A Smart Decision-Making Method for Soft Foundation Processing Based on MAUT and ANN

Abstract

This invention discloses an intelligent decision-making method for soft soil foundation treatment schemes based on MAUT and ANN, comprising the following steps: S1: Determine the identification parameters of soft soil foundation treatment schemes and construct a sample database; S2: Establish a numerical model of undisturbed soft soil foundation; S3: Conduct single-factor numerical simulations of relevant parameters; S4: Construct individual utility functions and weights based on settlement response results, and calculate the comprehensive utility value; S5: Establish a soft soil foundation treatment scheme classification model, input the comprehensive utility value and soft soil depth into the soft soil foundation treatment scheme classification model, and output the prediction results and recommended schemes for each soft soil foundation treatment scheme; S6: Train and test the output results of the soft soil foundation treatment scheme classification model in step S5 to obtain a relatively robust evaluation result of the soft soil foundation treatment scheme classification model. This invention can reduce the reliance on subjective experience in the scheme selection process, reduce input parameter redundancy while retaining key engineering information, and improve the accuracy and stability of soft soil foundation treatment scheme classification and identification.

Description

Technical Field

[0001] This invention belongs to the field of soft soil foundation treatment and intelligent decision-making technology in geotechnical engineering, specifically involving an intelligent decision-making method for soft soil foundation treatment schemes based on MAUT and ANN. Background Technology

[0002] Soft soil is widely distributed in inland rivers, lakes, and coastal areas of my country. Due to its high water content, high compressibility, low strength, and poor permeability, it has poor engineering properties. Many road construction projects involve soft soil foundation treatment. Improper soft soil foundation treatment can easily cause excessive deformation of the roadbed, affecting the stability and safety of the project. Therefore, scientifically selecting a soft soil foundation treatment scheme is a key link to ensure the quality of road construction and its performance in later service.

[0003] Currently, the selection of soft foundation treatment schemes mainly relies on engineering analogy, empirical judgment, multi-index comprehensive evaluation methods, and machine learning methods such as artificial neural networks. These methods provide a certain reference for the comparison and intelligent identification of soft foundation treatment schemes. However, existing technologies still have the following shortcomings: (1) Existing soft soil treatment schemes mainly rely on engineering analogy, designer experience and on-site survey data for judgment; engineering technicians usually combine existing engineering experience and survey results to select treatment measures. Although this method has certain engineering practicality, it is greatly affected by subjective experience, lacks unified quantitative analysis basis, and is difficult to achieve comprehensive evaluation under multiple factors, resulting in a large uncertainty in the scheme selection results. (2) Existing multi-index comprehensive evaluation methods can be used to compare and select soft soil treatment schemes. However, existing methods still have strong human setting characteristics in the process of assigning index weights, classifying evaluation levels and making decisions. Different evaluators may obtain different evaluation results. At the same time, this type of method has limited ability to express the complex nonlinear relationship under the coupling effect of multiple factors on soft soil foundations, and it is difficult to fully reflect the actual impact of each influencing factor on the engineering properties. (3) Machine learning methods such as artificial neural networks can establish a nonlinear mapping relationship between foundation parameters and treatment schemes, providing a new technical approach for intelligent identification of soft foundation treatment schemes. However, existing methods usually directly use the original foundation parameters as model inputs. When there are many input parameters and there is correlation between variables, feature redundancy problems are easy to occur, which affects the model training stability and generalization ability, thereby reducing the scheme identification effect. Summary of the Invention

[0004] (1) Technical problems to be solved To address the shortcomings of existing technologies, the present invention aims to provide an intelligent decision-making method for soft foundation processing schemes based on MAUT and ANN, which can reduce the reliance on subjective experience in the scheme selection process, reduce input parameter redundancy while retaining key engineering information, and improve the accuracy and stability of soft foundation processing scheme classification and identification.

