CPTU soil body layering method based on machine learning and uncertainty analysis

Abstract

The invention relates to a CPTU soil body layering method based on machine learning and uncertainty analysis, and the method comprises the following steps: S1, obtaining CPTU data of a marine soil body and soil body drilling data adjacent to a CPTU exploration position; s2, training a classification model; s3, inputting target position CPTU data into the classification model trained in the S2 to obtain a classification result, and evaluating the classification result by adopting a model accuracy evaluation index; s4, evaluating the confidence coefficient of each classification result in combination with uncertainty analysis to obtain an uncertainty analysis result; and S5, according to an uncertainty analysis result, determining the number of soil layering and a boundary position, and outputting a layering result. According to the method, CPTU high-resolution data and a machine learning algorithm are combined, and an automatic layering method for uncertainty analysis and prediction category confidence optimization is introduced to optimize a classification result, so that a soil layer division result with an engineering reference value is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of soil layer classification in marine engineering surveys, and particularly to a CPTU soil layer classification method based on machine learning and uncertainty analysis. Background Art

[0002] Marine surveys are indispensable in the fields of marine engineering, infrastructure construction, and seabed resource development. Accurate soil layer classification and categorization can provide necessary basis for design and construction. As an efficient means of marine soil exploration, the piezocone penetration test (CPTU) can quickly obtain high-resolution soil parameters by measuring parameters such as cone tip resistance, side friction resistance, and pore water pressure, providing important data support for soil layer classification and identification.

[0003] However, CPTU soil layer classification in marine soils faces more challenges. Due to the complexity and heterogeneity of marine soil layers, traditional CPTU classification methods (such as the Robertson chart) often cannot accurately determine soil types. In addition, although existing machine learning methods can perform soil type determination, the soil classification results obtained based on these determinations usually result in overly fine soil layer divisions and excessive numbers of layers, making it difficult to obtain engineering-effective and practically applicable layer division results. Summary of the Invention

[0004] In view of the deficiencies in the background art, the technical problem to be solved by the present invention is to provide a CPTU soil layer classification method based on machine learning and uncertainty analysis. This method combines CPTU high-resolution data and machine learning algorithms to effectively classify marine soils. By introducing uncertainty analysis, an automatic layer classification method that optimizes prediction class confidence is used to optimize the classification results, thereby obtaining layer division results with engineering reference value.

[0005] The present invention is achieved by adopting the following technical solutions: A CPTU soil layer classification method based on machine learning and uncertainty analysis, including the following steps:

[0006] S1. Obtain CPTU data of marine soils and soil drilling data adjacent to the CPTU exploration location to obtain soil types at some depths;

[0007] S2. Use the labeled data in the CPTU data in S1 to train a basic model, and combine the unlabeled data in the CPTU data to construct a semi-supervised learning model;

[0008] S3. Input the CPTU data at the target location into the classification model trained in S2 to obtain a classification result. Use model accuracy evaluation indicators to evaluate the classification result;

[0009] S4. Evaluate the confidence of each classification result in combination with uncertainty analysis, quantitatively analyze the uncertainty of the prediction results, and obtain the uncertainty analysis results;

[0010] S5. Determine the number and boundary positions of soil layers according to the uncertainty analysis results, and output the layering results.

[0011] Furthermore, the specific steps of S2 are as follows.

[0012] S21. Train a basic random forest classification model based on labeled data, input the unlabeled data into the model, and obtain the prediction results of each decision tree;

[0013] S22. Determine the predicted class as the class with the most votes through a voting mechanism;

[0014] S23. For each predicted sample, calculate the proportion of the votes of its predicted class in all decision trees, and use this proportion as the classification confidence. If the classification confidence exceeds a predetermined threshold, consider the prediction result reliable. If the confidence is lower than the threshold, skip this sample;

[0015] S24. Use the reliable prediction results as pseudo-labels and add them to the training data set to form a new training set;

[0016] S25. Through multiple rounds of iterative training, use the labeled data and the added pseudo-labeled data to gradually optimize the classification model and obtain a trained classification model.

