A prediction method for the spatio-temporal vibration trend of guide walls

Abstract

The present invention discloses a prediction method for the spatio-temporal vibration trend of a guide wall in the technical field of time series prediction, which includes inputting data with a correlation exceeding a threshold with the transverse vibration response characteristics of the guide wall into a CNN-BiLSTM-CBAM model to obtain the predicted values of the transverse vibration modal components of the guide wall; adding the predicted values of the transverse vibration modal components of the guide wall to obtain the predicted transverse vibration signal of the guide wall. The input features of the CNN-BiLSTM-CBAM model are data with a correlation exceeding a threshold with the transverse vibration response characteristics of the guide wall, and the output features are multiple modal components obtained by decomposing the transverse vibration response data of the guide wall. The present invention decomposes the spatio-temporal vibration response data of the guide wall into multiple modal components, reduces the complexity of the guide wall vibration data, and obtains the predicted transverse vibration signal of the guide wall based on the spatio-temporal vibration data and the transverse vibration modal components of the guide wall, solving the problem of inaccurate prediction accuracy of the spatio-temporal vibration trend of the guide wall vibration.

Description

Technical Field

[0001] The present invention relates to a method for predicting the spatio-temporal vibration trend of a guide wall, belonging to the technical field of time series prediction. Background Art

[0002] The water discharge structure is an important part of the water conservancy project, undertaking the key responsibility of discharging excess water and ensuring the safety of key hydraulic structures such as dams. As an important component of the water discharge structure, the guide wall is used to separate the discharged water flow from the water flow discharged from the power station behind the dam, so as to reduce the impact of the ejected water flow and the tail water fluctuation on the power station during the high-speed water discharge process. However, the light and thin-walled guide wall structure is subjected to the action of transient alternating water flow loads. When the amplitude exceeds the allowable value specified for the guide wall structure, continuous strong vibration is very likely to cause fatigue damage to the structure, thus posing a great threat to the project and the safety of people's lives and property. Therefore, accurately predicting the vibration response of the guide wall is crucial for evaluating the safety of the guide wall and effectively controlling the vibration.

[0003] Existing vibration prediction methods in the field of hydraulic engineering have a series of defects. For example, the guide wall vibration prediction method based on the finite element method has many artificial assumptions, high calculation costs, and low efficiency; most of the research methods on the vibration of hydraulic structures based on machine learning do not consider the spatio-temporal coupling characteristics and evolution trends between different parts; existing intelligent prediction models for the vibration of hydraulic structures often show complex structures, poor interpretability, and the inability to fully capture the inherent complex nonlinear relationships in the vibration sequence characteristics, so that the prediction accuracy of the spatio-temporal vibration trend of the guide wall vibration is inaccurate.

[0004] The information disclosed in this background art section is only intended to increase the understanding of the overall background of the present invention, and should not be regarded as an admission or any form of implication that this information constitutes the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: how to solve the problem of inaccurate prediction accuracy of the spatio-temporal vibration trend of the guide wall vibration.

[0006] To solve the above technical problem, the present invention is implemented by adopting the following technical solution.

[0007] The present invention discloses a method for predicting the spatio-temporal vibration trend of a guide wall, including:

[0008] Collecting the spatio-temporal vibration data of the guide wall and the area adjacent to the guide wall during the flood discharge period;

[0009] Obtaining the transverse vibration response characteristics of the guide wall in the spatio-temporal vibration data;

[0010] Determine the spatio-temporal vibration data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the set threshold, and sequentially input the determined spatio-temporal vibration data into the pre-trained CNN-BiLSTM-CBAM model;

[0011] Use the convolutional neural network CNN of the CNN-BiLSTM-CBAM model to perform convolution and pooling on the input spatio-temporal vibration data to obtain spatio-temporal features;

[0012] Use the convolutional block attention module CBAM of the CNN-BiLSTM-CBAM model to focus on the channel and spatial dimensions of the spatio-temporal features to obtain data features, and reassemble and fit the data features to obtain the first prediction value;

[0013] Use the bidirectional long short-term memory BiLSTM network of the CNN-BiLSTM-CBAM model to capture the output result with a bidirectional long-term dependence relationship with the input spatio-temporal vibration data to obtain the second prediction value;

[0014] Use the splicing layer of the CNN-BiLSTM-CBAM model to splice the first prediction value and the second prediction value to obtain multiple prediction values of the transverse vibration modes of the guide wall;

[0015] Add the multiple prediction values of the transverse vibration modes of the guide wall to obtain the predicted transverse vibration signal of the guide wall;

[0016] Predict the spatio-temporal vibration trend of the guide wall based on the predicted transverse vibration signal of the guide wall.

