Deep mixed artificial intelligence-based method for predicting compressibility coefficient of soft soil of beach-overtopping facies

Abstract

The invention discloses a deep mixing artificial intelligence-based method for predicting the compression coefficient of overtoll phase soft soil, which comprises the following steps of: acquiring test data obtained by testing an overtoll phase soft soil sample, generating a data set by utilizing the test data, and establishing an overtoll phase soft soil compression coefficient prediction model by utilizing a CNN-CatBoost deep mixing model. And inputting the data set into an overtopping facies soft soil compression coefficient prediction model for training, and carrying out interpretive analysis on the overtopping facies soft soil compression coefficient prediction model. According to the method, data features can be automatically learned and extracted in a complex environment, the superiority of the prediction capability can be proved by comparing with an actual value, the black box problem of a deep learning model is solved through interpretability analysis, and the influence rule of each feature on the influence output variable is obtained.

Description

Technical Field

[0001] The present invention relates to the field of floodplain soft soil performance research and digital modeling, and in particular to a floodplain soft soil compression coefficient prediction method based on deep hybrid artificial intelligence. Background Art

[0002] There are a large number of floodplain soft soils distributed on both sides of the Yangtze River. These soft soils usually have the characteristics of high water content, high compressibility, high sensitivity, low density, low strength, expansibility, poor consolidation properties, high organic matter content, low viscosity and friction coefficient of the soil, which brings challenges of insufficient bearing capacity and differential foundation settlement to geotechnical engineers.

[0003] Understanding the compression deformation characteristics of soft soil is helpful for accurately calculating the settlement and designing a reasonable foundation structure. The compression coefficient can be obtained through indoor tests (one-dimensional consolidation tests) and in-situ tests (static penetration tests). Traditional methods for obtaining compression parameters rely on indoor and field experiments, which may lead to a waste of cost and time, thus affecting the progress of the project. In addition, the experimental results are easily affected by factors such as test equipment and improper transportation. Therefore, it is of great significance to develop a model with strong generalization ability to predict the compression characteristics of floodplain soft soil.

[0004] With the rapid development of computer technology and data science, machine learning methods have gradually become a new and potential tool in the field of geotechnical engineering for the efficient prediction of soil mechanical properties, such as ANN (Artificial Neural Network), RF (Random Forest) and SVM (Support Vector Machine) algorithms, MLP (Multilayer Perceptron) neural network models optimized by PSO (Particle Swarm Optimization) meta-heuristic algorithm, spatial-geological stratigraphic maps and SGGAT (Spatial-Geological Gragh Attention Network) solutions, and ANN-GMDH (ANN-Group Method of Data Handling) models.

[0005] However, soft soil in the natural environment can be regarded as a complex system. The interactions of its components (such as soil moisture content, porosity and particle density) at different levels jointly affect the compression characteristics of soft soil. Previous studies usually only considered the prediction effects of a small number of models, lacked a comprehensive evaluation framework combined with experiments, and the internal mechanism of the model was insufficiently explainable, which may lead to contradictions between the prediction results and objective laws.

[0006] Therefore, how to enhance the fitting and generalization ability of the floodplain soft soil compression coefficient prediction model, improve the prediction accuracy and reduce the error, so as to help engineers better understand the mechanical properties of soft soil, avoid excessive or uneven settlement of the foundation, and ensure the stability and safety of the building, is a technical problem that needs to be solved urgently. Summary of the invention

[0007] In order to address the shortcomings of the prior art, the present invention provides a method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence, collects sample data, constructs a CNN-CatBoost (CNN-categorical boosting, convolutional neural network-classification boosting method) hybrid model, trains and optimizes the model, and finally uses SHAP (Shapley Additive Explanations, global explanatory analysis) and PDP (Partial Dependence Plot, partial dependence plot) to analyze the importance of feature variables and reveal the influence of feature variables on model output.