[0017] Furthermore, the CPTU data includes cone tip resistance, side friction resistance, and pore water pressure parameters. The labeled data is CPTU data with corresponding soil types, and the unlabeled data is CPTU data without corresponding soil types.

[0018] Furthermore, when training the basic model with the labeled data of CPTU data in S2, the basic learner is a random forest, the input parameters are cone tip resistance, side friction resistance, and pore water pressure, and the output feature is the soil type.

[0019] Furthermore, the model accuracy evaluation indicators in S3 are accuracy, recall rate, and F1 score.

[0020] Furthermore, the hyperparameters of the semi-supervised algorithm include three hyperparameters: the depth of the random forest decision tree max_depth, the number of random forest decision trees n_estimators, and the confidence threshold threshold. The selection of the depth of the random forest decision tree max_depth and the number of random forest decision trees n_estimators is optimized through Bayesian hyperparameter tuning and cross-validation. Among them, the depth of the random forest decision tree max_depth and the number of random forest decision trees n_estimators are used in the supervised model random forest to control the depth and number of decision trees, and the confidence threshold threshold is used to control the generation of pseudo-labels.

[0021] Furthermore, in S5, by combining the predicted classification results in S3 and the uncertainty analysis results in S4, the optimal boundary position is automatically determined based on the predicted class confidence for automatic stratification to achieve soil stratification. After multiple iterations, the final soil stratification result is obtained.

[0022] Furthermore, the automatic stratification based on the optimization of the predicted class confidence includes the following steps:

[0023] 1) Divide the target area into two soil layers and traverse the candidate boundary positions;

[0024] 2) In each sub-interval, by calculating the mean confidence of each soil layer, and determining the soil layer category corresponding to the maximum confidence as the soil layer category of this layer;

[0025] 3) When the two stratification categories are different, the target value is the weighted mean of the overall confidence of this boundary position combination; when the stratification categories are the same, the target value is the difference in the mean confidence between the two intervals;

[0026] 4) Finally, select the boundary position corresponding to the maximum value of the objective function as the stratification result of this iteration.

[0027] Furthermore, parameters such as the maximum number of iterations, the minimum layer thickness, and the lowest confidence are set in the iterative process. When the maximum number of iterations is reached, the iteration stops. If the soil layer thickness of a certain interval is lower than the minimum layer thickness in a certain iteration, the subdivision of this interval is stopped and the current stratification result is retained. If the mean confidence of a certain soil layer is lower than the lowest confidence, it is regarded as an uncertain soil layer. After a specified number of iterations, the soil layers with the same adjacent soil layer categories are regarded as the same soil layer, and the overly thin layers and uncertain soil layers are removed to obtain the final soil stratification result.

[0028] In the present invention, by combining the high-resolution characteristics of CPTU data and a random forest classification model, marine soil classification based on CPTU data is achieved. The semi-supervised learning algorithm is used to fully utilize unlabeled data to enhance the robustness and prediction effect of the model, and uncertainty analysis is introduced for quantitative evaluation of classification results. Through an automatic stratification method optimized by confidence, the soil layer boundary position is accurately determined, and overly thin layers and uncertain soil layers are effectively removed, thereby outputting a stratification result with engineering reference value. This method can provide a scientific and reliable basis for soil layer division in marine engineering design and construction. Description of the Drawings

[0029] Figure 1 It is a flowchart of a CPTU soil stratification method based on machine learning and uncertainty analysis;

[0030] Figure 2 It is a schematic diagram of CPTU4 parameters in an embodiment of the present invention;

[0031] Figure 3 It is a schematic diagram of CPTU6 parameters in an embodiment of the present invention;

[0032] Figure 4 It is a schematic diagram of the soil classification results of CPTU4 and CPTU6 models in an embodiment of the present invention;

[0033] Figure 5 It is a schematic diagram of the uncertainty analysis results in an embodiment of the present invention;

[0034] Figure 6 It is a schematic diagram of the comparison between the stratification result and the drilling result in an embodiment of the present invention. Detailed Embodiment

[0035] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in conjunction with the drawings and preferred embodiments, details the specific embodiments, structures, features, and their effects of the present invention as follows.