[0017] Further, before sequentially inputting the corresponding spatio-temporal vibration data into the pre-trained CNN-BiLSTM-CBAM model for prediction, convert the corresponding spatio-temporal vibration data into continuous sub-time series data based on the sliding window iterative prediction mode.

[0018] Further, the collection of the spatio-temporal vibration data of the guide wall and the area adjacent to the guide wall during flood discharge includes:

[0019] Collect the vertical, longitudinal, and transverse acceleration time history curves of eight different parts of the guide wall and the area adjacent to the guide wall respectively;

[0020] Obtain spatio-temporal vibration data according to the vertical, longitudinal, and transverse acceleration time history curves of eight different parts of the guide wall and the area adjacent to the guide wall;

[0021] Among them, the eight different parts of the guide wall and the area adjacent to the guide wall include: the right guide wall, the left guide wall, the right side of the grouting gallery, the left side of the grouting gallery, the junction of the right guide wall gallery and the ground, the extension of the right guide wall gallery to a depth of 100 meters, the junction of the left guide wall gallery and the ground, and the extension of the left guide wall gallery to a depth of 100 meters.

[0022] Further, the transverse vibration response characteristics of the guide wall include the transverse acceleration vibration response of the guide wall.

[0023] The spatio-temporal vibration data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the set threshold includes:

[0024] Taking the transverse acceleration vibration response sequence in the spatio-temporal vibration data as the second feature, and the remaining acceleration vibration response sequences as the first feature;

[0025] Forming an ordered pair set from the first feature and the second feature, so that each ordered pair contains a first feature value and a corresponding second feature value; wherein, the first feature value refers to the maximum mutual information coefficient value of the first feature; the second feature value refers to the maximum mutual information coefficient value of the second feature;

[0026] On a two-dimensional plane, with the first feature as the axis and the second feature as the axis, constructing a feature space;

[0027] Along the axis direction, dividing the feature space into equally wide intervals, and along the axis direction, dividing the feature space into equally high intervals, forming a grid ;

[0028] According to the number of points in the ordered pair set falling into each grid of the grid , obtaining the mutual information estimation value of the first feature and the second feature;

[0029] According to the mutual information estimation value of the first feature and the second feature, using the maximum mutual information coefficient algorithm, calculating the maximum mutual information coefficient value of each pair of the first feature and the second feature to characterize the correlation;

[0030] Taking the first feature whose maximum mutual information coefficient value with the second feature exceeds the set threshold as the spatio-temporal vibration data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the set threshold.

[0031] Further, when the maximum mutual information coefficient value between the first feature and the second feature is larger, the correlation between the first feature and the second feature is stronger; conversely, when the maximum mutual information coefficient value is smaller, the correlation between the first feature and the second feature is smaller.

[0032] Further, according to the number of points in the ordered pair set that fall within each grid of the grid the calculation formula for the mutual information estimate value of the first feature and the second feature is expressed as:

[0033] (1);

[0034] In the formula, represents the first feature and the second feature forming an ordered pair set in the grid with the number of cells being the maximum mutual information estimate value in represents the probability distribution of the points in the set falling on the grid ; represents taking the maximum value of the data under the distribution of the grid ; is the first feature and the second feature forming an ordered pair set in the grid with the number of cells being the mutual information estimate value in

[0035] Further, the calculation formula for the maximum mutual information coefficient is expressed as:

[0036] (2);

[0037] In the formula, represents the maximum mutual information coefficient value of the first feature and the second feature with a value range of [0, 1]; represents the exponential function of the total number of samples N of the first feature and the second feature, ; represents the number of cells in the grid ; represents the logarithmic function with base 2, represents taking and the minimum value of represents the maximum value function, represents the maximum mutual information estimate value of the ordered pair set formed by the first feature X and the second feature Y in the grid with the number of cells being in

[0038] Furthermore, the CNN-BiLSTM-CBAM model is trained by using spatio-temporal vibration data whose correlation with the cross-river vibration response characteristics of the guide wall exceeds a set threshold as input and the corresponding cross-river vibration modal components of the guide wall as output;

[0039] Among them, the cross-river vibration modal components of the guide wall are obtained by decomposing the cross-river vibration response characteristics of the guide wall into intrinsic mode functions with different complexities by using the variational mode decomposition algorithm.

