A deep learning-based optimization design method for slope pile-anchor support structures

Abstract

This invention discloses a deep learning-based optimization design method for slope pile-anchor support structures, comprising: acquiring parameters of different slope pile-anchor support structures and dynamic response data under seismic conditions; preprocessing the data, dividing it into training and test sets, and reconstructing time-series data; constructing a prediction model based on a bidirectional long short-term memory network, and optimizing the hyperparameters of the prediction model by combining it with an improved virus colony search; training the model with data from the training set to obtain the optimal prediction model, and using the prediction model to predict the dynamic response results of the pile-anchor support structure, and optimizing and adjusting the pile-anchor support structure design scheme based on the results. This method significantly shortens the design time and cost, and provides a more detailed understanding of the effects of different seismic loads on the pile-anchor support structure, improving the reliability of the design results, and has good socio-economic benefits and engineering application prospects.

Description

Technical Field

[0001] This invention relates to the field of slope support structure design technology, specifically to a deep learning-based optimization design method for slope pile-anchor support structures. Background Technology

[0002] Compared with traditional single anti-slide piles or anchored structures, pile-anchored support structures have superior seismic performance and are widely used. However, due to their structural complexity, their dynamic response under seismic loads is more nonlinear than that of traditional single support structures, which increases the design difficulty.

[0003] In current seismic design of slope engineering, time history analysis is a crucial method for dynamic response analysis. However, due to the inherent complexity of the problem, traditional time history analysis methods, employing mathematical and physical approaches, struggle to accurately describe the strongly nonlinear relationship between seismic loads and dynamic responses. Furthermore, the complex and variable nature of seismic loads makes large-scale, multi-condition centrifugal shaking table model tests for single projects time-consuming, expensive, and difficult to justify within design budgets. Therefore, utilizing previous centrifugal shaking table model test data and existing slope support structure monitoring data to obtain the dynamic response characteristics of pile-anchor supported structures under different seismic loads is key to optimizing structural design, reducing design costs, and shortening design time.

[0004] In recent years, artificial intelligence (AI) methods have made significant progress. By utilizing sophisticated algorithms and massive datasets, AI methods have demonstrated substantial advantages in modeling complex nonlinear relationships and adaptively handling the dynamic interactions of uncertain systems. Deep learning, in particular, has shown remarkable advantages in processing complex time-series analysis data. Introducing deep learning technology into the optimization design of slope pile-anchor support structures allows for the full utilization of field monitoring data and historical engineering data to establish a nonlinear mapping relationship between slope response and support parameters. This overcomes the limitations of traditional methods and improves the scientific rigor and accuracy of the design. It enables efficient evaluation and optimization of support structure design schemes under different geological conditions, ultimately determining the optimal support structure design scheme and providing a scientific, precise, and efficient design tool for slope support engineering. Summary of the Invention

[0005] The purpose of this invention is to provide a deep learning-based optimization design method for slope pile-anchor support structures. This method learns from previous centrifugal shaking table model test data of pile-anchor support structures and monitoring data of slope support structures to predict the dynamic response results of pile end displacement, anchor bolt tension, etc. of the pile-anchor support structure under different seismic loads.

[0006] To achieve the above functions, this invention designs a deep learning-based optimization design method for slope pile-anchor support structures. For the slope soil and pile-anchor support structure, the following steps S1-S5 are performed to complete the optimization of the slope pile-anchor support structure:

[0007] Step S1: Obtain dynamic response data of different slope pile-anchor support structures under seismic motion through centrifugal shaking table tests and on-site monitoring of slope pile-anchor support structures, as well as the corresponding slope soil strength parameters and geometric parameters, and pile-anchor support structure design parameters; construct a sample database of slopes and pile-anchor support structures.

[0008] Step S2: Preprocess the dynamic response data, including using discrete wavelet transform to reduce noise and normalize the dynamic response data; divide the sample database of slope and pile anchor support structure into 80% training set and 20% test set; reconstruct the dynamic response data using the step reconstruction method;

[0009] Step S3: Construct and train a deep learning prediction model based on optimization algorithms to optimize hyperparameters;

[0010] Step S4: Use ground motion data, slope soil strength parameters and geometric parameters, and pile anchor support structure design parameters as input data for the deep learning prediction model, and use the preprocessed dynamic response data as output data for the deep learning prediction model.

