A method for interpreting ground mechanical property parameters based on impact penetration sounding

Abstract

This invention provides a method for interpreting ground mechanical property parameters based on impact penetration testing. It utilizes an unmanned aerial vehicle (UAV) carrying an impact penetration tester to remotely survey the target ground. A radial basis function neural network model and a multi-objective genetic algorithm are then used to interpret the dynamic response data collected by the tester, enabling rapid and accurate acquisition of ground mechanical property parameters. This invention is conceptually sound and can quickly and accurately interpret soil ground mechanical property parameters. These parameters can be used to assess the mobility of equipment in complex geological environments, thereby improving the ability to cope with such environments.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, specifically to a method for interpreting surface mechanical property parameters based on impact penetration testing. Background Technology

[0002] Frequent geological disasters pose a significant challenge to my country. Particularly due to the influence of earthquakes, climate change, and human activities, landslides, mudslides, and other disasters are becoming increasingly frequent, seriously threatening people's lives and property. These disasters are characterized by their suddenness, insidious nature, and destructive power, making prediction, forecasting, and prevention extremely difficult, often catching people off guard. When disasters occur, the lack of geological and topographical information in the affected area hinders the rapid arrival of rescue personnel and equipment, significantly reducing the timeliness of rescue efforts.

[0003] Existing geological hazard survey technologies mainly include remote sensing and in-situ soil penetration testing. While remote sensing can acquire geological information over long distances and wide areas, its non-contact measurement methods cannot accurately retrieve the mechanical information of the ground soil, and it is easily affected by weather conditions in complex disaster environments, resulting in low data reliability and resolution. In-situ soil penetration testing includes static and dynamic penetration testing. Static penetration testing techniques, such as topographic survey vehicles, vehicle-mounted ground-penetrating radar, and cone penetrators, can accurately identify soil and rock types and mechanical parameters, but these methods all require close-range manual exploration, failing to meet the needs of remote, real-time surveys. Dynamic penetration testing, such as... Figure 7 As shown, by using a drone equipped with an impact penetrometer to obtain the penetration velocity under the action of gravity, and interpreting the ground mechanical property parameters based on the penetration resistance and acceleration, the limitations of traditional technologies in conducting remote, wide-area, and rapid soil mechanics surveys in areas inaccessible to personnel can be overcome.

[0004] Ground mechanical properties cannot be directly measured by impact penetration testing. Currently, research on the interpretation of ground mechanical properties by researchers both domestically and internationally mainly focuses on static penetration testing, while research on the interpretation of ground mechanical properties based on impact penetration testing is rarely mentioned. Summary of the Invention

[0005] To address the technical problems existing in the background art, this invention proposes a ground mechanical property parameter interpretation method based on impact penetration testing. Its concept is reasonable and can quickly and accurately interpret the mechanical parameters of the soil. These parameters can be used for equipment mobility assessment in complex geological environments, thereby improving the ability to cope with complex geological environments.

[0006] To address the aforementioned technical problems, this invention provides a method for interpreting ground mechanical property parameters based on impact penetration testing. This method involves using an unmanned aerial vehicle (UAV) carrying an impact penetration instrument to remotely survey the target ground, and then using a radial basis function neural network model and a multi-objective genetic algorithm to interpret the dynamic response data collected by the penetration instrument, thereby quickly and accurately obtaining ground mechanical property parameters.

[0007] The method for interpreting ground mechanical property parameters based on impact penetration testing specifically includes the following steps:

[0008] (1) Data Acquisition

[0009] First, an impact penetration test is conducted on the target ground. The UAV carrying the impact penetrator flies to the airspace above the target ground. After the UAV reaches the predetermined altitude, the impact penetrator is dropped into the ground in a free fall manner to measure the time history data of the cone tip drag and acceleration during the penetration process. Then, the data acquisition and processing module embedded in the impact penetrator performs preliminary data processing to obtain high-quality time history data of cone tip drag and acceleration.

[0010] (2) Database Construction

[0011] The Latin hypercube sampling method is used to generate ground mechanical property parameter samples in the soil parameter space. The CEL finite element method is used to simulate the parameters in the ground mechanical property parameter samples to generate dynamic response data of the penetrometer under different soil conditions, thereby constructing a database covering a wide range of soil conditions.

