the current stage of the project. The number of monitoring points; Indicates the current stage of the project. Monitoring points , Indicates the current stage of the project. The number of monitoring points; , , , , and , respectively representing the monitoring points The on-site monitoring data includes prior monitoring predictions, monitoring residuals, monitoring point locations, monitoring point monitoring data, and missing measurement masks.

[0103] The monitoring encoder adopts a set coding structure, which can adapt to different numbers of monitoring points, different spatial arrangements, different monitoring types, and monitoring points with missing measurements, and outputs sparse monitoring correction features to correct prior latent variables.

[0104] Step S3: Use a pre-trained shared VAE encoder to process the prior stress-deformation field. Encoded as prior latent variables in a low-dimensional latent space The expression is: , Indicates a shared VAE encoder;

[0105] Based on the sparse monitoring correction features, a pre-trained latent variable updater is used in the low-dimensional latent space. Correcting the prior latent variables , obtain the posterior latent variables The expression is: , This represents a latent variable updater. As an optional implementation, the latent variable updater first obtains the latent variable corrections. Then use the formula Obtain posterior latent variables .

[0106] This step involves updating latent variables within the latent space. The fusion update within the latent space is achieved through a latent variable updater, which takes prior latent variables and sparse monitoring correction features as inputs, updating the prior latent variables to posterior latent variables. Therefore, this invention ensures that the correction of the complete stress-deformation field by monitoring data occurs within the latent space. The core of this step lies in the sparse monitoring correction features extracted from the monitoring residuals. Correcting prior latent variables Instead of generating a complete stress-deformation field directly from sparse monitoring points.

[0107] Step S4, decode and reconstruct to obtain the monitoring assimilated complete finite element stress-deformation field as the posterior stress-deformation field: input the posterior latent variable into the pre-trained shared VAE decoder, the shared VAE decoder reconstructs the posterior latent variable into the monitoring assimilated complete finite element stress-deformation field;

[0108] In this step, the shared VAE decoder can adopt at least one of the following: direct decoding structure, residual decoding structure, or modal residual decoding structure.

[0109] In the direct decoding structure, the decoding and reconstruction process can be represented as follows: ,in, This indicates that the complete stress-deformation field of the finite element after monitoring and assimilation is obtained by inputting the posterior latent variables into the decoder of the shared variational autoencoder.

[0110] In the residual decoding structure, the shared VAE decoder first outputs the field correction. , This indicates a shared VAE decoder; the field correction is then fused with the prior stress-deformation field to obtain the monitored and assimilated complete finite element stress-deformation field, expressed as: In this way, while maintaining the overall physical structure of the finite element field, local monitoring information can produce a globally consistent correction effect on the complete field.

[0111] Step S5, Autoregressive Construction-Level Time Series Prediction: The monitored and assimilated finite element complete stress-deformation field is used as the posterior stress-deformation field of the current engineering stage. Combined with the engineering conditions of the next engineering stage, the finite element complete stress-deformation field of the next engineering stage is predicted by the autoregressive propagation model. Then, return to step S2 to realize online assimilation and rolling prediction.

[0112] As one implementation method, the posterior stress-deformation field is used as the initial state. Combined with engineering conditions such as the next construction stage, water level, temperature, dam age, and material parameters, an autoregressive propagation model is used to predict the stress-deformation field in future construction or operational stages. The autoregressive propagation model can be any one or a combination of GRU, Transformer, Temporal Convolution, Neural ODE, or PINN state-space models. In this embodiment, autoregressive propagation is performed directly within the complete field space, and its working principle is as follows: ; This represents the conditions for the next project; The complete stress-deformation field of the finite element represents the next stage of the project; This represents the autoregressive propagation model.

[0113] In practical applications, the autoregressive propagation model can also be implemented using another method proposed in this embodiment of the invention. In this embodiment, the posterior latent variable... As the initial state, autoregressive propagation is carried out in the latent space to obtain the prior latent variables for the next engineering stage. : Then, the complete finite element stress-deformation field for the next engineering stage is recovered by using a shared VAE decoder. .

[0114] When new monitoring data emerges in the future, the monitoring residual calculation, monitoring encoding, latent space posterior update, and decoding processes are repeated to achieve continuous state assimilation and online rolling prediction.

