Dynamic compaction construction parameter optimization method based on XGBoost-PSO

Abstract

The invention relates to the technical field of foundation treatment, in particular to a dynamic compaction construction parameter optimization method based on XGBoost-PSO, and the method comprises the steps: obtaining a plurality of groups of compaction settlement data through a compaction test; the method comprises the following steps: selecting original features from actual measurement parameters of a test, constructing interactive features by adopting the actual measurement parameters, selecting optimal interactive features according to the importance ranking of XGboost features, and analyzing the correlation between the features; based on the feature set obtained through screening, a compaction settlement prediction model is established; and performing dynamic compaction parameter optimization on the compaction settlement prediction model by using a particle swarm algorithm to finally obtain an optimal compaction parameter combination. According to the method, the tamping parameters are automatically searched and optimized on the premise that the tamping stopping standard is met by building the tamping settlement prediction model, and therefore the purposes of guaranteeing the construction quality, improving the construction efficiency and reducing the construction cost are achieved; the method is suitable for various foundation construction design stages needing dynamic compaction treatment, and has good engineering adaptability and popularization value.

Description

Technical Field

[0001] The present invention relates to the technical field of ground treatment, and particularly relates to a method for optimizing dynamic compaction construction parameters based on XGBoost-PSO. Background Art

[0002] In dynamic compaction construction, the settlement amount is an important index of the foundation response and is generally considered to be able to directly reflect the reinforcement effect of the soil mass. In actual engineering design, the settlement amounts of the last two blows are often used as the judgment criterion for stopping compaction. Some other studies dynamically evaluate the reinforcement degree and saturation state of the soil mass by monitoring the change trend of the settlement amount. Therefore, the settlement amount is not only an important parameter for construction process control but also an important basis for parameter optimization and compaction determination. In current engineering practice, determining the combination of dynamic compaction construction parameters (such as the number of blows, the weight of the rammer, the lifting height, the diameter of the rammer, etc.) often relies on the on-site trial compaction method, that is, the construction parameters of the entire site are deduced from the results of small-scale trial compaction. Although this method has engineering practicality, it has the disadvantages of relying on manual and empirical judgment, being unable to cover the combined design space of multiple parameters and multiple working conditions, and having a high cost and low efficiency in the trial compaction process, making it difficult to carry out a large amount of work in the early design stage.

[0003] With the rise of artificial intelligence and machine learning, more and more studies have applied data-driven methods to the fields of engineering prediction and parameter optimization. Among them, some studies have tried to predict the dynamic compaction settlement amount or the reinforcement depth. For example: Zhang Yuchuan et al. constructed a settlement amount prediction model based on the BP neural network, using variables such as the tamping energy per unit area, the number of blows, and the water content; Xu Yongbing et al. compared the performance of models such as BP, SVM, RF, and XGBoost in predicting the effective reinforcement depth and considered that RF and XGBoost are more suitable for dynamic compaction data modeling under small sample conditions. However, limited by factors such as the difficulty of obtaining dynamic compaction engineering data and the high cost of experiments, there is currently a lack of high-quality and diverse dynamic compaction parameter-response datasets, as well as a systematic solution that combines "prediction model + parameter optimization".

[0004] In terms of settlement prediction, the performance of the model highly depends on the quality of feature engineering. Although existing studies have selected input variables based on engineering experience, there is a lack of a systematic method for mining parameter interaction terms. In addition, due to many interference factors on-site and strong data inconsistency, it also limits the further improvement of the model performance. For this reason, Bian Haiding et al. used indoor reduced-scale model tests to carry out multiple groups of repeatable indoor model tests by controlling variables such as the rammer mass, the drop height, and the water content, providing a controllable means for exploring the influencing factors of settlement.

[0005] In terms of parameter optimization, the Particle Swarm Optimization (PSO) algorithm has been widely applied to various engineering optimization tasks due to its natural support for real - number coding, few parameters, and fast convergence speed. Existing research has attempted to combine it with prediction models. For example, Hu Changming et al. optimized the surface settlement and propulsion parameters during shield tunneling construction based on LSTM - PSO.