[0008] The embodiment of the present invention provides the following solution:

[0009] The embodiment of the present invention provides a method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence, the method comprising:

[0010] Step 1: Obtain test data of soft soil samples in floodplain phase after testing;

[0011] Step 2: Generate a data set using the test data. Each sample in the data set includes influencing factors and influencing targets. Influencing factors include soil specific gravity G s , moisture content w, wet density ρ, dry density porosity ρ d , Saturation S r 、Liquid limit w L , plastic limit w P , liquid index I L , Plasticity Index I p And the depth of soil sample H, the influencing target is the compression coefficient a; select the depth of soil sample H, water content w, plastic limit w from the influencing factors P , Plasticity Index I p Liquidity index I L As output variable;

[0012] Step 3: Use the CNN-CatBoost deep hybrid model to establish a floodplain soft soil compression coefficient prediction model;

[0013] Step 4: input the input variables of the data set into the floodplain soft soil compression coefficient prediction model for training, optimize the parameters of the floodplain soft soil compression coefficient prediction model, and obtain the optimized floodplain soft soil compression coefficient prediction model;

[0014] Step 5: Conduct an explanatory analysis on the floodplain soft soil compression coefficient prediction model.

[0015] In an optional embodiment, the test described in step 1 is performed using a compression tester to obtain the compression coefficient of the floodplain soft soil sample under different influencing factors.

[0016] In an optional embodiment, the floodplain soft soil compression coefficient prediction model includes a CNN module and a CatBoost module.

[0017] In an optional embodiment, the CNN module includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer in sequence.

[0018] In an optional embodiment, the processing flow of the floodplain soft soil compression coefficient prediction model is as follows:

[0019] S3.1. Take the influencing factors in the data set as input variables, input them through the input layer, and perform convolution and pooling processing in turn through the convolution layer and pooling layer to extract the feature values;

[0020] S3.2, map the extracted feature values ​​to the output space through the fully connected layer, and input them into the CatBoost module through the output layer;

[0021] S3.3, CatBoost module processes the data from the output layer to obtain the prediction results.

[0022] In an optional embodiment, in the process of optimizing the parameters of the floodplain soft soil compression coefficient prediction model described in step 4, the root mean square error, mean absolute error and goodness of fit are used as evaluation criteria.

[0023] In an optional embodiment, in the data set, 75% of the samples are used as a training data set, and the remaining 25% of the samples are used as a testing data set, and the data in the data set are normalized.

[0024] In an optional embodiment, the explanatory analysis described in step five includes SHAP analysis and PDP analysis.

[0025] In an optional embodiment, the SHAP analysis includes global explanatory analysis and feature interaction analysis.

[0026] The beneficial effects of the present invention based on its technical solution are:

[0027] (1) The present invention provides a method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence, selecting wet density ρ, plastic limit w P , liquid index I L , Plasticity Index I p As for the depth H, as the influencing factor of the compression coefficient, it is in line with the characteristics of soft soil in floodplain phase, and it is reliable to use it as the input variable of the model to predict the compression coefficient;

[0028] (2) The present invention provides a method for predicting the compression coefficient of soft soil in floodplain phase based on deep hybrid artificial intelligence, and selects the CNN-CatBoost hybrid model as the prediction model. The convolution layer of the CNN module associates each neuron in two adjacent layers with the neuron in the convolution window of the previous layer. Through the parameter sharing mechanism, the convolution layer can greatly reduce the number of parameters that the model needs to learn, thereby reducing the computational complexity and preventing overfitting. The pooling layer downsamples the data through local connections to achieve feature selection and further avoid overfitting. The CatBoost module is based on a new gradient algorithm of symmetric decision trees and introduces an Orderedboosting mechanism. Compared with the traditional Boosting framework, this improvement of CatBoost helps to avoid target leakage and cleverly process classification features, thereby improving the accuracy and generalization ability of the model. The prediction model constructed by deep hybridization of the CNN module and the CatBoost module can automatically learn and extract data features in a complex environment, and its superiority in prediction ability can be confirmed by comparison with the actual value.

[0029] (3) The present invention provides a method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence for interpretability analysis, which can solve the black box problem of the deep learning model and derive the influence of each feature on the output variable. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 The present invention is a flow chart of a method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence.