[0012] (3) Radial basis neural network model training

[0013] The radial basis function neural network model was trained using the constructed database to establish a mapping relationship between ground mechanical property parameters and penetrator dynamic response data;

[0014] (4) Interpretation of Ground Mechanical Properties

[0015] By combining a radial basis function neural network model with a multi-objective genetic algorithm, accurate interpretation of ground mechanical property parameters can be achieved.

[0016] The ground mechanical property parameter interpretation method based on impact penetration testing, wherein the specific process of generating samples of ground mechanical property parameters using the Latin hypercube sampling method in step (2) is as follows:

[0017] (2.1) Determining the parameter range

[0018] Determine the value range of each ground mechanical property parameter, assuming there are P parameters in total, forming a P-dimensional parameter space;

[0019] (2.2) Divide the parameter range into equal parts

[0020] The range of values ​​for each ground mechanical property parameter is divided into N equally spaced intervals along its dimension, that is, the range of values ​​corresponding to each ground mechanical property parameter is divided into N sub-intervals.

[0021] (2.3) Randomly select sample values ​​for each interval

[0022] For each dimension of ground mechanical property parameter, a value is randomly selected in each sub-interval to ensure that a sample value can be obtained for all intervals in that dimension.

[0023] (2.4) Sample arrangement

[0024] After randomly arranging N samples of each ground mechanical property parameter, P-dimensional samples are formed. The final generated sample points will cover the entire ground mechanical property parameter space, and the samples in each dimension will be evenly distributed.

[0025] The ground mechanical property parameter interpretation method based on impact penetration testing, wherein the specific process of training the radial basis function neural network model using the constructed database in step (3) is as follows:

[0026] (3.1) Input the ground mechanical property parameter sample points generated by Latin hypercube sampling into the input layer of the radial basis neural network model;

[0027] (3.2) Using CEL finite element simulation, the dynamic response data of the penetrator is obtained by simulating the sample points of each group of ground mechanical property parameters, and these data are used as the output of the radial basis neural network model.

[0028] (3.3) Activation function for constructing the structure of the radial basis function neural network model

[0029] The structure of a radial basis function neural network model consists of an input layer, hidden layers, and an output layer. The hidden layers use radial basis functions as activation functions.

[0030]

[0031] In equation (1) above, r is the Euclidean distance from the input to the center of the hidden layer; σ is the width parameter of the Gaussian function.

[0032] The ground mechanical property parameter interpretation method based on impact penetration test, wherein: the preliminary data processing in step (1) includes filtering and noise reduction.

[0033] The ground mechanical property parameter interpretation method based on impact penetration testing, wherein the process of training the radial basis function neural network model in step (3) is as follows: the radial basis function neural network model takes the ground mechanical property parameters as input and the time history data of cone tip drag and acceleration as output. At the same time, in order to verify the accuracy of the radial basis function neural network model, a portion of the samples are randomly selected from the database as the test set, and the remaining portion is used as the training set; normalized root mean square error and coefficient of determination R are introduced. 2 To evaluate the prediction accuracy of the radial basis function neural network (RBN) model, the following two evaluation metrics are used:

[0034] One evaluation metric is the normalized root mean square error (RAAE):

[0035]

[0036] In equation (2) above, y pred,i It is the model prediction value, y ture,i y is the true value, n is the number of samples, and y is the true value. max and y min These are the maximum and minimum values ​​of the true value, respectively.

[0037] Another evaluation metric is the coefficient of determination R. 2 :

[0038]

[0039] In equation (3) above, y is the mean of the true values;

[0040] If the prediction accuracy of the radial basis function neural network model on the test set does not meet the requirements, increase the training data, adjust the structure and parameters of the radial basis function neural network model, until the prediction accuracy of the radial basis function neural network model reaches the expected standard.

[0041] The ground mechanical property parameter interpretation method based on impact penetration testing, wherein the specific process of step (4) is as follows: using the difference between the cone tip drag and acceleration time history data predicted by the radial basis function neural network model and the experimental observation data as the objective function, the ground mechanical property parameters are determined and interpreted by minimizing the objective function through a multi-objective genetic algorithm; in order to interpret the ground mechanical property parameters, it is first necessary to establish an objective function to minimize the difference between the radial basis function neural network model prediction value and the experimental observation value, the expression of which is:

[0042]

[0043] In equation (4), pred_a t and pred_F tThese are the time histories of cone tip drag and acceleration predicted by the radial basis function neural network model; tar_a t and tar_F t These are experimental observation data for cone tip drag and acceleration, respectively.