[0006] Generally speaking, there are still the following deficiencies in the field of dynamic compaction construction design: The determination of construction parameters still mainly relies on trial compaction and experience, lacking a parameter - response model that can be popularized and predicted; existing models mostly focus on prediction and are not coupled with optimization algorithms, making it difficult to achieve automatic optimal design; the data source mainly depends on on - site collection, which is greatly affected by the outside world, and it is difficult to build a systematic database. Summary of the Invention

[0007] The purpose of the present invention is to provide a dynamic compaction parameter optimization method that integrates a scaled - down model test, a machine - learning regression model (XGBoost), and a particle swarm optimization algorithm (PSO). The present invention obtains systematic data through experiments, constructs a high - precision settlement prediction model, and combines the optimization algorithm to search for the optimal parameter combination on the premise of meeting the compaction stop standard, so as to realize the automatic design of construction parameters and improve the scientific nature of construction efficiency and foundation quality control.

[0008] To achieve the above - mentioned purpose, the present invention provides a dynamic compaction construction parameter optimization method based on XGBoost - PSO, including the following steps: S1. Dynamic compaction model test: Design and fabricate a dynamic compaction model, and conduct compaction tests using test rammers of various different specifications and the dynamic compaction model to obtain settlement data under several different parameter combinations. S2. Feature selection: First, select original features from the actual measured parameters of the dynamic compaction model test; then construct interaction features using the actual measured parameters, and select the optimal interaction features according to the XGboost feature importance ranking; then conduct a correlation analysis on the original features and the optimal interaction features. S3. Based on the feature set screened in step S2, establish an XGBoost settlement prediction model. S4. Use the particle swarm algorithm to optimize the dynamic compaction parameters of the XGBoost settlement prediction model, and finally obtain the optimal compaction parameter combination.

[0009] Furthermore, step S1 specifically includes: S1.1. Determine the similarity criteria and scale ratio: According to the three types of similarity criteria of geometric similarity, physical similarity, and mechanical similarity, and combined with the actual engineering background, select the model control variables, and determine the similarity coefficient system required for the tamping test through the second similarity theorem and dimensional analysis method; the model control variables include the ram mass M, the ram lifting height H, the ram diameter D, the number of tamping blows N, the tamping energy E, the tamping settlement s, the water content w, and the dry density d of the soil. S1.2. Fabricate the dynamic compaction model box, configure a triangular support for the dynamic compaction model box, and install an electromagnetic adsorption device at the top of the triangular support to simulate the free fall behavior of the heavy object during the dynamic compaction process. S1.3. Configure test rams of different specifications; some of the ram bodies are perforated on the surface to reduce air resistance and be close to the actual engineering conditions. S1.4. Arrange the measuring devices and perform the tamping operation: Design different tamping parameters for multiple groups of tamping tests. For each group of tests, set the number of tamping blows to 8 - 18 times. After each tamping blow, read the tamping settlement data and record it in the database.

[0010] Further, in step S1.4, after each group of tamping is completed, the soil is re - disturbed and loosened and leveled to maintain the consistency of the initial state of the soil.

[0011] Further, step S2 specifically includes: S2.1. Selection of original features: Based on the parameters actually measured in the model test data, select the ram mass, lifting height, ram diameter, and number of tamping blows as the original feature variables; use the grey relational analysis method to calculate the correlation degree between each original feature variable and the target variable, and regard the original feature variables with a correlation degree greater than the correlation degree threshold as the key influencing features and include them in the subsequent modeling; among them, the correlation degree threshold is 0.6. S2.2. Selection of optimal interaction features: The specific steps are as follows: (1). Determine the combination form of interaction features: Select the ram mass M and the lifting height H as the basic variables for constructing interaction features, and set three types of combination forms: MH a , M b H and M c H d ; where a, b, c, and d are power parameters used to adjust the non - linear form. (2). Set the parameter search range and step size: To ensure the search accuracy and calculation efficiency, set a, b, c, d ∈ [0.1, 2, 0.1]. (3) Construct interaction features and perform K-fold cross-validation: Use K-fold cross-validation to train the XGBoost model for the interaction features of three types of combination forms, and count the feature importance indicators of each interaction feature in each round of training. Obtain the stable importance of each interaction feature by using the weighted average method; where K = 5; (4) Calculate the total feature importance and select the optimal power combination: Add the average importance values of the MH a , M b H and M c H d three interaction features, and define it as the total importance of the current combination. After traversing all power combinations, select the power combination with the largest total importance as the final interaction feature construction strategy; (5) Construct the optimal interaction features and incorporate them into model training: According to the evaluation results, formally incorporate the interaction features corresponding to the optimal power parameters into the model input feature set, and jointly participate in the XGBoost model training with the original features; S2.3 Combine the original features selected in step S2.1 and the optimal interaction features selected in step S2.2 as variables for Spearman calculation and analysis. If the correlation coefficient ρ between variables > 0.8, it is considered that the collinearity is too high and feature elimination is performed; Subsequently, use the variance inflation factor method to calculate vif to ensure that the vif value of the remaining features < 10.