[0032] Figure 2 Schematic diagram of the Pearson correlation matrix of sample space distribution.

[0033] Figure 3 It is a regression distribution diagram between predicted values ​​and actual values.

[0034] Figure 4 is the prediction result of the training sample.

[0035] Figure 5 is the prediction result of the test sample.

[0036] Figure 6 Summary plot of SHAP values ​​and feature importance bar chart for all input feature variables.

[0037] Figure 7 Heat map of SHAP values ​​for different input feature variables.

[0038] Figure 8 It is the PDP diagram of each input feature variable versus compression coefficient.

[0039] Fig. 9 is the PDP diagram between variables. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention belong to the scope of protection of the embodiments of the present invention.

[0041] The embodiment of the present invention provides a method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence, referring to Figure 1 , the method comprising:

[0042] Step 1: Obtain test data of floodplain soft soil samples.

[0043] The test is conducted by using a compression tester to obtain the compression coefficient of the floodplain soft soil sample under different influencing factors. In this embodiment, 699 groups of physical and mechanical parameter data of floodplain soft soil samples are collected.

[0044] Step 2: Generate a data set using the test data. Each sample in the data set includes influencing factors and influencing targets. Influencing factors include soil specific gravity G s , moisture content w, wet density ρ, dry density porosity ρ d , Saturation S r 、Liquid limit w L , plastic limit wP , liquid index I L , Plasticity Index I p And the soil sample depth H, the influencing target is the compression coefficient a. The distribution statistics of each parameter are shown in the following table:

[0045]

[0046]

[0047] It can be seen that the coefficient of variation of different parameters varies significantly. s , wet density ρ, saturation S r The coefficient of variation is low, and the dry density ρ d , water content w, porosity e, liquid limit w L , plastic limit w p The coefficient of variation is medium, and the liquid index I L , Plasticity Index I p The coefficient of variation of the compression coefficient a is high, while the coefficient of variation of the depth H is the highest. The average specific gravity of soil particles in the floodplain soft soil G s The average saturation S is 2.72, and the coefficient of variation is only 0.0015, indicating that the weight of soil particles is almost completely consistent. r The average natural moisture content w is 38.83%, the coefficient of variation is 0.103, and the natural wet density ρ is 1.768 g / cm 3 , the coefficient of variation is 0.019, the dry density ρ d 1.275g / cm 3 , the coefficient of variation is 0.040, the dry density ρ d It is calculated by natural wet density ρ and natural water content w, so its coefficient of variation is between the coefficients of variation of natural wet density ρ and natural water content w. The porosity e is 1.136, the coefficient of variation is 0.079, and the average liquid limit w is L is 36.550%, the coefficient of variation is 0.083, and the plastic limit w p The average flow index I L is 1.160, the coefficient of variation is less than 0.114, and the average plasticity index I p The average depth H is 18.507m, and the coefficient of variation is 0.680, indicating that the depth range of the soil sample is relatively large. The average compression coefficient a is 0.618MPa. -1It can be seen that the stiffness of the soft soil in the floodplain phase is relatively low and it is easy to deform under load. The maximum coefficient of variation is 0.211, which means that the compression coefficient is relatively discrete. Further analysis is needed to find out the main influencing factors of the compression coefficient and realize the accurate prediction of the compression coefficient.

[0048] The Pearson correlation matrix of the sample space distribution is calculated for the 11 variables used in the data set. Figure 2 As shown in the figure, it can be seen that the correlation between soil sample depth and water content and saturation is high, which is mainly because the water content and saturation of soil samples are mainly affected by groundwater level in the natural state; and the change of water content will change the fluidity of the soil, thereby affecting the size of liquid limit and plastic limit, so there is a high correlation between liquid limit and plastic limit; the plasticity index is affected by the particle structure and arrangement of the soil, so it also has a high correlation with the specific gravity of soil particles; the wet density has a high correlation with water content and porosity, because the wet density represents the mass of unit volume of soil in the natural state, including the mass of soil particles and the mass of natural water in the pores.