[0005] (2) Technical solution To address the aforementioned technical problems, this invention provides an intelligent decision-making method based on a soft-foundation processing scheme using MAUT and ANN, comprising the following steps: S1. Determine the identification parameters of soft soil treatment schemes and construct a sample database: Select the compression modulus, soft soil layer thickness, hard shell layer thickness, fill height and soft soil burial depth as identification parameters of soft soil treatment schemes, and use the actual soft soil treatment method as the sample output category. Unify the units and dimensions of each parameter to form a sample database for calculating comprehensive utility value and training the soft soil treatment scheme classification model. S2. Establish a numerical model of the original soft soil foundation; S3. Conduct single-factor numerical simulations of relevant parameters to analyze the foundation settlement response under different index changes; S4. Construct individual utility functions and weights based on the settlement response results, and calculate the comprehensive utility value; S5. Establish an ANN soft soil treatment scheme classification model, input the comprehensive utility value and soft soil burial depth into the soft soil treatment scheme classification model, and output the prediction results and recommended schemes for each soft soil treatment scheme. S6. Train and test the output results of the soft foundation treatment scheme classification model in step S5 to obtain a more robust evaluation result of the soft foundation treatment scheme classification model.

[0006] Preferably, the compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height in step S1 are used to characterize the main influencing factors of settlement and deformation of soft soil foundation, and to construct a single utility function and a comprehensive utility value λ under the numerical model of undisturbed soft soil foundation; the soft soil burial depth is used to characterize the spatial location of the soft soil layer, does not participate in the calculation of the comprehensive utility value λ, but is used as an independent input variable to be input together with the comprehensive utility value λ into the soft soil foundation treatment scheme classification model.

[0007] Furthermore, step S2 specifically includes: A numerical model of the undisturbed soft soil foundation was established using PLAXIS two-dimensional finite element software. First, the geological structure, material parameters, boundary conditions, and loading conditions were set in the numerical model of the undisturbed soft soil foundation. Then, settlement monitoring points were selected in the original soft soil foundation numerical model, and foundation settlement was used as the response index to provide a computational basis for subsequent single-factor numerical simulation and settlement response analysis.

[0008] Furthermore, step S3 specifically includes: On the original soft soil foundation numerical model, the control variable method was used to carry out single-factor numerical simulations of compression modulus, soft soil layer thickness, hard crust layer thickness and fill height respectively; Obtain the foundation settlement response results under various identification parameter changes, and establish the correspondence between compression modulus, soft soil layer thickness, hard crust layer thickness, and fill height and foundation settlement.

[0009] Furthermore, step S4 specifically includes: Based on the foundation settlement response results of each identified parameter obtained from the single-factor numerical simulation in step S3, a single-item utility function corresponding to the compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height is constructed. Based on the foundation settlement response results of each identification parameter obtained from the single-factor numerical simulation in step S3, the index weight of each identification parameter is determined by the elastic sensitivity coefficient method. Based on the multi-attribute utility theory, the comprehensive utility value λ is calculated from the individual utility values ​​obtained by the individual utility functions of each identification parameter and the index weights of each identification parameter, thereby realizing the comprehensive characterization of compression modulus, soft soil layer thickness, hard shell layer thickness and fill height and the dimensionality reduction of input features.

[0010] Furthermore, the soft foundation processing scheme classification model in step S5 is constructed using a feedforward backpropagation neural network. The soft foundation processing scheme classification model includes an input layer, a hidden layer, and an output layer. The hidden layer uses a sigmoid activation function, and the output layer uses a softmax function.

[0011] Furthermore, step S5 specifically involves: inputting the comprehensive utility value λ and the soft soil burial depth as input variables into the input layer of the trained soft soil treatment scheme classification model; enhancing the model's ability to fit complex nonlinear relationships through a hidden layer; and then outputting a class of soft soil treatment schemes through an output layer.

[0012] Furthermore, the predicted results and recommended scheme categories include: replacement of mountain stone cushion layer, drainage consolidation, cement-soil mixing piles, CFG piles, and prestressed pipe piles.

[0013] Furthermore, in step S6, K-fold cross-validation is used to train and test the ANN soft-foundation classification model, and the model performance is evaluated using the confusion matrix, accuracy, precision, recall, and F1 score.