[0012] Further, step S3 specifically includes: Step S3.1 Data division and standardization: Divide the original data set into a training set and a test set according to a set ratio; Use normalization preprocessing for the model input features to ensure data consistency between different dimensions; S3.2 XGBoost model initialization: Select the XGBoost regression model to construct a non-linear mapping relationship f: X → y, where X is the feature vector and y is the ramming settlement; S3.3 Hyperparameter grid search and tuning: Use grid search to optimize the combination of key hyperparameters of the XGBoost model, with the goal of minimizing the cross-validation error. Use five-fold cross-validation to calculate the negative mean squared error under each set of hyperparameters, and select the model combination with the best performance; S3.4 Model prediction and performance evaluation: Perform predictions on the optimized optimal model on the training set and the test set respectively, and comprehensively evaluate the model performance using the following indicators: RMSE (root mean square error), MAE (mean absolute error), and R² (coefficient of determination); S3.5 Output the optimal model and parameters: Finally, retain and output the tuned best model and its corresponding optimal hyperparameter combination, providing a core support tool for subsequent parameter optimization based on the prediction model.

[0013] Further, step S4 specifically includes: S4.1. Start the optimization system and prepare for dynamic compaction parameter search and prediction evaluation: Load the XGBoost settlement prediction model, which is a trained XGBoost model based on historical compaction test data, and is used to predict the settlement response under the input parameter combination. S4.2. Set the initial parameters of PSO, including the particle swarm size, maximum number of iterations, inertia factor, individual / global learning factors, and search space boundaries. S4.3. Initialize the positions and velocities of the particles, and randomly generate the positions (i.e., parameter combinations) and initial velocities of the particle swarm individuals in the parameter space. S4.4. Use the XGBoost model to predict the settlement response, input the parameter combination of each particle into the XGBoost model, and obtain the predicted settlement amount. S4.5. Construct an objective function that combines construction cost and settlement response, in the form of P = α×N + β×M×H + penalty; α and β are weights, representing the importance of the number of blows and construction cost respectively, and need to be set according to engineering experience and actual conditions; penalty is a penalty term for violating the settlement standard. If the settlement amount exceeds the corresponding energy level threshold, a positive penalty is imposed; N, M, and H represent the number of blows, hammer mass, and lift height respectively. S4.6. Update the individual best value P_best, record the current historical best fitness value of each particle and its corresponding position; update the global best value G_best, select the particle with the smallest fitness value among all current particles as the global optimal solution, record its position, and determine whether the stop compaction standard is met. If the current predicted settlement amount meets the stop compaction condition under the corresponding compaction energy, continue; otherwise, update the particle position and velocity and then continue the iteration. S4.7. Output the optimal compaction parameter combination: Finally, output the dynamic compaction parameter combination that meets the stop compaction standard and has the minimum fitness, and the optimization process is completed.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention first constructs a systematic test data set through a reduced-scale model test. On the basis of controlling multiple variables such as hammer mass, lift height, number of blows, and hammer diameter, the settlement responses under different parameter combinations are collected. The model test data provides a reliable physical basis and data support for subsequent machine learning modeling and intelligent optimization. Subsequently, a dynamic compaction parameter optimization method combining XGBoost regression prediction and PSO particle swarm optimization algorithm is proposed. By constructing a settlement prediction model, automatic optimization of compaction parameters is realized under the premise of meeting the stop compaction standard, so as to achieve the purpose of ensuring construction quality, improving construction efficiency, and reducing construction cost. The method of the present invention is applicable to the construction design stage of various foundations that need to be treated by dynamic compaction, and has good engineering adaptability and popularization value.