[0115] In this embodiment, during the continuous state assimilation and rolling prediction process, when the on-site monitoring data and engineering conditions of each engineering stage are updated, the relevant parameters of the prediction model of this invention are updated synchronously. The observation operator mapping, monitoring residual calculation, monitoring encoding, latent space posterior update, and decoding reconstruction processes are repeatedly executed, and the updated posterior state is continued to be input into the autoregressive propagation model to achieve continuous state assimilation and rolling prediction. See reference. Figure 3 It demonstrates the closed-loop process of current construction-level assimilation, post-construction status generation, future construction-level prediction, and re-assimilation of new monitoring data.

[0116] The training principle of the offline training phase is introduced below:

[0117] In embodiments of the present invention, such as Figure 1 As shown, this mainly involves offline training of a shared variational autoencoder and a latent variable correction model. The shared variational autoencoder includes a shared VAE encoder and a shared VAE decoder; the latent variable correction model includes a monitoring encoder and a latent variable updater.

[0118] The offline training method mainly includes steps A1 to A3:

[0119] Step A1: Based on the finite element model of the earth-rock dam, construct a complete stress-deformation field sample library for multiple working conditions using finite element methods; the complete stress-deformation field sample library for multiple working conditions using finite element methods is characterized as follows: Wherein: the sample library Finite element complete stress-deformation field sample Characterization Engineering Conditions and engineering phase Finite element complete stress-deformation field ; The number of samples representing the complete stress-deformation field of a finite element method; engineering conditions. , Indicates the geometric parameters of the dam body. Indicates material parameters, Indicate boundary conditions, Indicates the construction loading path. Indicates the path of water level change; dam geometric parameters Including dam type, dam height, and dam body zoning; engineering stages This indicates the construction or operation phase.

[0120] As one example, a multi-condition finite element complete stress-deformation field sample library is established for different engineering stages of earth-rock dams, including the construction period, impoundment period, and initial operation period. The sample library includes complete finite element stress-deformation fields under different dam types, dam heights, dam body partitions, material parameters, construction loading paths, water level change paths, and boundary conditions. The complete finite element stress-deformation field for each construction or operation stage may include horizontal displacement field, vertical settlement field, stress field, and pore pressure field. This sample library provides the numerical prior distribution of the complete field and serves as a sample basis for training variational autoencoders, monitoring encoders, and latent variable updaters.

[0121] As one example, the construction method of the complete stress-deformation field sample library of multi-condition finite element method is as follows:

[0122] Finite element models were established for earth-rock dams under different engineering conditions during the construction, impoundment, and initial operation phases. Multi-condition finite element calculation samples were generated by changing the dam type, dam height, dam body zoning, material parameters, construction loading path, water level change path, and boundary conditions. A complete stress-deformation field corresponding to each construction stage or operation phase was extracted. This complete stress-deformation field includes at least one of a horizontal displacement field, a vertical settlement field, a stress field, and a pore pressure field. This complete stress-deformation field was used as a numerical prior distribution for subsequent latent space modeling and monitoring assimilation updates.

[0123] Step A2: Train the shared variational autoencoder using the multi-condition finite element complete stress-deformation field sample library;

[0124] Through training, the shared VAE encoder is used to map the complete finite element stress-deformation field into latent variables in a low-dimensional latent space, including: mapping the complete finite element stress-deformation field into latent variable distribution parameters in the low-dimensional latent space, including the mean and standard deviation; obtaining the latent variables using a reparameterization method based on the mean and standard deviation; and through training, the shared VAE decoder is used to reconstruct the latent variables in the low-dimensional latent space to obtain the reconstructed complete finite element stress-deformation field.

[0125] During training, the shared VAE encoder will generate a complete finite element stress-deformation field. Mapped to latent variables in a low-dimensional latent space The latent variable distribution parameters, including the mean and standard deviation The expression is , Indicates a shared VAE encoder; based on the mean. and standard deviation The latent variables are obtained by reparameterization. The expression is: ;in, This represents a random perturbation vector sampled from a standard multivariate normal distribution. This represents a standard multivariate normal distribution with a mean of 0 and a covariance matrix of identity matrix I, where I represents the identity matrix. This represents element-wise multiplication; during training, the shared VAE decoder will store latent variables in the low-dimensional latent space. Reconstruction yields the complete finite element stress-deformation field. The expression is , Indicates a shared VAE decoder;

[0126] Therefore, the shared VAE encoder maps the high-dimensional stress-deformation field, i.e. the finite element complete stress-deformation field, to the latent variables in the low-dimensional latent space, and the shared VAE decoder then reconstructs the latent variables in the low-dimensional latent variable space into the finite element complete stress-deformation field.

