A dam structure deformation prediction method and system based on digital twinning

Abstract

The application discloses a kind of based on digital twin's dam structure deformation prediction method and system, the method includes: obtaining dam digital twin configured with measuring point space mapping relationship, fluid-structure interaction finite element model and dam deformation state estimate value;The multi-source monitoring data of dam to be measured is time-space alignment, and the seepage driving sequence and dam deformation sequence are obtained;Nonlinear time series features of seepage driving sequence are extracted, hidden state vector is obtained, and the current input condition of fluid-structure interaction finite element model is determined accordingly, and the additional deformation increment at the current time is calculated;Kalman filter correction is carried out in combination with actual monitoring deformation value, and the dam deformation prediction result at the current time is obtained;According to residual error, judge seepage anomaly, and generate abnormal feedback update amount to update model when abnormal, and then the prior dam deformation prediction result at the next time is obtained.The application can improve the accuracy of dam deformation prediction under abnormal seepage working condition.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy project safety monitoring technology, and in particular relates to a method and system for predicting dam structural deformation based on digital twins. Background Technology

[0002] During long-term operation, dams are affected by various factors such as reservoir water level changes, dam seepage, temperature changes, external loads, and changes in dam foundation conditions, which may cause varying degrees of deformation in the dam body and foundation structure. To ensure the safe operation of the dam, it is usually necessary to continuously collect dam operation status data using monitoring equipment such as piezometers, displacement gauges, and water level gauges, and to analyze and predict the deformation trend of the dam structure based on the monitoring data.

[0003] Existing methods for predicting dam deformation typically include statistical regression methods based on historical monitoring data, data-driven prediction methods based on machine learning models, and physical simulation analysis methods based on finite element models. Among these, statistical regression methods rely heavily on historical data patterns and struggle to accurately reflect the coupling relationship between seepage changes and dam deformation under complex conditions such as rapid water level changes, local seepage channel evolution, or variations in dam material parameters. While data-driven prediction methods can extract nonlinear patterns from time-series data, their predictions often lack clear physical constraints, making it difficult to explain the specific impact of local seepage anomalies on dam structural deformation. Finite element analysis can describe the physical relationship between the dam structure and the seepage field; however, in actual operation, the boundary conditions and model parameters of the finite element model are often difficult to update in a timely manner with changes in monitoring data, leading to discrepancies between the model calculation results and the actual monitoring conditions.

[0004] Especially under abnormal conditions such as accelerated seepage, cracking of the contact surface, or expansion of seepage channels in a dam, the seepage state may change significantly in a short period of time, further altering the pore pressure distribution and the dam's deformation response. Existing methods typically only issue alarms based on the deviation between predicted and monitored values, or simply correct the predicted results, making it difficult to feed back the abnormal seepage state into the physical model to update the fluid-structure interaction calculation process. Therefore, existing technologies still suffer from problems such as delayed model updates under abnormal conditions, insufficient characterization of the impact of seepage anomalies on subsequent deformation predictions, and low accuracy in predicting deformation at the next moment. Summary of the Invention

[0005] To address the problems of existing dam deformation prediction methods, such as difficulty in timely updating of physical models under abnormal seepage conditions and insufficient accuracy of subsequent deformation prediction, this invention provides a dam structure deformation prediction method and system based on digital twins. This method can accurately predict dam deformation.

[0006] In a first aspect, the present invention provides a method for predicting dam structural deformation based on digital twins, comprising:

[0007] Obtain a pre-constructed digital twin of the dam body, which contains the spatial mapping relationship of the measuring points of the dam to be measured, the fluid-structure interaction finite element model, and the estimated value of the dam body deformation state.

[0008] Multi-source monitoring data of the dam under test were acquired, and the multi-source monitoring data were spatiotemporally aligned based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence.

[0009] The nonlinear temporal features of the seepage-driven sequence are extracted to obtain the hidden state vector corresponding to each time step in the seepage-driven sequence.

[0010] Based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence, the current input conditions of the fluid-structure interaction finite element model are determined, and the additional deformation increment at the current moment is calculated.

[0011] The estimated value of dam deformation state at the previous moment is updated based on the additional deformation increment at the current moment to obtain the a priori dam deformation prediction result at the current moment. Kalman filtering correction is then performed on the actual monitored deformation value at the current moment in the dam deformation sequence to obtain the dam deformation prediction result at the current moment.

[0012] Based on the residual between the predicted deformation of the dam body at the current moment and the actual monitored deformation value at the current moment, it is determined whether there is an abnormal seepage.

[0013] If seepage anomalies exist, local feature enhancement is performed by combining the local seepage driving sub-sequences associated with the current time and the hidden state vector corresponding to the current time in the seepage driving sequence to obtain enhanced dynamic change characterization. Based on the enhanced dynamic change characterization, anomaly feedback update amount is generated to update the fluid-structure interaction finite element model to obtain the target model.

[0014] If there is no seepage anomaly, the fluid-structure interaction finite element model will be used as the target model.

[0015] The additional deformation increment at the next moment is calculated based on the target model, and the dam deformation prediction result at the current moment is updated based on the additional deformation increment at the next moment to obtain the a priori dam deformation prediction result at the next moment.

[0016] Optionally, the multi-source monitoring data includes: seepage pressure data, dam displacement data, and water level data; the spatiotemporal alignment processing of the multi-source monitoring data based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence includes:

[0017] The multi-source monitoring data is filled with missing values ​​and outliers are removed to obtain cleaned multi-source monitoring data;

[0018] Based on the spatial coordinates and acquisition timestamps of each measuring point, time alignment and spatial matching are performed on the cleaned multi-source monitoring data.

[0019] Based on the time-aligned and spatially matched seepage pressure data, a seepage driving sequence was determined to characterize seepage changes.

[0020] Based on the dam displacement data after time alignment and spatial matching, the dam deformation sequence is determined.

[0021] Optionally, the step of extracting the nonlinear temporal features of the seepage-driven sequence to obtain the hidden state vector corresponding to each time step in the seepage-driven sequence includes:

[0022] The seepage-driven sequence is input into the time-series feature extraction model;

[0023] The time-series feature extraction model is used to extract time-series features from the seepage-driven data at different sampling times to obtain the hidden state vectors corresponding to each time. The hidden state vectors are used to characterize the dynamic changes in seepage at the corresponding time and the influence of historical seepage changes.

[0024] Optionally, determining the current input conditions of the fluid-structure interaction finite element model based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence includes:

[0025] Based on the seepage driving data at the current moment, the seepage boundary conditions of the fluid-structure interaction finite element model are determined.

