A method for obtaining soil strength index based on soil layer geophysical data

Abstract

The application discloses a method for obtaining soil strength indexes based on soil layer geophysical prospecting data, and comprises the following steps: S1, obtaining the test depth and equivalent shear wave velocity of different measuring points in a target site; obtaining the drilling samples of each measuring point at the corresponding depth, obtaining the corresponding internal friction angle and soil type; preprocessing the collected data to obtain a comprehensive data set; S2, selecting the test depth and equivalent shear wave velocity in the comprehensive data set as basic features, constructing derived features, performing feature standardization, and obtaining a data set; dividing the data set into a training set and a test set; S3, based on the training set obtained in S2, optimizing and training the soil classification model and the internal friction angle prediction model; S4, obtaining the soil strength index internal friction angle based on the soil classification model and the internal friction angle prediction model obtained by the training in S3. The method of the application realizes high-precision prediction of soil strength indexes by constructing derived features and introducing physical range constraints.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering investigation, specifically to a method for obtaining soil strength indicators based on soil geophysical data. Background Technology

[0002] In marine engineering, due to the complex and variable natural conditions, determining geotechnical engineering parameters and making reasonable designs often become the main challenges in engineering construction. Soil strength indicators are the core basis for calculating the bearing capacity of marine geotechnical engineering foundations and assessing construction safety. Traditional methods of obtaining these indicators rely on borehole sampling and laboratory testing, which have inherent drawbacks compared to in-situ testing: First, borehole sampling inevitably disturbs the soil, affecting the accuracy of laboratory test results; second, test results based on soil sample units cannot continuously reflect the soil condition. Most importantly, this method is costly in terms of manpower and resources, and has a long testing cycle, making it difficult to meet the needs of rapid site surveys in complex environments.

[0003] In recent years, with the rapid development of machine learning, applying machine learning methods to the field of intelligent geotechnical engineering exploration has become a worthwhile direction to explore. Compared with traditional empirical formulas that can only describe linear or simple nonlinear relationships, machine learning models can capture the complex nonlinear mapping between geophysical data and soil strength indicators, improving the prediction accuracy of complex data. At the same time, machine learning has the ability to fuse and analyze multi-source data, integrating different types of geophysical parameters to achieve comprehensive prediction. Furthermore, after model training, it can quickly output prediction results, significantly improving the efficiency of engineering exploration.

[0004] However, existing applications still have technical shortcomings. On the one hand, the feature engineering is too simple, relying solely on the direct input of the original geophysical parameters without combining geotechnical mechanics theory to explore the physical relationship between the parameters, resulting in a lack of interpretability of the model. On the other hand, the physical constraints of soil mechanics are not embedded in the model training process, and the prediction results may exceed the reasonable range of engineering, which greatly reduces the practicality of engineering. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method for obtaining soil strength indicators based on soil geophysical data, thereby achieving high-precision prediction of soil strength indicators.

[0006] Therefore, the present invention adopts the following technical solution:

[0007] A method for obtaining soil strength indicators based on soil geophysical exploration data, the specific steps of which are as follows:

[0008] S1, Data Acquisition and Preprocessing, including:

[0009] S11, conduct shear wave velocity tests on the target site to obtain geophysical data at different measuring points at each site, including test depth and equivalent shear wave velocity; obtain borehole samples at corresponding depths at each measuring point, conduct indoor triaxial test data on the borehole samples, and obtain the corresponding internal friction angle; conduct soil particle analysis tests on the borehole samples to obtain the corresponding soil type.

[0010] S12, preprocesses the data collected in S11 to obtain the preprocessed comprehensive dataset. ;

[0011] S2, Selecting the comprehensive dataset Using the test depth and equivalent shear wave velocity as basic features, derived features are constructed, and feature standardization is performed to obtain the dataset. ; the dataset The dataset is divided into a training set and a test set; the derived features include depth-wave velocity coupling features and wave velocity-depth ratio features.

[0012] S3, based on the training set obtained in S2, optimizes and trains the soil classification model and the internal friction angle prediction model.

[0013] S4, based on the soil classification model and internal friction angle prediction model trained by S3, obtains the soil strength index internal friction angle.