[0127] The training loss of the shared variational autoencoder Represented as:

[0128] (4)

[0129] (5)

[0130] in: This represents the reconstruction loss of the complete stress-deformation field in the finite element method. Latent space regularization constraint; Indicates the regularization weight; Indicates KL divergence; This represents the complete stress-deformation field of the input finite element model as determined by the shared VAE encoder. latent variables obtained The approximate posterior distribution, This represents the pre-defined prior distribution of the latent variables.

[0131] Therefore, by training a shared variational autoencoder using reconstruction loss and latent space regularization constraints, the low-dimensional latent space can characterize the main stress-deformation response modes of earth-rock dams under different dam types, material parameters, construction paths, and water level conditions. After training, a shared VAE encoder and a shared VAE decoder are obtained and used for prior field encoding and posterior field reconstruction in the online phase.

[0132] Step A3: Take the complete finite element stress-deformation field of each sample in the multi-condition finite element complete stress-deformation field sample library as the target field, construct a biased prior field paired with each target field through a perturbation algorithm, and then obtain a sample library of paired biased prior fields and target fields; based on the trained shared variational autoencoder, use the paired sample library to train the latent variable correction model offline.

[0133] To enable the predictive model of this invention to learn from the systematic errors that may occur in the finite element model of earth-rock dams in real engineering projects, a biased prior field and a target field are paired. The biased prior field can be generated by pairing high- and low-precision finite element models, dam height scale deviation, material parameter disturbance, construction path deviation, water level response sensitivity deviation, creep or wetting parameter deviation, and latent space disturbance. Among these, the dam height scale deviation can be used to simulate systematic errors such as underestimating the calculated displacement of high dams and overestimating the calculated displacement of low dams.

[0134] It should be noted that the paired samples of the target field and the biased prior field are only used in the offline training phase to enable the monitoring encoder and the latent variable updater to learn the mapping relationship of the prior latent variables corrected by the monitoring residuals. In the online application phase, the pseudo-field monitoring data matrix obtained by mapping the target field to the sparse field monitoring space is replaced by the field monitoring data. The monitoring residual is directly formed by the field monitoring data and the current prior monitoring prediction value to update the current prior latent variables posteriorly, so as to realize the assimilation of monitoring data to update the prior latent variables of the prior stress-deformation field mapping, thereby correcting the errors of the finite element model.

[0135] Step A3 includes steps A31 to A35:

[0136] Step A31: Construct paired samples of the biased prior field and the target field;

[0137] Each finite element complete stress-deformation field sample in the multi-condition finite element complete stress-deformation field sample library Finite element complete stress-deformation field As the target field, it is re-represented as ;in, Indicates engineering conditions , This represents the engineering phase; the target field is constructed using a perturbation algorithm. Paired biased prior fields ;

[0138] Biased a priori field and target field The relationship between them can be represented as , This indicates a systematic deviation introduced by material parameters, construction path, water level conditions, scale effects, or latent space disturbances.

[0139] As an example, an engineering condition perturbation algorithm or a latent space perturbation algorithm can be used to obtain the target field. Paired biased prior fields :

[0140] Engineering condition perturbation algorithm: for engineering conditions According to the set deviation strength parameters To perform the perturbation, use the deviation generation function. Generate system deviation The expression is ; to system bias Acting on the target field , obtain the target field Paired biased prior fields ;

[0141] The latent space perturbation algorithm is expressed as:

[0142]

[0143]

[0144] target field The input is sent to the shared VAE encoder, which will then input the target field. Encoding target latent variables in a low-dimensional latent space ;

[0145] target latent variables Perturbation is performed to obtain the perturbed target latent variables. The expression is: ;in, Indicates the intensity of the disturbance; Let represent the latent space perturbation term, and let represent the random perturbation vector sampled from the multivariate normal distribution. This indicates that the mean is 0 and the covariance matrix is... The multivariate normal distribution, It is an identity matrix.