[0011] By dividing the training set into 80% and the test set into 20% and then using 5-fold cross-validation, the evaluation index of each validation is calculated and obtained to determine the optimal deep learning prediction model.

[0012] Step S5: Based on the optimal deep learning prediction model, input the design parameters of the pile-anchor support structure, as well as the slope soil strength parameters and geometric parameters. According to the seismic protection design objectives, input the ground motion data to predict the dynamic response results of the pile-anchor support structure under ground motion. Based on the dynamic response results output by the deep learning model, optimize and adjust the design scheme of the pile-anchor support structure.

[0013] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0014] This invention presents a deep learning-based optimization design method for slope pile-anchor support structures. This method provides an assessment of the parameter design of pile-anchor support structures and optimizes and adjusts the design parameters based on the dynamic response results. Compared to traditional centrifuge model tests for design result verification, this method significantly shortens the design time and reduces design costs. Furthermore, it allows for a more detailed understanding of the effects of different seismic loads on the pile-anchor support structure. This improves the reliability of the design results and has significant socio-economic benefits and promising engineering application prospects. Attached Figure Description

[0015] Figure 1 This is a flowchart of a deep learning-based optimization design method for slope pile-anchor support structures, provided by an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of the step-reconstruction method provided in an embodiment of the present invention;

[0017] Figure 3 This is a structural diagram of a deep learning prediction model provided according to an embodiment of the present invention;

[0018] Figure 4 This is a schematic diagram illustrating the training convergence of a deep learning prediction model according to an embodiment of the present invention;

[0019] Figure 5 This is a diagram showing the results of 5-fold cross-validation of a deep learning prediction model provided in an embodiment of the present invention;

[0020] Figure 6 This is a test model diagram of the pile-anchor support model provided in the embodiments of the present invention;

[0021] Figure 7 This is a diagram showing the predicted bending moment of the pile body using a deep learning prediction model provided in an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0023] This invention provides a deep learning-based optimization design method for slope pile-anchor support structures, targeting both the slope soil and the pile-anchor support structure, referring to... Figure 1 Perform the following steps S1-S5 to complete the optimization of the slope pile anchor support structure:

[0024] Step S1: Obtain dynamic response data of different slope pile-anchor support structures under seismic motion through centrifugal shaking table tests and on-site monitoring of slope pile-anchor support structures, as well as the corresponding slope soil strength parameters and geometric parameters, and pile-anchor support structure design parameters; construct a sample database of slopes and pile-anchor support structures.

[0025] The slope soil strength parameters include soil elastic modulus E, Poisson's ratio μ, unit weight γ, cohesion c, and internal friction angle φ; slope geometric parameters include slope height and slope angle; the dynamic response data of the slope pile-anchor support structure includes anchor angle and diameter, pile length, width, and thickness, coordinates of the pile-anchor connection point, and pile elastic modulus E. p The elastic modulus E of the anchor bolt a The time t, the anchor bolt tension T, and the pile top displacement D.

[0026] Step S2: Preprocess the dynamic response data, including using discrete wavelet transform to reduce noise and normalize the dynamic response data; divide the sample database of slope and pile anchor support structure into 80% training set and 20% test set; reconstruct the dynamic response data using the step reconstruction method;

[0027] The discrete wavelet transform method described above reduces high-frequency components in dynamic response data, thereby highlighting low-frequency components. This helps to smooth signals and improve the ability of deep learning prediction models to identify and analyze long-term changes or periodic trends in dynamic response data. The outputs of the high-frequency and low-frequency components of the discrete wavelet transform method are as follows:

[0028]

[0029] x = x hp +x lp

[0030] In the formula, X represents the dynamic response input data of the original pile-anchor support structure, x hp and x lp The outputs are for high-frequency and low-frequency components, respectively. and These are the filtering parameters that adjust the ratio of high-frequency and low-frequency components, respectively, and x is the output data after noise reduction.

[0031] The normalization mentioned is a normalization within the range (-1, 1), as shown in the following formula:

[0032]

[0033] Where, x n It is the normalized data; x min and x max These are the minimum and maximum values ​​of the entire feature variable in the dataset, respectively; y min and y max These are the minimum and maximum values ​​in each data set, respectively.

[0034] Reference Figure 2 The aforementioned step-by-step reconstruction method selects 10% of the data set as the step size and 1 data point as the step distance, thereby reconstructing the dynamic response data of the slope pile anchor support structure in the form of a 2×1000 time series into a 901×2×100 reconstructed data set; that is, expanding the 1000-length time series data into 901 sets of sequences, each set of sequences having a length of 100.