[0014] In step S12 above:

[0015] The preprocessing includes data merging, outlier removal, and missing value handling, wherein:

[0016] (1) Data merging: The geophysical data is merged with the corresponding soil type and internal friction angle in a unified format to construct a comprehensive dataset. :

[0017] ,

[0018] in, The sample number. The total number of samples; For the first Soil type for each sample; For the first The test depth of each sample. For the first Equivalent shear wave velocity of each sample For the first The internal friction angle of each sample;

[0019] (2) Outlier removal: using 3 Criteria identify and eliminate the comprehensive dataset The formula for abnormal samples in the data is:

[0020] ,

[0021] in, For the parameter values ​​of the sample, This is the average of this parameter across all samples. The standard deviation of this parameter for all samples;

[0022] (3) Handling missing values: If the comprehensive dataset The Middle Missing parameters for each sample KNN interpolation method is used for filling. The calculation formula is:

[0023] ,

[0024] in, For distance from the first The parameter values ​​of the nearest sample to the current sample. .

[0025] In step S12 above: =3.

[0026] In step S2 above:

[0027] The formula for the depth-wave velocity coupling feature is:

[0028] ,

[0029] The formula for the wave velocity-depth ratio characteristic is:

[0030] .

[0031] The specific operation of feature standardization in step S2 above is as follows:

[0032] The basic and derived features are standardized using StandardScaler, with the following formula:

[0033] ,

[0034] in, To test the mean values ​​of depth, equivalent shear wave, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics, The standard deviation of the test depth, equivalent shear wave, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics.

[0035] In step S2 above: the dataset The dataset is divided into training and test sets in an 8:2 ratio.

[0036] In step S3 above: the soil classification model is the XGBoost classification model; the internal friction angle prediction model is the random forest regression model.

[0037] In step S3 above:

[0038] During the training process of the random forest regression model, a physical range constraint on the internal friction angle is added. For different soil types, the physical range of the internal friction angle is set separately. The improved loss function is as follows:

[0039] ,

[0040] in, This is the actual internal friction angle. The internal friction angle predicted by the model. The penalty coefficient is... This represents the lower limit of the physical range of the internal friction angle. This represents the lower limit of the physical range of the internal friction angle.

[0041] The physical range of the internal friction angle mentioned in step S3 above is as follows: cohesive soil 0°~30°; sandy soil 30°~35°; coarse sand and gravel 35°~40°; gravelly soil 40°~45°; gravel 45°~50°.

[0042] The specific operation of step S4 above is as follows:

[0043] Using the test depth and equivalent shear wave velocity of the measuring points obtained in engineering applications as basic features, we further obtain the depth-wave velocity coupling feature and the wave velocity-depth ratio feature, and standardize the test depth, equivalent shear wave velocity, depth-wave velocity coupling feature and wave velocity-depth ratio; use the soil classification model trained by S3 to classify the soil type of the measuring points.

[0044] Based on the classification results and standardized test depth, equivalent shear wave velocity, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics, the internal friction angle prediction model trained by S3 is used to predict the soil strength index, internal friction angle.

[0045] In summary, this invention proposes a method for obtaining soil strength indicators based on soil geophysical data. This method can achieve non-destructive, high-precision, and interactive on-site prediction of soil strength indicators.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. Based on the correlation between soil mechanics and geophysical data, two types of core derived features are constructed based on basic characteristics to explore the physical correlation between parameters and improve the prediction accuracy of the model.

[0048] 2. The internal friction angle, a strength indicator, was added during model training. The physical range constraint is defined by setting a physical range for the internal friction angle for different soil types, and a penalty is imposed for exceeding the internal friction angle by a loss function regularization term. The prediction results for the physical scope further improved the accuracy of the model's predictions. Attached Figure Description

[0049] Figure 1 This is a flowchart of a method for obtaining soil strength indicators based on soil geophysical data in an embodiment of the present invention;

[0050] Figure 2 This is a comparison chart of the prediction results of 25 typical measuring points obtained based on the internal friction angle prediction model in this embodiment of the invention and the correct results. Detailed Implementation

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the following embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0052] Example

[0053] This example uses three sites along the Shenyang-Baihe Railway during the preliminary survey phase as research objects to provide a detailed explanation of the method for obtaining soil strength indicators based on soil geophysical exploration data, as described in this invention. The geological conditions of the three sites are continuous and representative.