[0146] Using the shared VAE decoder, the perturbed target latent variables are... Reconstructed into a complete finite element stress-deformation field, which serves as the target field. Paired biased prior fields The expression is ;, This indicates a shared VAE decoder.

[0147] Step A32, using the same observation operator , respectively target field and biased a priori fields Mapping to the sparse field monitoring space yields the target monitoring prediction matrix for the sparse field monitoring space. And biased monitoring prediction matrix The expression is and ;

[0148] Step A33, generate the target monitoring prediction value matrix. As a pseudo-site monitoring data matrix, the biased monitoring prediction matrix will be used. As the prior monitoring and prediction value matrix, the formula is used. Obtain the pseudo-monitoring residual matrix The data is then input into the monitoring encoder to obtain pseudo-sparse monitoring correction features. ;

[0149] Step A34, respectively target fields and biased a priori fields The input is sent to the shared VAE encoder, which will then input the target field. Encoding target latent variables in a low-dimensional latent space There will be a biased prior field Encoded as biased prior latent variables in a low-dimensional latent space ;

[0150] Step A35, the pseudo-sparse monitoring correction feature and the biased prior latent variables The latent variable updater is input, and the latent variable updater learns the pseudo-sparse monitoring and correction features. For the biased prior latent variables The latent variable correction amount is used to update the biased prior latent variable. Obtain biased posterior latent variables Through training, the biased posterior latent variables are made... Continuously approaching the target latent variable The monitoring encoder and the latent variable updater are trained to obtain the trained latent variable correction model.

[0151] The present invention discloses a method and system for time-series prediction of stress-deformation field of earth-rock dams based on finite element prior and monitoring data assimilation. As one implementation method, the method includes:

[0152] Offline training phase:

[0153] Construct a complete finite element stress-deformation field sample library for earth-rock dams, including multiple dam types, multiple dam heights, multiple dam body zones, multiple material parameters, multiple construction loading paths, multiple water level change paths, and multiple boundary conditions;

[0154] A shared variational autoencoder is trained based on a complete stress-deformation field sample library of earth-rock dams. The shared variational autoencoder includes an encoder for mapping the complete stress-deformation field to latent variables and a decoder for reconstructing the complete stress-deformation field from the latent variables. By training the shared variational autoencoder, a low-dimensional latent space of the complete finite element field is established, which represents the main stress-deformation response modes under different dam types, material parameters, construction conditions and water levels.

[0155] Based on a complete finite element stress-deformation field sample library for earth-rock dams, a paired sample of biased prior fields and target fields was constructed. This sample was used to train a latent variable correction model composed of a monitoring encoder and a latent variable updater. This enabled the latent variable updater to generate posterior latent variables based on prior latent variables and sparse monitoring correction features extracted from monitoring residuals. The biased prior fields were generated through at least one of the following methods: pairing high- and low-precision finite element models, dam height scale bias, material parameter perturbation, construction path bias, water level response sensitivity bias, long-term deformation parameter perturbation, creep or wetting parameter bias, and latent space perturbation. Specifically, the dam height scale bias was used to simulate the scale effect error of underestimating the calculated displacement of high dams or overestimating the calculated displacement of low dams.

[0156] The biased prior field and the corresponding target field are paired to form samples, so that the model can learn the systematic deviations that may exist between the finite element calculation results and the actual engineering response.

[0157] Online assimilation and prediction phase:

[0158] Based on the observation operator, the prior stress-deformation field is mapped to the actual monitoring point to obtain the prior monitoring prediction value that is comparable to the field monitoring data.

[0159] Acquire field monitoring data and construct monitoring residuals based on the difference between field monitoring data and prior monitoring predictions; the monitoring residuals are used to characterize the deviation between the current prior stress-deformation field and the actual engineering response.

[0160] Sparse monitoring correction features are extracted from the monitoring residuals by monitoring encoders; the prior stress-deformation field is encoded into prior latent variables by a shared VAE encoder of a shared variational autoencoder. The prior latent variables characterize the stress-deformation response mode of earth-rock dams, and the prior latent variables and sparse monitoring correction features are fused in the latent space to obtain posterior latent variables.