[0044] By employing non-dominated sorting, crowding calculation, and elite retention strategies, the global optimality of the Pareto optimal solution set formed after multi-objective genetic algorithm iteration is ensured, thereby accurately interpreting the ground mechanical property parameters. The specific interpretation process is as follows:

[0045] (4.1) Initialize the population for the genetic algorithm

[0046] An initial population is randomly generated, with each individual representing a set of ground mechanical property parameters; the genes of each individual are represented as a parameter vector X = [φ, c, E, ν, ρ].

[0047] (4.2) Calculation of the objective function

[0048] For each individual, the radial basis function neural network model is used to predict the cone tip drag and acceleration time history data, and the objective function value corresponding to that individual is calculated by the following equation (5):

[0049]

[0050] (4.3) Non-dominated sorting

[0051] Individuals in the population are ranked according to Pareto superiority to form multiple non-dominant ranks, with individuals in lower ranks being superior to those in higher ranks.

[0052] (4.4) Crowding Calculation

[0053] The crowding level of each individual is calculated to measure its density within the target space. Crowding level is used to maintain population diversity, and the specific formula is as follows:

[0054]

[0055] in, and It is the function value of the adjacent solution of the i-th solution on the j-th objective function. and These are the maximum and minimum values ​​of the j-th objective function, respectively;

[0056] (4.5) Elite Retention Strategy

[0057] In each generation of the population in the genetic algorithm, non-dominated solutions are retained first, and individuals are selected in areas with high crowding to ensure that the population develops towards the Pareto front.

[0058] (4.6) Crossover and Mutation

[0059] New individuals are generated through crossover and mutation operations in genetic algorithms, and these new individuals will replace the inferior individuals in subsequent generations of iteration.

[0060] (4.7) Termination Conditions

[0061] When the population reaches the maximum number of generations or the convergence change of the objective function value is less than a preset threshold, the genetic algorithm terminates and selects the optimal ground mechanical property parameter vector according to the decision criteria.

[0062] By adopting the above technical solution, the present invention has the following beneficial effects:

[0063] This invention presents a well-designed method for interpreting ground mechanical property parameters based on impact penetration testing. It can rapidly and accurately interpret soil ground mechanical property parameters, including cohesion (c), internal friction angle (φ), elastic modulus (E), Poisson's ratio (ν), and density (ρ), with an average interpretation error of less than 25% across multiple conditions. Conventional computer-based single-condition parameter interpretation takes less than 1 minute. This invention can be applied to remote and rapid ground mechanical surveys in disaster areas and areas difficult for personnel to access.

[0064] This invention constructs a physical model of soil by impact penetration and develops a method for interpreting ground mechanical property parameters based on the radial basis function neural network (RBF) model and the multi-objective genetic algorithm (NSGA-II). Impact penetration experiments on sandy soil and silty clay at different moisture contents show that, according to the method of this invention, the ground mechanical property parameters of the soil, including cohesion, internal friction angle, elastic modulus, Poisson's ratio and density, can be accurately interpreted by the cone tip resistance and acceleration time history curves. Attached Figure Description

[0065] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0066] Figure 1 This is a flowchart of the ground mechanical property parameter interpretation method based on impact penetration testing according to the present invention.

[0067] Figure 2 This is a flowchart of the radial basis function neural network model training involved in the ground mechanical property parameter interpretation method based on impact penetration probing of the present invention.

[0068] Figure 3This is a diagram showing the verification results of the radial basis neural network model involved in the ground mechanical property parameter interpretation method based on impact penetration probing in this invention.

[0069] Figure 4 The present invention relates to a method for interpreting ground mechanical property parameters based on impact penetration testing, and includes a flowchart of the ground mechanical property parameter interpretation process.

[0070] Figure 5 The flowchart of the NSGA-II algorithm involved in the ground mechanical property parameter interpretation method based on impact penetration probing in this invention;

[0071] Figure 6 This is a schematic diagram of the CEL model involved in the ground mechanical property parameter interpretation method based on impact penetration testing in this invention.

[0072] Figure 7 This is a field photo of the free-fall dynamic penetration test technology. Detailed Implementation

[0073] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] The present invention will be further explained below with reference to specific embodiments.