[0026] Based on the hidden state vector corresponding to the current moment, determine at least one of the pore pressure boundary correction, permeability coefficient correction, and elastic modulus correction of the fluid-structure interaction finite element model.

[0027] Based on the seepage boundary conditions and at least one of the pore pressure boundary correction, permeability coefficient correction, and elastic modulus correction, the current input conditions of the fluid-structure interaction finite element model are determined.

[0028] Optionally, calculating the additional deformation increment at the current moment includes:

[0029] The current input conditions are loaded into the fluid-structure interaction finite element model, and the fluid flow equations in the fluid-structure interaction finite element model are solved to obtain the preliminary pore pressure distribution at the current moment.

[0030] By combining the property calibration data obtained from the dam material test, the preliminary pore pressure distribution is calibrated to obtain the pore pressure distribution at the current moment;

[0031] Obtain the water level data at the current moment, and determine the degree of coupling between the current seepage and the dam deformation based on the deviation between the pore pressure distribution at the current moment and the water level data;

[0032] If the coupling degree does not exceed the preset coupling threshold, the pore pressure distribution at the current moment is determined as the target pore pressure distribution, and the additional deformation increment at the current moment is calculated based on the target pore pressure distribution and the solid deformation equation in the fluid-structure interaction finite element model.

[0033] If the degree of coupling exceeds the preset coupling threshold, the elastic modulus and permeability coefficient in the fluid-structure interaction finite element model are adjusted according to the degree of coupling, and the pore pressure distribution at the current moment is corrected based on the adjusted permeability coefficient to obtain the target pore pressure distribution.

[0034] Based on the target pore pressure distribution and the solid deformation equation in the adjusted fluid-structure interaction finite element model, the additional deformation increment at the current moment is calculated.

[0035] Optionally, the step of updating the estimated dam deformation state value of the previous moment based on the additional deformation increment at the current moment to obtain the prior dam deformation prediction result at the current moment, and then performing Kalman filtering correction in combination with the actual monitored deformation value at the current moment in the dam deformation sequence to obtain the dam deformation prediction result at the current moment, includes:

[0036] The estimated value of the dam deformation state at the previous moment is superimposed with the additional deformation increment at the current moment to obtain the a priori dam deformation prediction result at the current moment.

[0037] Based on the deviation between the a priori dam deformation prediction result and the actual monitored deformation value at the current moment, the deformation observation residual at the current moment is determined;

[0038] The Kalman correction weights are determined based on the prediction error estimate corresponding to the a priori dam deformation prediction result at the current moment and the monitoring error estimate corresponding to the actual monitored deformation value at the current moment.

[0039] Based on the Kalman correction weights and the deformation observation residuals, the a priori dam deformation prediction results at the current moment are corrected to obtain the dam deformation prediction results at the current moment.

[0040] The predicted deformation result of the dam body at the current moment is used as the estimated value of the dam body deformation state at the current moment in the digital twin of the dam body.

[0041] Optionally, determining whether there is an anomaly in seepage based on the residual between the predicted dam deformation at the current moment and the actual monitored deformation value at the current moment includes:

[0042] Calculate the difference between the predicted deformation of the dam body at the current moment and the actual monitored deformation value at the current moment, and determine the residual at the current moment based on the absolute value or norm of the difference;

[0043] Obtain the residuals within a preset anomaly detection time window, including the current time, and form a residual change sequence;

[0044] The residual at the current moment is compared with a preset residual threshold, and the residual change trend is determined based on the residual change sequence;

[0045] If the residual at the current moment exceeds the preset residual threshold, and the residual change sequence satisfies the preset abnormality persistence condition, then it is determined that there is a seepage anomaly.

[0046] The preset abnormality persistence condition includes: the residuals at a preset number of consecutive sampling times all exceed the preset residual threshold, and the residual change sequence shows an increasing trend.

[0047] Optionally, the step of combining the local seepage driving sub-sequence extracted from the seepage driving sequence and associated with the current time with the hidden state vector corresponding to the current time for local feature enhancement to obtain a strengthened dynamic change representation includes:

[0048] Using the current moment as the endpoint or center, a local seepage driving sub-sequence is extracted from the seepage driving sequence according to a preset anomaly analysis time window;

[0049] Obtain the hidden state vector corresponding to the current time from the hidden state vectors corresponding to each time step;

[0050] The hidden state vector corresponding to the current moment is subjected to dimension matching processing, and the dimension-matched hidden state vector is fused with the local seepage driving subsequence to obtain the enhanced input sequence;

[0051] The enhanced input sequence is input into the time-series feature extraction model. Based on the seepage change amplitude at each sampling time in the enhanced input sequence, the corresponding feature update weights are determined. The enhanced input sequence is then subjected to time-series feature extraction according to the feature update weights to obtain a representation of enhanced dynamic changes.

[0052] The magnitude of the seepage change is determined based on the difference or norm of the seepage-driven data at adjacent sampling times.

[0053] Optionally, the temporal feature extraction model may employ any one of the following: Long Short-Term Memory Network, Gated Recurrent Unit, or Temporal Convolutional Network.

[0054] Secondly, the present invention provides a dam structural deformation prediction system based on digital twins, comprising:

[0055] The digital twin acquisition module is used to acquire a pre-constructed digital twin of the dam body, which contains the spatial mapping relationship of the measurement points of the dam to be measured, the fluid-structure interaction finite element model, and the estimated value of the dam body deformation state.

[0056] The data processing module is used to acquire multi-source monitoring data of the dam under test, and perform spatiotemporal alignment processing on the multi-source monitoring data based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence.

[0057] The temporal feature extraction module is used to extract the nonlinear temporal features of the seepage-driven sequence and obtain the hidden state vector corresponding to each time step in the seepage-driven sequence.

[0058] The deformation increment acquisition module is used to determine the current input conditions of the fluid-structure interaction finite element model based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence, and to calculate the additional deformation increment at the current moment.

[0059] The deformation prediction module is used to update the estimated value of the dam deformation state at the previous time based on the additional deformation increment at the current time, to obtain the a priori dam deformation prediction result at the current time, and to perform Kalman filtering correction in combination with the actual monitored deformation value at the current time in the dam deformation sequence to obtain the dam deformation prediction result at the current time.

[0060] The anomaly detection module is used to determine whether there is an anomaly in seepage based on the residual between the predicted deformation result of the dam body at the current moment and the actual monitored deformation value at the current moment.

