A seawall settlement intelligent prediction method and system

Abstract

The application discloses a seawall settlement intelligent prediction method and system, relates to the technical field of settlement prediction, and comprises the following steps: reading fixed-interval settlement data of a settlement meter and performing pretreatment; collecting basic data to establish a seawall numerical model; selecting uncertainty parameters, generating parameter-settlement data sets based on the seawall numerical model, and constructing a seawall agent model; updating the uncertainty parameters based on a multiple data assimilation ensemble smoother, and outputting an updated parameter posterior ensemble; inputting each member of the updated parameter posterior ensemble into the seawall agent model to predict a seawall settlement value; and triggering a hierarchical early warning mechanism when the settlement prediction value exceeds an alarm threshold. The prediction method firstly constructs a seawall agent model based on numerical software, and then calculates a settlement prediction result containing a mean value, a standard deviation and a confidence interval. The application realizes the full-process automation of monitoring-analysis-prediction-early warning, can effectively guide seawall construction and operation and maintenance, and improves engineering safety and stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of settlement prediction, in particular to a seawall settlement intelligent prediction method and system. BACKGROUND

[0002] Seawall engineering plays a key role in coastal protection system, and its structural stability directly affects regional flood safety. Influenced by complex environmental factors, including the consolidation characteristics of soft soil foundation, periodic tidal load action, wave dynamic impact and seawater chemical corrosion, etc., the seawall structure is prone to uneven settlement problems. Traditional monitoring methods mainly rely on manual periodic measurement, which has defects such as low monitoring frequency and poor data continuity, making it difficult to discover structural safety hazards in time and take countermeasures. In addition, the conventional data analysis method has insufficient adaptability to the settlement evolution law under complex environment, and the prediction result has limited reliability, making it difficult to provide effective decision support for engineering maintenance.

[0003] In recent years, with the innovative development of monitoring technology, the intelligent monitoring scheme based on Internet of Things provides a new solution for seawall safety monitoring. Modern monitoring system realizes the automatic collection and remote transmission of settlement data by laying multi-element sensor network including high-precision settlement sensor, pore water pressure monitoring device, etc., combined with satellite positioning technology. The intelligent analysis platform uses data cleaning algorithm and machine learning model to accurately predict the deformation trend of seawall, and intuitively displays the monitoring results through visual interface. This technology fusion significantly improves the monitoring efficiency and data reliability, and provides a scientific decision basis for engineering management personnel.

[0004] However, the current intelligent monitoring technology still faces some challenges in practical application: first, the prediction model generalization ability under complex marine environment needs to be strengthened; second, the collaborative optimization of system modules still needs to be improved. In view of these problems, it is particularly urgent to develop a new generation of monitoring system with high integration and intelligence, which is of great significance to improve the safety management level of seawall engineering. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] Therefore, the present application provides a seawall settlement intelligent prediction method to solve the problems of insufficient prediction model generalization ability under complex marine environment and imperfect collaborative optimization of system modules.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides an intelligent seawall settlement prediction method, which comprises reading fixed-interval settlement data of a settlement gauge and pre-processing to obtain a multi-measurement-point time-settlement data set; collecting basic data to establish a seawall numerical model; selecting uncertainty parameters, generating a parameter-settlement data set based on the established seawall numerical model, and constructing a seawall surrogate model; updating the uncertainty parameters based on a multi-data assimilation ensemble smoother, and outputting an updated parameter posterior ensemble; inputting each member of the updated parameter posterior ensemble into the seawall surrogate model to predict a seawall settlement value; and triggering a hierarchical early warning mechanism when the settlement prediction value exceeds an alarm threshold.

[0009] As a preferred scheme of the seawall settlement intelligent prediction method, the pre-processing to obtain a multi-measurement-point time-settlement data set comprises the following steps:

[0010] Abnormal settlement values of each monitoring point are identified and removed based on a quartile range method;

[0011] The time series data of each monitoring point are aligned and verified to ensure that the collection times of the measurement points are synchronized; the cumulative settlement and settlement rate of each measurement point are calculated, and finally a standardized multi-measurement-point time-settlement data set is output.

[0012] As a preferred scheme of the seawall settlement intelligent prediction method, the collecting basic data to establish a seawall numerical model comprises the following steps:

[0013] Basic data of geological survey reports, existing engineering data, seawall design documents and construction loading plans are collected;

[0014] Based on the geological survey report, the soil layers are divided, and the physical and mechanical parameters of each soil layer are determined;

[0015] A seawall numerical model is established according to the actual design size, the Mohr-Coulomb and Modified Cam-clay constitutive models are used to simulate the behavior of the soil body, and the load is applied step by step according to the construction loading plan to simulate the whole construction process.

