Dam multi-point deformation prediction method and system

Abstract

The invention discloses a dam multi-point deformation prediction method and system, and relates to the technical field of dam safety monitoring, and the method comprises the steps: constructing a finite element analysis model, calculating a hydraulic component displacement feature set, introducing measuring point space coordinates to construct a dam deformation feature factor set, carrying out the feature screening through employing an improved BorutaShap algorithm, and building a simple and efficient deformation prediction BorutaShap model. And capturing a complex nonlinear relationship contained in a residual error of a model predicted value and a real value by using an iTransform deep learning model, predicting a residual error value, superposing the residual error value with a BorutaShap model predicted value, outputting a final deformation predicted value, and establishing a BorutaShap-iTransform model. According to the dam multi-point deformation prediction method, through feature screening and residual error correction, the model complexity is remarkably reduced, effective information in residual errors is effectively mined, and the dam multi-point deformation prediction precision is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dam safety monitoring, in particular to a dam multi-point deformation prediction method and system. BACKGROUND

[0002] Dam deformation is a key indicator reflecting the safety state of its structure. Traditional deformation prediction models (such as statistical models) are mostly constructed for a single measuring point, and it is difficult to effectively depict the multi-point correlation deformation characteristics formed by the overall structure of the dam in cooperative bearing.

[0003] The dam deformation prediction analysis method is the core project of concrete dam safety monitoring. Although the introduction of spatial coordinates can reflect the position difference of the measuring points and enhance the multi-point prediction ability of the dam, it will significantly increase the input feature dimension of the model, and a large number of redundant features not only increase the complexity of the model, but also reduce the prediction accuracy and generalization ability. At the same time, the linear model established cannot accept all the deformation characteristics of the dam, and there is a moving residual between the predicted value and the true value, and the residual often contains complex nonlinear information (such as environmental factor mutation effect, material nonlinearity, etc.) that has not been fully captured.

[0004] Therefore, in order to meet the needs of concrete dam safety monitoring, the spatial coordinate information factor is introduced to realize the prediction of multiple points, and the spatial correlation of multiple measuring points is established, and the feature redundancy and residual capture are effectively solved. On this basis, a more effective deformation prediction method is constructed, which is the problem to be solved by the present application.

[0005] In order to solve these problems, a dam multi-point deformation prediction method and system are urgently needed. SUMMARY

[0006] To solve the above problems, the present application proposes a dam multi-point deformation prediction method and system, which aims to reduce the input dimension of the model and improve the efficiency through an improved feature selection algorithm.

[0007] A dam multi-point deformation prediction method, comprising the following steps: Step 1: constructing a dam section finite element analysis model through historical data, and constructing an initial dam deformation characteristic factor set based on the dam section finite element analysis model ; Step 2: screening an important characteristic factor set from the initial dam deformation characteristic factor set based on the BorutaShap algorithm ; Step 3: establishing a dam deformation prediction BorutaShap model based on the important characteristic factor set , outputting a dam deformation prediction value , and calculating the residual of the dam deformation prediction BorutaShap model ; Step 4: inputting the residual The input is trained and predicted based on the iTransformer deep learning model of inverse architecture, and the output is residual prediction value ; The residual prediction value is superimposed with the prediction value of the BorutaShap model in step 2 , to establish a dam deformation prediction BorutaShap-iTransformer model. Based on the dam deformation prediction BorutaShap-iTransformer model, the dam multi-point deformation is predicted.

[0008] Preferably, in step 1, a dam section finite element analysis model is constructed based on historical data, and an initial dam deformation feature factor set is constructed based on the dam section finite element analysis model. The specific content of includes: H The finite element grid model is established by historical data and the load is applied to establish the dam section finite element analysis model. The water level before the dam is substituted into the dam section finite element analysis model to calculate the water pressure component displacement factor . Taking a reference point of the dam as a reference, the relative spatial coordinate values of each measuring point are introduced, and the water pressure component displacement factor is combined to construct an initial dam deformation feature factor set

[0009] . The expression of the initial dam deformation feature factor set is: ; ; ; Wherein, is the number of monitoring days, is the period parameter, is the power parameter, is the finite element feature set, is the temperature related feature set, is the aging related feature set.