[0161] The complete stress-deformation field after monitoring assimilation is recovered by a shared VAE decoder with a shared variational autoencoder, and used as the posterior stress-deformation field.

[0162] Online rolling update: Taking the a priori stress-deformation field or a priori latent variables as the initial state, and combining conditional variables such as construction level, water level, temperature, dam age and material parameters, when new field monitoring data exists in the future stage, the monitoring residual calculation, monitoring coding, latent variable update, decoding reconstruction and autoregressive prediction process are repeatedly executed, thereby realizing the rolling prediction of stress-deformation field in the future construction or operation stage.

[0163] This invention also provides a time-series prediction system for the stress-deformation field of earth-rock dams based on finite element prior and monitoring data assimilation, comprising:

[0164] The finite element model construction and simulation module for earth-rock dams is used to determine the current engineering conditions and current engineering stage of earth-rock dams, and input them into the constructed finite element model of earth-rock dams to perform complete stress-deformation response simulation, and obtain a high-dimensional complete finite element stress-deformation field as the a priori stress-deformation field for the current engineering stage.

[0165] An observation operator is used to map the prior stress-deformation field to a sparse field monitoring space to obtain a prior monitoring prediction matrix that is comparable to the field monitoring data matrix.

[0166] A monitoring residual matrix construction module is used to construct a monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction value matrix.

[0167] A monitoring encoder is used to extract features from the monitoring residual matrix to obtain sparse monitoring correction features;

[0168] A shared VAE encoder is used to encode the prior stress deformation field into prior latent variables in a low-dimensional latent space;

[0169] A latent variable updater is used to correct the prior latent variable based on the sparse monitoring correction feature in a low-dimensional latent space to obtain the posterior latent variable;

[0170] A shared VAE decoder is used to reconstruct the posterior latent variables into a complete finite element stress-deformation field after monitoring assimilation;

[0171] The autoregressive propagation model is used to take the monitored and assimilated finite element complete stress-deformation field as the posterior stress-deformation field of the current engineering stage, combine it with the engineering conditions of the next engineering stage, predict the finite element complete stress-deformation field of the next engineering stage, and take the finite element complete stress-deformation field of the next engineering stage as the prior stress-deformation field of the next engineering stage, and transmit it to the observation operator and the shared VAE encoder. This process is repeated to achieve online assimilation and rolling prediction.

[0172] This invention enables dynamic correction and future state prediction of finite element full-field prior under sparse monitoring conditions, improving the spatial integrity, real-time performance, and accuracy of safety state evaluation during the construction and initial operation phases of earth-rock dams. It has significant practicality and broad application prospects.