[0061] The model update module is used to perform local feature enhancement by combining the local seepage driving sub-sequence that is extracted from the seepage driving sequence and associated with the current time and the hidden state vector corresponding to the current time if seepage anomalies exist, so as to obtain enhanced dynamic change characterization, and generate anomaly feedback update amount based on the enhanced dynamic change characterization to update the fluid-structure interaction finite element model to obtain the target model.

[0062] The model update module is also used to use the fluid-structure interaction finite element model as the target model if there is no seepage anomaly.

[0063] The next moment prediction module is used to calculate the additional deformation increment at the next moment based on the target model, and to update the dam deformation prediction result at the current moment based on the additional deformation increment at the next moment, so as to obtain the a priori dam deformation prediction result at the next moment.

[0064] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0065] This invention provides a method and system for predicting dam structural deformation based on digital twins. First, the method uses a pre-constructed digital twin of the dam body to carry the spatial mapping relationship of the measuring points, the fluid-structure interaction (FSI) finite element model, and the estimated deformation state of the dam body. This allows actual monitoring data to be loaded into the corresponding digital twin calculation model according to the spatial location of the measuring points. Thus, multi-source monitoring data such as seepage pressure, dam displacement, and water level are no longer isolated participants in trend analysis, but can be correlated with the boundary conditions, pore pressure distribution, and deformation state estimation process in the FSI finite element model, thereby improving the consistency between the prediction process and the actual operating state of the dam. Second, based on the seepage driving data and the corresponding hidden state vector at the current moment in the seepage driving sequence, the current input conditions of the FSI finite element model are determined. This allows the FSI finite element model to not only receive the physical driving information corresponding to the current seepage state, but also to incorporate the dynamic influence of historical seepage changes on the current moment. This allows for a more accurate characterization of the impact of seepage state changes over time on pore pressure distribution and additional dam deformation, avoiding prediction lag caused by relying solely on finite element calculations based on single-moment monitoring values. Then, the estimated dam deformation state from the previous moment is updated using the current additional deformation increment, yielding the current moment's prior dam deformation prediction. This prediction is then corrected using Kalman filtering, combining the current moment's actual monitored deformation value to integrate the physical prediction from the finite element model with the actual monitoring results. This approach preserves the seepage-structure coupling physical constraints while using real-time monitoring data to correct prediction biases, improving the reliability of the current moment's dam deformation prediction. Furthermore, after obtaining the current moment's dam deformation prediction, the residual between the prediction and the actual monitored deformation value is used to determine if seepage anomalies exist. When seepage anomalies are present, this invention does not merely output an alarm; instead, it enhances the local seepage driving sub-sequence associated with the current moment and the corresponding hidden state vector, obtaining a strengthened dynamic change characterization. Based on this, an anomaly feedback update is generated to update the fluid-structure interaction finite element model. Therefore, the anomaly identification results can have a reverse effect on the subsequent physical model calculation process, enabling the model to be dynamically updated in response to local seepage anomalies. Finally, by determining the target model under both the presence and absence of seepage anomalies, and calculating the additional deformation increment for the next moment based on the target model, the current dam deformation prediction result is updated to obtain the a priori dam deformation prediction result for the next moment. Thus, this invention forms a closed-loop prediction mechanism of "current prediction - residual anomaly judgment - anomaly feedback update - a priori prediction for the next moment," which can improve the real-time performance and accuracy of dam structural deformation prediction under abnormal seepage conditions. Attached Figure Description

[0066] Figure 1The diagram shown is a flowchart illustrating a dam structure deformation prediction method based on digital twins in one embodiment of the present invention. Detailed Implementation

[0067] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0068] Example 1

[0069] like Figure 1 As shown in the figure, this invention provides a method for predicting dam structural deformation based on digital twins, including:

[0070] S1: Obtain the pre-constructed digital twin of the dam body, which contains the spatial mapping relationship of the measuring points of the dam to be measured, the fluid-structure interaction finite element model, and the estimated value of the dam body deformation state.

[0071] S2: Acquire multi-source monitoring data of the dam under test, and perform spatiotemporal alignment processing on the multi-source monitoring data based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence;

[0072] S3: Extract the nonlinear temporal features of the seepage-driven sequence to obtain the hidden state vector corresponding to each time step in the seepage-driven sequence;

[0073] S4: Based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence, determine the current input conditions of the fluid-structure interaction finite element model and calculate the additional deformation increment at the current moment.

[0074] S5: Update the estimated value of the dam deformation state at the previous moment based on the additional deformation increment at the current moment, obtain the a priori dam deformation prediction result at the current moment, and perform Kalman filtering correction in combination with the actual monitored deformation value at the current moment in the dam deformation sequence to obtain the dam deformation prediction result at the current moment.

[0075] S6: Based on the residual between the predicted deformation of the dam body at the current moment and the actual monitored deformation value at the current moment, determine whether there is any seepage anomaly;

[0076] S7: If seepage anomaly exists, local feature enhancement is performed by combining the local seepage driving sub-sequence associated with the current time and the hidden state vector corresponding to the current time extracted from the seepage driving sequence to obtain enhanced dynamic change characterization, and anomaly feedback update amount is generated based on the enhanced dynamic change characterization to update the fluid-structure interaction finite element model to obtain the target model; if there is no seepage anomaly, the fluid-structure interaction finite element model is used as the target model.

[0077] S8: Calculate the additional deformation increment at the next moment based on the target model, and update the dam deformation prediction result at the current moment based on the additional deformation increment at the next moment to obtain the a priori dam deformation prediction result at the next moment.

[0078] In this embodiment, the dam to be monitored can be a concrete gravity dam, arch dam, earth-rock dam, or other dam structure requiring seepage monitoring and structural deformation prediction. The dam to be monitored can be equipped with monitoring points such as seepage pressure measuring points, dam body displacement measuring points, and water level measuring points. Seepage pressure measuring points can be used to collect seepage pressure data at different locations within the dam body or dam foundation; dam body displacement measuring points can be used to collect horizontal, vertical, or three-dimensional displacement data at different measuring point locations within the dam body; water level measuring points can be used to collect data on upstream water level, downstream water level, or reservoir water level changes. The above monitoring data can be collected in real-time or periodically by the dam safety monitoring system and transmitted to computing equipment used to run the dam body digital twin.