[0054] The specific steps are as follows:

[0055] S1, Data Acquisition and Preprocessing, including:

[0056] S11, Data Acquisition:

[0057] Based on the "Code for Seismic Design of Railway Engineering", the single-hole layer detection method was adopted, the seismic source was the tapping method, and an imported NZ72 seismograph and a three-component geophone were used to receive the signal. The transverse spacing of the measuring points was 1.0m. The longitudinal shear wave velocity test was carried out uniformly on three sites, and the geophysical data of different measuring points at each site were obtained, including the test depth and equivalent shear wave velocity, as shown in Table 1.

[0058] Table 1

[0059]

[0060] In addition, borehole samples at corresponding depths are obtained at each measuring point, and indoor triaxial test data are performed on them to obtain the corresponding internal friction angle; soil particle analysis tests are performed on the borehole samples to obtain the corresponding soil type.

[0061] S12, Data preprocessing, including:

[0062] (1) Data merging: The geophysical data is merged with the corresponding soil type and internal friction angle in a unified format to construct a comprehensive dataset. :

[0063] ,

[0064] in, The sample number. The total number of samples; For the first Soil type for each sample; For the first The test depth of each sample. For the first Equivalent shear wave velocity of each sample For the first The internal friction angle of each sample.

[0065] In this embodiment, a comprehensive dataset is obtained from geophysical data and triaxial test data from various sites. The study included 96 valid samples, covering shear wave velocities from 100 to 1863.9 m / s and depths from 0.0 to 40.0 m, indicating strong data representativeness.

[0066] (2) Outlier removal: using 3 Criteria for identifying and eliminating aggregate datasets The formula for abnormal samples in the data is:

[0067] ,

[0068] in, For the parameter values ​​of the sample, This is the average of this parameter across all samples. The standard deviation of this parameter is for all samples.

[0069] In this embodiment, the comprehensive dataset The equivalent shear wave velocity and internal friction angle in the data are both within a reasonable range, and they are not considered abnormal samples, so no data was removed.

[0070] (3) Handling missing values: If the comprehensive dataset The Middle Missing parameters for each sample The KNN interpolation method was used to fill in the missing values. The calculation formula is:

[0071] ,

[0072] in, For distance from the first The parameter values ​​of the nearest sample to the current sample. In engineering data processing, =3.

[0073] In this embodiment, the data from the three sites are complete, so there is no need for interpolation, and all 96 sets of samples are directly retained.

[0074] S2, construct derived features and partition the dataset, including:

[0075] S21, Select the comprehensive dataset The test depth and equivalent shear wave velocity are used as basic features. Based on the correlation between soil mechanics and geophysical data, two types of derived features are constructed to explore the physical correlation between parameters and improve the model's prediction accuracy. These two types of derived features include:

[0076] Deep-wave velocity coupling characteristics:

[0077] ,

[0078] Wave velocity-depth ratio characteristics:

[0079] ,

[0080] S22, Feature Standardization:

[0081] The basic and derived features are standardized using StandardScaler to eliminate the influence of dimensions. The standardization formula is:

[0082] ,

[0083] in, To test the mean values ​​of depth, equivalent shear wave, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics, The standard deviation of the test depth, equivalent shear wave, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics.

[0084] In this embodiment, all features after feature standardization meet the distribution requirements of mean 0 and variance 1, thus adapting to the input format of the model.

[0085] After standardization, a dataset containing the basic features and derived features is obtained. In this embodiment, the dataset It contains a total of 96 samples.

[0086] S23, Dataset Partitioning: Dividing the dataset The samples were divided into training and test sets in an 8:2 ratio to ensure consistent sample distribution and improve the model's generalization ability with small sample sizes. The training set consisted of 77 samples, and the test set consisted of 19 samples. During the partitioning, it was ensured that the training and test sets maintained consistency in soil distribution and the range of internal friction angles.

[0087] S3, based on the training set obtained in S23, optimizes and trains the soil classification model and the internal friction angle prediction model, including:

[0088] S31, Train the soil classification model based on the training set obtained in S23. In this embodiment, the soil classification model is an XGBoost classification model. Save the trained soil classification model as a reusable file.

