Post-earthquake retaining wall surface displacement prediction model training method and device, and electronic equipment

Abstract

The application relates to the technical field of prediction, in particular to a wall surface displacement prediction model training method and device for a post-earthquake retaining wall and electronic equipment, wherein the method comprises the following steps: acquiring retaining wall soil body parameters, retaining wall geometric parameters and seismic motion data; performing numerical simulation on the retaining wall soil body parameters, the retaining wall geometric parameters and the seismic motion data to obtain wall surface displacement of the retaining wall and generate a data set; dividing the data set into multiple training samples and true value labels of each training sample; and training a pre-constructed wall surface displacement prediction model by using the multiple training samples and the true value labels of each training sample, wherein the input of the wall surface displacement prediction model is the retaining wall soil body parameters, the retaining wall geometric parameters and the seismic motion data, and the output of the wall surface displacement prediction model is the wall surface displacement. Therefore, the problem that only the soil body parameters and the geometric parameters are considered for wall surface displacement prediction in the related art, and the accuracy is low is solved.

Description

Technical Field

[0001] This application relates to the field of prediction technology, and in particular to a method, apparatus and electronic equipment for training a prediction model of wall displacement of a retaining wall after an earthquake. Background Technology

[0002] Retaining walls, as an important engineering structure, are widely used in various fields such as transportation, water conservancy, construction, and environmental engineering. For example, in highway and railway engineering, they are used for roadbed slope stabilization and bridge connection section and tunnel entrance protection, effectively preventing landslides and riverbed compression. In municipal engineering, the use of stone materials and greenery achieves a unity of function and aesthetics while protecting ancient buildings. However, earthquakes pose a significant challenge to the stability of retaining walls, causing changes in soil pressure behind the walls, resulting in wall displacement, posing huge risks to the project, and seriously threatening people's lives and property.

[0003] The retaining wall displacement prediction methods in related technologies only consider soil parameters and geometric parameters, and rely on models such as random forest and XGBoost for prediction, resulting in low accuracy of prediction results. Moreover, these models exhibit low efficiency and large memory consumption when processing high-dimensional and sparse data. Summary of the Invention

[0004] This application provides a method, device, and electronic equipment for training a wall displacement prediction model for a retaining wall after an earthquake, which solves the problem that the accuracy of wall displacement prediction is low when only soil parameters and geometric parameters are considered in related technologies.

[0005] The first aspect of this application provides a method for training a prediction model of wall displacement of a retaining wall after an earthquake, comprising the following steps: acquiring soil parameters, geometric parameters, and ground motion data of the retaining wall; performing numerical simulation on the soil parameters, geometric parameters, and ground motion data to obtain the wall displacement of the retaining wall, and generating a dataset from the soil parameters, geometric parameters, ground motion data, and wall displacement; dividing the dataset into multiple training samples and ground truth labels for each training sample, wherein the training samples include soil parameters, geometric parameters, and ground motion data, and the ground truth labels are wall displacements; training a pre-constructed wall displacement prediction model using the multiple training samples and the ground truth labels for each training sample, wherein the input of the wall displacement prediction model is the soil parameters, geometric parameters, and ground motion data of the retaining wall, and the output of the wall displacement prediction model is the wall displacement.

[0006] Optionally, a pre-built wall displacement prediction model is trained using multiple training samples and the ground truth labels of each training sample, including: dividing the multiple training samples and the ground truth labels of each training sample into a training set and a validation set; iteratively training the wall displacement prediction model using the training set, and optimizing the hyperparameters of the wall displacement prediction model based on the output data and ground truth labels of the wall displacement prediction model during the training process; and validating the performance of the wall displacement prediction model using the validation set.

[0007] Optionally, the wall displacement prediction model is iteratively trained using the training set, including: feature binding of the data in the training set; sampling of training samples in the training set after feature binding based on gradients; generating feature histograms based on the sampled training samples; iteratively training the wall displacement prediction model using the sampled training samples; calculating the training loss based on the output data and ground truth labels of the wall displacement prediction model during training; correcting the feature histograms during training; calculating the gain based on the corrected feature histograms; and optimizing the hyperparameters of the wall displacement prediction model based on the gain until the training loss reaches the training objective.