[0014] Beneficial effects Compared with the prior art, the beneficial effects of the present invention are as follows: This invention uses PLAXIS numerical simulation to analyze the influence of indicators such as soft soil layer thickness, compression modulus, hard shell layer thickness, and fill height on foundation settlement. The settlement response results are used to construct individual utility functions and determine indicator weights, so that the selection of soft soil treatment schemes is shifted from simple empirical judgment to quantitative analysis based on deformation response laws, reducing the reliance on subjective experience in the scheme selection process.

[0015] This invention utilizes MAUT to comprehensively characterize multiple indicators such as soft soil layer thickness, compression modulus, hard crust layer thickness, and fill height into a comprehensive utility value λ. This allows foundation parameters with different dimensions and different influence directions to be expressed under a unified utility scale, reducing the dimensionality of model input while retaining key engineering information and reducing the impact of original parameter redundancy on model training stability.

[0016] This invention uses the comprehensive utility value λ and the soft soil burial depth as input variables of an ANN model to establish a classification model for soft soil foundation treatment schemes. This model can characterize the nonlinear mapping relationship between foundation parameters and treatment scheme categories, thereby improving the accuracy and generalization ability of soft soil foundation treatment scheme identification.

[0017] This invention combines numerical simulation, MAUT comprehensive utility characterization, and ANN classification to form a complete technical process from foundation parameter analysis and comprehensive utility value construction to treatment scheme classification. This process enables the soft soil treatment scheme identification process to have both engineering mechanism basis and nonlinear learning capability, which can improve the overall reliability and engineering applicability of intelligent decision-making for soft soil treatment schemes under complex foundation conditions. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of an intelligent decision-making method for a soft-foundation processing scheme based on MAUT and ANN provided in an embodiment of the present invention; Figure 2 This is a network structure diagram of the soft foundation processing scheme classification model in an embodiment of the present invention; Figure 3 This is a flowchart of the training process for the classification model of the soft foundation processing scheme in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described 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.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] This specific implementation method is an intelligent decision-making method for soft soil foundation treatment schemes based on MAUT and ANN. It is used for intelligent identification, classification prediction, and decision support of soft soil foundation treatment schemes. It combines soft soil foundation numerical simulation, the comprehensive utility representation of the MAUT undisturbed soft soil foundation numerical model, and the classification prediction of the ANN soft soil foundation treatment scheme classification model. The MAUT undisturbed soft soil foundation numerical model is used to comprehensively represent the utility of multiple soft soil foundation engineering parameters affecting the selection of soft soil foundation treatment schemes, constructing a comprehensive utility value λ. The ANN soft soil foundation treatment scheme classification model is then used to establish a nonlinear mapping relationship between the comprehensive utility value λ, the soft soil depth, and the soft soil foundation treatment scheme category, achieving classification prediction and recommendation output for soft soil foundation treatment schemes. The method specifically includes the following steps: S1. Determine the identification parameters of the soft foundation treatment scheme and construct a sample database.

[0022] Collect engineering survey data and actual treatment data to form cross-sectional soil layer parameters required for soft soil treatment scheme identification. The cross-sectional soil layer parameters include compression modulus, soft soil layer thickness, hard crust layer thickness, fill height and soft soil burial depth as identification parameters for soft soil treatment schemes. The actual soft soil treatment method used is used as the sample output category. The units and dimensions of each parameter are unified to form a sample database for calculating comprehensive utility value and training ANN soft soil treatment scheme classification model. Among them, the compression modulus, soft soil layer thickness, hard shell layer thickness and fill height are used to characterize the main engineering factors affecting the settlement and deformation of soft soil foundation. They are suitable for constructing single utility functions and performing unified utility characterization under the MAUT undisturbed soft soil foundation numerical model. Therefore, they are used to calculate the comprehensive utility value λ to realize the comprehensive expression and dimensionality reduction of multiple input features. In addition, the soft soil depth is used to characterize the spatial location and distribution characteristics of the soft soil layer. Although it also affects the foundation settlement and the selection of treatment schemes, its index attributes are different from the above parameters, making it difficult to uniformly characterize it according to the same utility function form. Therefore, the soft soil depth does not participate in the construction of the comprehensive utility value λ, but is used as an independent input variable to input the comprehensive utility value λ into the ANN soft soil treatment scheme classification model to establish a nonlinear mapping relationship between the comprehensive characteristics of foundation parameters and the treatment scheme category.