[0016] As a preferred scheme of the seawall settlement intelligent prediction method, the selecting uncertainty parameters and generating a parameter-settlement data set based on the established seawall numerical model comprises the following steps:

[0017] The uncertainty parameters are selected, and the parameter distribution range of the uncertainty parameters is determined based on the established seawall numerical model;

[0018] Based on the parameter distribution range, a representative parameter combination is generated by using a Latin hypercube sampling method;

[0019] The API of the seawall numerical model is called by the Python program for batch calculation to obtain the seawall settlement response values corresponding to each group of parameters, and finally a complete parameter-settlement data set is constructed.

[0020] As a preferred scheme of the seawall settlement intelligent prediction method, the seawall agent model is constructed by dividing the constructed parameter-settlement data set into a training set and a test set, wherein the training set is used to construct the seawall agent model, and the test set is used to verify the accuracy of the seawall agent model.

[0021] As a preferred scheme of the seawall settlement intelligent prediction method, the updating of the uncertainty parameters based on the multi-data assimilation ensemble smoother includes the following steps,

[0022] Setting the total number of iterations of data assimilation ;

[0023] Defining the inflation coefficient of the first iteration , which satisfies the constraint condition;

[0024] Taking the uncertainty parameters as target updating variables, determining the prior distribution of the target updating variables according to the information collected from the geological exploration report and the existing engineering data, and generating an initial parameter set;

[0025] For each member in the current parameter set of the first iteration , the seawall agent model is used to calculate the settlement prediction value;

[0026] Based on the current parameter set and the settlement prediction value, the ensemble statistics are calculated;

[0027] The Kalman gain matrix is calculated based on the ensemble statistics, and the initial parameter set is updated;

[0028] If the current iteration number , the iteration is terminated, and the final updated parameter posterior set is output.

[0029] As a preferred scheme of the seawall settlement intelligent prediction method, the prediction of the seawall settlement value includes the following steps,

[0030] Each member of the parameter posterior set is input into the seawall agent model to obtain the settlement prediction value in the future period;

[0031] The statistical characteristics of the prediction mean, the prediction standard deviation and the confidence interval are calculated.

[0032] As a preferred scheme of the seawall settlement intelligent prediction method, the triggering of the hierarchical early warning mechanism when the settlement prediction value exceeds the alarm threshold is,​

[0033] When the height of the embankment H is less than or equal to 4.0 m, and the single-day settlement rate is greater than or equal to 20 mm / d, stop loading; when the continuous average settlement is less than or equal to 3 mm / d, allow loading.

[0034] When the height of the embankment H is greater than 4.0 m, and the single-day settlement rate is greater than or equal to 10 mm / d, stop loading; when the continuous average settlement is less than or equal to 2 mm / d, allow loading.

[0035] In a second aspect, the present application provides an intelligent prediction system for embankment settlement, comprising: a data acquisition module for reading fixed-interval settlement data of a settlement gauge; a data processing module for preprocessing to obtain a multi-measurement-point time-settlement dataset; a data collection module for collecting basic data to establish an embankment numerical model; a model construction module for selecting uncertainty parameters, generating a parameter-settlement dataset based on the constructed embankment numerical model, and constructing an embankment surrogate model; a parameter updating module for updating the uncertainty parameters based on a multi-data assimilation ensemble smoother, and outputting an updated parameter posterior ensemble; a settlement detection module for inputting each member of the updated parameter posterior ensemble into the embankment surrogate model to predict an embankment settlement value; and an early warning mechanism module for triggering a hierarchical early warning mechanism when the settlement prediction value exceeds an alarm threshold.

[0036] The present application has the following beneficial effects: based on the quartile range method, abnormal values of each monitoring point are identified and removed, and time series data are aligned and verified to ensure data synchronization. This step effectively cleans up noise in the data, improving the accuracy and reliability of data analysis. Through the calculation of cumulative settlement and settlement rate, a standardized multi-measurement-point time-settlement dataset is generated, which not only improves the accuracy of settlement trend analysis, but also provides high-quality data support for subsequent settlement prediction; through the three steps of establishing an embankment surrogate model, updating uncertainty parameters, and finally predicting settlement, the future settlement trend is accurately predicted, greatly enhancing the generalization ability and accuracy of the prediction model, providing strong decision support for engineering maintenance, and helping to prevent potential safety hazards. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0038] Figure 1 A flowchart of an intelligent prediction method for embankment settlement provided by the embodiments of the present application.