[0036] Referring to Figure 1-6 As shown, the present invention provides a CPTU soil stratification method based on machine learning and uncertainty analysis, including the following steps:

[0037] S1. Obtain CPTU data of marine soil and soil drilling data adjacent to the CPTU exploration location to obtain the soil types at some depths. Among them, the CPTU data includes parameters such as cone tip resistance, side friction resistance, and pore water pressure.

[0038] S2. Train a basic model using the labeled data in S1 and construct a semi - supervised learning model by combining the unlabeled data of CPTU data. The specific steps of the learning method of the semi - supervised learning model are as follows:

[0039] S21. Train a basic random forest classification model based on the labeled data, and input the unlabeled data into this model to obtain the prediction results of each decision tree. Among them, the labeled data is the CPTU data with corresponding soil types, and the unlabeled data is the CPTU data without corresponding soil types. When training the basic random forest classification model with the labeled data of CPTU data, the basic learner is a random forest, the input parameters are tip resistance, side friction resistance, and pore water pressure, and the output feature is the soil type.

[0040] S22. Determine the predicted class as the class with the most votes through a voting mechanism.

[0041] S23. For each predicted sample, calculate the proportion of the votes of its predicted class in all decision trees, and use this proportion as the classification confidence. If the classification confidence exceeds a predetermined threshold, consider this prediction result as reliable; if the confidence is lower than this threshold, skip this sample.

[0042] S24. Use the reliable prediction results as pseudo - labels and add them to the training data set to form a new training set.

[0043] S25. Through multiple rounds of iterative training, use the labeled data and the added pseudo - label data to gradually optimize the classification model and obtain a trained classification model.

[0044] Among them, semi - supervised algorithm hyperparameters are used in the semi - supervised learning method. The semi - supervised algorithm hyperparameters include three hyperparameters: the depth of the random forest decision tree max_depth, the number of random forest decision trees n_estimators, and the confidence threshold threshold. The selection of hyperparameters is optimized through Bayesian hyperparameter tuning and cross - validation. Among them, the depth of the random forest decision tree max_depth and the number of random forest decision trees n_estimators are used in the supervised model random forest to control the depth of the decision tree and the number of decision trees. The confidence threshold threshold is used to control the generation of pseudo - labels.

[0045] S3. Input the CPTU data of the target location into the trained classification model in S2 to obtain the classification result. Use the model accuracy evaluation indicators to evaluate the classification result in S2, where the model accuracy evaluation indicators are accuracy, recall rate, and F1 - score.

[0046] S4. Evaluate the confidence of each classification result in combination with uncertainty analysis, and quantitatively analyze the uncertainty of the prediction results to obtain the uncertainty analysis results. Among them, the uncertainty analysis and evaluation refers to, for each classification result, calculating the proportion of the voting number of its predicted category in all decision trees, and taking this proportion as the classification confidence.

[0047] S5. Determine the number and boundary positions of soil layers according to the uncertainty analysis results, and output the layering results.

[0048] Specifically, combining the predicted classification results in S3 and the uncertainty analysis results in S4, the automatic layering method optimized based on the predicted category confidence automatically determines the optimal boundary position and realizes soil layering. After multiple iteration processes, the final soil layering results are obtained. Among them, parameters such as the maximum number of iterations, the minimum layer thickness, and the lowest confidence are set in the iteration process. When the maximum number of iterations is reached, the iteration stops. If the soil layer thickness in a certain interval is lower than the minimum layer thickness during a certain iteration, the subdivision of this interval is stopped, and the current layering results are retained. If the average confidence of a certain soil layer is lower than the lowest confidence, it is regarded as an uncertain soil layer. After the specified number of iterations, the soil bodies with the same adjacent soil body categories are regarded as the same soil layer, and the overly thin layers and uncertain soil layers are removed to obtain the final soil layering results.