[0035] Step S3: Construct and train a deep learning prediction model based on optimization algorithms to optimize hyperparameters;

[0036] In step S3, a deep learning prediction model is constructed based on a bidirectional long short-term memory network. The structure diagram is shown below. Figure 3 The convergence process of the training model is as follows Figure 4 As shown in Table 1, the parameter values ​​for the bidirectional long short-term memory network are as follows:

[0037] Table 1. Parameter values ​​of bidirectional long short-term memory network

[0038]

[0039] In one embodiment, the optimization algorithm upon which the deep learning prediction model is based employs an improved Virus Colony Search (IVCS) algorithm. This improvement targets the immune response phase of the VCS algorithm, thereby avoiding its limitations and enhancing its solution capability. Specifically, the improvement introduces a stochastic process based on the mutation principle. This process defines a mutation rate mr ranging from 0 to 1 for a solution vector X of dimension n in the VCS algorithm. For each element x in the solution vector X… i Generate a random number r in the range [0, 1]; if r < mr, then for x i Perform a mutation operation; the mutation operation is to move x i Add a random perturbation, which is generated by a normal distribution with a mean of 0 and a set standard deviation ms (the intensity of the mutation), where mr and ms are defined as 0.1 and 0.01, respectively.

[0040] The optimization algorithm in step S3 can be flexibly selected. Taking the dung beetle optimization algorithm as an example, the parameter settings of the optimization algorithm are shown in Table 2 below:

[0041] Table 2. Parameter settings for the dung beetle optimization algorithm.

[0042]

[0043] To avoid unnecessary calculations, a stopping criterion is defined: the root mean square error (RMSE) of the objective function result fluctuates by less than 0.01 over 10 consecutive iterations. The value of the stopping criterion should be adjusted appropriately for different computational scenarios. The RMSE is calculated as follows:

[0044]

[0045] In the formula, n represents the amount of data; y i and These are the predicted and actual values ​​of the dynamic response of the i-th pile-anchor support structure, respectively.

[0046] During the training process of a deep learning model based on optimization algorithms, the hyperparameters of the deep learning model are adjusted according to the loss function. These hyperparameters include the number of hidden layer neurons (n). h), learning rate (LR), learning rate decrease factor, and regularization coefficient (L2).

[0047] Step S4: Use ground motion data, slope soil strength parameters and geometric parameters, and pile anchor support structure design parameters as input data for the deep learning prediction model, and use the preprocessed dynamic response data as output data for the deep learning prediction model.

[0048] By dividing the training set into 80% and the test set into 20% and then using 5-fold cross-validation, the evaluation index of each validation is calculated and obtained to determine the optimal deep learning prediction model.

[0049] Five-fold cross-validation is used to evaluate the impact of different data partitions on model training, reflecting the model's stability and determining the optimal deep learning prediction model through fluctuations in multiple evaluation metrics; the evaluation metrics include regression coefficients (R²). 2 ), Root Mean Square Error (RMSE), Absolute Error (MAE), and Mean Squared Error (MSE). See the graph for the 5-fold cross-validation results of the deep learning prediction model. Figure 5 .

[0050] Step S5: Based on the optimal deep learning prediction model, input the design parameters of the pile-anchor support structure, as well as the slope soil strength parameters and geometric parameters. According to the seismic protection design objectives, input the ground motion data to predict the dynamic response results of the pile-anchor support structure under ground motion. Based on the dynamic response results output by the deep learning model, optimize and adjust the design scheme of the pile-anchor support structure.

[0051] This invention also provides a deep learning-based optimization design system for slope pile-anchor support structures, comprising:

[0052] One or more processors;

[0053] The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, specifically for the slope soil and pile-anchor support structure, executing the following steps S1-S5 to complete the optimization of the slope pile-anchor support structure:

[0054] Step S1: Obtain dynamic response data of different slope pile-anchor support structures under seismic motion through centrifugal shaking table tests and on-site monitoring of slope pile-anchor support structures, as well as the corresponding slope soil strength parameters and geometric parameters, and pile-anchor support structure design parameters; construct a sample database of slopes and pile-anchor support structures.