[0173] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A soil and rockfill dam stress and deformation field time series prediction method based on finite element prior and monitoring data assimilation, characterized in that, include: Step S1: Determine the current engineering conditions and current engineering stage of the earth-rock dam, and input them into the constructed earth-rock dam finite element model to perform a complete stress-deformation response simulation, and obtain a high-dimensional finite element complete stress-deformation field as the prior stress-deformation field of the current engineering stage. Step S2: Based on the observation operator, the prior stress deformation field is mapped to the sparse field monitoring space to obtain a prior monitoring prediction matrix that is comparable to the field monitoring data matrix. A monitoring residual matrix is ​​constructed based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix. The monitoring residual matrix is ​​then used to extract features from the monitoring residual matrix through a pre-trained monitoring encoder to obtain sparse monitoring correction features. Step S3: The prior stress deformation field is encoded into prior latent variables in a low-dimensional latent space using a pre-trained shared VAE encoder; the prior latent variables are corrected in the low-dimensional latent space based on the sparse monitoring and correction features using a pre-trained latent variable updater to obtain posterior latent variables. Step S4: Input the posterior latent variable into the pre-trained shared VAE decoder, and the shared VAE decoder reconstructs the posterior latent variable into the monitored and assimilated complete finite element stress-deformation field; Step S5: The monitored and assimilated finite element complete stress-deformation field is used as the posterior stress-deformation field of the current engineering stage. Combined with the engineering conditions of the next engineering stage, the finite element complete stress-deformation field of the next engineering stage is predicted by the autoregressive propagation model. Then, return to step S2 to realize online assimilation and rolling prediction. When constructing the monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix, the missing field monitoring data is considered, and a missing data mask matrix is ​​introduced. Formula (1) is used to obtain the monitoring residual matrix: （1） wherein: denotes a missing mask matrix, and denotes whether the field monitoring data of each monitoring point in the sparse field monitoring space is valid. If valid, the corresponding element in the missing mask matrix takes the value of 1, otherwise takes the value of 0; denotes element-wise multiplication; denotes the field monitoring data matrix of the current engineering stage ; denotes the prior monitoring prediction value matrix of the current engineering stage ; denotes the monitoring residual error matrix; The monitoring residual matrix is ​​used to extract features through a pre-trained monitoring encoder to obtain sparse monitoring correction features, including: Monitor residual matrix Monitoring point location matrix Monitoring point type matrix Missing test mask matrix Prior monitoring prediction matrix and on-site monitoring data matrix A common input is used to pre-train a monitoring encoder, which obtains sparse monitoring correction features. The expression is: （2） in: Indicates monitoring encoder; As online assimilation and rolling prediction continue to advance, the number of monitoring points with on-site monitoring data in the sparse field monitoring space is constantly increasing. The monitoring encoder adopts an ensemble coding structure, firstly extracting features from each monitoring point, and then performing pooling aggregation on the features extracted from each monitoring point. The expression is as follows: （3） in: The feature extraction function represents a single monitoring point. Indicates the current stage of the project. The number of monitoring points; Indicates the current stage of the project. Monitoring points , Indicates the current stage of the project. The number of monitoring points; , , , , and , respectively representing the monitoring points The on-site monitoring data includes prior monitoring predictions, monitoring residuals, monitoring point locations, monitoring point monitoring data, and missing measurement masks.

2. The method for time-series prediction of stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation as described in claim 1, characterized in that, The engineering conditions include at least one of the following: geometric parameters of the earth-rock dam body, material parameters, construction loading path, water level change path, and boundary conditions; the engineering stages include the construction period, the impoundment period, and the operation period; the complete finite element stress-deformation field includes at least one of the following: horizontal displacement field, vertical settlement field, stress field, and pore pressure field; the observation operator includes at least one of the following: interpolation operator, projection operator, and integration operator.

3. The method for time-series prediction of stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation as described in claim 1, characterized in that, The process of mapping the prior stress-deformation field to a sparse field monitoring space based on the observation operator to obtain a prior monitoring prediction matrix comparable to the field monitoring data matrix includes: The sparse field monitoring space has multiple monitoring points with corresponding field monitoring data. Based on the observation operator, the prior stress-deformation field is mapped to the sparse field monitoring space to obtain the prior monitoring prediction value for each monitoring point, thereby obtaining the prior monitoring prediction value matrix of the sparse field monitoring space, specifically including: For displacement or settlement monitoring points, the prior stress deformation field is interpolated based on the element shape function of the finite element unit where the monitoring point is located to obtain the prior monitoring prediction value at the corresponding monitoring point. For inclinometer, plumb line or stratified settlement monitoring data, the prior stress-deformation field is projected or extracted in layers according to the monitoring path or monitoring depth to obtain the prior monitoring prediction value at the corresponding depth position. For fiber optic monitoring data, the prior monitoring prediction values ​​are linearly sampled or path integrated according to the fiber optic deployment path to obtain the prior monitoring prediction values ​​for each monitoring point along the line.

4. The method for time-series prediction of stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation as described in claim 1, characterized in that, The shared VAE encoder and the shared VAE decoder constitute a shared variational autoencoder; the monitoring encoder and the latent variable updater constitute a latent variable correction model; the training method for the shared variational autoencoder and the latent variable correction model during the offline training phase is as follows: Step A1: Based on the finite element model of the earth-rock dam, construct a complete stress-deformation field sample library for multiple working conditions using finite element methods. Step A2: Train the shared variational autoencoder using the multi-condition finite element complete stress-deformation field sample library; Through training, the shared VAE encoder is used to map the complete finite element stress-deformation field into latent variables in a low-dimensional latent space, including: mapping the complete finite element stress-deformation field into latent variable distribution parameters in the low-dimensional latent space, including the mean and standard deviation; obtaining the latent variables using a reparameterization method based on the mean and standard deviation; and through training, the shared VAE decoder is used to reconstruct the latent variables in the low-dimensional latent space to obtain the reconstructed complete finite element stress-deformation field. Step A3: Take the complete finite element stress-deformation field of each sample in the multi-condition finite element complete stress-deformation field sample library as the target field, construct a biased prior field paired with each target field through a perturbation algorithm, and then obtain a sample library of paired biased prior fields and target fields; based on the trained shared variational autoencoder, use the paired sample library to train the latent variable correction model offline.