[0010] Preferably, in step 2, based on the BorutaShap algorithm, the important feature factor set is screened from the initial dam deformation feature factor set , which specifically includes: The BorutaShap algorithm is used to screen the initial dam deformation feature factor set Perform feature screening to generate the initial dam deformation feature factor set Shadow features with the same statistical distribution ; Shadow Features The expression is: ; in, is the random rearrangement function, is the shadow feature index, for F A single original feature in is the index of the original feature; Shadow Features Merge to get shadow feature set ; The initial dam deformation characteristic factor set With shadow feature set Merge to form an extended feature set ; Extended feature set The expression is: ; Compute extended feature set The Shapley value of the .

[0011] Preferably, the extended feature set is calculated The Shapley value of the , specifically including: Based on extended feature set Build a training prediction model and calculate the Shapley value of each feature based on the SHAP (Shapley Additive exPlanations) framework, including the Shapley values ​​of the original features and shadow features. The Shapley values ​​of the original features and shadow features are used as their feature importance scores. Filter important feature sets based on the size of the Sharpley value : for Features, its The maximum value of all shadow features Compare values; If the original feature The value is higher than the maximum value in the shadow feature The value and statistical significance test p The value is less than the significance level , then the original feature is determined to be an important feature ; Otherwise, it is determined to be a redundant feature and removed; for The filter expression for the feature is: ; in p For statistical significance test p value.

[0012] Preferably, the Shapley calculation method is as follows: ; in, For the Sharpley value of the feature; No features A subset of features; F is the set of all features; Add features to the model The predicted value after Defeature the model The predicted value after is the factorial symbol, is a feature subset S The number of elements.

[0013] Preferably, in step 2, based on the important feature factor set Establish the BorutaShap model for dam deformation prediction and calculate the residual of the BorutaShap model for dam deformation prediction , specifically including: Based on important feature sets , establish important features The linear mapping with dam deformation is used to construct the BorutaShap model for dam deformation prediction; The expression of the BorutaShap model for dam deformation prediction is: ; in is the predicted value of the BorutaShap model, for A single key feature in b is the bias term, Features The weight of Express Sum the calculation results of all features in; Calculate the residual between the predicted result and the true value of the dam linear prediction model , the expression is: .

[0014] wherein is the displacement value.

[0015] Preferably, the residual error in step 3 is The input is trained and predicted based on the iTransformer deep learning model of inverse architecture, and the output is the residual prediction value The residual prediction value is superimposed with the prediction value of the BorutaShap model in step 2 to establish a dam deformation prediction BorutaShap-iTransformer model, which specifically includes: Step 3.1, the residual error is input into the embedding layer of the iTransformer model and inverted, and the expression of the inversion operation is: ; ; ; wherein is the embedding bias vector, is the embedding weight matrix, and the dimension is T × d model , T is the time step, d model is the number of model points, is the feature matrix after embedding, dimension permutation operation, is the feature matrix after inversion, P is the position encoding matrix, is the feature matrix after fusing position information; Step 3.2, the feature matrix after fusing position information is input into the multi-element attention module of the iTransformer model to realize dynamic weighted focusing on key features, and the expression is: ; ; ; wherein, Z is the input matrix of the self-attention mechanism, wherein , K , V respectively represent the query vector (Q query vector ), the key vector (K key vector ) and the value vector (V value vector ). denotes the projection dimension, linear transformation matrix of query, key, value, normalization function, output matrix of self-attention mechanism; Step 3.3, the output matrix of the self-attention mechanism iTransformer model layer normalization module, standardize the feature distribution to enhance the training stability, the expression is: ; wherein, is the initial layer-normalized output feature matrix; is the layer normalization operation; Step 3.4, the initial layer-normalized output feature matrix is input into the iTransformer feedforward network module, and the feature nonlinear expression is enhanced, and the expression is: ; wherein, W1, b1, W2, b2 represent the learnable weight matrix and bias item respectively, and max(0,...) represents the ReLU rectified linear unit activation function, denotes the output of the feedforward neural network; Step 3.5, the output feature of the feedforward neural network is subjected to secondary layer normalization, and the feature distribution is constrained, and the expression is: ; wherein, is the output matrix of the secondary layer normalization; Step 3.6, output residual correction value : the layer normalization result is mapped, and the prediction result is output, and the expression is: ; wherein, L is the number of stacked layers, b is the prediction step, the final feature after L layer stacking, the weight and bias of the output layer, is the residual prediction value, t is the current time step; Step 3.7, the residual prediction value is superimposed with the BorutaShap model prediction value, a dam multi-point prediction BorutaShap-iTransformer model is established, and the calculation formula and modeling expression are: ; ; wherein, is the model prediction result, Outputs the predicted value for the sequence after training with the iTransformer deep learning model.