[0039] Figure 2 A sea wall in an embodiment of the present application.

[0040] Figure 3 A sea wall filling plan in an embodiment of the present application.

[0041] Figure 4 A finite element model grid chart of a sea wall in an embodiment of the present application.

[0042] Figure 5 A sea wall settlement calculation result comparison chart of a sea wall agent model and a finite element model in an embodiment of the present application.

[0043] Figure 6 Prior and posterior distributions of uncertainty parameters in an embodiment of the present application.

[0044] Figure 7 Settlement prediction using different amounts of observation data in an embodiment of the present application.

[0045] Figure 8 Effect of monitoring data on settlement prediction in an embodiment of the present application.

[0046] Figure 9 Day 1013 settlement prediction of different monitoring data amounts in an embodiment of the present application.

[0047] Figure 10 Error between all predicted values and observed values in an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0049] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details given herein, that the present application can be practiced with other than the described embodiments and that variations from the particular embodiments described herein can be made and still be within the scope of the present application.

[0050] Secondly, the "one embodiment" or "embodiment" referred to herein means that a specific feature, structure or characteristic can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is independent or alternative to other embodiments.

[0051] Embodiment 1, Reference Figure 1 , the first embodiment of the present application provides a sea wall settlement intelligent prediction method, comprising the following steps:

[0052] S1. Read the fixed interval settlement data of the settlement meter and pre-process to obtain a multi-point time-settlement data set;

[0053] Further, the periodically collected settlement data of multiple monitoring points are cleaned and analyzed, specifically including: identifying and removing abnormal settlement values of each monitoring point based on the interquartile range method; aligning and checking the time series data of each monitoring point to ensure that the collection times of each point are synchronized; calculating the cumulative settlement and settlement rate of each point; and finally outputting the standardized multi-point time-settlement data set to provide a reliable data basis for subsequent differential settlement analysis and overall settlement trend prediction.

[0054] S2. Settlement prediction specifically includes the following steps:

[0055] Step 1: Collect basic data to establish a seawall proxy model;

[0056] Step 2: Update the uncertainty parameters based on the Ensemble Smoother with Multiple Data Assimilation (ES-MDA);

[0057] Step 3: Predict the seawall settlement value.

[0058] Further, step 1 specifically includes the following steps:

[0059] Collect basic data such as geological exploration reports, existing engineering data, seawall design documents, and construction loading plans to build a seawall numerical model. Specifically, first divide the soil layers based on the geological survey report to determine the physical and mechanical parameters of each soil layer; then establish a seawall numerical model according to the actual design size, use Mohr-Coulomb and Modified Cam-clay constitutive models to simulate soil behavior, and apply loads step by step according to the construction loading plan to simulate the entire construction process.

[0060] Based on the constructed seawall numerical model, determine the value range of key uncertainty parameters through literature research. According to existing research results and engineering specifications, select soil layer permeability coefficient, compression modulus, etc. as uncertainty parameters. Based on the parameter distribution range recommended by the literature, use the Latin hypercube sampling method to generate representative parameter combinations, use Python program to call the seawall numerical model API for batch calculation, obtain the seawall settlement response value corresponding to each group of parameters, and finally build a complete parameter-settlement data set. This parameter-settlement data set will provide a basis for subsequent reliability analysis and optimization design, all parameter values are strictly referred to relevant literature and specification requirements to ensure the engineering applicability of the analysis.

[0061] The construction parameter-settlement data set is divided into a training set and a test set, wherein the training set is used to build the seawall proxy model, and the test set is used to verify the accuracy of the seawall proxy model; wherein the seawall proxy model includes a long short-term memory network, a random forest model, a polynomial regression model, a radial basis function model, a support vector regression model, a Kriging model, etc., and the long short-term memory network is selected in this scheme.

[0062] Step 2 specifically includes the following steps:

[0063] Determine the data assimilation parameters:

[0064] Set the total number of iterations of data assimilation as ;

[0065] Define the inflation coefficient of the first iteration , which needs to meet the constraint condition:

[0066] ;

[0067] Parameter initialization:

[0068] The uncertainty parameters (including soil compression parameters , vertical permeability coefficient , and horizontal permeability coefficient ) are taken as target update variables, the prior distribution of the target update variables is determined according to the information collected from the geological exploration report and existing information, and an initial parameter set is generated:

[0069] ;

[0070] wherein is the number of set members; each member is a parameter combination independently drawn from the prior distribution.