[0023] S2. Establish a numerical model of the original soft soil foundation; The MAUT original soft soil foundation numerical model is established using PLAXIS two-dimensional finite element software. From top to bottom, it includes a hard shell layer, a soft soil layer and an underlying soil layer. An embankment filling area is set on the surface. First, the stratum structure, material parameters, boundary conditions and loading conditions are set in the original soft soil foundation numerical model. Then, settlement monitoring points were selected in the original soft soil foundation numerical model, and foundation settlement was used as the response index to provide a computational basis for subsequent single-factor numerical simulation and settlement response analysis.

[0024] Step S2 specifically involves: establishing a plane strain numerical model of the MAUT undisturbed soft soil foundation using PLAXIS two-dimensional finite element software; the MAUT undisturbed soft soil foundation plane strain numerical model has a width of 60m and a height of 20m. The MAUT undisturbed soft soil foundation plane strain numerical model is generalized based on the general layered structural characteristics of soft soil foundations, and includes, from top to bottom, a hard shell layer, a soft soil layer, and an underlying soil layer, and an embankment filling area is set on the surface.

[0025] Each soil layer was simulated using the Mohr-Coulomb constitutive model, with parameters such as unit weight, compression modulus, cohesion, internal friction angle, Poisson's ratio, and permeability coefficient input. The MAUT undisturbed soft soil foundation plane strain numerical model simultaneously constrained horizontal and vertical displacements at the bottom boundary, and constrained horizontal displacements at the two side boundaries; the top and right boundaries were set as drained boundaries, and the bottom and left symmetrical boundaries were set as undrained boundaries. The embankment load was applied using a layered filling method. Each layer of fill was used to simulate the gradual increase of fill load and the development of foundation consolidation during actual construction. Each layer of fill was 1m thick, loaded for 10 days, and then consolidated for 10 days. During the calculation, the surface point A at the center of the roadbed was selected as the monitoring point to provide basic data for subsequent single-factor numerical simulation and comprehensive utility value construction.

[0026] S3. Conduct single-factor numerical simulations of relevant parameters to analyze the ground settlement response under different index changes.

[0027] Step S3 specifically involves: on the undisturbed soft soil foundation numerical model, using the controlled variable method to conduct single-factor numerical simulations of compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height; obtaining the foundation settlement response results under the condition of changing identification parameters, and establishing the correspondence between compression modulus, soft soil layer thickness, hard shell layer thickness, fill height, and foundation settlement.

[0028] During the simulation, only one index parameter was changed each time, while the other parameters remained unchanged under the baseline conditions, in order to identify the influence of a single factor on foundation settlement. Specifically, the thickness of the soft soil layer was taken in 0.5m increments within the range of 0–5m; the compression modulus was taken in 1MPa increments within the range of 1–10MPa; the thickness of the hard crust layer was taken in 0.5m increments within the range of 0–5m; and the fill height was taken in 1m increments within the range of 1–7m. During the calculation, the settlement value of the roadbed center point A after the filling was completed and fully consolidated was extracted as the main response index.

[0029] S4. Based on the settlement response results, construct individual utility functions and weights, and calculate the MAUT comprehensive utility value; based on the foundation settlement response results of each identified parameter obtained from the single-factor numerical simulation in step S3, construct individual utility functions corresponding to the compression modulus, soft soil layer thickness, hard shell layer thickness and fill height. Based on the foundation settlement response results of each identification parameter obtained from the single-factor numerical simulation in step S3, the index weight of each identification parameter is determined by the elastic sensitivity coefficient method. Based on the multi-attribute utility theory, the comprehensive utility value λ is calculated from the individual utility values ​​obtained by the individual utility functions of each identification parameter and the index weights of each identification parameter, thereby realizing the comprehensive characterization of compression modulus, soft soil layer thickness, hard shell layer thickness and fill height and the dimensionality reduction of input features.