[0008] Optionally, generating a feature histogram based on the sampled training samples includes: calculating the first-order gradient of each training sample; sorting the calculated first-order gradient values, taking the first preset number of training samples as high-gradient samples, taking a portion of the remaining training samples as low-gradient samples, and adding a weight compensation factor to the loss function of the low-gradient samples; and generating a feature histogram based on the high-gradient samples and the compensated low-gradient samples.

[0009] Optionally, generating a feature histogram based on high-gradient samples and compensated low-gradient samples includes: for each feature split point, calculating the gradient and second derivative sum of the left and right child nodes of the split point respectively; during the calculation, the original gradient is used to calculate the high-gradient samples, and the compensated gradient is used to calculate the high-gradient samples; generating a feature histogram based on the gradient and second derivative sum of the left and right child nodes of the split point.

[0010] Optionally, the formula for calculating the gain value is:

[0011]

[0012] Where, ∑g left ∑g represents the sum of the first-order gradients of all samples in the left child node. right ∑g represents the sum of the first-order gradients of all samples in the right child node. parent ∑h represents the sum of the first-order gradients of all samples of the parent node. left Let ∑h represent the sum of the second-order gradients of all samples in the left child node. right Let ∑h represent the sum of the second-order gradients of all samples in the right child node. parentλ represents the sum of the second-order gradients of all samples of the parent node, and λ represents the L2 regularization coefficient.

[0013] A second aspect of this application provides a training device for a post-earthquake retaining wall displacement prediction model, comprising: an acquisition module for acquiring retaining wall soil parameters, retaining wall geometric parameters, and seismic ground motion data; a generation module for performing numerical simulation on the retaining wall soil parameters, retaining wall geometric parameters, and seismic ground motion data to obtain the retaining wall displacement, thereby generating a dataset from the retaining wall soil parameters, retaining wall geometric parameters, seismic ground motion data, and wall displacement; a partitioning module for partitioning the dataset into multiple training samples and ground truth labels for each training sample, wherein the training samples include retaining wall soil parameters, retaining wall geometric parameters, and seismic ground motion data, and the ground truth labels are wall displacements; and a training module for training a pre-constructed wall displacement prediction model using the multiple training samples and the ground truth labels for each training sample, wherein the input to the wall displacement prediction model is the retaining wall soil parameters, retaining wall geometric parameters, and seismic ground motion data, and the output of the wall displacement prediction model is the wall displacement.

[0014] Optionally, the training module is further used to divide multiple training samples and the ground truth labels of each training sample into a training set and a validation set; to iteratively train the wall displacement prediction model using the training set, and to optimize the hyperparameters of the wall displacement prediction model based on the output data and ground truth labels of the wall displacement prediction model during the training process; and to validate the performance of the wall displacement prediction model using the validation set.

[0015] Optionally, the training module is further used to perform feature binding on the data in the training set; to sample training samples in the training set after feature binding is completed based on gradients; to generate feature histograms based on the sampled training samples; to iteratively train the wall displacement prediction model using the sampled training samples; during the training process, to calculate the training loss based on the output data and ground truth labels of the wall displacement prediction model; during the training process, to correct the feature histograms; to calculate the gain based on the corrected feature histograms; and to optimize the hyperparameters of the wall displacement prediction model based on the gain until the training loss reaches the training objective.

[0016] Optionally, the training module is further used to calculate the first gradient of each training sample; sort the calculated first gradient values, take the first preset number of training samples as high gradient samples, take a portion of the remaining training samples as low gradient samples, and add a weight compensation factor to the loss function of the low gradient samples; generate a feature histogram based on the high gradient samples and the compensated low gradient samples.

[0017] Optionally, the training module is further used to calculate the gradient and second derivative sum of the left and right child nodes of each feature split point. During the calculation, the original gradient is used for high gradient samples and the compensated gradient is used for high gradient samples. A feature histogram is generated based on the gradient and second derivative sum of the left and right child nodes of the split point.

[0018] Optionally, the formula for calculating the gain value is:

[0019]

[0020] Where, ∑g left ∑g represents the sum of the first-order gradients of all samples in the left child node. right ∑g represents the sum of the first-order gradients of all samples in the right child node. parent ∑h represents the sum of the first-order gradients of all samples of the parent node. left Let ∑h represent the sum of the second-order gradients of all samples in the left child node. right Let ∑h represent the sum of the second-order gradients of all samples in the right child node. parent λ represents the sum of the second-order gradients of all samples of the parent node, and λ represents the L2 regularization coefficient.