[0071] Forward simulation calculation:

[0072] For the current parameter set of the first iteration: for each member , the settlement prediction value is calculated by the seawall proxy model, and the expression is:

[0073] ;

[0074] wherein is a pre-trained seawall proxy model (such as a long short-term memory network); the settlement prediction value belongs to the real number space ( ​​​The number of monitoring points and the number of time steps are multiplied.

[0075] Data assimilation update:

[0076] Based on the current parameter set and the predicted settlement value, the ensemble statistics are calculated.

[0077] Specifically, the ensemble statistics are calculated:

[0078] The parameter mean is:

[0079] ;

[0080] The predicted settlement mean is:

[0081] ;

[0082] The parameter-predicted covariance matrix is:

[0083] ;

[0084] The predicted covariance matrix is:

[0085] ;

[0086] The observation error covariance matrix is:

[0087] ;

[0088] wherein, is the standard deviation of the th observation value, and the value of the standard deviation is referred to the relevant literature.

[0089] Based on the ensemble statistics, the Kalman gain matrix is calculated, and the initial parameter set is updated.

[0090] Specifically, the Kalman gain matrix is calculated.

[0091] ;

[0092] The expression for updating the initial parameter set is:

[0093] ;

[0094] wherein, is the actual observed settlement data; is the added disturbance noise.

[0095] Iteration end value judgment:

[0096] If the current iteration number = , the iteration is terminated.

[0097] The result is outputted as:

[0098] Output the final updated parameter posterior set .

[0099] Step 3 specifically includes the following steps:

[0100] After obtaining the final updated parameter posterior set , the settlement prediction is performed as follows:

[0101] Settlement prediction calculation: input each member of the parameter posterior set into the seawall agent model to obtain the settlement prediction results in the future period (such as the construction period, operation period):

[0102] ;

[0103] wherein is the predicted settlement sequence corresponding to the th member (N is the number of prediction time points).

[0104] Calculate the prediction mean, prediction standard deviation, and statistical characteristics of the confidence interval;

[0105] Prediction mean (best estimate):

[0106] ;

[0107] Prediction standard deviation (uncertainty quantification):

[0108] ;

[0109] Confidence interval (such as 95% confidence level):

[0110] .

[0111] S3. Trigger a hierarchical warning mechanism when the settlement prediction value exceeds the warning threshold;

[0112] Specifically, when the seawall loading height H≤4.0m, the single-day settlement rate ≥20mm / d, stop loading, when the continuous average settlement amount ≤3mm / d, allow loading;

[0113] When the seawall loading height H>4.0m, the single-day settlement rate ≥10mm / d, stop loading, when the continuous average settlement amount ≤2mm / d, allow loading.

[0114] The embodiment also provides a seawall settlement intelligent prediction system, comprising: a data acquisition module configured to read fixed-interval settlement data of a settlement gauge; a data processing module configured to perform preprocessing to obtain a multi-measurement-point time-settlement data set; a data collection module configured to collect basic data to establish a seawall numerical model; a model construction module configured to select uncertainty parameters, generate a parameter-settlement data set based on the constructed seawall numerical model, and construct a seawall surrogate model; a parameter updating module configured to update the uncertainty parameters based on a multi-data assimilation ensemble smoother, and output an updated parameter posterior ensemble; a settlement detection module configured to input each member of the updated parameter posterior ensemble into the seawall surrogate model to predict a seawall settlement value; and an early warning mechanism module configured to trigger a hierarchical early warning mechanism when the settlement prediction value exceeds an alarm threshold.

[0115] To sum up, the application achieves the following effects: based on the quartile range method, the abnormal values of each monitoring point are identified and removed, and the time series data is aligned and verified to ensure data synchronization. This step effectively cleans up the noise in the data, improving the accuracy and reliability of data analysis. Through the calculation of cumulative settlement and settlement rate, a normalized multi-measurement-point time-settlement data set is generated, which not only improves the accuracy of settlement trend analysis, but also provides high-quality data support for subsequent settlement prediction; through the three steps of establishing a seawall surrogate model, updating uncertainty parameters, and finally predicting settlement, the future settlement trend is accurately predicted, greatly enhancing the generalization ability and accuracy of the prediction model, providing strong decision support for engineering maintenance, and helping to prevent potential safety hazards.