[0030] Specifically, based on the results of single-factor numerical simulation, the settlement value of the roadbed center point during the stage when the settlement tends to be gradual after the filling is completed is uniformly extracted as the basis for fitting the single-factor utility function when the single-factor changes of each index. The settlement at this location can better characterize the overall deformation level of the foundation and has strong representativeness. Subsequently, the range standardization method is used to normalize each index, and the negative index of settlement is mapped to the [0,1] interval by index inversion. The larger the obtained utility value, the smaller the settlement of the original foundation under the index condition, which is more beneficial to the project.

[0031] Positive indicators are converted according to formula (1): The negative index is converted according to formula (2): in, These are the standardized indicator values; Its original value; and These are the maximum and minimum values ​​of the indicator across all data samples, respectively.

[0032] Based on the interaction mechanism between each indicator and the central point settlement, different functional forms are selected to construct individual utility functions, which are then fitted using the least squares method. After the above fitting, each original indicator can be uniformly transformed into a dimensionless function value with consistent utility meaning, thus providing a basis for subsequent indicator weight determination and comprehensive utility value calculation.

[0033] To objectively characterize the influence of each input parameter on the model output settlement, the index weights are determined using the elastic sensitivity coefficient method. This involves calculating the relative change in settlement caused by a unit relative change in the input parameters, quantifying the sensitivity of each index to the output results, and determining the index weights accordingly.

[0034] Let the model output be yThe input parameters are Then the first i The elastic sensitivity coefficient of each parameter is defined as follows: In the formula, For parameters The amount of disturbance. For the reason The change in output caused by the disturbance. It represents the relative change in output caused by a unit relative change in input.

[0035] When the model is continuously differentiable, the following equation can also be expressed in differential form: This formula reflects the relative sensitivity of the model output to the input variables, i.e., the elasticity of the parameters. The larger the elasticity sensitivity coefficient, the more sensitive the output is to the parameter, and the more significant its role in the system.

[0036] To satisfy the comparability between weights and the constraint that their sum is 1, the sensitivity coefficients of each parameter are normalized to obtain the weight of the i-th parameter: In the formula: For parameters Objective weighting.

[0037] The weights of each indicator are calculated using the formulas above. It should be noted that the settlement data used in the weight calculation is consistent with that used in the fitting stage of the individual utility function, and is based on the settlement value at the center point of the roadbed.

[0038] Based on the MAUT weighted integration concept, a comprehensive utility value λ is constructed. Let the individual utility value corresponding to the i-th indicator be and its weight be . Then the comprehensive utility value can be expressed as the sum of the products of each individual utility value and its weight, i.e.: In the formula, λ This is a comprehensive utility value used to comprehensively characterize the engineering condition of undisturbed soft soil foundations. n For the number of indicators, For the first i The weight of each indicator, This is the utility function value of the indicator.

[0039] Based on the above method, the individual utility values ​​of compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height can be calculated separately, and the comprehensive utility value λ can be calculated by combining the weights of each indicator. By constructing the comprehensive utility value λ, the four engineering parameters of compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height are uniformly represented as a single comprehensive utility index, realizing the dimensionality reduction from four-dimensional input features to one-dimensional comprehensive features. Subsequently, the comprehensive utility value λ and the soft soil burial depth, which represents the spatial location of the soft soil layer, are used together as input variables of the ANN model, providing a basis for intelligent identification and decision-making of subsequent soft soil foundation treatment schemes.

[0040] S5. Establish an ANN soft soil treatment scheme classification model. Input the comprehensive utility value and the soft soil burial depth into the ANN soft soil treatment scheme classification model, and output the prediction results and recommended schemes for each soft soil treatment scheme. The ANN soft soil treatment scheme classification model in step S5 is constructed using a feedforward backpropagation neural network. The ANN soft soil treatment scheme classification model includes an input layer, a hidden layer and an output layer. The hidden layer uses a sigmoid activation function and the output layer uses a softmax function.