[0049] In addition, the complex relationship between each variable and the compression coefficient can be further confirmed through analysis. In the natural state, the surface soil is relatively loose, while the deep soil becomes denser due to the long-term pressure of the upper soil layer, and its value is usually negatively correlated with the compression coefficient. Wet density refers to the mass of soil per unit volume in the natural state. A higher wet density indicates that the soil particle structure is denser, which helps to increase the stiffness of the soil, thereby reducing the compression coefficient. The plastic limit refers to the limit moisture content when the soil transitions from the plastic state to the semi-solid state; the liquid index is an index to judge the softness and hardness of the soil. These two items are usually positively correlated with the compression coefficient. The plasticity index indicates the plasticity of the soil. A higher plasticity index indicates that the soil is more likely to deform, usually due to a higher water content. These variables jointly affect the compression coefficient and provide insights into the physical properties, particle distribution, engineering properties, and actual application behavior of the soil. Therefore, it is reliable to select the above variables to accurately predict the compression coefficient.

[0050] In this embodiment, 75% of the samples are set as the training data set in the model training process, and 25% are used as the prediction data set. Considering that the input variables have different sizes and units, which will affect the results of data analysis, the input variables are normalized according to the normalization procedure so that all data results are between 0 and 1.

[0051] Step 3: Use the CNN-CatBoost deep hybrid model to establish a floodplain soft soil compression coefficient prediction model. The floodplain soft soil compression coefficient prediction model includes a CNN module and a CatBoost module. The CNN module includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer in sequence. This embodiment adopts a one-dimensional convolutional CNN structure, 512 filters are set in the convolution layer, the convolution kernel size is 1, and the activation function adopts a rectified linear unit (ReLU) function.

[0052] The processing flow of the floodplain soft soil compression coefficient prediction model is as follows:

[0053] S3.1, input variables are input through the input layer, and then convolution and pooling are performed in sequence through the convolution layer and the pooling layer to extract feature values;

[0054] S3.2, map the extracted feature values ​​to the output space through the fully connected layer, and input them into the CatBoost module through the output layer;

[0055] S3.3, CatBoost module processes the data from the output layer to obtain the prediction results.

[0056] The floodplain soft soil compression coefficient prediction model extracts deep features from the signal through convolution, pooling, and activation operations, and passes the obtained final feature vector to the CatBoost module. After the CatBoost module obtains the input features, it performs regression prediction on the compression coefficient through feature fusion operation. The addition of the CNN module effectively improves the generalization ability of the model. The CatBoost method reduces the overfitting problem by introducing ordered boosting technology and symmetric decision trees, and effectively retains data features. The present invention combines the respective advantages of the CatBoost algorithm and convolutional neural network (CNN) to construct a model that can automatically learn and extract data features in a complex environment.

[0057] Step 4: Use the data set to input the floodplain soft soil compression coefficient prediction model for training, optimize the parameters of the floodplain soft soil compression coefficient prediction model, and obtain the optimized floodplain soft soil compression coefficient prediction model. In the data set, 75% of the samples are used as the training data set, and the remaining 25% of the samples are used as the test data set, and the data in the data set are normalized.

[0058] To evaluate the performance of the model, the root mean square error (RMSE), mean absolute error (MAE) and goodness of fit (R 2 ) as the evaluation criteria for the prediction model. The root mean square error and mean absolute error values ​​are close to zero, indicating that the model's prediction is more accurate, while the goodness of fit R 2 It represents the correlation between the characteristics of the model and the prediction target. Figure 3 As shown in the figure, the R2 of the hybrid model in the training set and test set is 0.965 and 0.933, which shows that the model shows high accuracy in prediction and can effectively capture the trend of data changes. In addition, the RMSE of the model on the training set is 0.0240 and the MAE is 0.0198; the RMSE on the test set is 0.0347 and the MAE is 0.0255. Figure 4 The actual values ​​of different samples and the predicted values ​​of the model are shown. The training data and the prediction model show good consistency. This high overlap reflects that the model can accurately capture the basic patterns in the training data. These results show that the hybrid model maintains relatively consistent prediction accuracy on the training set and the test set, reflecting the good performance of the model in practical applications.