[0055] Step S2: Preprocess the dynamic response data, including using discrete wavelet transform to reduce noise and normalize the dynamic response data; divide the sample database of slope and pile anchor support structure into 80% training set and 20% test set; reconstruct the dynamic response data using the step reconstruction method;

[0056] Step S3: Construct and train a deep learning prediction model based on optimization algorithms to optimize hyperparameters;

[0057] Step S4: Use ground motion data, slope soil strength parameters and geometric parameters, and pile anchor support structure design parameters as input data for the deep learning prediction model, and use the preprocessed dynamic response data as output data for the deep learning prediction model.

[0058] By dividing the training set into 80% and the test set into 20% and then using 5-fold cross-validation, the evaluation index of each validation is calculated and obtained to determine the optimal deep learning prediction model.

[0059] Step S5: Model diagram of pile-anchor support test. Figure 6 As shown, based on the optimal deep learning prediction model, the design parameters of the pile-anchor support structure, as well as the slope soil strength and geometric parameters, are input. According to the seismic protection design objectives, seismic ground motion data is input to predict the dynamic response of the pile-anchor support structure under seismic ground motion. Based on the dynamic response results output by the deep learning model, the design scheme of the pile-anchor support structure is optimized and adjusted. Taking the pile bending moment as an example, the input seismic ground motion and dynamic response prediction results are as follows... Figure 7 As shown.

[0060] This invention also provides a computer-readable medium for storing software, the readable medium including instructions executable by one or more computers, the instructions, when executed by the one or more computers, performing the operation of the deep learning-based slope pile-anchor support structure optimization design method.

[0061] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A deep learning-based optimization design method for slope pile-anchor support structures, characterized in that, For the slope soil and pile-anchor support structure, perform the following steps S1-S5 to complete the optimization of the slope pile-anchor support structure: Step S1: Obtain dynamic response data of different slope pile-anchor support structures under seismic motion through centrifugal shaking table tests and on-site monitoring of slope pile-anchor support structures, as well as the corresponding slope soil strength parameters and geometric parameters, and pile-anchor support structure design parameters; construct a sample database of slopes and pile-anchor support structures. Step S2: Preprocess the dynamic response data, including using discrete wavelet transform to reduce noise and normalize the dynamic response data; divide the sample database of slope and pile anchor support structure into 80% training set and 20% test set; reconstruct the dynamic response data using the step reconstruction method; Step S3: Construct and train a deep learning prediction model based on optimization algorithms to optimize hyperparameters; A deep learning prediction model is constructed based on a bidirectional long short-term memory network. The optimization algorithm underlying the deep learning prediction model adopts an improved virus colony search algorithm, which is improved in the immune response stage of the virus colony search algorithm. The specific improvement method is to introduce... A stochastic process based on the principle of mutation is described, specifically, a mutation rate mr ranging from 0 to 1 is defined for a solution vector X of dimension n in a virus colony search algorithm; for each element x in the solution vector X... i Generate a random number r in the range [0, 1]; if r < mr, then for x i Perform a mutation operation; the mutation operation is to move x i Add a random perturbation, which is generated by a normal distribution with a mean of 0 and a set standard deviation ms, where mr and ms are defined as 0.1 and 0.01, respectively; Step S4: Use ground motion data, slope soil strength parameters and geometric parameters, and pile anchor support structure design parameters as input data for the deep learning prediction model, and use the preprocessed dynamic response data as output data for the deep learning prediction model. By dividing the training set into 80% and the test set into 20% and then using 5-fold cross-validation, the evaluation index of each validation is calculated and obtained to determine the optimal deep learning prediction model. Step S5: Based on the optimal deep learning prediction model, input the design parameters of the pile-anchor support structure, as well as the slope soil strength parameters and geometric parameters, and according to the seismic protection design objectives, input the ground motion data to predict the dynamic response results of the pile-anchor support structure under ground motion. Based on the dynamic response results output by the deep learning model, the design scheme of the pile-anchor support structure is optimized and adjusted.

2. The method for optimizing the design of slope pile-anchor support structure based on deep learning according to claim 1, characterized in that, The slope soil strength parameters include soil elastic modulus E, Poisson's ratio μ, unit weight γ, cohesion c, and internal friction angle φ; slope geometric parameters include slope height and slope angle; the dynamic response data of the slope pile-anchor support structure includes anchor angle and diameter, pile length, width, and thickness, coordinates of the pile-anchor connection point, and pile elastic modulus E. p The elastic modulus E of the anchor bolt a The time t, the anchor bolt tension T, and the pile top displacement D.