[0040] Furthermore, the expression of the variational mode decomposition algorithm is:

[0041] (3);

[0042] In the formula, represents the th modal component decomposed by the variational mode decomposition algorithm, represents the th central frequency decomposed by the variational mode decomposition algorithm, , represents the total number of modal components decomposed by the variational mode decomposition algorithm, represents the imaginary unit, represents the Dirac function at time represents the partial derivative operation at time represents the convolution operation, represents the cross-river vibration response data of the guide wall at time represents the constraint condition; represents the minimum value operation, represents the th modal component decomposed by the variational mode decomposition algorithm at time

[0043] Furthermore, the convolutional neural network CNN includes a convolutional layer, an activation function, and a pooling layer connected in sequence;

[0044] The convolutional block attention module CBAM includes a channel attention mechanism CAM and a spatial attention mechanism SAM, which are used to respectively focus on the channel and spatial dimensions of the spatio-temporal features to obtain data features;

[0045] The bidirectional long short-term memory BiLSTM network includes a forward long short-term memory LSTM network and a backward long short-term memory LSTM network.

[0046] Compared with the prior art, the beneficial effects achieved by the present invention:

[0047] The present invention decomposes the complex cross-river vibration response data of the guide wall into multiple modal components, significantly reducing the complexity of the spatio-temporal vibration response data of the guide wall. Compared with directly predicting the vibration signal of the guide wall, the present invention decomposes the complex cross-river vibration response data of the guide wall into simpler modal components, and then inputs the data whose correlation with the cross-river vibration response characteristics of the guide wall exceeds the threshold into the CNN-BiLSTM-CBAM model for prediction to obtain the predicted values of the cross-river vibration modal components of the guide wall, and then integrates the predicted values of the cross-river vibration modal components of the guide wall to obtain the predicted vibration signal of the guide wall, which can better improve the accuracy and stability of the overall prediction.

[0048] The CNN-BiLSTM-CBAM model provided by the present invention overcomes the shortcomings of existing time prediction models, such as weak ability to capture non-linearity and data features and low prediction accuracy, by integrating the convolutional neural network CNN, the convolutional block attention module CBAM, and the bidirectional long short-term memory BiLSTM network, provides a technical reference for improving the ability to capture non-linear features of the guide wall vibration response and the data prediction accuracy. At the same time, the present invention also solves the problem of inaccurate prediction accuracy of the spatio-temporal vibration trend of the guide wall vibration in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a schematic flow chart of a method for predicting the spatio-temporal vibration trend of a guide wall provided by an embodiment of the present invention;

[0050] Figure 2 is a schematic diagram of the network architecture of the CNN-BiLSTM-CBAM model provided by an embodiment of the present invention;

[0051] Figure 3 is a schematic diagram of the network structure of the convolutional block attention module CBAM provided by an embodiment of the present invention;

[0052] Figure 4 is a schematic diagram of the calculation result of the maximum mutual information coefficient value of the spatio-temporal vibration response data of the guide wall provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] The technical solution of the present invention will be described in detail below through the drawings and specific embodiments. It should be understood that the specific features in the embodiments of the present invention and the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0054] Embodiment 1

[0055] This embodiment introduces a method for predicting the spatio-temporal vibration trend of a guide wall, as Figure 1 shown, including:

[0056] Step 101: Collect the spatio-temporal vibration data of the guide wall and the area adjacent to the guide wall during flood discharge;

[0057] In the embodiment of the present invention, the spatio-temporal vibration data of the guide wall and its adjacent area during flood discharge can be obtained through devices such as sensors. The spatio-temporal vibration data includes the vibration information of the guide wall at different time points and spatial positions. In this embodiment, the area adjacent to the guide wall can be determined according to the collected guide wall area and in combination with the preset determination conditions for the adjacent area. As an embodiment, the area adjacent to the guide wall can be determined as the right side of the grouting gallery, the left side of the grouting gallery, the junction of the right guide wall gallery and the ground, the extension of the right guide wall gallery to a depth of 100 meters, the junction of the left guide wall gallery and the ground, and the extension of the left guide wall gallery to a depth of 100 meters. The guide wall area may include: the right guide wall and the left guide wall.

[0058] Step 102: Obtain the cross-river vibration response characteristics of the guide wall in the spatio-temporal vibration data;

[0059] The so-called cross-river vibration response characteristics of the guide wall refer to the vibration response characteristics in the direction perpendicular to the river flow direction.