[0079] In this embodiment, the dam digital twin is a virtual computational model corresponding to the dam under test. The dam digital twin includes a spatial mapping relationship of measurement points, a fluid-structure interaction (FSI) finite element model, and estimated values ​​of the dam's deformation state. The spatial mapping relationship of measurement points characterizes the correspondence between actual monitoring points and the mesh nodes, mesh elements, or model regions in the FSI finite element model, enabling actual monitoring data to be loaded into the corresponding positions in the digital twin based on their spatial locations. The FSI finite element model describes the coupling relationship between the dam's seepage field and its structural deformation field. The estimated values ​​of the dam's deformation state characterize the dam's deformation state at the previous or current time, and may include deformation values ​​corresponding to one or more measurement points, or deformation field data corresponding to the dam's finite element mesh nodes.

[0080] In this embodiment, steps S1-S2 first acquire a pre-constructed digital twin of the dam body and then acquire multi-source monitoring data of the dam under test. The multi-source monitoring data may include seepage pressure data, dam displacement data, and water level data. Since the sampling periods, acquisition timestamps, and deployment locations of different types of sensors may vary, it is necessary to perform spatiotemporal alignment processing on the multi-source monitoring data based on the spatial mapping relationship of the measuring points to obtain the seepage-driven sequence and the dam deformation sequence.

[0081] Specifically, missing value imputation and outlier removal are performed on the multi-source monitoring data to obtain cleaned multi-source monitoring data. Missing value imputation can be performed using methods such as forward imputation, linear interpolation, or estimation based on neighboring monitoring points. Outlier removal can be performed using the three-standard-deviation criterion, box plot threshold, or threshold methods set based on engineering experience. Then, based on the spatial coordinates and acquisition timestamps of each monitoring point, the cleaned multi-source monitoring data is time-aligned and spatially matched. For data with inconsistent timestamps, interpolation or resampling can be performed according to a preset unified time scale; for data with different spatial locations, the monitoring data is mapped to the corresponding finite element mesh nodes or mesh regions according to the spatial mapping relationship of the monitoring points. Based on the time-aligned and spatially matched seepage pressure data, the seepage driving sequence used to characterize seepage changes can be determined; based on the time-aligned and spatially matched dam displacement data, the dam deformation sequence can be determined.

[0082] For example, seepage measuring point A is located upstream of the dam body at coordinates (10, 5, 20), and displacement measuring point C is located downstream of the dam body at coordinates (60, 8, 18). Within the same monitoring period, seepage pressure data is collected at 8:00 and 10:00, and dam displacement data is collected at 9:00. If the preset timestamp difference threshold is 1 hour, linear interpolation can be performed on the seepage pressure data. Assuming the seepage pressure value p1 at 8:00 is 180 kPa and the pressure value p2 at 10:00 is 200 kPa, then the interpolated pressure value at 9:00 is: p = 180 + (200 - 180) × (9 - 8) / (10 - 8) = 190 kPa. Thus, seepage pressure data aligned with the dam displacement data in time can be obtained; the dam displacement data can also be aligned to a preset unified time scale in the same way.

[0083] For example, if the distance *d* between seepage measuring points A and B is 20 m, and the seepage pressure difference Δp between them at 9:00 is 40 kPa, then the pressure gradient can be calculated as 2 kPa / m based on Δp / d. This pressure gradient can be used to characterize the seepage change trend between measuring points A and B and serve as seepage driving data. When seepage velocity data is needed, it can be further converted using the permeability coefficient of the corresponding region. For example, if the permeability coefficient of the corresponding region is 1.0 × 10⁻⁵ m / s, the corresponding seepage velocity characterization value can be determined based on this permeability coefficient and the pressure gradient. For dam displacement data, assuming the horizontal displacement of measuring point C is 12.0 mm at 8:00 and 13.2 mm at 10:00, then the dam deformation change during this time period can be determined to be 1.2 mm, and the corresponding dam deformation change rate can be determined to be 0.6 mm / h.

[0084] Subsequently, the seepage-driven data and dam deformation data are interpolated or resampled according to a preset uniform time scale. For example, the target prediction step size can be set to 1 hour, and 1 hour can be used as the uniform time scale; for cases where there are multiple sampled values ​​within an hour, the corresponding data value for that hour can be determined by averaging, weighted averaging, or linear interpolation. Through the above processing, the seepage-driven sequence and dam deformation sequence arranged according to the uniform time scale can be obtained.

[0085] In this embodiment, step S3 extracts the nonlinear temporal features of the seepage-driven sequence to obtain the hidden state vector corresponding to each time step in the seepage-driven sequence, including:

[0086] Input the seepage-driven sequence into the time-series feature extraction model;

[0087] The temporal feature extraction model is used to extract temporal features from the seepage-driven data at different sampling times to obtain the hidden state vectors corresponding to each time. The hidden state vectors are used to characterize the dynamic changes of seepage at the corresponding time and the influence of historical seepage changes.

[0088] Specifically, the temporal feature extraction model can employ Long Short-Term Memory (LSTM) networks, gated recurrent units (ROUs), or temporal convolutional networks. Taking LSM networks as an example, this model can retain historical information related to current seepage changes through a gating mechanism and reduce the influence of historical information that is weakly related to the current prediction, thereby extracting nonlinear temporal features from the seepage-driven sequence.

[0089] Furthermore, as a specific implementation of this embodiment, seepage pressure data can be obtained from multiple seepage measurement points of the dam under test, such as upstream measurement point P1, downstream measurement point P2, and measurement point P3 at the dam foundation. Based on the seepage pressure difference between adjacent seepage measurement points and the distance between measurement points, the pressure gradient or hydraulic gradient corresponding to each sampling time can be determined and used as seepage driving data. The seepage driving data from multiple sampling times form a seepage driving sequence. For example, if the distance between measurement points P1 and P2 is 20m, and the seepage pressure differences at 8:00, 9:00, and 10:00 are 30kPa, 36kPa, and 44kPa, respectively, then the corresponding pressure gradients are 1.5kPa / m, 1.8kPa / m, and 2.2kPa / m, respectively. This continuous pressure gradient data can constitute part of the seepage driving sequence.

[0090] The seepage-driven sequence is input into a temporal feature extraction model to extract nonlinear temporal features from the seepage-driven sequence. The temporal feature extraction model can employ a Long Short-Term Memory (LSTM) network, a gated recurrent unit (GTU), or a temporal convolutional network. Taking an LSM network as an example, this model can generate hidden state vectors corresponding to each sampling time based on the seepage-driven data from previous sampling times and the seepage-driven data from the current sampling time, enabling the hidden state vectors to simultaneously represent the influence of current seepage changes and historical seepage changes.