[0021] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the post-earthquake retaining wall displacement prediction model training method as described in the above embodiments.

[0022] The fourth aspect of this application provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed, are used to implement the post-earthquake retaining wall displacement prediction model training method as described in the above embodiments.

[0023] The fifth aspect of this application provides a computer program product, including: a computer program or instructions, which, when executed, implement the post-earthquake retaining wall displacement prediction model training method as described in the above embodiments.

[0024] Therefore, this application has at least the following beneficial effects:

[0025] This application's embodiments not only consider soil parameters and retaining wall geometric parameters but also incorporate seismic ground motion data to more comprehensively reflect the impact of earthquakes on the retaining wall. Through numerical simulation, a dataset containing retaining wall soil parameters, geometric parameters, seismic parameters, and corresponding wall displacements is constructed using the collected data. This dataset is then used to train a pre-built wall displacement prediction model, enabling the trained model to accurately output wall displacements. This significantly improves the accuracy and practicality of the prediction results while increasing computational efficiency. Therefore, it solves the problem of low accuracy in related technologies that only consider soil and geometric parameters for wall displacement prediction.

[0026] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0027] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0028] Figure 1 This is a flowchart of a method for training a post-earthquake retaining wall displacement prediction model according to an embodiment of this application;

[0029] Figure 2 This is an example diagram of training a post-earthquake retaining wall displacement prediction model according to an embodiment of this application;

[0030] Figure 3 This is a flowchart illustrating model training and optimization according to one embodiment of this application;

[0031] Figure 4 This is an example diagram of a training device for a post-earthquake retaining wall displacement prediction model provided in an embodiment of this application;

[0032] Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0033] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0034] The following description, with reference to the accompanying drawings, outlines a method, apparatus, electronic device, and storage medium for training a post-earthquake retaining wall displacement prediction model according to embodiments of this application. Addressing the problems mentioned in the background section, this application provides a method for training a post-earthquake retaining wall displacement prediction model. This method considers not only soil parameters and retaining wall geometric parameters but also incorporates seismic motion data to more comprehensively reflect the impact of earthquakes on the retaining wall. Through numerical simulation, a dataset containing retaining wall soil parameters, geometric parameters, seismic parameters, and corresponding wall displacements is constructed using the collected data. This dataset is then used to train a pre-constructed wall displacement prediction model, enabling the trained model to accurately output wall displacements. This significantly improves the accuracy and practicality of the prediction results while increasing computational efficiency. Therefore, this method solves the problem of low accuracy in related technologies that only consider soil and geometric parameters for wall displacement prediction.

[0035] Specifically, Figure 1 This is a flowchart illustrating a method for training a wall displacement prediction model for a retaining wall after an earthquake, as provided in an embodiment of this application.

[0036] like Figure 1 As shown, the training method for the wall displacement prediction model of the retaining wall after the earthquake includes the following steps:

[0037] In step S101, the soil parameters of the retaining wall, the geometric parameters of the retaining wall, and the ground motion data are obtained.

[0038] The soil parameters include: soil cohesion (c), with a data range of 0-30 kPa; and soil internal friction angle. The data range is 15-45°; the elastic modulus (E) ranges from 5-100 MPa; the Poisson's ratio (v) ranges from 0.25-0.5. The geometric parameter is the retaining wall height (H), which ranges from 0-10 m.

[0039] The seismic ground motion database obtained in this application provides a rich data source for slope stability analysis, through which seismic parameters can be extracted. Specifically, the selected seismic parameters include: magnitude (M), distance (R), peak ground acceleration (PGA), peak ground velocity (PGV), and Arias intensity (I). a ).

[0040] The peak ground acceleration (PGA) is calculated as follows: PGA = max|a(t)|;

[0041] a(t) is the time history of seismic acceleration;

[0042] Peak ground speed (PGV) calculation formula:

[0043] Arias intensity (I) a Calculation formula:

[0044] In step S102, numerical simulation is performed on the retaining wall soil parameters, retaining wall geometric parameters, and ground motion data to obtain the wall surface displacement of the retaining wall. The retaining wall soil parameters, retaining wall geometric parameters, ground motion data, and wall surface displacement are used to generate a dataset.