[0060] Step 103: Determine the spatio-temporal vibration data whose correlation with the cross-river vibration response characteristics of the guide wall exceeds the set threshold, and sequentially input the determined spatio-temporal vibration data into a pre-trained CNN-BiLSTM-CBAM model. Use the CNN-BiLSTM-CBAM model to predict the modal components of the cross-river vibration response of the guide wall, and obtain multiple predicted values of the cross-river vibration modal components of the guide wall;

[0061] The purpose of determining the spatio-temporal vibration data whose correlation with the cross-river vibration response characteristics of the guide wall exceeds the set threshold is to screen out the data with high correlation with the cross-river direction of the guide wall from the collected spatio-temporal vibration data. To improve the prediction accuracy, the spatio-temporal vibration data can also be preprocessed, for example: by removing the data that is not directly related to the prediction target to reduce data redundancy.

[0062] The CNN-BiLSTM-CBAM model integrates the advantages of the convolutional neural network CNN, the bidirectional long short-term memory BiLSTM network, and the convolutional block attention module CBAM, and can capture the spatial features, temporal dependencies, and important feature information of the data simultaneously. Specifically, using the convolutional neural network CNN of the CNN-BiLSTM-CBAM model, convolutional and pooling operations are performed on the input spatio-temporal vibration data to obtain spatio-temporal features; using the convolutional block attention module CBAM of the CNN-BiLSTM-CBAM model, the channels and spatial dimensions of the spatio-temporal features are focused to obtain data features, and the data features are reassembled and fitted to obtain a first prediction value; using the bidirectional long short-term memory BiLSTM network of the CNN-BiLSTM-CBAM model, the output result with bidirectional long-term dependencies with the input spatio-temporal vibration data is captured to obtain a second prediction value; using the concatenation layer of the CNN-BiLSTM-CBAM model to concatenate the first prediction value and the second prediction value to obtain multiple prediction values of the transverse vibration modes of the guide wall in the cross-river direction;

[0063] Through the prediction of the CNN-BiLSTM-CBAM model, the prediction values of each modal component at future time points can be obtained, and the prediction values reflect the vibration trends of the guide wall at different frequencies or time scales.

[0064] Step 104: Add the multiple prediction values of the transverse vibration modes of the guide wall in the cross-river direction to obtain the predicted vibration signal of the guide wall in the cross-river direction;

[0065] Step 105: Predict the spatio-temporal vibration trend of the guide wall based on the predicted vibration signal of the guide wall in the cross-river direction.

[0066] In summary, in the embodiment of the present invention, by integrating the prediction values of each transverse vibration mode component of the guide wall in the cross-river direction into an overall prediction result, the overall predicted vibration signal of the guide wall at future time points can be obtained.

[0067] According to the predicted vibration signal of the guide wall, the stability and safety of the guide wall can be evaluated or other related engineering analyses can be performed.

[0068] In step 103, before applying the CNN-BiLSTM-CBAM model, a training set should be constructed first. The training set includes input features and output features. In the embodiment of the present invention, the input features are data whose correlation with the transverse vibration response features of the guide wall exceeds a threshold, and the output features are multiple modal components obtained by decomposing the transverse vibration response data of the guide wall.

[0069] Using signal processing techniques, the transverse vibration response data of the guide wall is decomposed into multiple modal components, which are used as the target values of the prediction model. Each modal component represents the vibration characteristics of the transverse vibration response data of the guide wall at different frequencies or time scales, helping to capture the non-linear and non-stationary characteristics in the transverse vibration response data of the guide wall.

[0070] Embodiment 2

[0071] Based on the same inventive concept as Embodiment 1, this embodiment introduces a method for predicting the spatio-temporal vibration trend of the guide wall, which specifically includes the following steps:

[0072] Step 1: Input feature selection:

[0073] Collect the spatio-temporal vibration data of the guide wall and the adjacent area during flood discharge, including:

[0074] Collect the vertical, longitudinal, and transverse acceleration time history curves of eight different parts of the guide wall and the adjacent area of the guide wall respectively;

[0075] According to the vertical, longitudinal, and transverse acceleration time history curves of eight different parts of the guide wall and the adjacent area of the guide wall, obtain the spatio-temporal vibration data.