[0041] To verify the effectiveness of the MAUT comprehensive utility value in input feature dimensionality reduction and to compare the impact of different input feature organization methods on the classification and recognition results of soft soil treatment schemes, two ANN soft soil treatment scheme classification models were constructed for comparative analysis. The first ANN soft soil treatment scheme classification model uses the comprehensive utility value λ and the soft soil burial depth as input layers, where λ is comprehensively characterized by four original indicators: soft soil layer thickness, compression modulus, hard shell layer thickness, and fill height. The soft soil burial depth is used to supplement the spatial distribution information of the soft soil layer. The second ANN soft soil treatment scheme classification model directly uses five original indicators: soft soil layer thickness, compression modulus, hard shell layer thickness, fill height, and soft soil burial depth as input. This setting can not only test the effectiveness of the comprehensive utility value in the dimensionality reduction process, but also compare the impact of different input feature organization methods on the classification and recognition results.

[0042] The model output layer is used to characterize the soft soil treatment schemes. Combining commonly used engineering treatment technologies, five treatment methods are used as output categories: mountain stone cushion replacement, drainage consolidation, cement-soil mixing piles, CFG piles, and prestressed pipe piles. Each treatment method corresponds to an output neuron, which is used to establish the classification mapping relationship between geological conditions and treatment schemes.

[0043] To ensure fairness in the model comparison, except for the difference in the number of input layer nodes due to different feature dimensions, the two ANN soft-foundation processing scheme classification models are consistent in terms of network structure, training algorithm, and main training parameter settings. Both ANN soft-foundation processing scheme classification models are constructed using a feedforward backpropagation neural network, in which the hidden layer is used to extract nonlinear features between input variables, and the output layer is used to realize the classification of processing schemes.

[0044] Regarding activation function settings, the hidden layer employs a sigmoid activation function to enhance the model's ability to fit complex nonlinear relationships; the output layer uses a softmax function to achieve probabilistic outputs for various soft-base processing methods under multi-class conditions. Through this structural design, the ANN soft-base processing scheme classification model can establish a stable nonlinear mapping relationship between input variables and processing methods, thereby meeting the needs of intelligent recognition of soft-base processing schemes. During model training, the optimizer uses the stochastic gradient descent (SGD) algorithm, with a maximum number of iterations set to 2000, a gradient penalty coefficient of 10, and a learning rate of 0.002. The above network structure and training parameters are kept consistent in both types of ANN soft-base processing scheme classification models to ensure the fairness of the comparison between ANN soft-base processing scheme classification models under different input feature organization methods.

[0045] Since the output of the ANN soft foundation processing scheme classification model needs to be used in training in numerical form, this application expresses the processing method categories numerically before training the ANN soft foundation processing scheme classification model, and uses one-hot encoding to construct the output labels. That is, the five soft foundation processing methods are mapped to five-dimensional binary vectors respectively, where the position corresponding to the target category is 1 and the other positions are 0. Through one-hot encoding, the artificial order relationship between category labels can be avoided, and it is also beneficial for the Softmax output layer to perform multi-class training.

[0046] After the above processing, the model input variables and output labels together form the "feature-label" training samples, thereby realizing supervised learning modeling from foundation parameters to treatment scheme categories. This encoding method can be well adapted to multi-class classification tasks and provides a unified data expression form for subsequent model training and evaluation.

[0047] S6. Train and test the output results of the soft foundation treatment scheme classification model in step S5 to obtain a more robust evaluation result of the soft foundation treatment scheme classification model.

[0048] In step S6, the K-fold cross-validation method is used to train and test the output results of the ANN soft foundation processing scheme classification model, and the performance of the soft foundation processing scheme classification model is evaluated using the confusion matrix, accuracy, precision, recall and F1 score. This method divides all samples into 5 subsets, selects one subset as the test set each time, and uses the remaining subsets as the training set. This process is repeated for 5 rounds of training and testing. Finally, the results of each round are statistically analyzed to obtain a more robust model evaluation result.

[0049] This application employs K-fold cross-validation, which can improve data utilization under limited sample conditions and reduce the random error caused by a single random partition. For the two types of ANN soft-foundation processing scheme classification models constructed above, this training method can more objectively compare the differences in model classification ability under different input feature organization methods while maintaining sufficient sample coverage, providing a reliable basis for subsequent performance analysis.