[0045] It is understandable that, such as Figure 2 As shown, this embodiment of the application can utilize the collected data to predict the wall displacement of a retaining wall under seismic loading. Numerical simulation is typically performed using specialized geotechnical engineering software. During the simulation, specific parameters of the retaining wall (soil parameters, geometric parameters) and seismic ground motion data are input to simulate the behavior of the retaining wall under seismic influence, particularly the changes in wall displacement. After the numerical simulation is completed, the wall displacement of the retaining wall under each specific combination of conditions is obtained. Then, these results are combined with the corresponding input parameters (i.e., soil parameters, geometric parameters, and seismic ground motion data) to form a dataset. This dataset contains all factors affecting the wall displacement of the retaining wall and their corresponding results.

[0046] In step S103, the dataset is divided into multiple training samples and ground truth labels for each training sample. The training samples include retaining wall soil parameters, retaining wall geometric parameters, and seismic ground motion data, while the ground truth labels are wall surface displacements.

[0047] In step S104, a pre-built wall displacement prediction model is trained using multiple training samples and the ground truth labels of each training sample. The input of the wall displacement prediction model is the retaining wall soil parameters, retaining wall geometric parameters, and ground motion data, and the output of the wall displacement prediction model is the wall displacement.

[0048] The wall displacement prediction model can be the lightGBM model. LightGBM, as an efficient gradient boosting framework, has advantages such as fast training speed, low memory consumption, and support for parallel computing; however, its specific applications in geotechnical engineering have not been fully explored. This application's embodiment predicts post-earthquake retaining wall displacement based on the lightGBM model, effectively solving the problems of large workload, low efficiency, high memory consumption, and generally low accuracy.

[0049] It is understood that, by using a large dataset containing multiple parameter combinations for training, the model can learn the complex nonlinear relationship between soil parameters, geometric parameters and ground motion data, thereby more accurately predicting wall displacement.

[0050] In one embodiment of this application, a pre-constructed wall displacement prediction model is trained using multiple training samples and the ground truth labels of each training sample, including: dividing the multiple training samples and the ground truth labels of each training sample into a training set and a validation set; iteratively training the wall displacement prediction model using the training set, and optimizing the hyperparameters of the wall displacement prediction model based on the output data and ground truth labels of the wall displacement prediction model during the training process; and validating the performance of the wall displacement prediction model using the validation set.

[0051] It is understandable that, such as Figure 3 As shown, in this embodiment of the application, multiple training samples and the true value of each training sample can be divided into a training set and a validation set, typically divided into a 70% training set and a 30% validation set. Let data = [x1, x2, ..., x...]. i The dataset is divided into two parts: 70% as the training set (I) and 30% as the validation set (T). The training set contains the majority of the data samples and is used to train the wall displacement prediction model. The validation set is used to evaluate the model's generalization ability, ensuring that the model performs well not only on the training data but also accurately predicts unseen data.

[0052] In one embodiment of this application, iterative training of a wall displacement prediction model using a training set includes: feature binding of data in the training set; sampling of training samples in the training set after feature binding based on gradients; generating a feature histogram based on the sampled training samples; iteratively training the wall displacement prediction model using the sampled training samples; calculating the training loss based on the output data and ground truth labels of the wall displacement prediction model during the training process; correcting the feature histogram during the training process; calculating the gain based on the corrected feature histogram; and optimizing the hyperparameters of the wall displacement prediction model based on the gain until the training loss reaches the training objective.

[0053] Specifically, during the training process, such as Figure 3 As shown, the data in the training set is first bundled, merging mutually exclusive features (features that do not simultaneously take non-zero values) into a single "super feature." This reduces feature dimensionality and memory usage, significantly reduces the number of features, and improves training efficiency, especially suitable for datasets containing a large number of sparse features (such as certain discrete variables in seismic motion parameters or soil parameters). In actual implementation, this embodiment first performs feature mutual exclusion detection: identifying mutually exclusive feature groups through sparse matrix analysis; secondly, it employs a dynamic bundling strategy: setting a maximum conflict threshold max_conflict_rate = 0.05, allowing the overlap rate of non-zero values ​​of bundled features to be ≤5%. Then, iteratively merging features based on a greedy algorithm prioritizes bundling the combination with the highest memory saving rate; finally, it bundles feature encoding, mapping the academician feature values ​​to the same dimensional space, for example, Bundled-Feature = x i ×K+xj , where K is the feature x j The upper limit of the value is used to ensure that the bundled features can be reversed to solve the original features.