[0076] The transverse vibration response characteristics of the guide wall include the transverse acceleration vibration response of the guide wall. Further, the determination of the spatio-temporal vibration data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the set threshold includes:

[0077] Take the transverse acceleration vibration response sequence in the spatio-temporal vibration data as the second feature, and the remaining acceleration vibration response sequences as the first feature;

[0078] Form an ordered pair set by combining the first feature and the second feature, so that each ordered pair contains a first feature value and a corresponding second feature value; among them, the first feature value refers to the maximum mutual information coefficient value of the first feature; the second feature value refers to the maximum mutual information coefficient value of the second feature;

[0079] On the two-dimensional plane, with the first feature as axis, the second feature as axis, construct a feature space;

[0080] Along axis direction, divide the feature space into equally wide intervals, along axis direction, divide the feature space into equally high intervals, forming grid ;

[0081] According to the points in the ordered pair set falling into the grid The quantity within each grid is obtained to estimate the mutual information value between the first feature and the second feature;

[0082] Based on the estimated mutual information value between the first feature and the second feature, the maximum mutual information coefficient value for each pair of the first feature and the second feature is calculated using the maximum mutual information coefficient algorithm, which is used to characterize the correlation;

[0083] The first feature with the maximum mutual information coefficient value exceeding the set threshold with respect to the second feature is taken as the spatio-temporal vibration data whose correlation with the transverse vibration response feature of the guide wall exceeds the set threshold.

[0084] Among them, the calculation formula for the estimated mutual information value between the first feature and the second feature is expressed as:

[0085] (1);

[0086] In the formula, represents the first feature and the second feature forming an ordered pair set In a grid with the number of cells being the maximum estimated mutual information value, represents the probability distribution of the points in the set falling on the grid represents taking the maximum value of the data under the distribution of the grid

[0087]

[0088]

[0089] (2);

[0089] In the formula, represents the maximum mutual information coefficient value between the first feature and the second feature with a value range of [0, 1]; represents the exponential function of the total number of samples N of the first feature and the second feature, ; represents the number of cells within the grid ​represents the logarithmic function with base 2, represents taking and the minimum value of, represents the maximum value function, represents the set of ordered pairs formed by the first feature X and the second feature Y in a grid with the number of cells being the maximum mutual information estimate value in.

[0090] Take the first feature whose maximum mutual information coefficient value with the second feature exceeds the threshold as the data whose correlation with the transverse vibration response feature of the guide wall exceeds the threshold.

[0091] When the first feature and the second feature the larger the maximum mutual information coefficient value, the first feature and the second feature the stronger the correlation between them. Conversely, the smaller the maximum mutual information coefficient value, the first feature and the second feature the smaller the correlation between them.

[0092] In this embodiment, the threshold of the maximum mutual information coefficient is defined as 0.86.

[0093] As Figure 4 shown, the data whose correlation with the transverse vibration response feature of the guide wall exceeds the threshold are: Feature 14, Feature 19, Feature 20, Feature 22, Feature 23, Feature 24 as input features. It should be noted that the maximum mutual information coefficient is dimensionless data.

[0094] Step 2: Data decomposition:

[0095] Since the complexity of the spatio-temporal vibration data is relatively high, in this embodiment, the variational mode decomposition algorithm is used to perform stationary processing on the transverse vibration response data of the guide wall.

[0096] The CNN-BiLSTM-CBAM model is trained using the spatio-temporal vibration data whose correlation with the transverse vibration response feature of the guide wall exceeds the set threshold as input and the corresponding transverse vibration mode components of the guide wall as output;

[0097] Among them, the transverse vibration mode components of the guide wall are obtained by decomposing the transverse vibration response feature of the guide wall into intrinsic mode functions with different complexities using the variational mode decomposition algorithm.

[0098] The variational mode decomposition algorithm can be expressed by the following formula:

[0099] (3);

[0100] Wherein, represents the th modal component decomposed by the variational mode decomposition algorithm, represents the th central frequency decomposed by the variational mode decomposition algorithm, , represents the total number of modal components decomposed by the variational mode decomposition algorithm, represents the imaginary unit, represents the Dirac function at time represents the partial derivative operation at time represents the convolution operation, represents the transverse vibration response data of the guide wall in the river direction at time represents the constraint condition; represents the operation of finding the minimum value, represents the th modal component decomposed by the variational mode decomposition algorithm at time

[0101] Step 3: Data preprocessing:

[0102] Before inputting the corresponding spatio-temporal vibration data into the pre-trained CNN-BiLSTM-CBAM model for prediction in sequence, based on the sliding window iterative prediction mode, the corresponding spatio-temporal vibration data is converted into continuous sub-time series data, and the obtained continuous sub-time series data is divided into a training set and a test set according to a ratio of 0.8:0.2.

[0103] In this embodiment, batch normalization is adopted, the length of the sliding window is taken as 10, and data preprocessing is performed on the data whose correlation with the transverse vibration response characteristics of the guide wall in the river direction exceeds the threshold. Among them, the formula for batch normalization is:

[0104] (4);

[0105] Wherein, represents the th data whose correlation with the transverse vibration response characteristics of the guide wall in the river direction exceeds the threshold 's normalized value, represents the mean of all data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the threshold, σ represents the standard deviation of all data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the threshold, γ and β respectively represent the weight scaling parameter and the weight offset parameter, which are used to adjust the output of batch normalization. Usually, γ is initialized to 1 and β is initialized to 0, and ε represents a very small value to prevent division by zero.