[0116] Embodiment 2, referring to Tables 1-4, is a second embodiment of the application, which provides experimental simulation data of the seawall settlement intelligent prediction method to further verify the technical solution of the application.

[0117] Step 1: Collect basic data to establish a surrogate model.

[0118] In this study, a real soft foundation project in Shaoxing, China was selected as the verification object, and a settlement prediction method based on a multi-data assimilation ensemble smoother (ES-MDA) was used for evaluation. According to the drilling exploration and cone penetration test (CPT) data, the soil profile can be divided into six layers of soil structure on the weathered bedrock, including a 1-meter-thick top weathered layer, a 3.8-meter-thick silty clay layer, a 9.5-meter-thick extremely soft silt clay layer, a 3.9-meter-thick silt silty clay layer, a 4.8-meter-thick silty clay layer, and a 2-meter-thick clayey sand layer. A 0.5-meter-thick sand cushion and a 5.38-meter-thick filling layer are set above the ground surface.

[0119] The monitoring system is deployed as follows: Settlement plates (SP) are installed at the center, sides, and toe of the slope; pore water pressure gauges (P) are arranged at different depths along the centerline. Precast vertical drainage boards (PVDs) are arranged in a triangular pattern at 1.5-meter intervals, with a depth of 19 meters. The engineering geometric profile and the filling construction sequence are shown below. Figure 2 and Figure 3 As shown.

[0120] The PLAXIS 2D V20 software was used to perform a half-section finite element simulation of the soft soil foundation project: the left and right boundaries were fixed with horizontal displacement, and the bottom boundary was fixed with both horizontal and vertical displacement; the surface and bottom were set as permeable boundaries, and the left and right sides were set as impermeable boundaries (finite element model see...). Figure 4 The five cohesive soil layers were modeled using the Modified Cambridge (MCC) model, while the fill layer and clayey sand layer were modeled using the Mohr-Coulomb (MC) model. Soil parameters (Table 1) include unit weight γ, initial void ratio e0, elastic modulus E', Poisson's ratio ν, and compression index C. C Compression parameters , rebound parameter κ, slope of critical state line Horizontal permeability coefficient and vertical permeability coefficient The compression index C C The permeability coefficient was determined through consolidation tests. and The results were obtained through specialized permeability tests on soil samples in both the horizontal and vertical directions. Convert the compression index to C C Compression parameters And set the springback parameter κ to .

[0121] Table 1 Soil parameters

[0122]

[0123] To simultaneously reflect the vertical drainage effect of the natural foundation and the radial drainage effect caused by the installation of PVDs, an equivalent vertical permeability coefficient is used in the finite element model. This method simplifies the drainage behavior of PVD-reinforced foundations to that of untreated foundations. Equivalent vertical permeability coefficient. The calculation formula is as follows:

[0124] ;

[0125] in, The horizontal permeability coefficient; The vertical permeability coefficient; D is the length of the drainage path.e For the influence zone diameter, its value is 1.05S L (S L For the PVDs spacing); For the PVDs geometry factor, its expression is:

[0126] ;

[0127] Where, , For the drain diameter, its expression is ( For the drain width, d is the drain thickness); , Represents the smear zone diameter; For the smear zone horizontal hydraulic conductivity; For the drain flow. The parameters related to PVDs are summarized in Table 2.

[0128] Table 2. Parameters related to PVDs behavior

[0129]

[0130] The soil parameters that need to be updated are the compression index of the five layers of soil below the fill layer and the equivalent vertical hydraulic conductivity of the four layers of soil below the fill layer This choice is based on the following considerations: the compression index directly reflects the compressibility of the soil, has a significant impact on the consolidation settlement, and the soil modulus itself has a high degree of uncertainty. The permeability coefficient parameter is also sensitive to consolidation calculations, and its variability has been confirmed by many studies. Since the total thickness of the SC2 layer is 4.8 meters, and the insertion depth of the PVDs is only 0.8 meters, the drainage path does not penetrate the entire layer, so the influence of PVDs on the equivalent permeability coefficient of the SC2 layer is ignored. The prior distribution of parameters and is listed in Table 3, with the mean values determined by consolidation tests and permeability tests, and the coefficient of variation (COV) set to 0.5 ( ) and 1 ( ), respectively.