[0059] Step 5: Conduct an explanatory analysis on the floodplain soft soil compression coefficient prediction model, including SHAP analysis (global explanatory analysis and feature interaction analysis) and PDP analysis.

[0060] (1) Global explanatory analysis

[0061] Figure 6 The SHAP value summary graph and feature importance bar chart of all feature parameters are displayed. The positive or negative SHAP value on the horizontal axis indicates the influence of the feature parameter on the output result. The degree of influence is proportional to the absolute value. The feature parameters are sorted in descending order according to their contribution to the compression coefficient. The color indicates the value of the feature parameter, where red indicates a lower value and blue indicates a higher value. It can be seen from the figure that the plasticity index has an important influence on the prediction of the model output, and its characteristic value is positively correlated with the SHAP value. Increasing the plasticity index will enhance the positive effect on the output result. Another key parameter is wet density, which has a negative effect, and the negative effect on the dynamic shear modulus increases with the increase of the parameter value. Although the plastic limit and liquid index have little effect on the model output, they still show a certain positive effect, while the depth shows a negative effect. The above phenomenon is consistent with the law of preliminary analysis in feature selection.

[0062] (2) Feature interaction analysis

[0063] The SHAP values ​​of different features for each data point are displayed in detail through heat maps, where high values ​​of f(x) represent high predictions and low values ​​represent low predictions. Heat maps can be used to effectively explore and understand the importance of features in complex models and the relationships between features, thereby further improving the interpretability of the model. Figure 7 Medium and high predictions are associated with high plasticity index and low wet density (blue), while low predictions are associated with low plasticity index (red) and high wet density. In addition, all data points in the features Depth, Plastic Limit, and Liquidity Index show low SHAP values.

[0064] (3) PDP analysis

[0065] The PDP plot analyzes the impact of a feature on the model's predictions by changing the value of only one feature while fixing the values ​​of the other features. This univariate analysis helps to establish a simple relationship between the feature variable and the model's predictions. The PDP plot also allows for the assessment of the significance of a feature on the model's predictions. The more significant the impact of a feature on the model's predictions, the more significant the change between its PDP curves, indicating that the feature plays an important role in the prediction.

[0066] Figure 8 It reveals how the content of each characteristic variable affects the prediction of the model output. The horizontal axis represents the value of the characteristic variable, and the vertical axis represents the value of the compressibility coefficient. It is observed that depth and wet density have a negative effect on the prediction results of the model, while the other three indicators show a positive effect. This finding is consistent with the law in the SHAP global explanatory analysis. The curves in wet density and plasticity index change more obviously with the characteristic value, which indicates that wet density and plasticity index play a more important role in the prediction.

[0067] In order to clearly compare the effects of the remaining three indicators on the model output prediction, two-dimensional PDP diagrams between depth and plastic limit, plastic limit and plasticity index variables were drawn, as shown in Fig. 9 As shown. The depth range is 1.3m-58.3m. When the depth exceeds 10m, the change trend of the compression coefficient in the prediction results slows down. This may be because the deep soil is relatively more stable, resulting in the change of the compression coefficient in the deep soil not being as obvious as in the shallow soil. On the contrary, the change trend of the compression coefficient in the prediction results becomes more significant with the increase of the plastic limit. This may be because the increase of the plastic limit enhances the plasticity of the soil, making the soil more sensitive to external pressure. The influence of the liquid index on the prediction results shows a nonlinear change. As the liquid index increases, the compression coefficient in the prediction results increases first, and then the change tends to be flat after reaching the peak. This is because as the liquid index increases, the soil becomes softer and the compressibility increases. When the liquid index reaches a certain value, the soil approaches a fluid, and its compressibility changes are no longer significant.