[0015] In addition to the purposes, features, and advantages described above, the present invention has other purposes, features, and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. Description of the Drawings

[0016] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the accompanying drawings: Figure 1 is the overall flow of the dynamic compaction construction parameter optimization method based on XGBoost-PSO in the present invention; Figure 2 is the structural schematic diagram of the model test device in the present invention; Figure 3 is the schematic diagram of the rammer model in the present invention; Figure 4 is the flow chart of optimizing dynamic compaction parameters using the PSO algorithm in the present invention. Specific Embodiments

[0017] The present invention will be described in detail below in conjunction with the various embodiments shown in the accompanying drawings. However, it should be noted that these embodiments are not limitations to the present invention, and any equivalent transformation or substitution in terms of function, method, or structure made by those of ordinary skill in the art based on these embodiments shall fall within the protection scope of the present invention.

[0018] Please refer to Figure 1 , this embodiment provides a dynamic compaction construction parameter optimization method based on XGBoost-PSO, including the following steps: S1. Dynamic compaction model test: Design and fabricate a dynamic compaction model, and conduct compaction tests using test rammers of various different specifications and the dynamic compaction model to obtain the settlement data under several different parameter combinations. Specifically, it includes: S1.1. Determine the similarity criteria and scale ratio. According to the three types of similarity criteria of geometric similarity, physical similarity, and mechanical similarity, and in combination with the actual engineering background, select parameters such as rammer mass M, rammer lifting height H, rammer diameter D, number of blows N, compaction energy E, settlement s, water content w, and dry density d of the soil as model control variables, and determine the similarity coefficient system required for the compaction test through the second theorem of similarity and dimensional analysis method to guide the design of the scaled model.

[0019] Step S1.2: Design a dynamic compaction model. Fabricate a dynamic compaction model box 1 with a cuboid structure, sized 1000mm × 1000mm × 800mm. The box is made of wood and meets the boundary requirement that "the acting boundary of the dynamic compaction model box is at least 4 times the diameter of the rammer" to prevent the influence of boundary effects on the test. Equip the dynamic compaction model box with a triangular support 2 and install an electromagnetic adsorption device 3 at the top of the triangular support. This dynamic compaction model can achieve a free fall height of up to 1.5m to simulate the free fall behavior of heavy objects during dynamic compaction. The dynamic compaction model, triangular support, and electromagnetic adsorption device together constitute the test device, and the test device is as shown in Figure 2 shown.

[0020] S1.3: Configure test rammers of different specifications. Specifically, five circular iron rammers with masses of 5kg, 8kg, 11kg, 17kg, and 22kg can be selected. Each rammer of each mass includes three specifications with diameters of 120mm, 150mm, and 180mm. Some of the rammer bodies have holes on the surface to reduce air resistance and approximate actual engineering conditions. The schematic of the rammer is as shown in Figure 3 shown.

[0021] S1.4: Layout the measuring device and perform the tamping operation. Design multiple groups of tamping tests with different tamping parameters (such as rammer weight M, lifting height H, rammer diameter D, and number of tamping times N, etc.). Record the vertical distance from the geometric center of the rammer to the soil surface with a millimeter scale as the measured value of the tamping settlement. Set the number of tamping times for each group of tests to be 8 - 18 times; preferably, set the number of tamping times for each group of tests to be 12 times. During the test, use the electromagnetic adsorption device to control the free fall of the rammer. After each tamping, read the tamping settlement data and record it in the database. After each tamping, the settlement of the soil needs to be measured with a ruler. To ensure the independence of the test, the soil is re-disturbed and loosened and leveled after each group of tamping to maintain the consistency of the initial state of the soil for multiple groups of tamping tests.

[0022] S2: Feature selection: First, select the original features from the actual measured parameters of the dynamic compaction model test; then construct interaction features using the actual measured parameters and select the optimal interaction features according to the XGboost feature importance ranking; then perform a correlation analysis on the original features and the optimal interaction features, and finally obtain the input features for the subsequent XGboost prediction model. The specific steps are as follows: S2.1. Original feature selection: Based on the parameters actually measured in the model test data, the ram mass M, lift height H, ram diameter D, and number of tamping blows N are selected as the original feature variables. The grey relational analysis (GRA) is used to calculate the correlation degree between each original feature variable and the target variable (tamping settlement), and the original feature variables with a correlation degree greater than the threshold of 0.6 are regarded as key influencing features and included in the subsequent modeling.