5. The method for time-series prediction of stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation as described in claim 4, characterized in that, The multi-condition finite element complete stress-deformation field sample library is characterized as follows: ; Wherein: the sample library Finite element complete stress-deformation field sample Characterization Engineering Conditions and engineering phase Finite element complete stress-deformation field ; The number of samples representing the complete stress-deformation field of a finite element method; engineering conditions. , Indicates the geometric parameters of the dam body. Indicates material parameters, Indicate boundary conditions, Indicates the construction loading path. Indicates the path of water level change; dam geometric parameters Including dam type, dam height, and dam body zoning; During training, the shared VAE encoder will generate a complete finite element stress-deformation field. Mapped to latent variables in a low-dimensional latent space The latent variable distribution parameters, including the mean and standard deviation The expression is , Indicates a shared VAE encoder; based on the mean. and standard deviation The latent variables are obtained by reparameterization. The expression is: ;in, This represents a random perturbation vector sampled from a standard multivariate normal distribution. This represents a standard multivariate normal distribution with a mean of 0 and a covariance matrix of identity matrix I, where I represents the identity matrix. This represents element-wise multiplication; During training, the shared VAE decoder will store latent variables in the low-dimensional latent space. Reconstruction yields the complete finite element stress-deformation field. The expression is , Indicates a shared VAE decoder; The training loss of the shared variational autoencoder Represented as: （4） （5） in: This represents the reconstruction loss of the complete stress-deformation field in the finite element method. Latent space regularization constraint; Indicates the regularization weight; Indicates KL divergence; This represents the complete stress-deformation field of the input finite element model as determined by the shared VAE encoder. latent variables obtained The approximate posterior distribution, This represents the pre-defined prior distribution of the latent variables.

6. The method for time-series prediction of stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation as described in claim 4, characterized in that, Step A3 includes: Step A31: Construct paired samples of the biased prior field and the target field; Each finite element complete stress-deformation field sample in the multi-condition finite element complete stress-deformation field sample library Finite element complete stress-deformation field As the target field, it is re-represented as ;in, Indicates engineering conditions , This represents the engineering phase; the target field is constructed using a perturbation algorithm. Paired biased prior fields ; Step A32, using the same observation operator , respectively target field and biased a priori fields Mapping to the sparse field monitoring space yields the target monitoring prediction matrix for the sparse field monitoring space. And biased monitoring prediction matrix The expression is and ; Step A33, generate the target monitoring prediction value matrix. As a pseudo-site monitoring data matrix, the biased monitoring prediction matrix will be used. As the prior monitoring and prediction value matrix, the formula is used. Obtain the pseudo-monitoring residual matrix The data is then input into the monitoring encoder to obtain pseudo-sparse monitoring correction features. ; Step A34, respectively target fields and biased a priori fields The input is sent to the shared VAE encoder, which will then input the target field. Encoding target latent variables in a low-dimensional latent space There will be a biased prior field Encoded as biased prior latent variables in a low-dimensional latent space ; Step A35, the pseudo-sparse monitoring correction feature and the biased prior latent variables The latent variable updater is input, and the latent variable updater learns the pseudo-sparse monitoring and correction features. For the biased prior latent variables The latent variable correction amount is used to update the biased prior latent variable. Obtain biased posterior latent variables Through training, the biased posterior latent variables are made... Continuously approaching the target latent variable The monitoring encoder and latent variable updater are trained to obtain the trained latent variable correction model.