[0075] like Figure 1 As shown in the figure, this embodiment provides a ground mechanical property parameter interpretation method based on impact penetration testing. It uses an UAV carrying an impact penetration instrument to conduct remote surveys of the target ground (soil), and uses a radial basis function neural network model (i.e., RBF model) and a multi-objective genetic algorithm (NSGA-II) to interpret the dynamic response data collected by the penetration instrument in order to quickly and accurately obtain the ground mechanical property parameters.

[0076] S100, Data Acquisition

[0077] First, an impact penetration test is conducted on the target ground (soil). A drone carrying an impact penetrator flies to the airspace above the target ground. After reaching the predetermined altitude, the impact penetrator plunges into the ground in a free-fall manner, measuring the cone tip drag and acceleration time history data during the penetration process. Then, the data acquisition and processing module embedded in the impact penetrator performs preliminary data processing, including filtering and noise reduction, to obtain high-quality cone tip drag and acceleration time history data.

[0078] S200, Database Construction

[0079] To establish the relationship between penetrometer dynamic response data and ground mechanical property parameters, this invention employs the Latin hypercube sampling method to generate ground mechanical property parameter samples within the soil parameter space (Latin hypercube sampling is a statistical method for generating sample points in a multidimensional parameter space, ensuring a uniform distribution of sample values ​​in each dimension, thus avoiding sample clustering problems that may occur in traditional random sampling). The CEL finite element method is then used to simulate the parameters in the ground mechanical property parameter samples to generate dynamic response data of the penetrometer under different soil conditions, thereby constructing a database covering a wide range of soil conditions. The specific steps are as follows:

[0080] S201, Parameter Range Determination

[0081] For each ground mechanical property parameter, such as cohesion c, internal friction angle φ, elastic modulus E, Poisson's ratio ν, and density ρ, determine their value range; assume that the number of parameters is P, that is, P-dimensional parameter space.

[0082] S202, Divide the parameter range into equal parts

[0083] The range of values ​​for each ground mechanical property parameter is divided into N equally spaced intervals along its dimension, that is, the range of values ​​corresponding to each ground mechanical property parameter is divided into N sub-intervals.

[0084] S203. Randomly select sample values ​​for each interval.

[0085] For each dimension of ground mechanical property parameter, a value is randomly selected within each sub-interval to ensure that a sample value can be obtained for all intervals in that dimension.

[0086] S204, Sample Arrangement

[0087] After randomly arranging N samples of each ground mechanical property parameter, P-dimensional samples are formed respectively; the final generated sample points will cover the entire ground mechanical property parameter space, and the samples in each dimension are evenly distributed.

[0088] The purpose of using Latin hypercube sampling is to ensure that sample points are distributed as evenly as possible in the parameter space, avoiding sample point clustering problems caused by random sampling. Subsequently, the CEL (Coupled Eulerian-Lagrangian) finite element method is used to simulate each group of soil parameters (this invention uses the Coupled Eulerian-Lagrangian (CEL) finite element method, which is particularly suitable for handling large deformation problems and can accurately simulate the behavior of soil under dynamic loading conditions; the CEL finite element model used in this invention is as follows...). Figure 6As shown, the cone in the model uses Lagrange elements with a diameter of 35.7 mm; while the soil uses Eulerian elements. The soil model dimensions are identical to those of the laboratory soil tank: 1 meter wide, 1.5 meters long, and 1.2 meters high. To ensure numerical stability and computational accuracy, the Eulerian elements near the cone are selected with a size of 2 mm. Dynamic response data of the penetrometer under different soil conditions are generated, thus constructing a database covering a wide range of soil conditions for subsequent training and validation of the radial basis function neural network model.

[0089] S300, Radial Basis Intermediate Neural Network Model Training

[0090] The radial basis function neural network model is trained using the constructed database; the specific training process is as follows (e.g.) Figure 2 As shown):

[0091] S301. Input data: Input the ground mechanical property parameter sample points generated by Latin hypercube sampling into the input layer of the radial basis neural network model.

[0092] S302. Output data: Using CEL finite element simulation, the dynamic response data of the penetrator is obtained by simulating the sample points of each group of ground mechanical property parameters, and these data are used as the output of the radial basis neural network model.

[0093] S303. Activation Functions for Constructing the Radial Basis Function Neural Network Model: The structure of a radial basis function neural network model consists of an input layer, hidden layers, and an output layer. The hidden layers use radial basis functions (Gaussian functions) as activation functions.