3. The method for optimizing the design of slope pile-anchor support structure based on deep learning according to claim 1, characterized in that, The discrete wavelet transform method described in step S2 outputs high-frequency and low-frequency components as follows: x=x hp +x lp In the formula, X represents the dynamic response input data of the original pile-anchor support structure, x hp and x lp The outputs are for high-frequency and low-frequency components, respectively. and These are the filtering parameters that adjust the ratio of high-frequency and low-frequency components, respectively, and x is the output data after noise reduction.

4. The method for optimizing the design of slope pile-anchor support structure based on deep learning according to claim 1, characterized in that, The normalization mentioned in step S2 is a normalization within the range of (-1, 1), as shown in the following formula: Where, x n It is the normalized data; x min and x max These are the minimum and maximum values ​​of the entire feature variable in the dataset, respectively; y min and y max These are the minimum and maximum values ​​in each data set, respectively.

5. The method for optimizing the design of slope pile-anchor support structure based on deep learning according to claim 1, characterized in that, The step-by-step reconstruction method described in step S2 selects 10% of the data set as the step size and 1 data point as the step distance, thereby reconstructing the dynamic response data of the slope pile anchor support structure in the form of 2×1000 time series into a reconstructed data set of 901×2×100.

6. The method for optimizing the design of slope pile-anchor support structure based on deep learning according to claim 1, characterized in that, In step S3, during the training of the deep learning model based on the optimization algorithm, the hyperparameters of the deep learning model are adjusted according to the loss function. The hyperparameters include the number of hidden layer neurons, the learning rate, the learning rate reduction factor, and the regularization coefficient.

7. The method for optimizing the design of slope pile-anchor support structure based on deep learning according to claim 1, characterized in that, The evaluation indicators mentioned in step S4 include regression coefficient, root mean square error, absolute error, and mean square error.

8. A deep learning-based optimization design system for slope pile-anchor support structures, characterized in that, include: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, specifically for the slope soil and pile-anchor support structure, executing the following steps S1-S5 to complete the optimization of the slope pile-anchor support structure: Step S1: Obtain dynamic response data of different slope pile-anchor support structures under seismic motion through centrifugal shaking table tests and on-site monitoring of slope pile-anchor support structures, as well as the corresponding slope soil strength parameters and geometric parameters, and pile-anchor support structure design parameters; construct a sample database of slopes and pile-anchor support structures. Step S2: Preprocess the dynamic response data, including using discrete wavelet transform to reduce noise and normalize the dynamic response data; divide the sample database of slope and pile anchor support structure into 80% training set and 20% test set; reconstruct the dynamic response data using the step reconstruction method; Step S3: Construct and train a deep learning prediction model based on optimization algorithms to optimize hyperparameters; A deep learning prediction model is constructed based on a bidirectional long short-term memory network. The optimization algorithm underlying the deep learning prediction model adopts an improved virus colony search algorithm, which is improved in the immune response stage of the virus colony search algorithm. The specific improvement method is to introduce... A stochastic process based on the principle of mutation is described, specifically, a mutation rate mr ranging from 0 to 1 is defined for a solution vector X of dimension n in a virus colony search algorithm; for each element x in the solution vector X... i Generate a random number r in the range [0, 1]; if r < mr, then for x i Perform a mutation operation; the mutation operation is to move x i Add a random perturbation, which is generated by a normal distribution with a mean of 0 and a set standard deviation ms, where mr and ms are defined as 0.1 and 0.01, respectively; Step S4: Use ground motion data, slope soil strength parameters and geometric parameters, and pile anchor support structure design parameters as input data for the deep learning prediction model, and use the preprocessed dynamic response data as output data for the deep learning prediction model. By dividing the training set into 80% and the test set into 20% and then using 5-fold cross-validation, the evaluation index of each validation is calculated and obtained to determine the optimal deep learning prediction model. Step S5: Based on the optimal deep learning prediction model, input the design parameters of the pile-anchor support structure, as well as the slope soil strength parameters and geometric parameters, and according to the seismic protection design objectives, input the ground motion data to predict the dynamic response results of the pile-anchor support structure under ground motion. Based on the dynamic response results output by the deep learning model, the design scheme of the pile-anchor support structure is optimized and adjusted.

9. A computer-readable medium for storing software, characterized in that, The readable medium includes instructions executable by one or more computers, which, when executed by the one or more computers, perform the operation of a deep learning-based slope pile-anchor support structure optimization design method as described in any one of claims 1-7.

Images