[0068] In order to further verify the effectiveness of the CNN-Catboost deep hybrid model used in the invention, four tree models, CatBoost, XGBoost (eXtreme Gradient Boosting), LGBM (Light Gradient Boosting Machine) and Random Forest, as well as a neural network model such as CNN, are used to predict the compression coefficient, and the prediction results of these models are compared with the prediction results of the CNN-Catboost deep hybrid model. The following table summarizes the prediction performance of each model on the training set and test set.

[0069] The following table shows the prediction performance of each model on the training set and test set.

[0070]

[0071] R of CNN-Catboost deep hybrid model in training and test sets 2 The value is the highest, and the RMSE value is also the lowest, which shows that the model performs well in predicting and fitting the data. In terms of mean absolute error (MAE), the hybrid model performs equally well, although it lags slightly behind the LGBM model and XGBoost model in the training set, but the overall gap is not significant.

[0072] The performance in the training set shows that the prediction accuracy of the CNN model is relatively low. This is because there are complex nonlinear relationships and specific feature interactions between feature variables when processing the floodplain soft soil dataset. Using CNN alone is not enough to capture these relationships, while CatBoost can effectively handle this multi-level feature relationship through the tree structure. This shows that when dealing with the nonlinear problem of predicting the compression coefficient of floodplain soft soil, the use of a hybrid model can achieve better prediction results.

[0073] In addition, the R of the hybrid model is significantly lower than that of the CatBoost model. 2 The higher the value, the lower the RMSE and MAE values, indicating that through the deep feature mining of the CNN module, the hybrid model shows better fitting and generalization capabilities in dealing with the problem of soil compression coefficient prediction.

[0074] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (modules, systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0076] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0078] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0079] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence, characterized in that: The method comprises: Step 1: Obtain test data of soft soil samples in floodplain phase after testing; Step 2: Generate a data set using the test data. Each sample in the data set includes influencing factors and influencing targets. Influencing factors include soil specific gravity G s , moisture content w, wet density ρ, dry density porosity ρ d , Saturation S r 、Liquid limit w L , plastic limit w P , liquid index I L , Plasticity Index I p and soil sample depth H, the influencing target is the compression coefficient a; select soil sample depth H, water content w, plastic limit w from the influencing factors P , Plasticity Index I p Liquidity index I L as input variables; Step 3: Use the CNN-CatBoost deep hybrid model to establish a floodplain soft soil compression coefficient prediction model; Step 4: input the input variables of the data set into the floodplain soft soil compression coefficient prediction model for training, optimize the parameters of the floodplain soft soil compression coefficient prediction model, and obtain the optimized floodplain soft soil compression coefficient prediction model; Step 5: Conduct an explanatory analysis on the floodplain soft soil compression coefficient prediction model.

2. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 1 is characterized by: The test described in step 1 is carried out using a compression tester to obtain the compression coefficient of the floodplain soft soil sample under different influencing factors.

3. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 1 is characterized by: The floodplain soft soil compression coefficient prediction model includes a CNN module and a CatBoost module.

4. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 3 is characterized by: The CNN module includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer in sequence.

5. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 4 is characterized in that: The processing flow of the floodplain soft soil compression coefficient prediction model is as follows: S3.1, input variables are input through the input layer, and then convolution and pooling are performed in sequence through the convolution layer and the pooling layer to extract feature values; S3.2, map the extracted feature values ​​to the output space through the fully connected layer, and input them into the CatBoost module through the output layer; S3.3, CatBoost module processes the data from the output layer to obtain the prediction results.

6. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 1 is characterized by: In the process of optimizing the parameters of the floodplain soft soil compression coefficient prediction model described in step 4, the root mean square error, mean absolute error and goodness of fit are used as evaluation criteria.

7. The method for predicting compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 1 is characterized by: In the data set, 75% of the samples are used as a training data set, and the remaining 25% of the samples are used as a test data set, and the data in the data set are normalized.

8. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 1 is characterized by: The explanatory analysis described in step 5 includes SHAP analysis and PDP analysis.

9. The method for predicting the compression coefficient of floodplain soft soil based on deep hybrid artificial intelligence according to claim 8 is characterized by: The SHAP analysis includes global explanatory analysis and feature interaction analysis.

Images