[0054] Furthermore, in order to further improve training efficiency, such as Figure 3 As shown, this embodiment employs a gradient-based sampling strategy to select representative samples from the training set after feature binding. After sampling, a feature histogram is generated based on the sampled training samples and continuously revised as training progresses. The revised feature histogram better reflects the current model state, thus guiding the selection of split points in the next step. In each iteration, the wall displacement prediction model is trained using the sampled training samples, and the training loss is calculated based on the model's output data and ground truth labels. During training, the contribution of different features to model performance is evaluated by calculating the gain of split points, and the model's hyperparameters are optimized accordingly. The model is considered sufficiently fitted to the training data when the training loss falls below a preset threshold. The preset threshold can be specifically set.

[0055] In one embodiment of this application, generating a feature histogram based on sampled training samples includes: calculating the first-order gradient of each training sample; sorting the calculated first-order gradient values, taking the first preset number of training samples as high-gradient samples, taking a portion of the remaining training samples as low-gradient samples, and adding a weight compensation factor to the loss function of the low-gradient samples; and generating a feature histogram based on the high-gradient samples and the compensated low-gradient samples.

[0056] It is understood that the embodiments of this application can sample samples from the training set after feature binding, sort the samples based on the absolute value of the gradient, retain the head samples with larger gradients (high information content), and randomly select the tail samples with smaller gradients, which significantly reduces the amount of computation while maintaining accuracy. For example, the top 20% of high gradient samples can be retained, and 10% of low gradient samples can be randomly selected.

[0057] Specifically, in this embodiment, the first gradient of all samples (the negative gradient direction of the prediction error) can be calculated; secondly, the samples are screened, sorted in descending order of absolute gradient value, the top 20% of samples are retained, and 10% are randomly selected from the remaining samples; finally, the weights are corrected, and a weight compensation factor ω = (1-top_rate) / other_rate is applied to the loss function of low gradient samples to avoid sampling bias.

[0058] The gradient is calculated as follows: first-order gradient Second gradient Where L(y) i ,F(x i )) is the loss function, y iLet F(x) be the true value of the i-th sample. i Let be the model's predicted value for the i-th sample. The mean squared error (MSE) loss function is used. Then g i =F(x) i )-y i h i =1.

[0059] In one embodiment of this application, generating a feature histogram based on high-gradient samples and compensated low-gradient samples includes: for each feature split point, calculating the gradient and second derivative sum of the left and right child nodes of the split point respectively; during the calculation process, the original gradient is used to calculate the high-gradient samples, and the compensated gradient is used to calculate the high-gradient samples; generating a feature histogram based on the gradient and second derivative sum of the left and right child nodes of the split point.

[0060] It is understandable that the original gradient g can be directly used for high-gradient samples. i For low-gradient samples, use a scaled gradient ωg i For each feature split point, calculate the sum of the gradients and second derivatives of the left and right child nodes. Left subtree: ∑g left =∑ i∈left g i ,∑h left =∑ i∈left h i Right subtree: ∑g right =∑ i∈rihgt g i ,∑h right =∑ i∈right h i Parent node: ∑g parent =∑g left +∑g right ,∑h parent =∑h left +∑h right .

[0061] Specifically, for each candidate split point, the dataset is divided into two parts: data falling to the left of the split point are designated as left child nodes, and data falling to the right are designated as right child nodes. For samples belonging to the high gradient category and falling into either the left or right child node, their original first-order gradient g is used. i and second-order gradient h i This is used to accumulate the gradient sum of the corresponding child nodes. For low-gradient samples that have been sampled and weighted, their contribution to the first-order gradient becomes ωg. i And the contribution of the second-order gradient becomes ωh iSimilarly, these values ​​are added to the gradient sum of the corresponding child nodes. Based on the gradient sums and second derivative sums of the left and right child nodes of each split point calculated above, a feature histogram can be constructed. In this embodiment, the gain can be calculated based on the modified feature histogram. The split point with the largest gain is selected to generate a new leaf node, and this process continues until the loss function reaches the expected threshold. Where ∑g left ∑g represents the sum of the first-order gradients of all samples in the left child node. right ∑g represents the sum of the first-order gradients of all samples in the right child node. parent ∑h represents the sum of the first-order gradients of all samples of the parent node. left Let ∑h represent the sum of the second-order gradients of all samples in the left child node. right Let ∑h represent the sum of the second-order gradients of all samples in the right child node. parent λ represents the sum of the second-order gradients of all samples of the parent node, and λ represents the L2 regularization coefficient, which corresponds to the LightGBM model parameter lambda_l2.