[0094]

[0095] In equation (1) above, r is the Euclidean distance from the input to the center of the hidden layer, and σ is the width parameter of the Gaussian function;

[0096] To establish a mapping relationship between ground mechanical property parameters and penetrator dynamic response data, a radial basis function (RBF) neural network model was developed. The RBF model used ground mechanical property parameters as input and cone drag and acceleration time history data as output. To verify the accuracy of the RBF model, a subset of samples was randomly selected from the database as the test set, and the remainder as the training set. Normalized root mean square error (RAAE) and the coefficient of determination R0 were introduced. 2 To evaluate the prediction accuracy of the radial basis function neural network (RBN) model, the following two evaluation metrics are used:

[0097] One evaluation metric is the normalized root mean square error (RAAE):

[0098]

[0099] In equation (2) above, y pred,i It is the model prediction value, y ture,i y is the true value, n is the number of samples, and y is the true value. max and y min These are the maximum and minimum values ​​of the true value, respectively.

[0100] Another evaluation metric is the coefficient of determination (R²). 2 ):

[0101]

[0102] In the above formula (3), The mean of the true values;

[0103] If the prediction accuracy of the radial basis function neural network model on the test set does not meet the requirements, increase the training data, adjust the structure and parameters of the radial basis function neural network model, until the prediction accuracy of the radial basis function neural network model reaches the expected standard.

[0104] In this embodiment, the CEL finite element method is used to generate 4,000 sets of time history data of the dynamic response of the penetrometer under different soil conditions through Latin hypercube sampling. Based on these data, the radial basis function neural network model is trained. At the same time, in order to evaluate the prediction performance of the trained RBF, an additional 100 CEL simulations are performed, and the simulation results are compared with the prediction results of the RBF model. Figure 3 The predicted results for maximum acceleration, maximum penetration resistance, and penetration depth are presented. The results show that the trained RBF model can effectively predict the dynamic response characteristics of the impact penetration process, thus establishing a mapping relationship between ground mechanical property parameters and the time history data of the penetrometer's dynamic response.

[0105] S400, Interpretation of Ground Mechanical Properties Parameters

[0106] This invention combines the radial basis function (RBF) neural network model with... Figure 5 The multi-objective genetic algorithm (NSGA-II) is combined to achieve accurate interpretation of ground mechanical property parameters. Specifically, the difference between the cone drag and acceleration time history data predicted by the RBF model and the experimental observation data is used as the objective function. The NSGA-II algorithm minimizes the objective function to determine the ground mechanical property parameters. To interpret the ground mechanical property parameters, an objective function needs to be established first to minimize the difference between the radial basis function neural network model predictions and experimental observations. Its mathematical expression is:

[0107]

[0108] In equation (4), pred_a tand pred_F t These are the time histories of cone tip drag and acceleration predicted by the radial basis function neural network model; tar_a t and tar_F t These are experimental observation data for cone tip drag and acceleration, respectively.

[0109] The NSGA-II algorithm, through non-dominated sorting, crowding calculation, and elite retention strategies, ensures the global optimality of the Pareto optimal solution set (i.e., the set containing all non-dominated individuals) formed after multi-objective genetic algorithm iterations, thereby accurately interpreting the ground mechanical property parameters. The specific process is as follows (e.g.) Figure 4 As shown):

[0110] S401. Initialize the population for the genetic algorithm.

[0111] An initial population is randomly generated, and each individual represents a set of soil surface mechanical property parameters (e.g., internal friction angle φ, cohesion c, elastic modulus E, Poisson's ratio ν, and density ρ, etc.); the genes of each individual are represented as parameter vector X = [φ, c, E, ν, ρ].

[0112] S402, Calculation of Objective Function

[0113] For each individual, the radial basis function neural network model is used to predict the cone tip drag and acceleration time history data, and the objective function value corresponding to that individual is calculated by the following equation (5):

[0114]

[0115] S403, Non-dominated sorting

[0116] Individuals in the population (including both the initial and subsequent generations) are ranked according to Pareto superiority, forming multiple non-dominant tiers. Individuals in lower tiers are superior to those in higher tiers.

[0117] S404, Congestion Calculation

[0118] To maintain population diversity, the crowding degree of each individual is calculated to measure its density in the target space; crowding degree is used to maintain population diversity, and the specific calculation formula is as follows:

[0119]

[0120] Among them, in the above formula (6) and It is the function value of the adjacent solution of the i-th solution on the j-th objective function. and These are the maximum and minimum values ​​of the j-th objective function, respectively.