[0091] For example, when processing the seepage drive sequences corresponding to measurement points P1, P2, and P3, the time-series feature extraction model can output the corresponding hidden state vectors at consecutive sampling times. Assuming that the hidden state vectors output by the model at times t-2, t-1, and t are [0.30, 0.42], [0.48, 0.57], and [0.76, 0.84], respectively, the change in this hidden state vector can reflect the increasing trend of the seepage drive data over consecutive time intervals. This increasing trend is not directly used as the result of seepage anomaly determination, but rather as a time-series feature basis for subsequently determining the current input conditions of the fluid-structure interaction finite element model.

[0092] In this embodiment, step S4: Based on the seepage driving data and the hidden state vector corresponding to the current moment in the seepage driving sequence, determine the current input conditions of the fluid-structure interaction finite element model, and calculate the additional deformation increment at the current moment, including:

[0093] Based on the seepage driving data at the current moment, the seepage boundary conditions of the fluid-structure interaction finite element model are determined.

[0094] Based on the hidden state vector corresponding to the current moment, determine at least one of the pore pressure boundary correction, permeability coefficient correction, and elastic modulus correction of the fluid-structure interaction finite element model.

[0095] Based on the seepage boundary conditions and at least one of the pore pressure boundary correction, permeability coefficient correction, and elastic modulus correction, determine the current input conditions of the fluid-structure interaction finite element model.

[0096] The current input conditions are loaded into the fluid-structure interaction finite element model, and the fluid flow equations in the fluid-structure interaction finite element model are solved to obtain the preliminary pore pressure distribution at the current moment.

[0097] By combining the property calibration data obtained from the dam material test, the preliminary pore pressure distribution is calibrated to obtain the pore pressure distribution at the current moment;

[0098] Obtain the water level data at the current moment, and determine the degree of coupling between the current seepage and the dam deformation based on the deviation between the pore pressure distribution at the current moment and the water level data;

[0099] If the coupling degree does not exceed the preset coupling threshold, the pore pressure distribution at the current moment is determined as the target pore pressure distribution, and the additional deformation increment at the current moment is calculated based on the target pore pressure distribution and the solid deformation equation in the fluid-structure interaction finite element model.

[0100] If the degree of coupling exceeds the preset coupling threshold, the elastic modulus and permeability coefficient in the fluid-structure interaction finite element model are adjusted according to the degree of coupling, and the pore pressure distribution at the current moment is corrected based on the adjusted permeability coefficient to obtain the target pore pressure distribution.

[0101] Based on the target pore pressure distribution and the solid deformation equation in the adjusted fluid-structure interaction finite element model, the additional deformation increment at the current moment is calculated.

[0102] As a specific implementation of this embodiment, in an actual dam monitoring system, seepage pressure data, water level data, and dam displacement data are acquired from sensors installed at different locations on the dam body. A seepage driving sequence is obtained based on the seepage pressure data. Simultaneously, the hidden state vector corresponding to each moment in the seepage driving sequence is obtained based on a time-series feature extraction model. For the current moment, the current input conditions of the fluid-structure interaction finite element model can be determined based on the seepage driving data and the hidden state vector corresponding to the current moment in the seepage driving sequence. These current input conditions are then loaded into the pre-established fluid-structure interaction finite element model. This model combines fluid flow equations such as Darcy's law with solid deformation equations such as linear elasticity equations. Specifically, the current input conditions are loaded into the meshed dam model as at least one of seepage boundary conditions, pore pressure boundary corrections, permeability coefficient corrections, and elastic modulus corrections. The preliminary pore pressure distribution at the current moment is obtained through iterative solution. Specifically, in a concrete dam project, after the current input conditions are loaded into the model, the initial pore pressure value of a certain grid node in the dam foundation area can be calculated to be approximately 1.60 MPa, thus reflecting the initial impact of seepage on the stability of the dam body.

[0103] Based on the preliminary pore pressure distribution described above, it is necessary to integrate the property calibration data obtained from dam material testing. This data includes experimentally measured parameters such as the permeability coefficient, elastic modulus, and porosity of the dam material or foundation material. The preliminary pore pressure distribution is calibrated using parameter calibration or weighted correction methods to obtain the pore pressure distribution at the current moment. Specifically, for example, if the preliminary calculation shows a pore pressure of 1.60 MPa at a certain grid node in the upstream dam foundation region, and the corresponding reference pore pressure after correction using the property calibration data is 1.50 MPa, and if the weights of the preliminary calculation result and the reference pore pressure are 0.6 and 0.4 respectively, then the calibrated pore pressure is 1.56 MPa. Using this pore pressure distribution, real-time water level data from a water level sensor is further obtained, for example, an upstream water level height of 150 m. This water level height can be converted into an equivalent water pressure benchmark, or the pore pressure can be converted into an equivalent head for comparison. Taking the equivalent water pressure of 150m water level as an example, which is about 1.47MPa, if the calibrated pore pressure is 1.56MPa, the deviation between the two is about 6.1%, which can be used to determine the degree of coupling between the current seepage and the dam deformation.

[0104] If the coupling degree does not exceed a preset coupling threshold, the current pore pressure distribution is determined as the target pore pressure distribution, and the additional deformation increment at the current moment is calculated based on the target pore pressure distribution and the solid deformation equation in the fluid-structure interaction finite element model. If the coupling degree exceeds the preset coupling threshold, for example, 15%, the elastic modulus and permeability coefficient in the fluid-structure interaction finite element model are adjusted. For example, when the equivalent pressure deviation is 22%, the local permeability coefficient can be adjusted from 1.0 × 10⁻⁵ m / s to 1.2 × 10⁻⁵ m / s according to a preset parameter update rule, and the corresponding region's elastic modulus can be adjusted from 30 GPa to 28 GPa. Subsequently, the current pore pressure distribution is corrected based on the adjusted permeability coefficient to obtain the target pore pressure distribution, and the additional deformation increment at the current moment is calculated based on the target pore pressure distribution and the solid deformation equation in the adjusted fluid-structure interaction finite element model. In actual dam simulation, the estimated value of dam deformation state at the previous moment can be updated based on the additional deformation increment to obtain the a priori dam deformation prediction result at the current moment, and then Kalman filtering correction can be performed in combination with the actual monitored deformation value at the current moment.

[0105] In this embodiment, after obtaining the additional deformation increment at the current moment, step S5 is executed:

[0106] The estimated value of the dam deformation state at the previous moment is superimposed with the additional deformation increment at the current moment to obtain the a priori dam deformation prediction result at the current moment.