[0131] The settlement monitoring data were collected from settlement plates at the center line, with 28 monitoring time points corresponding to the 131st, 148th, 154th, 155th, 162nd, 173rd, 181st, 196th, 212th, 228th, 237th, 257th, 265th, 281st, 301st, 314th, 344th, 381st, 412th, 461st, 496th, 537th, 580th, 629th, 693rd, 782nd, 860th, and 1013th days, respectively. The monitoring frequency was higher in the early stage and gradually decreased as the deformation rate tended to be stable. Assuming that the settlement observation error was Gaussian white noise with a standard deviation of 0.02 meters, the collection size of ES-MDA was set to 500, and the total number of iterations N iter was set to 8.

[0132] Table 3 Prior distribution of random variables

[0133]

[0134] Considering the time series characteristics of soil settlement, a long short-term memory (LSTM) neural network was used to construct a seawall proxy model: the input was the soil parameters listed in Table 3, and the output was the settlement value at each monitoring time. The training set and test set of the LSTM model used 2000 and 400 sets of numerical simulation data, respectively, and the input parameters were generated within the mean ± three times the standard deviation range through Latin hypercube sampling.

[0135] The LSTM model used mean square error (MSE) as the loss function, and the Adam optimizer with a default learning rate of 0.001 was used to balance convergence stability and training efficiency. The network used a double hidden layer structure with 128 neurons in each layer to effectively capture the time series characteristics. The hyperparameters were determined by grid search: by comparing different combinations of the number of hidden layers (1 / 2 / 3 layers) and the number of neurons (64 / 128 / 256), the double hidden layer with 128 neurons in each layer was finally selected, which had the highest coefficient of determination (R²) on the validation set. In addition, the grid search also optimized the batch size and the number of training rounds, and finally determined the batch size as 16 and the number of training rounds as 100. The LSTM model hyperparameters are summarized in Table 4.

[0136] Table 4 LSTM model parameter settings

[0137]

[0138] Figure 5 The settlement calculation results of the finite element model and the LSTM model at the 5th, 10th, 15th, 20th monitoring time were compared. The average coefficient of determination (R²) of the LSTM model reached 0.97, and most of the data points were closely distributed near the diagonal line, indicating that it had high prediction accuracy and could be used as a forward model to replace the finite element analysis in ES-MDA.

[0139] Step 2: Uncertainty parameter updating (ES-MDA)

[0140] Figure 6 The compression parameters (Cc) 1~ 5) and the equivalent permeability coefficients (Kz) 1~ 4) are shown in the probability density function (PDF) dynamic updating process in time series (stage = 0, 5, 10, 15, 20, 25). By comparing the prior (stage 0) and posterior distribution characteristics, it is found that, in general, the posterior PDF curve is more concentrated and narrower than the prior PDF curve. 2、 4 and 5's posterior distribution is shifted to the left as a whole, revealing that the mean of its prior distribution is too large, i.e., Cc divided by 2.3 is too large in this example. The PDF curve of the equivalent permeability coefficient Kz shows significant changes. The mean of the prior distribution at the initial time (stage 0) is generally smaller than that of the posterior distribution, meaning that the equivalent vertical permeability coefficient converted based on the test data is too large. In particular 2, its posterior distribution PDF curve is significantly shifted to the right compared to the prior distribution. As the amount of observation data increases, 1、 3 and 4's posterior PDF curve becomes more and more slender, indicating that the parameter uncertainty is decreasing. The PDF curve updating process shows that dynamic data assimilation can effectively correct the prior deviation of the parameters by continuously integrating observation data.

[0141] Step 3: Predicting seawall settlement values

[0142] Figure 7 The dynamic updating process of the settlement prediction results is shown as the amount of observation data increases. Figure 7The (a) to (f) in FIG. 6 respectively show the settlement predictions using the posterior samples of soil parameters after incorporating 1, 5, 10, 15, 20, and 25 observations, and the comparison with the predictions based on the prior distribution of soil parameters. The results show that the settlement predictions based on the prior distribution of soil parameters have a wide 95% confidence interval (CI), such as the 95% CI ranges from 1.7 m to 2.5 m when predicting the long-term settlement, such as the settlement at the 1000th day, and the prediction interval has a large uncertainty. There is a significant deviation between the mean value of the settlement prediction (indicated by the blue dashed line) obtained from the prior distribution of soil parameters and the measured value (circle point). Overall, as the observation data is gradually incorporated, the deviation between the posterior prediction mean value (indicated by the red solid line) and the measured value gradually decreases, and the 95% CI gradually narrows. Specifically, when 5 observation data points are fused, the deviation between the prediction mean value and the measured value significantly decreases; as the number of observation data points increases to 10, the prediction mean value is basically consistent with the measured value; however, when the number of observation data points increases to 15, the deviation between the prediction mean value and the measured value increases, mainly due to the sudden increase in the settlement rate of the 11th observation data point, resulting in an overestimation of the subsequent prediction results; as the number of observation data points continues to increase to 20 and 25, the deviation between the prediction mean value and the measured value decreases again, and the confidence interval further narrows. In addition, the posterior confidence interval is significantly narrower than the prior, and the uncertainty of the prediction is significantly reduced, verifying the effectiveness of the method in the dynamic updating process.