[0121] S405, Elite Retention Strategy

[0122] In each generation of the genetic algorithm population, non-dominated solutions are preferentially retained, and individuals are selected in areas with high crowding to ensure that the population develops towards the Pareto front.

[0123] S406, Crossover and Mutation

[0124] New individuals are generated through crossover and mutation operations in genetic algorithms, and these new individuals will replace inferior individuals in subsequent generations of iteration.

[0125] S407 Termination Conditions

[0126] The algorithm terminates when the population reaches its maximum number of generations or when the convergence change of the objective function value is less than a preset threshold; the optimal ground mechanical property parameter vector of the soil is selected based on the decision criteria.

[0127] The results of the interpretation of ground mechanical property parameters are as follows:

[0128] The internal friction angle and cohesion of the soil were calibrated using indoor undrained triaxial experiments, and the results were compared with the interpretation results to verify the reliability of the interpretation method. The results show that the average interpretation error of the method under multiple conditions is less than 25%, and the parameter interpretation time for a single condition using a conventional computer is less than 1 minute.

[0129] Table 1 Interpretation Verification

[0130]

[0131] This invention can quickly and accurately interpret ground mechanical property parameters, including density, internal friction angle, cohesion, elastic modulus and Poisson's ratio. These parameters can be used to assess the mobility of equipment in complex geological environments, thereby improving the ability to cope with complex geological environments.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for interpreting ground mechanical property parameters based on impact penetration testing, characterized in that: Remote ground surveys are conducted using an unmanned aerial vehicle (UAV) carrying an impact penetrator. The dynamic response data collected by the penetrator is then interpreted using a radial basis function neural network model and a multi-objective genetic algorithm to quickly and accurately obtain ground mechanical property parameters. The specific steps include: (1) Data Acquisition First, an impact penetration test is conducted on the target ground. The UAV carrying the impact penetrator flies to the airspace above the target ground. After the UAV reaches the predetermined altitude, the impact penetrator is dropped into the ground in a free fall manner to measure the time history data of the cone tip drag and acceleration during the penetration process. Then, the data acquisition and processing module embedded in the impact penetrator performs preliminary data processing to obtain high-quality time history data of cone tip drag and acceleration. (2) Database Construction The Latin hypercube sampling method is used to generate ground mechanical property parameter samples in the soil mechanical parameter space. The CEL finite element method is used to simulate the parameters in the ground mechanical property parameter samples to generate dynamic response data of the penetrometer under different soil conditions, thereby constructing a database covering a wide range of soil conditions. (3) Radial basis neural network model training The radial basis function neural network model was trained using the constructed database to establish a mapping relationship between ground mechanical property parameters and penetrator dynamic response data; (4) Interpretation of Ground Mechanical Properties A radial basis function (RBF) neural network model is combined with a multi-objective genetic algorithm to achieve accurate interpretation of ground mechanical property parameters, including cohesion c, internal friction angle φ, elastic modulus E, Poisson's ratio ν, and density ρ. The specific process is as follows: using the difference between the cone tip drag and acceleration time history data predicted by the RBF neural network model and the experimental observation data as the objective function, the multi-objective genetic algorithm minimizes the objective function to determine and interpret the ground mechanical property parameters. To interpret the ground mechanical property parameters, an objective function needs to be established first to minimize the difference between the RBF neural network model predictions and experimental observations; its expression is: In equation (4), pred_a t and pred_F t These are the time histories of cone tip drag and acceleration predicted by the radial basis function neural network model; tar_a t and tar_F t These are experimental observation data for cone tip drag and acceleration, respectively. By employing non-dominated sorting, crowding calculation, and elite retention strategies, the global optimality of the Pareto optimal solution set formed after multi-objective genetic algorithm iteration is ensured, thereby accurately interpreting the ground mechanical property parameters. The specific interpretation process is as follows: (4.1) Initialize the population for the genetic algorithm An initial population is randomly generated, with each individual representing a set of ground mechanical property parameters; the genes of each individual are represented as a parameter vector X = [φ, c, E, ν, ρ]. (4.2) Calculation of the objective function For each individual, the radial basis function neural network model is used to predict the cone tip drag and acceleration time history data, and the objective function value corresponding to that individual is calculated by the following equation (5): (4.3) Non-dominated sorting Individuals in the population are ranked according to Pareto superiority to form multiple non-dominant ranks, with individuals in lower ranks being superior to those in higher ranks. (4.4) Crowding Calculation The crowding level of each individual is calculated to measure its density in the target space; crowding level is used to maintain population diversity, and the specific calculation formula is as follows: in, and It is the function value of the adjacent solution of the i-th solution on the j-th objective function. and These are the maximum and minimum values ​​of the j-th objective function, respectively; (4.5) Elite Retention Strategy In each generation of the population in the genetic algorithm, non-dominated solutions are retained first, and individuals are selected in areas with high crowding to ensure that the population develops towards the Pareto front. (4.6) Crossover and Mutation New individuals are generated through crossover and mutation operations in genetic algorithms, and these new individuals will replace the inferior individuals in subsequent generations of iteration. (4.7) Termination Conditions When the population reaches the maximum number of generations or the convergence change of the objective function value is less than a preset threshold, the genetic algorithm terminates and selects the optimal soil surface mechanical property parameter vector according to the decision criteria.