7. The method for time-series prediction of stress-deformation field of earth-rock dam based on finite element prior and monitoring data assimilation as described in claim 6, characterized in that, The target field is constructed using a perturbation algorithm. Paired biased prior fields ,include: Engineering condition perturbation algorithm: for engineering conditions According to the set deviation strength parameters To perform the perturbation, use the deviation generation function. Generate system deviation The expression is ; to system bias Acting on the target field , obtain the target field Paired biased prior fields ; Latent space perturbation algorithm: target field The input is sent to the shared VAE encoder, which will then input the target field. Encoding target latent variables in a low-dimensional latent space ; For target latent variables By perturbing the target latent variable, we can obtain the perturbed target latent variable. The expression is: ;in, Indicates the intensity of the disturbance; Let represent the latent space perturbation term, and let represent the random perturbation vector sampled from the multivariate normal distribution. This indicates that the mean is 0 and the covariance matrix is... The multivariate normal distribution, It is the identity matrix; Using the shared VAE decoder, the perturbed target latent variables are... Reconstructed into a complete finite element stress-deformation field, which serves as the target field. Paired biased prior fields The expression is ;, This indicates a shared VAE decoder.

8. A time-series prediction system for the stress-deformation field of earth-rock dams based on finite element prior and monitoring data assimilation, characterized in that, include: The finite element model construction and simulation module for earth-rock dams is used to determine the current engineering conditions and current engineering stage of earth-rock dams, and input them into the constructed finite element model of earth-rock dams to perform complete stress-deformation response simulation, and obtain a high-dimensional complete finite element stress-deformation field as the a priori stress-deformation field for the current engineering stage. An observation operator is used to map the prior stress-deformation field to a sparse field monitoring space to obtain a prior monitoring prediction matrix that is comparable to the field monitoring data matrix. A monitoring residual matrix construction module is used to construct a monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction value matrix. When constructing the monitoring residual matrix based on the difference between the field monitoring data matrix and the prior monitoring prediction matrix, the missing field monitoring data is considered, and a missing data mask matrix is ​​introduced. Formula (1) is used to obtain the monitoring residual matrix: （1） in: The missing measurement mask matrix indicates whether the field monitoring data of each monitoring point in the sparse field monitoring space is valid. If it is valid, the corresponding element in the missing measurement mask matrix is ​​1; otherwise, it is 0. This represents element-wise multiplication; Indicates the current stage of the project. The on-site monitoring data matrix; Indicates the current stage of the project. The prior monitoring prediction matrix; Represents the monitoring residual matrix; The monitoring encoder is used to extract features from the monitoring residual matrix to obtain sparse monitoring correction features, including: Monitor residual matrix Monitoring point location matrix Monitoring point type matrix Missing test mask matrix Prior monitoring prediction matrix and on-site monitoring data matrix A common input is used to pre-train a monitoring encoder, which obtains sparse monitoring correction features. The expression is: （2） in: Indicates monitoring encoder; As online assimilation and rolling prediction continue to advance, the number of monitoring points with on-site monitoring data in the sparse field monitoring space is constantly increasing. The monitoring encoder adopts an ensemble coding structure, firstly extracting features from each monitoring point, and then performing pooling aggregation on the features extracted from each monitoring point. The expression is as follows: （3） in: The feature extraction function represents a single monitoring point. Indicates the current stage of the project. The number of monitoring points; Indicates the current stage of the project. Monitoring points , Indicates the current stage of the project. The number of monitoring points; , , , , and , respectively representing the monitoring points The on-site monitoring data includes prior monitoring predictions, monitoring residuals, monitoring point locations, monitoring point monitoring data, and missing measurement masks. A shared VAE encoder is used to encode the prior stress deformation field into prior latent variables in a low-dimensional latent space; A latent variable updater is used to correct the prior latent variable based on the sparse monitoring correction feature in a low-dimensional latent space to obtain the posterior latent variable; A shared VAE decoder is used to reconstruct the posterior latent variables into a complete finite element stress-deformation field after monitoring assimilation; The autoregressive propagation model is used to take the monitored and assimilated finite element complete stress-deformation field as the posterior stress-deformation field of the current engineering stage, combine it with the engineering conditions of the next engineering stage, predict the finite element complete stress-deformation field of the next engineering stage, and take the finite element complete stress-deformation field of the next engineering stage as the prior stress-deformation field of the next engineering stage, and transmit it to the observation operator and the shared VAE encoder. This process is repeated to achieve online assimilation and rolling prediction.

Images