[0107] Based on the deviation between the a priori dam deformation prediction result and the actual monitored deformation value at the current moment, the deformation observation residual at the current moment is determined;

[0108] The Kalman correction weights are determined based on the prediction error estimate corresponding to the a priori dam deformation prediction result at the current moment and the monitoring error estimate corresponding to the actual monitored deformation value at the current moment.

[0109] Based on the Kalman correction weights and the deformation observation residuals, the a priori dam deformation prediction results at the current moment are corrected to obtain the dam deformation prediction results at the current moment.

[0110] The predicted deformation result of the dam body at the current moment is used as the estimated value of the dam body deformation state at the current moment in the digital twin of the dam body.

[0111] Specifically, Kalman filtering correction is a technique well known to those skilled in the art, and the process will not be described in detail here.

[0112] In this embodiment, step S6, based on the residual between the predicted dam deformation at the current moment and the actual monitored deformation value at the current moment, determines whether there is an abnormal seepage flow, including:

[0113] Calculate the difference between the predicted deformation of the dam body at the current moment and the actual monitored deformation value at the current moment, and determine the residual at the current moment based on the absolute value or norm of the difference;

[0114] Obtain the residuals within a preset anomaly detection time window, including the current time, and form a residual change sequence;

[0115] The residual at the current moment is compared with a preset residual threshold, and the residual change trend is determined based on the residual change sequence;

[0116] If the residual at the current moment exceeds the preset residual threshold, and the residual change sequence satisfies the preset abnormality persistence condition, then it is determined that there is a seepage anomaly.

[0117] The preset abnormality persistence condition includes: the residuals at a preset number of consecutive sampling times all exceed the preset residual threshold, and the residual change sequence shows an increasing trend.

[0118] In this embodiment, step S7 combines the local seepage driving sub-sequence extracted from the seepage driving sequence and associated with the current time with the hidden state vector corresponding to the current time to perform local feature enhancement, thereby obtaining an enhanced dynamic change representation, including:

[0119] Using the current moment as the endpoint or center, a local seepage driving sub-sequence is extracted from the seepage driving sequence according to a preset anomaly analysis time window;

[0120] Obtain the hidden state vector corresponding to the current time from the hidden state vectors corresponding to each time step;

[0121] The hidden state vector corresponding to the current moment is subjected to dimension matching processing, and the dimension-matched hidden state vector is fused with the local seepage driving subsequence to obtain the enhanced input sequence;

[0122] The enhanced input sequence is input into the time-series feature extraction model. Based on the seepage change amplitude at each sampling time in the enhanced input sequence, the corresponding feature update weights are determined. The enhanced input sequence is then subjected to time-series feature extraction according to the feature update weights to obtain a representation of enhanced dynamic changes.

[0123] The magnitude of the seepage change is determined based on the difference or norm of the seepage-driven data at adjacent sampling times.

[0124] Furthermore, as a specific implementation of this embodiment, assuming that a reservoir dam is marked as having seepage anomalies in the early morning of July 15, 2025, the system, using the current time as the endpoint, extracts local seepage driving sub-sequences formed every 10 minutes between 00:00 on July 14 and 00:00 on July 15 from the seepage driving sequence, totaling 144 sampling points. These local seepage driving sub-sequences can be composed of seepage pressure gradients, hydraulic gradients, or converted seepage velocity characterization values ​​at different elevations of the dam foundation, used to reflect the seepage change trend in the period before the anomaly occurred. Simultaneously, the hidden state vector corresponding to the current time is extracted from the previously continuously running time-series feature extraction model, used to characterize the dynamic changes in seepage at the current time and the influence of historical seepage changes. The local seepage driving sub-sequences are combined with the hidden state vector corresponding to the current time to form an enhanced input sequence. Specifically, the hidden state vector can be copied and extended to the same dimension as the local seepage-driven subsequence after a linear transformation, and then concatenated with the local seepage-driven subsequence along the feature dimension, or the hidden state information can be added at each time step by using feature concatenation, thereby obtaining the dimension-enhanced input sequence.

[0125] The enhanced input sequence is then input into the temporal feature extraction model for local feature enhancement. Taking a Long Short-Term Memory (LSTM) network as an example, the LSM network can selectively retain important percolation-driven historical information related to anomalies through a forgetting gate. When the percolation change amplitude in a certain time period of the local percolation-driven subsequence increases significantly, the forgetting gate can reduce the memory weight of the early stable period while maintaining a high level of attention to the time period with increased change amplitude, thereby avoiding irrelevant historical information from interfering with the extraction of local anomaly features. The hidden state vector is updated through the input gate mechanism to further highlight local anomaly features.

[0126] In one possible implementation, the input gate calculates the update amount based on the time-step features of the current enhanced input sequence. When an increase in the magnitude of seepage change is detected across multiple consecutive sampling times in the local seepage-driven subsequence, the input gate increases its opening, allowing more current anomaly-related information to be incorporated into the new cell state, thereby enhancing the representation of mutation features. The Long Short-Term Memory (LSTM) network then employs an output gate mechanism to fuse and control the updated hidden states, determining the enhanced dynamic change representation. The output gate dynamically determines the proportion of dynamic information to be output based on the correlation between the current cell state and the input features. When the local seepage state changes significantly relative to historical states, the output gate tends to release more features containing local change information, ultimately forming an enhanced dynamic change representation. Based on this enhanced dynamic change representation, anomaly feedback update amounts can be further generated to update the fluid-structure interaction finite element model to obtain the target model.

[0127] Furthermore, anomaly feedback updates can be generated based on enhanced dynamic change characterization to update the fluid-structure interaction finite element model and obtain the target model. The anomaly feedback updates can include at least one of the following: pore pressure boundary update, permeability coefficient update, elastic modulus update, local seepage zone marker update, or local boundary condition update. If no seepage anomaly exists, the fluid-structure interaction finite element model can be used as the target model.

[0128] Finally, in this embodiment, step S8 is executed: the additional deformation increment at the next moment is calculated based on the target model, and the dam deformation prediction result at the current moment is updated based on the additional deformation increment at the next moment to obtain the a priori dam deformation prediction result at the next moment.

[0129] Understandably, the predicted deformation of the dam body at the next moment is obtained before the actual monitored deformation value at the next moment is corrected. Once the actual monitored deformation value at the next moment is acquired, the predicted deformation of the dam body at the next moment can be corrected by Kalman filtering, and the corrected result can be used as the estimated value of the dam body deformation state in the next round of prediction, thereby realizing the rolling update of the digital twin of the dam body.