[0049] The above automatic layering method optimized based on the predicted category confidence includes the following steps:

[0050] 1) Divide the target area into two soil layers and traverse the candidate boundary positions;

[0051] 2) In each sub-interval, by calculating the average confidence of each layer of soil body, and determining the soil body category corresponding to the maximum confidence as the soil body category of this layer;

[0052] 3) When the two layering categories are different, the target value is the weighted average of the overall confidence of this boundary position combination; when the layering categories are the same, the target value is the difference between the average confidences of the two intervals;

[0053] 4) Finally, select the boundary position corresponding to the maximum value of the objective function as the layering result of this iteration.

[0054] Example: Refer to Figure 2-6 As shown, for a certain marine tunnel project, the above CPTU soil layering method based on machine learning and uncertainty analysis is used, including the following steps:

[0055] S1. Seven groups of CPTU data obtained during the investigation stage of a certain marine tunnel project were collected. For each group of CPTU data, a set of cone tip resistance, side friction resistance, and pore water pressure were output at intervals of 0.01 m in the depth direction. Combining with the drilling results within 3 m around the CPTU investigation points, soil type information at some depths was obtained. The main soil types at the target site include silty clay, silt, and silty clay.

[0056] S2. A basic random forest classification model was trained based on the labeled data of CPTU data, and a semi-supervised learning model was constructed by combining the unlabeled data of CPTU data.

[0057] Specifically, the seven groups of CPTU data collected in S1 were distinguished as having labels or not, and a total of 2,635 groups of labeled data and 33,016 groups of unlabeled data were collected.

[0058] When training the basic random forest classification model with the labeled data, five groups of CPTU data (CPTU1, CPTU2, CPTU3, CPTU5, CPTU7) at the target site and their corresponding soil types were used for training, and the remaining two groups of CPTU data (CPTU4, CPTU6) were used as the target positions for model verification. (The parameters of CPTU4 and CPTU6 are as Figure 2 shown Figure 3 by

[0059] In the learning method of the semi-supervised learning model, three hyperparameters, namely the depth of the random forest decision tree max_depth, the number of random forest decision trees n_estimators, and the confidence threshold threshold, were used. The depth of the random forest decision tree max_depth and the number of random forest decision trees n_estimators were selected to be optimized through Bayesian hyperparameter tuning and cross-validation. Among them, in Bayesian hyperparameter tuning, the search range of the depth of the random forest decision tree max_depth was from 3 to 10, the search range of the number of random forest decision trees n_estimators was from 50 to 200, and the number of folds for cross-validation was 10 folds. The hyperparameter tuning results showed that the optimal hyperparameter combination was max_depth = 3 and n_estimators = 187. At the same time, according to experience, threshold = 0.90 was set.

[0060] Finally, through multiple rounds of iterative training, a total of 1,052 groups of unlabeled data were regarded as pseudo-labels with high confidence and added to the training set, and finally an optimized classification model was formed to obtain a trained classification model.

[0061] S3. The CPTU data at the target position was input into the trained classification model in S3 to obtain the classification results. The classification results in S2 were evaluated using the model accuracy evaluation index.

[0062] The prediction performance of the model is shown in Table 1. The data in the table show that the model has the best classification performance in the silty clay category, with an accuracy and recall rate of 0.84 each, and an F1 score of 0.86. For the silt category, due to the small sample size, the classification performance is relatively weak, with an accuracy of 0.63 and an F1 score of 0.71. Generally speaking, the average accuracy and recall rate of the model in all categories are 0.81, indicating good classification performance.

[0063] Table 1

[0064] Soil type Number of samples Accuracy Recall F1 score Silty silty clay 214 0.84 0.74 0.79 Silt 155 0.63 0.81 0.71 Silty clay 519 0.84 0.84 0.86 Overall 888 0.82 0.81 0.81

[0065] S4. Evaluate the confidence of each classification result in combination with uncertainty analysis, and quantitatively analyze the uncertainty of the prediction results.