[0050] To comprehensively evaluate the ANN model's ability to identify soft-foundation processing schemes, a confusion matrix is ​​introduced to analyze the classification results. Based on this, accuracy, precision, recall, and F1 score are selected as model performance evaluation metrics. These metrics can measure the model from aspects such as overall classification accuracy, reliability of positive class identification, completeness of class identification, and comprehensive discriminative ability. Let the true positives, true negatives, false positives, and false negatives in the confusion matrix be true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), respectively. Then, accuracy, precision, recall, and F1 score can be expressed as follows: Their calculation formulas can be expressed as: In the formula, TP This represents the number of samples correctly identified as the target category. TN This represents the number of samples correctly identified as not belonging to the target category. FP This represents the number of samples that were incorrectly identified as the target category. FN This indicates the number of samples in the target category that were not correctly identified.

[0051] In this embodiment, the first model contains 55 test samples, of which 47 are correctly identified, with an overall accuracy of 85.45%. The macro-average precision, macro-average recall, and macro-average F1 score are 86.52%, 85.36%, and 85.74%, respectively. Looking at the different processing methods, for the CFG class, TP, TN, FP, and FN are 8, 44, 1, and 2, respectively, with precision, recall, and F1 scores of 88.89%, 80.00%, and 84.21%, respectively. For the mountain stone replacement class, the following are... The percentages for the following types of piles are as follows: 10, 41, 2, 2, corresponding to 83.33%, 83.33%, and 83.33%; for drainage consolidation piles, the percentages are as follows: 9, 42, 2, 2, corresponding to 81.82%, 81.82%, and 81.82%; for cement-soil mixing piles, the percentages are as follows: 11, 40, 3, 1, corresponding to 78.57%, 91.67%, and 84.62%; and for prestressed pipe piles, the percentages are as follows: 9, 45, 0, 1, corresponding to 100.00%, 90.00%, and 94.74%.

[0052] The second model correctly identified 37 groups, with an overall accuracy of 67.27%. The macro-average precision, macro-average recall, and macro-average F1 score were 68.55%, 67.73%, and 67.83%, respectively. Looking at the different processing methods, for the CFG class, the TP, TN, FP, and FN scores were 7, 39, 6, and 3, respectively, with precision, recall, and F1 scores of 53.85%, 70.00%, and 60.87%. For the mountain stone replacement class, the scores were 8, 40, and 3, respectively. 4, corresponding to indicators of 72.73%, 66.67%, and 69.57%; drainage consolidation type: 7, 40, 4, 4, corresponding to indicators of 63.64%, 63.64%, and 63.64%; cement-soil mixing pile type: 7, 39, 4, 5, corresponding to indicators of 63.64%, 58.33%, and 60.87%; prestressed pipe pile type: 8, 44, 1, 2, corresponding to indicators of 88.89%, 80.00%, and 84.21%.

[0053] The comparison shows that the first model is superior to the second model in overall accuracy, macro average precision, macro average recall, and macro average F1 score. The classification results of each processing method are also more stable, indicating that the first model has a better ability to identify soft foundation processing schemes. Therefore, the above results show that the intelligent decision-making model of the MAUT-ANN soft foundation processing scheme constructed in this invention can improve the classification and recognition accuracy while reducing the redundancy of input features, and has better scheme recognition ability and model stability.

[0054] Through the implementation of the above six steps, this invention realizes a complete process from soft soil foundation parameter selection, establishment of undisturbed foundation numerical model, single-factor settlement response analysis, construction of MAUT comprehensive utility value, training of ANN classification model to evaluation of soft soil treatment schemes. This method obtains the influence law of key parameters on foundation settlement based on numerical simulation results, uses MAUT to comprehensively characterize the soft soil layer thickness, compression modulus, hard shell layer thickness and fill height into a comprehensive utility value λ, and uses the comprehensive utility value λ and soft soil burial depth as input variables of ANN model to realize the classification prediction of soft soil treatment scheme categories. Thus, this invention can retain key engineering information while reducing input dimensions, reduce subjective experience dependence in the scheme selection process, and improve the accuracy and engineering applicability of intelligent identification of soft soil treatment schemes.

[0055] The above embodiments are preferred implementations of the present invention. In addition, the present invention can be implemented in other ways. Any obvious substitutions without departing from the concept of the present technical solution are within the protection scope of the present invention.