[0089] The formula for evaluating the classification accuracy of a soil classification model is:

[0090] ,

[0091] in, For indicator functions, The number of samples in the test set. The first prediction of the soil classification model Soil type of each sample For the test set The actual soil type of each sample;

[0092] The soil classification model was tested using the test set obtained in S23. In this embodiment, the classification accuracy of the XGBoost classification model was [value missing]. That is, 18 out of 19 test samples were correctly identified.

[0093] S32, based on the training set obtained from S23, trains the internal friction angle prediction model through physical constraint embedding.

[0094] In this embodiment, the internal friction angle prediction model is a random forest regression model. During model training, a physical range constraint on the internal friction angle is incorporated. For different soil types, reasonable physical ranges for the internal friction angle are set. The loss function's regularization term penalizes predictions that exceed the physical range. The improved loss function is as follows:

[0095]

[0096] in, This is the actual internal friction angle. The internal friction angle predicted by the model. The penalty coefficient is... This represents the lower limit of the physical range of the internal friction angle. This represents the lower limit of the physical range of the internal friction angle.

[0097] The internal friction angle of soil under normal density conditions is constrained by the following physical ranges: cohesive soil 0°~30°; sandy soil 30°~35°; coarse sand and gravel 35°~40°; gravelly soil 40°~45°; gravel 45°~50°.

[0098] In this embodiment, a penalty coefficient is set based on all soil types involved in the three sites. .

[0099] Save the trained friction angle prediction model as a reusable file.

[0100] The evaluation metrics for assessing the performance of the internal friction angle prediction model are as follows:

[0101] Mean absolute error ( ):

[0102] ,

[0103] Root mean square error (RMSE) ):

[0104] ,

[0105] Coefficient of determination ( ):

[0106] ,

[0107] in, The mean internal friction angle of the test set.

[0108] The internal friction angle prediction model was tested using the test set obtained in S23. In this embodiment, the random forest regression model... , , The smaller the mean absolute error and root mean square error, the better; the larger the coefficient of determination, the better.

[0109] S4, based on the soil classification model and internal friction angle prediction model trained by S3, obtains the soil strength index internal friction angle.

[0110] Using the test depth and equivalent shear wave velocity of the measuring points obtained in engineering applications as basic features, we further obtain the depth-wave velocity coupling feature and the wave velocity-depth ratio feature, and standardize the test depth, equivalent shear wave velocity, depth-wave velocity coupling feature and wave velocity-depth ratio; use the soil classification model trained by S3 to classify the soil type of the measuring points.

[0111] Then, based on the classification results and the standardized test depth, equivalent shear wave velocity, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics, the internal friction angle prediction model trained by S3 is used to complete the prediction of the soil strength index, the internal friction angle.

[0112] In this embodiment, 25 representative measurement points from three sites along the same route that were not part of the training were selected. The test depth and equivalent shear wave velocity were input, and the predicted results were compared with the correct results. Figure 2 As shown, the method for obtaining soil strength indicators in this invention has a high accuracy rate.

[0113] Comparative Example 1

[0114] The soil classification model in the example was changed to a random forest classification model, and the accuracy was tested using a test set. =84.2%, meaning that 16 out of 19 test samples were correctly identified.

[0115] Comparative Example 2

[0116] The soil classification model in the example was changed to a support vector machine classification model, and the classification accuracy was tested using a test set. =78.9%, meaning that 15 out of 19 test samples were correctly identified.

[0117] Comparative Example 3

[0118] The internal friction angle prediction model in the example was changed to an XGBoost regression model, and tested using a test set. , , .

[0119] Comparative Example 4

[0120] The internal friction angle prediction model in the example was changed to a support vector machine regression model, and tested using a test set. , , .

[0121] Comparative Example 5

[0122] The derived features in the examples are removed, and the dataset contains only the basic features. The XGBoost classification model and the random forest regression model are trained. Then, the XGBoost classification model is tested using the test set. =65.0%; The random forest regression model was tested using the test set. , , .

[0123] In summary, by comparing the embodiments and comparative examples, it can be seen that in this invention, when data containing derived features and basic features is used for training, and the soil classification model is the XGBoost classification model and the internal friction angle prediction model is the random forest regression model, the prediction effect is better.