[0106] Step Four: Prediction by the CNN-BiLSTM-CBAM Model:

[0107] As Figure 2 shown, the convolutional neural network CNN includes a convolutional layer, an activation function, and a pooling layer connected in sequence;

[0108] The convolutional block attention module CBAM includes a channel attention mechanism CAM and a spatial attention mechanism SAM, which are used to respectively focus on the channels and spatial dimensions of the spatio-temporal features to obtain data features.

[0109] The bidirectional long short-term memory BiLSTM network includes a forward long short-term memory LSTM network and a backward long short-term memory LSTM network.

[0110] The convolutional neural network CNN, including a convolutional layer, an activation function, and a pooling layer, is used to perform convolution and pooling on the data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the threshold to obtain spatio-temporal features.

[0111] The convolutional block attention module CBAM, including a channel attention mechanism CAM and a spatial attention mechanism SAM, is used to focus on the channels and spatial dimensions of the spatio-temporal features to obtain data features, and reassemble and fit the data features to obtain a first prediction value. The schematic diagram of the network structure of the convolutional block attention module CBAM is as Figure 3 shown.

[0112] The bidirectional long short-term memory BiLSTM network, including a forward long short-term memory LSTM network and a backward long short-term memory LSTM network, is used to splice the forward long short-term memory LSTM network and the backward long short-term memory LSTM network, and capture the output result with bidirectional long-term dependence relationship of the data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the threshold to obtain a second prediction value.

[0113] The splicing layer is used to splice the first prediction value and the second feature value.

[0114] In this embodiment, the global parameters of the CNN-BiLSTM-CBAM model in this example are set as follows: time window T = 10, number of features F = 5; the training parameters of the CNN-BiLSTM-CBAM model are set as follows: batch size = 2, number of epochs Epoch = 100, optimizer Optimizer = Adam, loss function Loss function = MSE. The parameters of the convolutional neural network CNN are set as follows: number of filters Filters = 64, kernel size Kernel_size = 3, dropout rate Dropout rate = 0.1, activation function Activation = ReLU; the parameters of the bidirectional long short-term memory BiLSTM network are set as follows: number of units in the hidden layer Units = 64, number of network layers Layer = 1, dropout rate Dropout rate = 0.3, activation function Activation = Sigmoid.

[0115] After setting the above parameters, use the training set to train the CNN-BiLSTM-CBAM model to obtain a trained CNN-BiLSTM-CBAM model.

[0116] In this embodiment, after determining the spatio-temporal vibration data whose correlation with the transverse vibration response characteristics of the guide wall exceeds the set threshold, sequentially input the determined spatio-temporal vibration data into the pre-trained CNN-BiLSTM-CBAM model;

[0117] Use the convolutional neural network CNN of the CNN-BiLSTM-CBAM model to perform convolution and pooling on the input spatio-temporal vibration data to obtain spatio-temporal features.

[0118] Use the convolutional block attention module CBAM of the CNN-BiLSTM-CBAM model to focus on the channel and spatial dimensions of the spatio-temporal features to obtain data features, and reassemble and fit the data features to obtain a first prediction value.

[0119] Use the bidirectional long short-term memory BiLSTM network of the CNN-BiLSTM-CBAM model to capture the output result with a bidirectional long-term dependence relationship with the input spatio-temporal vibration data to obtain a second prediction value.

[0120] Use the concatenation layer of the CNN-BiLSTM-CBAM model to concatenate the first prediction value and the second prediction value to obtain multiple prediction values of the transverse vibration mode components of the guide wall.

[0121] Add the multiple prediction values of the transverse vibration mode components of the guide wall to obtain the predicted transverse vibration signal of the guide wall;

[0122] Predict the spatio-temporal vibration trend of the guide wall based on the vibration signal predicted in the transverse direction of the guide wall river

[0123] Step 5: Prediction performance evaluation:

[0124] Use the test set to test and evaluate the performance of the CNN-BiLSTM-CBAM model. Based on the evaluation metrics, compare the prediction performance with that of different baseline models to verify the generalization ability and accuracy of the CNN-BiLSTM-CBAM model.