[0016] A dam multi-point deformation prediction system, comprising: Data acquisition unit: Construct the dam section finite element analysis model through historical data, and construct the initial dam deformation characteristic factor set based on the dam section finite element analysis model ; Initial model construction unit: Based on the BorutaShap algorithm, the initial dam deformation characteristic factor set Screening to obtain important feature factor sets , based on the set of important feature factors Establish the BorutaShap model for dam deformation prediction and output the predicted value of dam deformation ; Residual prediction unit: Calculates the residual of the BorutaShap model for dam deformation prediction , the residual Input the iTransformer deep learning model based on the inverse architecture for training and prediction, and output the residual prediction value ; Model secondary construction unit: the residual prediction value Compared with the predicted value of BorutaShap model in step 2 Superposition was performed to establish the BorutaShap-iTransformer model for dam deformation prediction; Model prediction unit: Based on the dam deformation prediction BorutaShap-iTransformer model, multi-point deformation of the dam is predicted.

[0017] In summary, compared with traditional technologies, the method and system for predicting multi-point deformation of a dam of the present invention adopts the method of calculating and predicting residuals to overcome the problem of limited prediction accuracy caused by feature redundancy and insufficient utilization of residual information in existing multi-point deformation prediction methods of dams; and uses advanced deep learning models to deeply mine the correlation information in the residuals, significantly improving the accuracy and robustness of multi-point deformation prediction of dams, and is suitable for high-precision prediction of multi-point deformation of concrete dams considering spatial coordinate information.

[0018] The technical method of the present invention is further described in detail below through the accompanying drawings and examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a dam multi-point deformation prediction method according to the present invention; Figure 2 A finite element mesh of a concrete dam section in a dam multi-point deformation prediction method according to the present invention; Figure 3 a dam extension feature set for a dam multi-point deformation prediction method of the present application; Figure 4 a BorutaShap feature screening result based on a dam multi-point deformation prediction method of the present application; Figure 5 a BorutaShap model prediction and residual for a dam multi-point deformation prediction method of the present application, Figure 5 (a) in the above is a prediction and residual result of measuring point 1, Figure 5 (b) in the above is a prediction and residual result of measuring point 1; Figure 6 a iTransformer model flowchart for a dam multi-point deformation prediction method of the present application; Figure 7 a BorutaShap-iTransformer model prediction and residual prediction schematic diagram for a dam multi-point deformation prediction method of the present application, Figure 7 (a) in the above is a model prediction result of measuring point 1, Figure 7 (b) in the above is a residual prediction result of measuring point 1, Figure 7 (c) in the above is a model prediction result of measuring point 2, Figure 7 (d) in the above is a residual prediction result of measuring point 2; Figure 8 a module schematic diagram of a dam multi-point deformation prediction system of the present application. DETAILED DESCRIPTION

[0020] The technical method of the present application is further described below by means of drawings and examples. It should be noted that: unless otherwise specified, the relative arrangement, numerical expression and values of the components and steps set forth in these examples do not limit the scope of the present application.

[0021] The following description of at least one exemplary embodiment is merely exemplary in nature and is in no way intended to limit the present application or its application or uses.

[0022] Techniques, systems, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered as part of the specification, where appropriate.

[0023] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary, and not as a limitation. Thus, other examples of the exemplary embodiments can have different values.

[0024] Unless otherwise defined, technical and scientific terms used herein should have the meaning commonly understood by one of ordinary skill in the art to which this application belongs.

[0025] The application provides a dam multi-point deformation prediction method and system. First, a finite element analysis model is constructed, a water pressure component displacement feature set is calculated, and a dam deformation characteristic factor set is constructed by introducing a monitoring point space coordinate; then, an improved BorutaShap algorithm is applied for feature screening, redundant factors are removed, and a simplified and efficient deformation prediction BorutaShap model is established; finally, an iTransformer deep learning model based on an inverse architecture is used to capture the complex nonlinear relationship contained in the model residual error, accurately predict the residual error value, and superimpose the BorutaShap model prediction value to output the final deformation prediction value, thereby establishing a BorutaShap-iTransformer model. The specific implementation flowchart is shown in the accompanying Figure 1 .