[0023] S2.2. Optimal interaction feature selection: The specific steps are as follows: (1). Determine the interaction feature combination form: Select the ram mass M and lift height H as the basic variables for constructing interaction features, and set three types of combination forms: MH a , M b H and M c H d . Where a, b, c, and d are power parameters used to adjust the non-linear form; (2). Set the parameter search range and step size: To ensure the search accuracy and calculation efficiency, set a, b, c, d ∈ [0.1, 2, 0.1].

[0024] (3). Construct interaction features and perform K-fold cross-validation: Use K-fold cross-validation (K = 5) to train the XGBoost model for the interaction features of the three combination forms, and statistically calculate the "feature importance" index of each interaction feature in each round of training, and obtain the stable importance of each interaction feature by weighted average.

[0025] (4). Calculate the total feature importance and screen the optimal power combination: Add the average importance values of the three interaction features and define it as the "total importance" of the current combination. After traversing all power combinations, screen out the power combination with the largest total importance as the final interaction feature construction strategy.

[0026] (5). Construct the optimal interaction features and include them in the model training: According to the evaluation results, the interaction features corresponding to the optimal power parameters are formally included in the model input feature set and participate in the XGBoost model training jointly with the original features.

[0027] S2.3. Feature correlation analysis: To avoid the influence of multicollinearity among features, the Spearman rank correlation coefficient is used to analyze the correlation between variables; the original features selected in step S2.1 and the optimal interaction features selected in step S2.2 are combined as variables for Spearman calculation and analysis. If the correlation coefficient ρ between variables > 0.8, it is considered that the collinearity is too high, and feature elimination is carried out; subsequently, the variance inflation factor method is used to calculate vif to ensure that the vif value of the remaining features < 10, thereby ensuring the stability and generalization ability of model training. The variance inflation factor vif is an important indicator to test whether there is still collinearity.

[0028] S3. Based on the feature set screened in step S2, an XGBoost settlement prediction model is established; the specific steps are as follows: S3.1. Data division and standardization: The original data set (i.e., the settlement data set under different parameter combinations collected during the model experiment in step s1) is divided into a training set and a test set in a ratio of 8:2; StandardScaler is used to perform normalization preprocessing on the input features to ensure data consistency between different dimensions.

[0029] S3.2. XGBoost model initialization: The XGBoost regression model is selected to construct a non-linear mapping relationship f: X → y, where X is the feature vector and y is the settlement.

[0030] S3.3. Hyperparameter grid search and tuning: Grid search (GridSearchCV) is used to optimize the combination of key hyperparameters (such as n_estimators, max_depth, learning_rate, subsample, colsample_bytree, etc.) of the XGBoost model. The optimization goal is to minimize the cross-validation error. Five-fold cross-validation is used to calculate the negative mean squared error under each set of hyperparameters, and the model combination with the best performance is selected.

[0031] S3.4. Model prediction and performance evaluation: The optimized optimal model is used to make predictions on the training set and the test set respectively, and the following indicators are used to comprehensively evaluate the performance of the model: RMSE (root mean squared error), MAE (mean absolute error), and R² (coefficient of determination).

[0032] S3.5. Output the optimal model and parameters: Finally, the optimized best model and its corresponding optimal hyperparameter combination are retained and output, providing a core support tool for subsequent parameter optimization based on the prediction model.

[0033] S4. Use the particle swarm optimization algorithm to optimize the dynamic compaction parameters of the XGBoost settlement prediction model, and finally obtain the optimal dynamic compaction parameter combination; the flow chart of using the PSO algorithm for dynamic compaction parameter optimization is as Figure 4As shown below. Specifically, this step is as follows: S4.1. Start the optimization system and prepare for dynamic compaction parameter search and prediction evaluation. Load the XGBoost compaction prediction model, which is a pre-trained XGBoost model based on historical compaction test data, and is used to predict the compaction response under the input parameter combination.

[0034] S4.2. Set the initial parameters of PSO, including the particle swarm size, maximum number of iterations, inertia factor, individual / global learning factor, and search space boundaries.

[0035] S4.3. Initialize the positions and velocities of the particles, and randomly generate the positions (i.e., parameter combinations) and initial velocities of the particle swarm individuals in the parameter space.

[0036] S4.4. Use the XGBoost model to predict the compaction response. Input the parameter combination of each particle into the XGBoost model to obtain the predicted compaction amount (especially the average compaction amount of the last two blows).