[0143] Figure 8 FIG. 6 shows the influence of the number of monitoring data on settlement prediction. Figure 8 (a) in FIG. 6 is the mean value of the settlement prediction when assimilating different numbers of monitoring data. The mean value of the settlement prediction obtained from the prior distribution and the mean value of the settlement prediction obtained from the posterior distribution of stage 1 are significantly deviated from the measured value. Overall, as the monitoring data is continuously assimilated, the fitting degree of the prediction mean value curve and the measured value gradually increases, and the improvement of the settlement prediction is more obvious when the first few monitoring values are assimilated. Figure 8 (b) in FIG. 6 shows the 95% prediction interval at different numbers of monitoring data. When using the prior distribution of soil parameters or after incorporating the initial monitoring data, the interval is mainly located in the upper left corner of the diagonal, meaning that the actual settlement is generally underestimated. As the number of assimilated monitoring data increases, the prediction interval gradually approaches the diagonal, and the interval width is significantly smaller, and the prediction deviation and uncertainty significantly decrease.

[0144] Figure 9The convergence characteristics of the predicted deposition amount at the 28th monitoring time (1013 days) with the data assimilation process are quantitatively analyzed, and the mean value of the posterior distribution and its 95% confidence interval are shown. In the early assimilation stage (≤11 monitoring times), the posterior mean gradually approaches the measured value, but the posterior distribution shows a wide confidence interval; with the continuous increase of the data amount (≥11 monitoring times), the confidence interval width tends to be stable, but the mean gradually deviates from the measured value; and after 15 monitoring times, the posterior mean again gradually approaches the measured value. In particular, the mean deviation phenomenon that occurs at the 11th-15th monitoring time is consistent in space and time with the sudden increase in prediction error of the corresponding interval in the middle Figure 8 In addition, the posterior distribution confidence interval of the last stage increases, which may be due to data noise.

[0145] Figure 10 The influence of the amount of observation data on the prediction error of the deposition is shown. The mean absolute error (MAE) is used as the error evaluation index, and its calculation formula is:

[0146] ;

[0147] Wherein, ( ) is the total number of samples, is the observation number, is the predicted value, is the measured value. The results show that: in the initial stage (i.e. when the first observation data point is fused), the prediction error of the deposition is significantly reduced, which verifies the preliminary effectiveness of data assimilation. With the gradual increase of the amount of observation data, the prediction error of the deposition generally shows a decreasing trend, and after 11 observation data points are fused, the error tends to be stable. However, when the observation data point increases to 11, the error appears a short-term increase, mainly due to the sudden increase of the deposition rate at this point, which increases the deviation between the predicted value and the measured value. After 15 observation data points are fused, with the gradual stabilization of the deposition rate, the error again shows a decreasing trend.