[0026] The specific scheme is as follows: A dam multi-point deformation prediction method, comprising the following steps: Step 1: constructing a dam section finite element analysis model through historical data, and constructing an initial dam deformation characteristic factor set based on the dam section finite element analysis model .

[0027] Further, the specific content of step 1 of constructing a dam section finite element analysis model through historical data and constructing an initial dam deformation characteristic factor set based on the dam section finite element analysis model includes: A finite element grid model is established through historical data, and a dam section finite element analysis model is established by applying loads, wherein the schematic diagram of the finite element grid model is shown in Figure 2 .

[0028] The monitoring date dam front water level H is substituted into the dam section finite element analysis model to calculate the water pressure component displacement factor .

[0029] The relative spatial coordinate value of the monitoring point is introduced with reference to a certain reference point of the dam, and the water pressure component displacement factor is combined to construct an initial dam deformation characteristic factor set .

[0030] Further, the expression of the initial dam deformation characteristic factor set is: .

[0031] .

[0032] .

[0033] .

[0034] wherein, is the monitoring day, is the period parameter, is the power parameter, representing the nonlinear association of the feature with the spatial coordinates, is the finite element feature set, is the temperature-related feature set, is the aging-related feature set.

[0035] Step 2: Based on the BorutaShap algorithm, the initial dam deformation feature factor set is screened to obtain an important feature factor set , based on the important feature factor set a dam deformation prediction BorutaShap model is established, and the dam deformation prediction value is output , and the residual of the dam deformation prediction BorutaShap model is calculated .

[0036] Further, in step 2, based on the BorutaShap algorithm, the initial dam deformation feature factor set is screened to obtain an important feature factor set , which specifically includes: The BorutaShap algorithm is used to perform feature screening on the initial dam deformation feature factor set , and shadow features with the same statistical distribution as the initial dam deformation feature factor set are generated .

[0037] The expression of the shadow feature is: .

[0038] wherein, is a random rearrangement function, is the number of shadow features, is a single original feature in F , and is the index of the original feature.

[0039] The shadow features are combined to obtain a shadow feature set .

[0040] The initial dam deformation feature factor set is merged with the shadow feature set to form an expanded feature set .

[0041] wherein, the expanded feature set The schematic diagram of the initial dam deformation feature set is shown in Figure 3 , wherein, wherein , , F is an initial dam deformation feature set, is a single original feature in F , is an index of the original feature, is a finite element feature set, is a temperature-related feature set, is an aging-related feature set.

[0042] The expression of the extended feature set is: .

[0043] The Shapley value of the extended feature set is calculated to identify and retain an important feature factor set .

[0044] Further, the Shapley value of the extended feature set is calculated to identify and retain an important feature factor set , specifically including: Based on the extended feature set , a training prediction model is constructed, and the Shapley value of each feature is calculated based on the SHAP framework, including the Shapley value of the original feature and the shadow feature, and the Shapley value of the original feature and the shadow feature is taken as the feature importance score thereof.

[0045] Based on the size of the Sharpley value, an important feature set is screened, wherein the feature screening result is shown in Figure 4 .

[0046] For a feature, its value is compared with the maximum value of all shadow features.

[0047] If the value of the original feature is higher than the maximum value of the shadow feature and the p value of the statistical significance test is less than the significance level , it is determined that the original feature is an important feature .

[0048] Otherwise, it is determined as a redundant feature and is removed.

[0049] For the screening expression of a feature is: .

[0050] in p For statistical significance test p value.

[0051] Furthermore, the Shapley calculation method is as follows: .

[0052] in, For the Sharpley value of the feature; No features A subset of features; F is the set of all features; Add features to the model i The predicted value after Defeature the model i The predicted value after is the factorial symbol, is a feature subset S The number of elements.

[0053] Furthermore, in step 2, based on the important feature factor set Establish the BorutaShap model for dam deformation prediction and calculate the residual of the BorutaShap model for dam deformation prediction , specifically including: Based on important feature sets , establish the important features of original features and shadow features The linear mapping of the dam deformation and the dam deformation is used to construct the BorutaShap model for dam deformation prediction. The BorutaShap model prediction and residual diagram are shown in the figure below. Figure 5 shown.