[0037] S4.5. Construct an objective function that combines construction cost and compaction response, in the form of P = α×N + β×M×H + penalty. Here, α and β are weights, representing the importance of the number of blows and construction cost respectively, and need to be set according to engineering experience and actual conditions; penalty is the penalty term for violating the compaction standard. If the compaction amount exceeds the corresponding energy level threshold, a positive penalty is imposed; N, M, and H represent the number of blows, ram weight, and lift height respectively. The feasible ranges for each parameter are set as shown in Table 1 below.

[0038] Table 1 Feasible range table for each parameter setting

[0039] S4.6. Update the individual best value P_best, record the current historical best fitness value of each particle and its corresponding position; update the global best value G_best, select the particle with the smallest fitness value among all current particles as the global optimal solution, record its position, and determine whether the compaction stop standard is met. If the current predicted compaction amount meets the compaction stop condition for the corresponding compaction energy, continue; otherwise, update the particle position and velocity and then continue the iteration. Specifically, the compaction stop standard refers to the "Technical Code for Building Foundation Treatment": the average compaction amount of the last two blows for compaction energies less than 4000 kN·m, compaction energies between 4000 and 6000 kN·m, and compaction energies greater than 6000 kN·m should not be greater than 50 mm, 100 mm, and 200 mm respectively.

[0040] S4.7. Output the optimal compaction parameter combination: finally output the dynamic compaction parameter combination that meets the compaction stop standard and has the minimum fitness value (including the number of blows N, ram weight M, ram lift height H, and ram diameter D), the optimization process is completed, and the program operation ends.

[0041] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for optimizing dynamic compaction construction parameters based on XGBoost-PSO, characterized in that: The following steps are involved: S1. Dynamic compaction model test: design and manufacture a dynamic compaction model, and use a variety of different specifications of test rammers and the dynamic compaction model to carry out a compaction test to obtain compaction sinking data under several different parameter combinations; S2, feature selection: first select the original features from the actual measured parameters of the dynamic compaction model test; Then, the actual measured parameters are used to construct interactive features, and the optimal interactive features are selected based on the XGboost feature importance ranking; then, the correlation analysis is performed between the original features and the optimal interactive features; S3, based on the feature set screened in step S2, establishing an XGBoost tamping amount prediction model; S4. The particle swarm algorithm is used to optimize the compaction parameters of the XGBoost compaction settlement prediction model, and finally the optimal compaction parameter combination is obtained.

2. The method for optimizing dynamic compaction construction parameters according to claim 1, characterized in that: Step S1 specifically includes: S1.

1. Determine similarity criteria and scale ratio: According to the three similarity criteria of geometric similarity, physical similarity and mechanical similarity, and in combination with the actual engineering background, select model control variables, and determine the similarity coefficient system required for the tamping test through the second similarity theorem and dimensional analysis method; the model control variables include tamping hammer mass M, tamping hammer lifting height H, tamping hammer diameter D, tamping number N, tamping energy E, tamping sinking amount s, moisture content w and soil dry density d; S1.2, making a dynamic compaction model box, configuring a triangular bracket for the dynamic compaction model box, and installing an electromagnetic adsorption device on the top of the triangular bracket to simulate the free fall of a heavy object during the dynamic compaction process; S1.

3. Test rammers of different specifications are configured; some of the hammers have holes on their surfaces to reduce air resistance and to be close to actual engineering conditions; S1.

4. Arrange the measuring device and perform the tamping operation: Design different tamping parameters to conduct multiple groups of tamping tests. Set the number of tamping times for each group of tests to 8 to 18 times. Read the tamping amount data after each tamping and record it in the database.

3. The method for optimizing dynamic compaction construction parameters according to claim 2, characterized in that: In step S1.4, the soil is disturbed again and loosely spread and leveled after each set of tamping is completed to maintain the consistency of the initial state of the soil in multiple sets of tamping tests.

4. The method for optimizing dynamic compaction construction parameters according to claim 2, characterized in that: Step S2 specifically includes: S2.