[0130] Example 2

[0131] This embodiment provides a dam structure deformation prediction system based on digital twins, used to implement the dam structure deformation prediction method based on digital twins in Embodiment 1, including:

[0132] The digital twin acquisition module is used to acquire a pre-constructed digital twin of the dam body, which contains the spatial mapping relationship of the measurement points of the dam to be measured, the fluid-structure interaction finite element model, and the estimated value of the dam body deformation state.

[0133] The data processing module is used to acquire multi-source monitoring data of the dam under test, and perform spatiotemporal alignment processing on the multi-source monitoring data based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence.

[0134] The temporal feature extraction module is used to extract the nonlinear temporal features of the seepage-driven sequence and obtain the hidden state vector corresponding to each time step in the seepage-driven sequence.

[0135] The deformation increment acquisition module is used to determine the current input conditions of the fluid-structure interaction finite element model based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence, and to calculate the additional deformation increment at the current moment.

[0136] The deformation prediction module is used to update the estimated value of the dam deformation state at the previous time based on the additional deformation increment at the current time, to obtain the a priori dam deformation prediction result at the current time, and to perform Kalman filtering correction in combination with the actual monitored deformation value at the current time in the dam deformation sequence to obtain the dam deformation prediction result at the current time.

[0137] The anomaly detection module is used to determine whether there is an anomaly in seepage based on the residual between the predicted deformation result of the dam body at the current moment and the actual monitored deformation value at the current moment.

[0138] The model update module is used to perform local feature enhancement by combining the local seepage driving sub-sequence that is extracted from the seepage driving sequence and associated with the current time and the hidden state vector corresponding to the current time if seepage anomalies exist, so as to obtain enhanced dynamic change characterization, and generate anomaly feedback update amount based on the enhanced dynamic change characterization to update the fluid-structure interaction finite element model to obtain the target model.

[0139] The model update module is also used to use the fluid-structure interaction finite element model as the target model if there is no seepage anomaly.

[0140] The next moment prediction module is used to calculate the additional deformation increment at the next moment based on the target model, and to update the dam deformation prediction result at the current moment based on the additional deformation increment at the next moment, so as to obtain the a priori dam deformation prediction result at the next moment.

[0141] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0142] Example 3

[0143] This embodiment provides a computer-readable storage medium storing a computer program that, when executed, implements the dam structure deformation prediction method based on digital twins described in Embodiment 1.

[0144] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0145] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0146] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0147] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for predicting dam structural deformation based on digital twins, characterized in that, include: Obtain a pre-constructed digital twin of the dam body, which contains the spatial mapping relationship of the measurement points of the dam to be measured, the fluid-structure interaction finite element model, and the estimated value of the dam body deformation state. Multi-source monitoring data of the dam under test were acquired, and the multi-source monitoring data were spatiotemporally aligned based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence. The nonlinear temporal features of the seepage-driven sequence are extracted to obtain the hidden state vector corresponding to each time step in the seepage-driven sequence. Based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence, the current input conditions of the fluid-structure interaction finite element model are determined, and the additional deformation increment at the current moment is calculated. The estimated value of dam deformation state at the previous moment is updated based on the additional deformation increment at the current moment to obtain the a priori dam deformation prediction result at the current moment. Kalman filtering correction is then performed on the actual monitored deformation value at the current moment in the dam deformation sequence to obtain the dam deformation prediction result at the current moment. Based on the residual between the predicted deformation of the dam body at the current moment and the actual monitored deformation value at the current moment, it is determined whether there is an abnormal seepage. If seepage anomalies exist, local feature enhancement is performed by combining the local seepage driving sub-sequences associated with the current time and the hidden state vector corresponding to the current time in the seepage driving sequence to obtain enhanced dynamic change characterization. Based on the enhanced dynamic change characterization, anomaly feedback update amount is generated to update the fluid-structure interaction finite element model to obtain the target model. If there is no seepage anomaly, the fluid-structure interaction finite element model will be used as the target model. The additional deformation increment at the next moment is calculated based on the target model, and the dam deformation prediction result at the current moment is updated based on the additional deformation increment at the next moment to obtain the a priori dam deformation prediction result at the next moment.

2. The method for predicting dam structural deformation based on digital twins according to claim 1, characterized in that, The multi-source monitoring data includes: seepage pressure data, dam displacement data, and water level data; the spatiotemporal alignment processing of the multi-source monitoring data based on the spatial mapping relationship of the measuring points yields the seepage driving sequence and the dam deformation sequence, including: The multi-source monitoring data is filled with missing values ​​and outliers are removed to obtain cleaned multi-source monitoring data; Based on the spatial coordinates and acquisition timestamps of each measuring point, time alignment and spatial matching are performed on the cleaned multi-source monitoring data. Based on the time-aligned and spatially matched seepage pressure data, a seepage driving sequence was determined to characterize seepage changes. Based on the dam displacement data after time alignment and spatial matching, the dam deformation sequence is determined.

3. The method for predicting dam structural deformation based on digital twins according to claim 2, characterized in that, The extraction of nonlinear temporal features from the seepage-driven sequence yields the hidden state vectors corresponding to each time step in the seepage-driven sequence, including: The seepage-driven sequence is input into the time-series feature extraction model; The time-series feature extraction model is used to extract time-series features from the seepage-driven data at different sampling times to obtain the hidden state vectors corresponding to each time. The hidden state vectors are used to characterize the dynamic changes in seepage at the corresponding time and the influence of historical seepage changes.

4. The method for predicting dam structural deformation based on digital twins according to claim 3, characterized in that, The determination of the current input conditions for the fluid-structure interaction finite element model based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence includes: Based on the seepage driving data at the current moment, the seepage boundary conditions of the fluid-structure interaction finite element model are determined. Based on the hidden state vector corresponding to the current moment, determine at least one of the pore pressure boundary correction, permeability coefficient correction, and elastic modulus correction of the fluid-structure interaction finite element model. Based on the seepage boundary conditions and at least one of the pore pressure boundary correction, permeability coefficient correction, and elastic modulus correction, the current input conditions of the fluid-structure interaction finite element model are determined.