[0054] The expression of the BorutaShap model for dam deformation prediction is: .

[0055] in, is the predicted value of the BorutaShap model, for A single key feature in b is the bias term, Features The weight of right The calculation results of all features in are summed.

[0056] Calculate the residual between the predicted result and the true value of the dam linear prediction model , the expression is: .

[0057] wherein is the monitoring displacement value; Further, step 3: input the residual error into the iTransformer deep learning model based on the inverse architecture for training and prediction, and output the residual prediction value .

[0058] The residual prediction value is superimposed with the prediction value of the BorutaShap model in step 2 to establish a dam deformation prediction BorutaShap-iTransformer model.

[0059] Further, in step 3, the residual error is input into the iTransformer deep learning model based on the inverse architecture for training and prediction, and the residual prediction value is output. The residual prediction value is superimposed with the prediction value of the BorutaShap model in step 2 Figure 6 to establish a dam deformation prediction BorutaShap-iTransformer model, as shown in specifically comprising: Step 3.1, input the residual error into the embedding layer of the iTransformer model and perform inversion operation, and the expression of the inversion operation is: .

[0060] .

[0061] .

[0062] wherein is the embedding bias vector, is the embedding weight matrix, and the dimension is T × d model , T is the time step, d model is the number of model points, is the feature matrix after embedding, is the dimension permutation operation, is the feature matrix after inversion, P is the position encoding matrix, is the feature matrix after fusing position information.

[0063] Step 3.2, input the feature matrix after fusing position information The input iTransformer model multi-element attention module realizes dynamic weighted focusing on key features, and the expression is: .

[0064] .

[0065] .

[0066] wherein, Z The input matrix of the self-attention mechanism is where , K , V respectively represent the query vector ( query vector ), the key vector ( key vector ), and the value vector ( value vector ); represents the projection dimension, the linear transformation matrix of the query, the key, and the value, the normalization function, the output matrix of the self-attention mechanism.

[0067] Step 3.3, normalize the output matrix of the self-attention mechanism iTransformer model layer to standardize the feature distribution and enhance the training stability, and the expression is: .

[0068] wherein, is the first layer-normalized output feature matrix, is the layer normalization operation.

[0069] Step 3.4, input the first layer-normalized output feature matrix into the iTransformer feedforward network module to enhance the nonlinear expression of the features, and the expression is: .

[0070] wherein, W1, b1, W2, b2 respectively represent the learnable weight matrix and the bias term, max(0,...) represents the ReLU rectified linear unit activation function, represents the output of the feedforward neural network.

[0071] Step 3.5, perform secondary layer normalization on the output features of the feedforward neural network to constrain the feature distribution, and the expression is: .

[0072] wherein, is the output matrix of the secondary layer normalization.

[0073] Step 3.6: Output residual correction value : Map the layer normalization results and output the prediction results. The expression is: .

[0074] in, L is the number of stacking layers, b is the prediction step size, go through L The final feature after layer stacking, The weights and biases of the output layer, is the residual prediction value, t is the current time step.

[0075] Step 3.7: Superimpose the residual prediction value with the BorutaShap model prediction value to establish the BorutaShap-iTransformer model for dam multi-point prediction. The calculation formula and modeling expression are as follows: .

[0076] = .

[0077] in, is the model prediction result, Outputs the predicted value for the sequence after training with the iTransformer deep learning model.

[0078] Or input the initial dam deformation characteristic factor set and dam deformation monitoring data to select the important characteristic factor set Establish the BorutaShap model for dam deformation prediction and output the prediction results , calculate the model residual , and input the iTransformer model for training to output the residual correction value , and the prediction results of BorutaShap model Superposition to obtain the final deformation prediction value , where the BorutaShap-iTransformer model prediction and residual prediction diagram is as follows Figure 7 shown.

[0079] A dam multi-point deformation prediction system, such as Figure 8 Shown, including: Data acquisition unit: Construct the dam section finite element analysis model through historical data, and construct the initial dam deformation characteristic factor set based on the dam section finite element analysis model ; Initial model construction unit: Based on the BorutaShap algorithm, the initial dam deformation characteristic factor set Screening to obtain an important feature factor set , based on the important feature factor set Establish a dam deformation prediction BorutaShap model, output the dam deformation prediction value ; Residual prediction unit: calculate the residual of the dam deformation prediction BorutaShap model , input the residual into the iTransformer deep learning model based on the inverse architecture for training and prediction, and output the residual prediction value ; Model secondary construction unit: superimpose the residual prediction value and the prediction value of the BorutaShap model in step 2 , and establish a dam deformation prediction BorutaShap-iTransformer model Model prediction unit: based on the dam deformation prediction BorutaShap-iTransformer model, predict the dam multi-point deformation.