2. The method for interpreting ground mechanical property parameters based on impact penetration testing as described in claim 1, characterized in that, The specific process of generating samples of ground mechanical property parameters using the Latin hypercube sampling method in step (2) is as follows: (2.1) Determining the parameter range The range of values ​​for each ground mechanical property parameter is determined, resulting in a total of P parameters, forming a P-dimensional parameter space; (2.2) Divide the parameter range into equal parts The range of values ​​for each ground mechanical property parameter is divided into N equally spaced intervals along its dimension, that is, the range of values ​​corresponding to each ground mechanical property parameter is divided into N sub-intervals. (2.3) Randomly select sample values ​​for each interval For each dimension of ground mechanical property parameter, a value is randomly selected in each sub-interval to ensure that a sample value can be obtained for all intervals in that dimension. (2.4) Sample arrangement After randomly arranging N samples of each ground mechanical property parameter, P-dimensional samples are formed. The final generated sample points will cover the entire ground mechanical property parameter space, and the samples in each dimension will be evenly distributed.

3. The method for interpreting ground mechanical property parameters based on impact penetration testing as described in claim 1, characterized in that, The specific process of training the radial basis function neural network model using the constructed database in step (3) is as follows: (3.1) Input the ground mechanical property parameter sample points generated by Latin hypercube sampling into the input layer of the radial basis neural network model; (3.2) Using CEL finite element simulation, the dynamic response data of the penetrator is obtained by simulating the sample points of each group of ground mechanical property parameters, and these data are used as the output of the radial basis neural network model. (3.3) Activation function for constructing the structure of the radial basis function neural network model The structure of a radial basis function neural network model consists of an input layer, hidden layers, and an output layer. The hidden layers use radial basis functions as activation functions. In equation (1) above, r is the Euclidean distance from the input to the center of the hidden layer; σ is the width parameter of the Gaussian function.

4. The method for interpreting ground mechanical property parameters based on impact penetration testing as described in claim 1, characterized in that: The preliminary data processing in step (1) includes filtering and noise reduction.

5. The method for interpreting ground mechanical property parameters based on impact penetration testing as described in claim 1, characterized in that, The process of training the radial basis function neural network model in step (3) is as follows: the radial basis function neural network model takes ground mechanical property parameters as input and time history data of cone tip drag and acceleration as output. At the same time, in order to verify the accuracy of the radial basis function neural network model, a portion of samples are randomly selected from the database as the test set, and the remaining portion is used as the training set; normalized root mean square error and coefficient of determination R are introduced. 2 To evaluate the prediction accuracy of the radial basis function neural network (RBN) model, the following two evaluation metrics are used: One evaluation metric is the normalized root mean square error (RAAE): In equation (2) above, y pred,i It is the model prediction value, y ture,i y is the true value, n is the number of samples, and y is the true value. max and y min These are the maximum and minimum values ​​of the true value, respectively. Another evaluation metric is the coefficient of determination R. 2 : In equation (3) above, The mean of the true values; If the prediction accuracy of the radial basis function neural network model on the test set does not meet the requirements, increase the training data, adjust the structure and parameters of the radial basis function neural network model, until the prediction accuracy of the radial basis function neural network model reaches the expected standard.

Images