5. The method for predicting dam structural deformation based on digital twins according to claim 4, characterized in that, The calculation of the additional deformation increment at the current moment includes: The current input conditions are loaded into the fluid-structure interaction finite element model, and the fluid flow equations in the fluid-structure interaction finite element model are solved to obtain the preliminary pore pressure distribution at the current moment. By combining the property calibration data obtained from the dam material test, the preliminary pore pressure distribution is calibrated to obtain the pore pressure distribution at the current moment; Obtain the water level data at the current moment, and determine the degree of coupling between the current seepage and the dam deformation based on the deviation between the pore pressure distribution at the current moment and the water level data; If the coupling degree does not exceed the preset coupling threshold, the pore pressure distribution at the current moment is determined as the target pore pressure distribution, and the additional deformation increment at the current moment is calculated based on the target pore pressure distribution and the solid deformation equation in the fluid-structure interaction finite element model. If the degree of coupling exceeds the preset coupling threshold, the elastic modulus and permeability coefficient in the fluid-structure interaction finite element model are adjusted according to the degree of coupling, and the pore pressure distribution at the current moment is corrected based on the adjusted permeability coefficient to obtain the target pore pressure distribution. Based on the target pore pressure distribution and the solid deformation equation in the adjusted fluid-structure interaction finite element model, the additional deformation increment at the current moment is calculated.

6. The method for predicting dam structural deformation based on digital twins according to claim 5, characterized in that, The method involves updating the estimated dam deformation state from the previous moment based on the additional deformation increment at the current moment, obtaining the prior dam deformation prediction result at the current moment, and then performing Kalman filtering correction on the actual monitored deformation value at the current moment in the dam deformation sequence to obtain the dam deformation prediction result at the current moment, including: The estimated value of the dam deformation state at the previous moment is superimposed with the additional deformation increment at the current moment to obtain the a priori dam deformation prediction result at the current moment. Based on the deviation between the a priori dam deformation prediction result and the actual monitored deformation value at the current moment, the deformation observation residual at the current moment is determined; The Kalman correction weights are determined based on the prediction error estimate corresponding to the a priori dam deformation prediction result at the current moment and the monitoring error estimate corresponding to the actual monitored deformation value at the current moment. Based on the Kalman correction weights and the deformation observation residuals, the a priori dam deformation prediction results at the current moment are corrected to obtain the dam deformation prediction results at the current moment. The predicted deformation result of the dam body at the current moment is used as the estimated value of the dam body deformation state at the current moment in the digital twin of the dam body.

7. The method for predicting dam structural deformation based on digital twins according to claim 6, characterized in that, The step of determining whether there is an abnormal seepage based on the residual between the predicted dam deformation at the current moment and the actual monitored deformation value at the current moment includes: Calculate the difference between the predicted deformation of the dam body at the current moment and the actual monitored deformation value at the current moment, and determine the residual at the current moment based on the absolute value or norm of the difference; Obtain the residuals within a preset anomaly detection time window, including the current time, and form a residual change sequence; The residual at the current moment is compared with a preset residual threshold, and the residual change trend is determined based on the residual change sequence; If the residual at the current moment exceeds the preset residual threshold, and the residual change sequence satisfies the preset abnormality persistence condition, then it is determined that there is a seepage anomaly. The preset abnormality persistence condition includes: the residuals at a preset number of consecutive sampling times all exceed the preset residual threshold, and the residual change sequence shows an increasing trend.

8. The method for predicting dam structural deformation based on digital twins according to claim 7, characterized in that, The method of combining the local seepage driving sub-sequence extracted from the seepage driving sequence and associated with the current time with the hidden state vector corresponding to the current time for local feature enhancement yields an enhanced dynamic change representation, including: Using the current moment as the endpoint or center, a local seepage driving sub-sequence is extracted from the seepage driving sequence according to a preset anomaly analysis time window; Obtain the hidden state vector corresponding to the current time from the hidden state vectors corresponding to each time step; The hidden state vector corresponding to the current moment is subjected to dimension matching processing, and the dimension-matched hidden state vector is fused with the local seepage driving subsequence to obtain the enhanced input sequence; The enhanced input sequence is input into the time-series feature extraction model. Based on the seepage change amplitude at each sampling time in the enhanced input sequence, the corresponding feature update weights are determined. The enhanced input sequence is then subjected to time-series feature extraction according to the feature update weights to obtain a representation of enhanced dynamic changes. The magnitude of the seepage change is determined based on the difference or norm of the seepage-driven data at adjacent sampling times.

9. The method for predicting dam structural deformation based on digital twins according to claim 8, characterized in that, The temporal feature extraction model employs any one of the following: Long Short-Term Memory Network, Gated Recurrent Unit, or Temporal Convolutional Network.

10. A dam structural deformation prediction system based on digital twins, characterized in that, include: The digital twin acquisition module is used to acquire a pre-constructed digital twin of the dam body, which contains the spatial mapping relationship of the measurement points of the dam to be measured, the fluid-structure interaction finite element model, and the estimated value of the dam body deformation state. The data processing module is used to acquire multi-source monitoring data of the dam under test, and perform spatiotemporal alignment processing on the multi-source monitoring data based on the spatial mapping relationship of the measuring points to obtain the seepage driving sequence and the dam deformation sequence. The temporal feature extraction module is used to extract the nonlinear temporal features of the seepage-driven sequence and obtain the hidden state vector corresponding to each time step in the seepage-driven sequence. The deformation increment acquisition module is used to determine the current input conditions of the fluid-structure interaction finite element model based on the seepage driving data and the hidden state vector at the current moment in the seepage driving sequence, and to calculate the additional deformation increment at the current moment. The deformation prediction module is used to update the estimated value of the dam deformation state at the previous time based on the additional deformation increment at the current time, to obtain the a priori dam deformation prediction result at the current time, and to perform Kalman filtering correction in combination with the actual monitored deformation value at the current time in the dam deformation sequence to obtain the dam deformation prediction result at the current time. The anomaly detection module is used to determine whether there is an anomaly in seepage based on the residual between the predicted deformation result of the dam body at the current moment and the actual monitored deformation value at the current moment. The model update module is used to perform local feature enhancement by combining the local seepage driving sub-sequence that is extracted from the seepage driving sequence and associated with the current time and the hidden state vector corresponding to the current time if seepage anomalies exist, so as to obtain enhanced dynamic change characterization, and generate anomaly feedback update amount based on the enhanced dynamic change characterization to update the fluid-structure interaction finite element model to obtain the target model. The model update module is also used to use the fluid-structure interaction finite element model as the target model if there is no seepage anomaly. The next moment prediction module is used to calculate the additional deformation increment at the next moment based on the target model, and to update the dam deformation prediction result at the current moment based on the additional deformation increment at the next moment, so as to obtain the a priori dam deformation prediction result at the next moment.

Images