[0062] Furthermore, embodiments of this application may select root mean square error and mean absolute error (MSE) as evaluation metrics, calculate the evaluation metrics using a validation set, and ensure the stability of the results through cross-validation. Priority is given to adjusting tree complexity (num_leaves, max_depth), learning rate (learning_rate), and number of iterations (n_estimators); grid search or Bayesian optimization is used to efficiently search for parameter combinations, balancing accuracy and computational cost.

[0063] In summary, after the wall displacement prediction model is trained, the embodiments of this application can extract the corresponding seismic parameters from an impending earthquake, and combine them with soil parameters and geometric parameters to form a dataset, which is then input into the model. The dataset Input = [x1, x2, ..., x...]. i ], where x i The following parameters are, in order: soil cohesion (c), soil internal friction angle (φ), elastic modulus (E), soil Poisson's ratio (ν), retaining wall height (H), magnitude (M), distance (R), peak ground acceleration (PGA), peak ground velocity (PGV), and Arias intensity (I). a The wall displacement prediction model outputs the predicted displacement (D).

[0064] The method for training a post-earthquake retaining wall displacement prediction model proposed in this application not only considers soil parameters and retaining wall geometric parameters but also incorporates seismic ground motion data to more comprehensively reflect the impact of earthquakes on retaining walls. Through numerical simulation, a dataset containing retaining wall soil parameters, geometric parameters, seismic parameters, and corresponding wall displacements is constructed using the collected data. This dataset is then used to train a pre-built wall displacement prediction model, enabling the trained model to accurately output wall displacements. This significantly improves the accuracy and practicality of the prediction results while increasing computational efficiency. Therefore, it solves the problem of low accuracy in related technologies that only consider soil and geometric parameters for wall displacement prediction.

[0065] Compared to traditional methods, this invention can predict retaining wall displacement with less post-processing work, lower memory consumption, higher computational efficiency, and better computational accuracy.

[0066] Next, referring to the accompanying drawings, a training device for predicting the wall displacement of a retaining wall after an earthquake, according to an embodiment of this application, is described.

[0067] Figure 4 This is a block diagram of a training device for predicting the wall displacement of a retaining wall after an earthquake, according to an embodiment of this application.

[0068] like Figure 4 As shown, the wall displacement prediction model training device 10 for the post-earthquake retaining wall includes: an acquisition module 100, a generation module 200, a division module 300, and a training module 400.

[0069] The system comprises the following modules: an acquisition module 100 for acquiring retaining wall soil parameters, retaining wall geometric parameters, and ground motion data; a generation module 200 for performing numerical simulations on the retaining wall soil parameters, retaining wall geometric parameters, and ground motion data to obtain the wall surface displacement, thereby generating a dataset; a partitioning module 300 for partitioning the dataset into multiple training samples and ground truth labels for each training sample, wherein the training samples include retaining wall soil parameters, retaining wall geometric parameters, and ground motion data, and the ground truth labels are wall surface displacements; and a training module 400 for training a pre-built wall surface displacement prediction model using the multiple training samples and the ground truth labels for each training sample, wherein the input to the wall surface displacement prediction model is the retaining wall soil parameters, retaining wall geometric parameters, and ground motion data, and the output of the wall surface displacement prediction model is the wall surface displacement.

[0070] In one embodiment of this application, the training module 400 is further configured to divide multiple training samples and the ground truth label of each training sample into a training set and a validation set; use the training set to iteratively train the wall displacement prediction model, and during the training process, optimize the hyperparameters of the wall displacement prediction model based on the output data and ground truth label of the wall displacement prediction model; and use the validation set to verify the performance of the wall displacement prediction model.