[0066] The classification accuracy of the model in different confidence intervals is shown in Table 2. The results show that there is a significant correlation between the model prediction confidence and the actual accuracy. The higher the confidence of the prediction sample, the higher its classification accuracy. This further verifies the effectiveness of uncertainty analysis in the evaluation of soil classification results. In the classification results at two groups of target positions (as Figure 4 shown), the preliminary stratification results along the depth direction are relatively detailed, with more soil layers and poor distribution regularity. This indicates that the classification results need to be further optimized through uncertainty analysis and post-processing to obtain more valuable stratification results for engineering applications. Through uncertainty analysis (as Figure 5 shown), the credibility of the classification results can be further understood. Taking CPTU4 as an example, its shallow soil layer is mainly composed of silt with high confidence, while the shallow layer of CPTU6 is mainly composed of silty clay with high confidence. This confidence-based analysis not only improves the interpretability of the classification results but also provides support for subsequent stratification optimization.

[0067] Table 2

[0068] Confidence interval Number of samples Accuracy 0.9 to 1.0 10 1.00 0.7 to 0.9 531 0.92 0.5 to 0.7 310 0.70 0 to 0.5 46 0.26

[0069] S5. Combine the classification results with uncertainty analysis, and an automatic stratification method optimized based on the prediction category confidence automatically determines the best boundary position and realizes soil stratification. After multiple iteration processes, the final soil stratification result is obtained.

[0070] The maximum number of iterations in the iteration process is set to 3, the minimum layer thickness is 0.5 meters, and the minimum confidence is 0.4. During the stratification processes of CPTU4 and CPTU6, the stratification depth range, soil category, objective function, and average confidence in each iteration are shown in Tables 3 and 4.

[0071] Table 3

[0072]

[0073] Table 4

[0074]

[0075] After a specified number of iterations, the soil layers with adjacent layers of the same soil type are regarded as the same soil layer, and the overly thin layers and uncertain soil layers are removed to obtain the final soil layer classification result.

[0076] In the embodiment of the present invention, the soil layer classification result of CPTU4 is as follows: from a depth of 9.30 m to 19.50 m, it is all determined as silt and regarded as the same soil layer; from a depth of 19.50 m to 20.39 m, it is determined as silty clay, but due to the low confidence level, it is regarded as an uncertain soil layer and removed; from a depth of 20.39 m to 28.75 m, it is re-determined as silt; from a depth of 28.76 m to 68.30 m, it is determined as silty clay and regarded as a complete soil layer. Based on the above analysis, this method divides the soil at the location of CPTU4 into two main soil layers, namely the silt layer from a depth of 9.30 m to 28.76 m and the silty clay layer from a depth of 28.76 m to 68.30 m. Similarly, in the soil layer classification of CPTU6, a total of three soil layers are identified: the silt layer from a depth of 4.00 m to 6.42 m, the silty clay with silt layer from a depth of 6.42 m to 27.66 m, and the silty clay layer from a depth of 27.66 m to 72.00 m. By comparing with the actual drilling results (as Figure 6 shown), it can be seen that: the classification result of CPTU4 is highly consistent with the drilling result, and the boundary position of the soil layer is accurate, verifying the reliability and accuracy of this method. While CPTU6 is correctly judged at most positions, only misjudged as a silt layer in the shallower depth section. Generally speaking, this method can effectively realize soil layer classification, show good classification performance, and generate a more valuable stratification result for engineering by removing uncertain layers and overly thin layers.