[0080] Finally, it should be noted that: the above examples are only used to illustrate the technical method of the present application, but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand: the technical method of the present application can still be modified or replaced by the equivalent, and these modifications or equivalent replacements also cannot make the modified technical method deviate from the spirit and scope of the technical method of the present application.

Claims

1. A method for predicting multi-point deformation of a dam, characterized in that: The following steps are involved: Step 1: Construct a finite element analysis model of the dam section based on historical data, and construct an initial dam deformation characteristic factor set based on the finite element analysis model of the dam section ; Step 2: Based on the BorutaShap algorithm, the initial dam deformation characteristic factor set Screening to obtain important feature factor sets , based on the set of important feature factors Establish the BorutaShap model for dam deformation prediction and output the predicted value of dam deformation , and calculate the residual of the BorutaShap model for dam deformation prediction ; Step 3: Convert the residual Input the iTransformer deep learning model based on the inverse architecture for training and prediction, and output the residual prediction value ; The residual prediction value Compared with the predicted value of BorutaShap model in step 2 Superposition was performed to establish the BorutaShap-iTransformer model for dam deformation prediction; Based on the dam deformation prediction BorutaShap-iTransformer model, multi-point deformation of the dam is predicted.

2. A dam multi-point deformation prediction method according to claim 1, characterized in that: In step 1, the finite element analysis model of the dam section is constructed based on historical data, and the initial dam deformation characteristic factor set is constructed based on the finite element analysis model of the dam section. The specific contents include: Establish a finite element mesh model through historical data and apply loads to establish a finite element analysis model of the dam section; The water level in front of the dam will be monitored on the date H Substitute the finite element analysis model of the dam section to calculate the displacement factor of the water pressure component ; Taking a certain benchmark point of the dam as a reference, the relative spatial coordinate values ​​of each measuring point are introduced , combined with the water pressure component displacement factor , construct the initial dam deformation characteristic factor set .

3. A dam multi-point deformation prediction method according to claim 2, characterized in that: Initial dam deformation characteristic factor set The expression is: ； ； ； ； in, For monitoring days, is the period parameter, is the power parameter, is the finite element feature set, is the temperature-related feature set, is a time-dependent feature set.

4. A dam multi-point deformation prediction method according to claim 1, characterized in that: In step 2, the initial dam deformation characteristic factor set is calculated based on the BorutaShap algorithm. Screening to obtain important feature factor sets , specifically including: The initial dam deformation characteristic factor set is analyzed using the BorutaShap algorithm. Perform feature screening to generate the initial dam deformation feature factor set Shadow features with the same statistical distribution ; Shadow Features The expression is: ； in, is the random rearrangement function, is the shadow feature index, for F A single original feature in is the index of the original feature; Shadow Features Merge to get shadow feature set ; The initial dam deformation characteristic factor set With shadow feature set Merge to form an extended feature set ; Extended feature set The expression is: ； Compute extended feature set The Shapley value of the .

5. A dam multi-point deformation prediction method according to claim 4, characterized in that: Compute extended feature set The Shapley value of the , specifically including: Based on extended feature set Build a training prediction model and calculate the Shapley value of each feature based on the SHAP framework, including the Shapley value of the original feature and the shadow feature, and use the Shapley value of the original feature and the shadow feature as their feature importance score; Filter important feature sets based on the size of the Sharpley value : for feature, its Shapley value The maximum Shapley value of all shadow features Make comparisons; If the original feature The value is significantly higher than the maximum value of the shadow feature The value of the statistical significance test p The value is less than the significance level , then the original feature is determined to be an important feature ; Otherwise, it is determined to be a redundant feature and removed; for The filter expression for the feature is: ； in p For statistical significance test p value.