[0071] In one embodiment of this application, the training module 400 is further configured to perform feature binding on the data in the training set; sample training samples in the training set after feature binding is completed based on gradients; generate feature histograms based on the sampled training samples; iteratively train the wall displacement prediction model using the sampled training samples; calculate the training loss based on the output data and ground truth labels of the wall displacement prediction model during the training process; correct the feature histogram during the training process; calculate the gain based on the corrected feature histogram; and optimize the hyperparameters of the wall displacement prediction model based on the gain until the training loss reaches the training objective.

[0072] In one embodiment of this application, the training module 400 is further configured to calculate the first-order gradient of each training sample; sort the calculated first-order gradient values, take the first preset number of training samples as high-gradient samples, take a portion of the remaining training samples as low-gradient samples, and add a weight compensation factor to the loss function of the low-gradient samples; and generate a feature histogram based on the high-gradient samples and the compensated low-gradient samples.

[0073] In one embodiment of this application, the training module 400 is further configured to calculate the gradient and second derivative sum of the left and right child nodes of each feature split point. During the calculation process, the original gradient is used to calculate the gradient of high gradient samples and the compensated gradient is used to calculate the gradient of high gradient samples. A feature histogram is generated based on the gradient and second derivative sum of the left and right child nodes of the split point.

[0074] In one embodiment of this application, the formula for calculating the gain value is:

[0075]

[0076] Where, ∑g left ∑g represents the sum of the first-order gradients of all samples in the left child node. right ∑g represents the sum of the first-order gradients of all samples in the right child node. parent ∑h represents the sum of the first-order gradients of all samples of the parent node. left Let ∑h represent the sum of the second-order gradients of all samples in the left child node. right Let ∑h represent the sum of the second-order gradients of all samples in the right child node. parent λ represents the sum of the second-order gradients of all samples of the parent node, and λ represents the L2 regularization coefficient.

[0077] It should be noted that the explanation of the above-described embodiment of the method for training the prediction model of the wall displacement of the retaining wall after an earthquake also applies to the training device for the prediction model of the wall displacement of the retaining wall after an earthquake in this embodiment, and will not be repeated here.

[0078] The earthquake-induced retaining wall displacement prediction model training device proposed in this application not only considers soil parameters and retaining wall geometric parameters but also incorporates seismic motion data to more comprehensively reflect the impact of earthquakes on the retaining wall. Through numerical simulation, a dataset containing retaining wall soil parameters, geometric parameters, seismic parameters, and corresponding wall displacements is constructed using the collected data. This dataset is then used to train a pre-built wall displacement prediction model, enabling the trained model to accurately output wall displacements. This significantly improves the accuracy and practicality of the prediction results while increasing computational efficiency. Therefore, it solves the problem of low accuracy in related technologies that only consider soil and geometric parameters for wall displacement prediction.

[0079] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0080] The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0081] When the processor 502 executes the program, it implements the method for training the wall displacement prediction model of the retaining wall after the earthquake provided in the above embodiment.

[0082] Furthermore, electronic devices also include:

[0083] Communication interface 503 is used for communication between memory 501 and processor 502.

[0084] The memory 501 is used to store computer programs that can run on the processor 502.

[0085] The memory 501 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0086] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0087] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0088] Processor 502 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of this application.

[0089] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for training a post-earthquake retaining wall displacement prediction model.

[0090] This application also provides a computer program product, including: a computer program or instructions, which, when executed, implement the post-earthquake retaining wall displacement prediction model training method as described in the above embodiments.

[0091] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0093] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0094] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0095] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0096] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for training a prediction model of wall displacement of a retaining wall after an earthquake, characterized in that, Includes the following steps: Obtain soil parameters, geometric parameters, and ground motion data for the retaining wall; The wall surface displacement of the retaining wall is obtained by numerical simulation of the soil parameters, geometric parameters, and ground motion data of the retaining wall. The soil parameters, geometric parameters, ground motion data, and wall surface displacement of the retaining wall are used to generate a dataset. The dataset is divided into multiple training samples and ground truth labels for each training sample. The training samples include the retaining wall soil parameters, the retaining wall geometric parameters, and the seismic motion data. The ground truth labels are the wall surface displacements. A pre-constructed wall displacement prediction model is trained using the multiple training samples and the ground truth labels of each training sample. The inputs of the wall displacement prediction model are the retaining wall soil parameters, the retaining wall geometric parameters, and the ground motion data. The output of the wall displacement prediction model is the wall displacement.