[0077] The above is only a preferred embodiment of the present invention, and it is not intended to limit the present invention in any form. Although the present invention has been disclosed as above with the preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to form an equivalent embodiment within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A CPTU soil stratification method based on machine learning and uncertainty analysis, characterized by: The steps include: S1. Obtain CPTU data of marine soil and soil drilling data near the CPTU survey location to obtain soil types at some depths; S2. Use the labeled data of CPTU data in S1 to train the basic model, and build a semi-supervised learning model with the unlabeled data of CPTU data; S3. Input the target position CPTU data into the classification model trained in S2 to obtain the classification result, and evaluate the classification result using the model accuracy evaluation index; S4. Combine uncertainty analysis to evaluate the confidence of each classification result, and quantitatively analyze the uncertainty of the prediction result to obtain the uncertainty analysis result; S5. According to the uncertainty analysis results, determine the number and boundary positions of soil stratification and output the stratification results.

2. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 1 is characterized by: The specific steps of S2 are as follows: S21. Based on the labeled data, a basic random forest classification model is trained, and the unlabeled data is input into the model to obtain the prediction results of each decision tree; S22. through a voting mechanism, determining the predicted category as the category with the most votes; S23. For each predicted sample, calculate the proportion of votes of its predicted category in all decision trees, and use the proportion as the classification confidence. If the classification confidence exceeds a predetermined threshold, the prediction result is considered reliable. If the confidence is lower than the threshold, skip the sample. S24. Use the reliable prediction results as pseudo labels and add them to the training data set to form a new training set; S25. Through multiple rounds of iterative training, using labeled data and added pseudo-labeled data, the classification model is gradually optimized to obtain a trained classification model.

3. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 1 is characterized by: The CPTU data include cone tip resistance, side friction resistance and pore water pressure parameters. The labeled data are CPTU data with corresponding soil categories, and the unlabeled data are CPTU data without corresponding soil categories.

4. A CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 1, 2 or 3, characterized in that: S2 When the basic model is trained with labeled data of CPTU data in , the basic learner is a random forest, the input parameters are cone tip resistance, lateral friction resistance and pore water pressure, and the output feature is soil category.

5. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 1 is characterized by: The model accuracy evaluation indicators in S3 are accuracy, recall and F1 score.

6. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 2 is characterized by: The semi-supervised algorithm hyperparameters include three hyperparameters: the depth of random forest decision tree max_depth, the number of random forest decision trees n_estimators, and the confidence threshold threshold. The depth of random forest decision tree max_depth and the number of random forest decision trees n_estimators are optimized through Bayesian parameter adjustment and cross-validation. The depth of random forest decision tree max_depth and the number of random forest decision trees n_estimators are used in the supervised model random forest to control the depth and number of decision trees. The confidence threshold threshold is used to control the generation of pseudo labels.

7. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 1 is characterized by: In S5, the prediction classification results in S3 and the uncertainty analysis results in S4 are combined, and the automatic stratification based on the prediction category confidence optimization automatically determines the optimal boundary position and realizes soil stratification. After multiple iterations, the final soil stratification result is obtained.

8. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 7 is characterized by: The automatic stratification based on prediction category confidence optimization includes the following steps: 1) Divide the target area into two soil layers and traverse the candidate boundary positions; 2) In each sub-interval, the confidence mean of each layer of soil is calculated, and the soil category corresponding to the maximum confidence is determined as the soil category of this layer; 3) When the two stratification categories are different, the target value is the weighted mean of the overall confidence of the boundary position combination; when the stratification categories are the same, the target value is the difference between the confidence means of the two intervals; 4) Finally, the boundary position corresponding to the maximum value of the objective function is selected as the stratification result of this iteration.

9. The CPTU soil stratification method based on machine learning and uncertainty analysis according to claim 7 is characterized by: The iterative process sets parameters such as the maximum number of iterations, the minimum layer thickness and the minimum confidence level. When the maximum number of iterations is reached, the iteration stops. If the soil layer thickness of a certain interval in a certain iteration is lower than the minimum layer thickness, the subdivision of the interval is stopped and the current stratification result is retained. If the confidence mean of a certain soil layer is lower than the minimum confidence level, it is regarded as an uncertain soil layer. After the specified number of iterations, adjacent layers of the same soil category are regarded as the same soil layer, and the thin layers and uncertain soil layers are eliminated to obtain the final soil stratification result.

Images