Claims

1. A smart decision-making method for soft-foundation processing schemes based on MAUT and ANN, characterized in that, Includes the following steps: S1. Determine the identification parameters of soft soil treatment schemes and construct a sample database: Select the compression modulus, soft soil layer thickness, hard shell layer thickness, fill height and soft soil burial depth as identification parameters of soft soil treatment schemes, and use the actual soft soil treatment method as the sample output category. Unify the units and dimensions of each parameter to form a sample database for calculating comprehensive utility value and training the ANN soft soil treatment scheme classification model. S2. Establish a numerical model of the original soft soil foundation; S3. Conduct single-factor numerical simulations of relevant parameters to analyze the foundation settlement response under different index changes; S4. Construct individual utility functions and weights based on the settlement response results, and calculate the comprehensive utility value; S5. Establish a classification model for soft soil treatment schemes. Input the comprehensive utility value and the soft soil burial depth into the classification model for soft soil treatment schemes, and output the prediction results and recommended schemes for each soft soil treatment scheme. S6. Train and test the output results of the soft foundation treatment scheme classification model in step S5 to obtain a more robust evaluation result of the soft foundation treatment scheme classification model.

2. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 1, characterized in that, In step S1, the compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height are used to characterize the main influencing factors of settlement and deformation of soft soil foundations, and to construct individual utility functions and comprehensive utility values ​​λ under the numerical model of undisturbed soft soil foundations. The soft soil burial depth is used to characterize the spatial location of the soft soil layer, and does not participate in the calculation of the comprehensive utility value λ, but is used as an independent input variable to be input into the soft soil foundation treatment scheme classification model together with the comprehensive utility value λ.

3. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 2, characterized in that, Step S2 specifically involves: A numerical model of the undisturbed soft soil foundation was established using PLAXIS two-dimensional finite element software. First, the geological structure, material parameters, boundary conditions, and loading conditions were set in the numerical model of the undisturbed soft soil foundation. Then, settlement monitoring points were selected in the original soft soil foundation numerical model, and foundation settlement was used as the response index to provide a computational basis for subsequent single-factor numerical simulation and settlement response analysis.

4. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 3, characterized in that, Step S3 specifically involves: On the original soft soil foundation numerical model, the control variable method was used to carry out single-factor numerical simulations of compression modulus, soft soil layer thickness, hard crust layer thickness and fill height respectively; Obtain the foundation settlement response results under various identification parameter changes, and establish the correspondence between compression modulus, soft soil layer thickness, hard crust layer thickness, and fill height and foundation settlement.

5. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 4, characterized in that, Specifically, step S4 involves: Based on the foundation settlement response results of each identified parameter obtained from the single-factor numerical simulation in step S3, a single-item utility function corresponding to the compression modulus, soft soil layer thickness, hard shell layer thickness, and fill height is constructed. Based on the foundation settlement response results of each identification parameter obtained from the single-factor numerical simulation in step S3, the index weight of each identification parameter is determined by the elastic sensitivity coefficient method. Based on the multi-attribute utility theory, the comprehensive utility value λ is calculated from the individual utility values ​​obtained by the individual utility functions of each identification parameter and the index weights of each identification parameter, thereby realizing the comprehensive characterization of compression modulus, soft soil layer thickness, hard shell layer thickness and fill height and the dimensionality reduction of input features.

6. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 5, characterized in that, The soft foundation processing scheme classification model in step S5 is constructed using a feedforward backpropagation neural network. The soft foundation processing scheme classification model includes an input layer, a hidden layer, and an output layer. The hidden layer uses a sigmoid activation function, and the output layer uses a softmax function.

7. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 6, characterized in that, In step S6, the output results of the soft foundation processing scheme classification model are used for training and testing, and the performance of the soft foundation processing scheme classification model is evaluated using the confusion matrix, accuracy, precision, recall and F1 score.

8. The intelligent decision-making method for soft-foundation processing schemes based on MAUT and ANN according to claim 1, characterized in that, The predicted results and recommended scheme categories include: replacement of mountain stone cushion layer, drainage consolidation, cement-soil mixing piles, CFG piles and prestressed pipe piles.

Images