2. The method for training a post-earthquake retaining wall displacement prediction model according to claim 1, characterized in that, The process of training a pre-built wall displacement prediction model using the multiple training samples and the ground truth label of each training sample includes: The multiple training samples and the ground truth labels of each training sample are divided into a training set and a validation set; The wall displacement prediction model is iteratively trained using the training set. During the training process, the hyperparameters of the wall displacement prediction model are optimized based on the output data of the wall displacement prediction model and the ground truth labels. The performance of the wall displacement prediction model is verified using the validation set.

3. The method for training a post-earthquake retaining wall displacement prediction model according to claim 2, characterized in that, The iterative training of the wall displacement prediction model using the training set includes: Feature bundling is performed on the data in the training set; Training samples are sampled from the training set after feature binding is completed based on gradients; Feature histograms are generated based on the sampled training samples. The wall displacement prediction model is iteratively trained using the sampled training samples. During the training process, the training loss is calculated based on the output data of the wall displacement prediction model and the ground truth labels. During training, the feature histogram is corrected, the gain is calculated based on the corrected feature histogram, and the hyperparameters of the wall displacement prediction model are optimized based on the gain until the training loss reaches the training objective.

4. The method for training a post-earthquake retaining wall displacement prediction model according to claim 3, characterized in that, The generation of feature histograms based on sampled training samples includes: Calculate the first-order gradient for each training sample; The calculated first-order gradient values ​​are sorted, the first preset number of training samples are taken as high-gradient samples, and a portion of the remaining training samples are taken as low-gradient samples. A weight compensation factor is added to the loss function of the low-gradient samples. The feature histogram is generated based on the high-gradient samples and the compensated low-gradient samples.

5. The method for training a post-earthquake retaining wall displacement prediction model according to claim 4, characterized in that, The step of generating the feature histogram based on the high-gradient samples and the compensated low-gradient samples includes: For each feature split point, the gradient and second derivative of the left and right child nodes of the split point are calculated respectively. During the calculation, the high gradient sample is calculated using the original gradient and the high gradient sample is calculated using the compensated gradient. The feature histogram is generated based on the gradients and second derivatives of the left and right child nodes of the split point.

6. The method for training a post-earthquake retaining wall displacement prediction model according to claim 3, characterized in that, The formula for calculating the gain value is: Where, ∑g left ∑g represents the sum of the first-order gradients of all samples in the left child node. right ∑g represents the sum of the first-order gradients of all samples in the right child node; parent ∑h represents the sum of the first-order gradients of all samples of the parent node. left Let ∑h represent the sum of the second-order gradients of all samples in the left child node. right Let ∑h represent the sum of the second-order gradients of all samples in the right child node. parent λ represents the sum of the second-order gradients of all samples of the parent node, and λ represents the L2 regularization coefficient.

7. A training device for predicting wall displacement of a retaining wall after an earthquake, characterized in that, include: The acquisition module is used to acquire retaining wall soil parameters, retaining wall geometric parameters, and seismic motion data; The generation module is used to perform numerical simulation on the retaining wall soil parameters, the retaining wall geometric parameters and the ground motion data to obtain the wall surface displacement of the retaining wall, and the retaining wall soil parameters, the retaining wall geometric parameters, the ground motion data and the wall surface displacement generate a dataset; A partitioning module is used to partition the dataset into multiple training samples and ground truth labels for each training sample, wherein the training samples include the retaining wall soil parameters, the retaining wall geometric parameters, and the seismic motion data, and the ground truth label is the wall surface displacement; The training module is used to train a pre-built wall displacement prediction model using the multiple training samples and the ground truth labels of each training sample. The input of the wall displacement prediction model is the retaining wall soil parameters, the retaining wall geometric parameters, and the ground motion data. The output of the wall displacement prediction model is the wall displacement.

8. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for training a wall displacement prediction model for a retaining wall after an earthquake as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they implement the post-earthquake retaining wall displacement prediction model training method as described in any one of claims 1-6.

10. A computer program product comprising: A computer program or instruction, characterized in that, when executed, the computer program or instruction implements the post-earthquake retaining wall displacement prediction model training method as described in any one of claims 1-6.

Images