[0148] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A smart prediction method for seawall subsidence, characterized in that: include, Read the fixed-interval settlement data from the settlement gauge and preprocess it to obtain a multi-point time-settlement dataset; Collect basic data to establish a numerical model of the seawall; By selecting uncertainty parameters, a parameter-settlement dataset is generated based on the constructed seawall numerical model, and a seawall surrogate model is constructed. Based on a multi-data assimilation set smoother, the uncertain parameters are updated, and the updated parameter posterior set is output. Input each member of the updated parameter posterior set into the seawall proxy model to predict the seawall settlement value; A tiered early warning mechanism is triggered when the predicted settlement value exceeds the alarm threshold. The process of updating the uncertainty parameters and outputting the updated posterior set of parameters based on a multiple data assimilation set smoother specifically includes the following steps: Set the total number of iterations for data assimilation ; Definition of the first iteration coefficient of thermal expansion The constraints are satisfied: ； Using the uncertain parameters as target update variables, based on information from collected geological survey reports and existing engineering data, the prior distribution of the target update variables is determined, and an initial parameter set is generated: ； in, The number of members in the set; each member For the combination of parameters drawn independently from the prior distribution; For the Each member of the current parameter set in the next iteration The predicted settlement value is calculated using the seawall proxy model, and the expression is: ； in, The pre-trained seawall surrogate model is a long short-term memory network; the predicted settlement value is... Belongs to the real number space , It is the product of the number of monitoring points and the number of time steps; Calculate the ensemble statistics based on the current parameter set and the predicted settlement values; Calculate the Kalman gain matrix based on set statistics and update the initial parameter set; If the current iteration number = If the iteration terminates, the final updated posterior set of parameters is output. The specific steps for predicting seawall subsidence include the following: By inputting each member of the posterior parameter set into the seawall surrogate model, the predicted settlement values ​​for future time periods are obtained: ； in, For the first The predicted settlement sequence corresponding to each member. For the number of predicted time points; Calculate the statistical characteristics of the predicted mean, predicted standard deviation, and confidence intervals; The mechanism for triggering a tiered early warning when the predicted settlement value exceeds the alarm threshold is as follows: Loading shall be stopped when the surcharge height H of the seawall is ≤4.0m and the daily settlement rate is ≥20mm / d; loading shall be allowed when the continuous average settlement is ≤3mm / d. When the loading height H of the seawall is greater than 4.0m and the daily settlement rate is greater than or equal to 10mm / d, loading shall be stopped. When the continuous average settlement is less than or equal to 2mm / d, loading shall be permitted.

2. The intelligent prediction method for seawall subsidence as described in claim 1, characterized in that: The preprocessing to obtain a multi-point time-settlement dataset includes the following steps. Abnormal settlement values ​​at each monitoring point were identified and eliminated based on the IQR method. Alignment verification is performed on the time series data of each monitoring point to ensure that the acquisition time of each monitoring point is synchronized; the cumulative settlement and settlement rate of each monitoring point are calculated, and finally a standardized multi-monitoring point time-settlement dataset is output.

3. The intelligent prediction method for seawall subsidence as described in claim 1, characterized in that: The process of collecting basic data to establish a numerical model of the seawall includes the following steps. Collect basic data from geological survey reports, existing engineering data, seawall design documents, and construction loading plans; Soil layers are divided based on the geological survey report, and the physical and mechanical parameters of each soil layer are determined. A numerical model of the seawall was established based on the actual design dimensions. The Mohr-Coulomb and Modified Cam-clay constitutive models were used to simulate the soil behavior, and the loads were applied step by step according to the construction loading plan to simulate the entire construction process.

4. The intelligent prediction method for seawall subsidence as described in claim 3, characterized in that: The process of selecting uncertainty parameters and generating a parameter-settlement dataset based on the constructed seawall numerical model includes the following steps: Uncertainty parameters are selected, and the parameter distribution range of the uncertainty parameters is determined based on the constructed seawall numerical model; Based on the parameter distribution range, a representative parameter combination is generated using the Latin hypercube sampling method. By calling the seawall numerical model API through a Python program to perform batch calculations, the seawall settlement response values ​​corresponding to each set of parameters are obtained, and finally a complete parameter-settlement dataset is constructed.

5. The intelligent prediction method for seawall subsidence as described in claim 4, characterized in that: The construction of the seawall proxy model includes dividing the constructed parameter-settlement dataset into a training set and a test set, wherein the training set is used to construct the seawall proxy model and the test set is used to verify the accuracy of the seawall proxy model.

6. A smart prediction system for seawall subsidence, based on the smart prediction method for seawall subsidence according to any one of claims 1 to 5, characterized in that: The data acquisition module is used to read the settlement data of the settlement gauge at fixed intervals; The data processing module is used for preprocessing to obtain a multi-point time-settlement dataset; The data collection module is used to collect basic data to build a numerical model of the seawall. The model building module is used to select uncertainty parameters, generate parameter-settlement datasets based on the constructed seawall numerical model, and construct a seawall proxy model. A parameter update model is used to update uncertain parameters based on a multiple data assimilation set smoother, and output the updated parameter posterior set. The settlement detection module is used to input each member of the updated parameter posterior set into the seawall proxy model to predict the seawall settlement value. The early warning mechanism module is used to trigger a tiered early warning mechanism when the predicted settlement value exceeds the alarm threshold.

Images