[0125] The evaluation metrics selected are the mean square error MSE, root mean square error RMSE, mean absolute error MAE, and correlation coefficient R. The formulas are as follows:

[0126]

[0127] Where, represents the total number of samples of the spatio-temporal vibration response data of the guide wall, represents the true value of the spatio-temporal vibration response data of the th guide wall, represents the predicted value of the spatio-temporal vibration response data of the th guide wall, represents the true value of the spatio-temporal vibration response data of the th guide wall and the predicted value of the spatio-temporal vibration response data of the th guide wall represents the true value of the spatio-temporal vibration response data of the th guide wall represents the variance of the predicted value of the spatio-temporal vibration response data of the

[0128] The verification results are shown in Table 1:

[0129]

[0130] In summary, in the embodiments of the present invention, the complex spatio-temporal vibration response data of the guide wall is decomposed into multiple modal components, significantly reducing the complexity of the spatio-temporal vibration response data of the guide wall. Compared with directly predicting the vibration signal of the guide wall, in the embodiments of the present invention, the complex vibration data is decomposed into simpler modal components, and then the spatio-temporal data with the maximum mutual information coefficient exceeding the threshold is predicted through the CNN-BiLSTM-CBAM model to obtain the predicted value of the transverse vibration modal component of the guide wall. The predicted vibration signal of the guide wall obtained by integrating the predicted values of each transverse vibration modal component of the guide wall can improve the overall prediction accuracy and stability more effectively.

[0131] The CNN-BiLSTM-CBAM model provided by the embodiments of the present invention integrates the convolutional neural network CNN, the convolutional block attention module CBAM, and the bidirectional long short-term memory BiLSTM network, overcomes the disadvantages of existing time prediction models such as weak ability to capture non-linear and data features and low prediction accuracy, provides a technical reference for improving the ability to capture non-linear features of the guide wall vibration response and data prediction accuracy. At the same time, the embodiments of the present invention also solve the problem of inaccurate prediction accuracy of the spatio-temporal vibration trend of the guide wall vibration in the prior art.

[0132] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented 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.

[0133] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a machine for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for implementing the functions specified in one block or multiple blocks.

[0134] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for implementing the functions specified in one block or multiple blocks.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for implementing the functions specified in one block or multiple blocks.

[0136] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit of the present invention and the scope protected by the claims. All of these fall within the protection scope of the present invention.

Claims

1. A method for predicting the temporal and spatial vibration trend of a guide wall, characterized in that: include: Collect spatiotemporal vibration data of the guide wall and the area adjacent to the guide wall during flood discharge; Obtain the transverse vibration response characteristics of the guide wall in the spatiotemporal vibration data; Determine the spatiotemporal vibration data whose correlation with the transverse river vibration response characteristics of the guide wall exceeds a set threshold, and sequentially input the determined spatiotemporal vibration data into the pre-trained CNN-BiLSTM-CBAM model; Using the convolutional neural network (CNN) of the CNN-BiLSTM-CBAM model, the input spatiotemporal vibration data is convolved and pooled to obtain spatiotemporal features; Using the convolutional block attention module CBAM of the CNN-BiLSTM-CBAM model, focusing on the channel and spatial dimensions of the spatiotemporal features to obtain data features, reassembling and fitting the data features to obtain a first prediction value; Using the bidirectional long short-term memory BiLSTM network of the CNN-BiLSTM-CBAM model, the output result having a bidirectional long-term dependency relationship with the input spatiotemporal vibration data is captured to obtain a second prediction value; Using the splicing layer of the CNN-BiLSTM-CBAM model to splice the first prediction value and the second prediction value, a plurality of prediction values ​​of the transverse river vibration modal components of the guide wall are obtained; Adding the plurality of predicted values ​​of the transverse-river vibration modal components of the guide wall to obtain a predicted transverse-river vibration signal of the guide wall; Predict the temporal and spatial vibration trend of the guide wall based on the predicted vibration signal of the guide wall in the transverse direction; Wherein, the transverse river vibration response characteristics of the guide wall include the transverse river acceleration vibration response of the guide wall; The determining of the spatiotemporal vibration data having a correlation with the transverse vibration response characteristics of the guide wall exceeding a set threshold value comprises: The acceleration vibration response sequence of the guide wall in the transverse direction of the spatiotemporal vibration data is used as the second feature, and the other acceleration vibration response sequences are used as the first feature; The first feature and the second feature are combined into an ordered pair set, so that each ordered pair includes a first eigenvalue and a corresponding second eigenvalue; wherein the first eigenvalue refers to the maximum mutual information coefficient value of the first feature; and the second eigenvalue refers to the maximum mutual information coefficient value of the second feature; On a two-dimensional plane, the first feature is axis, the second feature is Axis, construct feature space; along The axis direction divides the feature space into intervals of equal width, along The axis direction divides the feature space into intervals of equal height, forming Grid ; According to the points in the ordered pair set fall on the grid The number of each grid is calculated to obtain the mutual information estimation value of the first feature and the second feature; According to the mutual information estimation value between the first feature and the second feature, using the maximum mutual information coefficient algorithm, calculate the maximum mutual information coefficient value of each pair of the first feature and the second feature, for characterizing the correlation; The first feature whose maximum mutual information coefficient value with the second feature exceeds the set threshold is taken as the spatiotemporal vibration data whose correlation with the transverse river vibration response characteristic of the guide wall exceeds the set threshold.

2. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: It also includes converting the corresponding spatiotemporal vibration data into continuous sub-time series data based on a sliding window iterative prediction mode before sequentially inputting the corresponding spatiotemporal vibration data into a pre-trained CNN-BiLSTM-CBAM model for prediction.

3. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: The collection of spatiotemporal vibration data of the guide wall and the area adjacent to the guide wall during flood discharge includes: The vertical, downstream and cross-river acceleration time history curves of eight different locations of the guide wall and the area adjacent to the guide wall were collected respectively; According to the vertical, downstream and cross-river acceleration time history curves of eight different parts of the guide wall and the area adjacent to the guide wall, the spatiotemporal vibration data are obtained; Among them, the eight different parts of the guide wall and the adjacent area of ​​the guide wall include: the right guide wall, the left guide wall, the right side of the grouting corridor, the left side of the grouting corridor, the junction of the right guide wall corridor and the ground, the right guide wall corridor extending to a depth of 100 meters, the junction of the left guide wall corridor and the ground, and the left guide wall corridor extending to a depth of 100 meters.

4. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: When the first feature With the second feature The larger the maximum mutual information coefficient value, the greater the first feature With the second feature The stronger the correlation between them, the smaller the maximum mutual information coefficient value is. With the second feature The smaller the correlation between them.

5. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: According to the points in the ordered pair set fall on the grid The number of each grid, the calculation formula for the mutual information estimation value of the first feature and the second feature is expressed as: （1）； In the formula, Indicates the first feature With the second feature Ordered pair set In the cell number Grid The maximum mutual information estimate in , Representing a collection The points in the grid fall on The probability distribution on Represents the grid The maximum value of the data under the distribution, The first feature With the second feature Ordered pair set In the cell number Grid The mutual information estimate in .

6. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: The calculation formula of the maximum mutual information coefficient is expressed as: （2）； In the formula, Indicates the first feature With the second feature The maximum mutual information coefficient value of is in the range of [0,1]; The exponential function representing the total number of samples N of the first feature and the second feature, ; Representation Grid The number of cells in represents the logarithmic function with base 2, Indicates taking and The minimum value of represents the maximum value function, Indicates that the first feature X and the second feature Y form an ordered pair set In the cell number Grid The maximum mutual information estimate in .

7. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: The CNN-BiLSTM-CBAM model uses the spatiotemporal vibration data with a correlation with the vibration response characteristics of the guide wall in the transverse direction exceeding a set threshold as input and the corresponding vibration modal component of the guide wall in the transverse direction as output for training and acquisition; The transverse-river vibration modal component of the guide wall is obtained by decomposing the transverse-river vibration response characteristics of the guide wall into eigenmode functions of different complexities using a variational mode decomposition algorithm.

8. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 7, characterized in that: The expression of the variational mode decomposition algorithm is: （3）； In the formula, represents the first modal components, represents the first The center frequency, , represents the total number of modal components decomposed by the variational mode decomposition algorithm, represents the imaginary unit, express The Dirac function at time , express The partial derivative operation at time, represents the convolution operation, express The transverse vibration response data of the guide wall at all times, Indicates constraints; It represents the minimum value operation. express The moment is decomposed by the variational mode decomposition algorithm. modal components.

9. The method for predicting the temporal and spatial vibration trend of a guide wall according to claim 1, characterized in that: The convolutional neural network CNN includes a convolutional layer, an activation function and a pooling layer connected in sequence; The convolutional block attention module CBAM includes a channel attention mechanism CAM and a spatial attention mechanism SAM, which are used to respectively focus on the channel and spatial dimensions of the spatiotemporal features to obtain data features; The bidirectional long short-term memory BiLSTM network includes a forward long short-term memory LSTM network and a backward long short-term memory LSTM network.

Images