6. A dam multi-point deformation prediction method according to claim 5, characterized in that: The Shapley calculation method is as follows: ； in, For the Sharpley value of each feature; No features A subset of features; F is the set of all features; Add features to the model The predicted value after Defeature the model The predicted value after is the factorial symbol, is a feature subset S The number of elements.

7. A dam multi-point deformation prediction method according to claim 1, characterized in that: In step 2, based on the important feature factor set Establish the BorutaShap model for dam deformation prediction and output the predicted value of dam deformation , and calculate the residual of the BorutaShap model for dam deformation prediction , specifically including: Based on important feature sets , establish important features The linear mapping with dam deformation is used to construct the BorutaShap model for dam deformation prediction; The expression of the BorutaShap model for dam deformation prediction is: ； in is the predicted value of the BorutaShap model, for A single key feature in b is the bias term, Characterized by The weight of Express Sum the calculation results of all features in; Calculate the residual between the predicted result and the true value of the dam linear prediction model , the expression is: ； in To monitor the displacement value.

8. A dam multi-point deformation prediction method according to claim 1, characterized in that: In step 3, the residual Input the iTransformer deep learning model based on the inverse architecture for training and prediction, and output the residual prediction value , the residual prediction value Compared with the predicted value of BorutaShap model in step 2 The superposition is performed to establish the BorutaShap-iTransformer model for dam deformation prediction, which specifically includes: Step 3.1: The residual Input the embedding layer of the iTransformer model and perform an inversion operation. The expression of the inversion operation is: ； ； ； in is the embedding bias vector, is the embedding weight matrix, with dimension T × d model , T is the time step, d model Number of model measurement points, is the feature matrix after embedding, Dimension permutation operation, is the inverted characteristic matrix, P Position encoding matrix, is the feature matrix after fusing position information; Step 3.2: The feature matrix after fusion of position information Input the multi-attention module of the iTransformer model to achieve dynamic weighted focus on key features. The expression is: ； ； ； in, Z The input matrix of the self-attention mechanism, where Q 、 K 、 V Represent the query vector, key vector and value vector respectively; represents the projection dimension, Linear transformation matrix of query, key, and value, Normalization function, The output matrix of the self-attention mechanism; Step 3.3: Normalize the output matrix of the self-attention mechanism iTransformer model layer to standardize the feature distribution to enhance training stability. The expression is: ； in, is the first layer-normalized output feature matrix, is the layer normalization operation; Step 3.4: Input the initial layer-normalized output feature matrix into the iTransformer feedforward network module to enhance the nonlinear expression of the features. The expression is: ； Among them, W1, b1, W2, b2 represent the learnable weight matrix and bias term respectively, max(0,...) represents the ReLU corrected linear unit activation function, Represents the output of the feedforward neural network; Step 3.5: Normalize the output features of the feedforward neural network in the second layer to constrain the feature distribution. The expression is: ； in, is the normalized output matrix of the secondary layer; Step 3.6: Output residual correction value : Map the layer normalization results and output the prediction results. The expression is: ； in, L is the number of stacking layers, b is the prediction step size, go through L The final feature after layer stacking, The weights and biases of the output layer, is the residual prediction value, t is the current time step; Step 3.7: Superimpose the residual prediction value with the BorutaShap model prediction value to establish the BorutaShap-iTransformer model for dam multi-point prediction. The calculation formula and modeling expression are as follows: ； = ； in, is the model prediction result, Outputs the predicted value for the sequence after training with the iTransformer deep learning model.

9. A dam multi-point deformation prediction system, characterized in that: include: Data acquisition unit: Construct the dam section finite element analysis model through historical data, and construct the initial dam deformation characteristic factor set based on the dam section finite element analysis model ; Initial model construction unit: Based on the BorutaShap algorithm, the initial dam deformation characteristic factor set Screening to obtain important feature factor sets , based on the set of important feature factors Establish the BorutaShap model for dam deformation prediction and output the predicted value of dam deformation ; Residual prediction unit: Calculates the residual of the BorutaShap model for dam deformation prediction , the residual Input the iTransformer deep learning model based on the inverse architecture for training and prediction, and output the residual prediction value ; Model secondary construction unit: the residual prediction value Compared with the predicted value of BorutaShap model in step 2 Superposition was performed to establish the BorutaShap-iTransformer model for dam deformation prediction; Model prediction unit: Based on the dam deformation prediction BorutaShap-iTransformer model, multi-point deformation of the dam is predicted.

Images