Claims

1. A method for obtaining soil strength indicators based on soil geophysical exploration data, characterized in that, The specific steps are as follows: S1, Data Acquisition and Preprocessing, including: S11, conduct shear wave velocity tests on the target site to obtain geophysical data at different measuring points at each site, including test depth and equivalent shear wave velocity; obtain borehole samples at corresponding depths at each measuring point, conduct indoor triaxial test data on the borehole samples, and obtain the corresponding internal friction angle; conduct soil particle analysis tests on the borehole samples to obtain the corresponding soil type. S12, preprocesses the data collected in S11 to obtain the preprocessed comprehensive dataset. ; S2, Selecting the comprehensive dataset Using the test depth and equivalent shear wave velocity as basic features, derived features are constructed, and feature standardization is performed to obtain the dataset. ; the dataset The dataset is divided into a training set and a test set; the derived features include depth-wave velocity coupling features and wave velocity-depth ratio features. S3, based on the training set obtained in S2, optimizes and trains the soil classification model and the internal friction angle prediction model. S4, based on the soil classification model and internal friction angle prediction model trained by S3, obtains the soil strength index internal friction angle.

2. The method according to claim 1, characterized in that, In step S12: The preprocessing includes data merging, outlier removal, and missing value handling, wherein: (1) Data merging: The geophysical data is merged with the corresponding soil type and internal friction angle in a unified format to construct a comprehensive dataset. : ， in, The sample number. The total number of samples; For the first Soil type for each sample; For the first Test depth of each sample, For the first Equivalent shear wave velocity of each sample For the first The internal friction angle of each sample; (2) Outlier removal: using 3 Criteria identify and eliminate the comprehensive dataset The formula for abnormal samples in the data is: ， in, For the parameter values ​​of the sample, This is the average of this parameter across all samples. The standard deviation of this parameter for all samples; (3) Handling missing values: If the comprehensive dataset The Middle Missing parameters for each sample KNN interpolation method is used for filling. The calculation formula is: ， in, For distance from the first The parameter values ​​of the nearest sample to the current sample. .

3. The method according to claim 2, characterized in that, In step S12: =3.

4. The method according to claim 2, characterized in that, In step S2: The formula for the depth-wave velocity coupling feature is: ， The formula for the wave velocity-depth ratio characteristic is: 。 5. The method according to claim 4, characterized in that, The specific operation of feature standardization in step S2 is as follows: The basic and derived features are standardized using StandardScaler, with the following formula: ， in, To test the mean values ​​of depth, equivalent shear wave, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics, The standard deviation of the test depth, equivalent shear wave, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics.

6. The method according to claim 5, characterized in that, In step S2: the dataset The dataset is divided into training and test sets in an 8:2 ratio.

7. The method according to claim 6, characterized in that, In step S3: the soil classification model is the XGBoost classification model; the internal friction angle prediction model is the random forest regression model.

8. The method according to claim 7, characterized in that, In step S3: During the training process of the random forest regression model, a physical range constraint on the internal friction angle is added. For different soil types, different physical ranges for the internal friction angle are set. The improved loss function is as follows: ， in, This is the actual internal friction angle. The internal friction angle predicted by the model. The penalty coefficient is... This represents the lower limit of the physical range of the internal friction angle. This represents the lower limit of the physical range of the internal friction angle.

9. The method according to claim 8, characterized in that, The physical range of the internal friction angle mentioned in step S3 is as follows: cohesive soil 0°~30°; sandy soil 30°~35°; coarse sand and gravel 35°~40°; gravelly soil 40°~45°; gravel 45°~50°.

10. The method according to claim 8, characterized in that, The specific operation of step S4 is as follows: Using the test depth and equivalent shear wave velocity of the measuring points obtained in engineering applications as basic features, we further obtain the depth-wave velocity coupling feature and the wave velocity-depth ratio feature, and standardize the test depth, equivalent shear wave velocity, depth-wave velocity coupling feature and wave velocity-depth ratio; use the soil classification model trained by S3 to classify the soil type of the measuring points. Based on the classification results and standardized test depth, equivalent shear wave velocity, depth-wave velocity coupling characteristics, and wave velocity-depth ratio characteristics, the internal friction angle prediction model trained by S3 is used to predict the soil strength index, internal friction angle.

Images