1. Original feature selection: Based on the parameters actually measured in the model test data, the rammer mass M, lifting height H, rammer diameter D and rammer number N are selected as original feature variables; the grey correlation analysis method is used to calculate the correlation between each original feature variable and the target variable, and the original feature variables with a correlation greater than the correlation threshold are regarded as key influencing features and included in the subsequent modeling; S2.2, optimal interactive feature selection, the specific steps are: (1) Determine the combination form of interactive features: Select the rammer mass M and the lifting height H as the basic variables for constructing interactive features, and set three types of combination forms MH a 、M b H and M c H d ; Where a, b, c and d are power parameters used to adjust the nonlinear morphology; (2) Set the parameter search range and step size: set a, b, c, d ∈ [0.1, 2, 0.1]; (3) Construct interactive features and perform K-fold cross validation: Use K-fold cross validation to train the XGBoost model for the three types of combined interactive features, and count the feature importance index of each interactive feature in each round of training, and use weighted average to obtain the stable importance of each interactive feature; where K=5; (4) Calculate the total importance of features and select the optimal power combination: MH a 、M b H and M c H d The average importance values ​​of the three interactive features are added together to define the total importance of the current combination. After traversing all power combinations, the power combination with the largest total importance is selected as the final interactive feature construction strategy. (5) Construct the optimal interaction features and incorporate them into model training: Based on the evaluation results, the interaction features corresponding to the optimal power parameters are formally incorporated into the model input feature set, and the original features are combined to participate in the XGBoost model training; S2.

3. Combine the original features selected in step S2.1 and the optimal interaction features selected in step S2.2 as variables for Spearman calculation analysis. If the correlation coefficient ρ between variables is greater than 0.8, the collinearity is considered to be too high and features are removed. Then, the variance inflation factor method is used to calculate vif to ensure that the remaining feature vif value is less than 10.

5. The method for optimizing dynamic compaction construction parameters according to claim 4, characterized in that: Step S3 specifically includes: S3.

1. Data division and standardization: Divide the original data set into training set and test set according to the set ratio; perform normalization preprocessing on the model input features to ensure data consistency between different dimensions; S3.2, XGBoost model initialization: Use the XGBoost regression model to construct a nonlinear mapping relationship f: X→y, where X is the feature vector and y is the tamping amount; S3.

3. Hyperparameter grid search and tuning: Grid search is used to optimize the key hyperparameters of the XGBoost model, with the goal of minimizing the cross-validation error. Five-fold cross-validation is used to calculate the negative mean square error under each set of hyperparameters, and the model combination with the best performance is selected. S3.4, Model prediction and performance evaluation: The optimized model is used to make predictions on the training set and test set respectively, and the root mean square error RMSE, mean absolute error MAE and determination coefficient R² are used to comprehensively evaluate the model performance; S3.

5. Output the optimal model and parameters: Finally, retain and output the best model after parameter adjustment and its corresponding optimal hyperparameter combination, providing a core support tool for subsequent parameter optimization based on the prediction model.

6. The method for optimizing dynamic compaction construction parameters according to claim 2, characterized in that: Step S4 specifically includes: S4.

1. Start the optimization system and prepare for the dynamic compaction parameter search and prediction evaluation: load the XGBoost compaction settlement prediction model, load the XGBoost model that has been trained based on historical compaction test data, and use it to predict the compaction settlement response under the input parameter combination; S4.2, set the initial parameters of PSO, including particle swarm size, maximum number of iterations, inertia factor, individual / global learning factor, and search space boundary; S4.3, initialize the position and velocity of the particles, and randomly generate the position and initial velocity of the individual particle swarms in the parameter space; S4.4, using the XGBoost model to predict the tamping response, inputting the parameter combination of each particle into the XGBoost model to obtain the predicted tamping amount; S4.

5. Construct an objective function that integrates construction cost and tamping response in the form of P=α×N+β×M×H+penalty; α and β are weights, representing the importance of the number of tamping times and construction cost, respectively, and need to be set based on engineering experience and actual conditions; penalty is a penalty item for violating the tamping standard. If the tamping amount exceeds the corresponding energy level threshold, a positive penalty is imposed; N, M, and H represent the number of tamping times, the mass of the tamping hammer, and the lifting height, respectively; S4.6, update the individual optimal value P_best, record the current historical optimal fitness value of each particle and its corresponding position; update the global optimal value G_best, select the particle with the smallest fitness value among all the current particles as the global optimal solution, record its position, and determine whether it meets the stop tamping standard. If the current predicted tamping amount meets the stop tamping condition under the corresponding tamping energy, continue; otherwise, update the particle position and speed and continue to iterate; S4.

7. Output the optimal compaction parameter combination: The final output is the compaction parameter combination that meets the stop-compaction standard and has the minimum fitness, and the optimization process is completed.

Images