Over-beach foundation pit excavation soil parameter random prediction assignment method

Abstract

The invention discloses a method for randomly predicting and assigning parameters of an excavated soil body of a beach-overtopping foundation pit. The method comprises the following steps of: collecting geological survey data and construction information of an excavated area of the beach-overtopping foundation pit, screening redundant parts of characteristic parameters, retaining variables which have influences on mechanical parameters of the soil body, and constructing a complete original database; collected data is divided into a training set and a verification set, the training set is used for preliminary training of the model, and the verification set is used for optimization and verification of the model. According to the method, the nonlinear relation of the complex variables is captured through the random forest model, dynamic monitoring and real-time optimization are combined, the soil parameter prediction precision is remarkably improved, and the construction uncertainty risk is reduced; the dynamic monitoring function quickly responds to soil body changes in construction, the prediction result is optimized, and the construction safety is improved; the random field assignment method effectively reflects the spatial variability of soil parameters, supports multi-scene risk assessment, and provides scientific basis and reliable guarantee for engineering design and construction.

Description

Technical Field

[0001] The invention relates to the technical field of floodplain foundation pit excavation, and in particular to a random prediction and value assignment method for floodplain foundation pit excavation soil parameters. Background Art

[0002] Due to its special geological and environmental conditions, floodplain areas face many challenges during foundation pit excavation. The soil in this area has the characteristics of low strength, high compressibility, and easy deformation, which makes the surrounding soil prone to plastic flow during excavation. In addition, the soil has poor permeability and high sensitivity, and the soil layer has shown obvious deformation sensitivity before excavation. The geological characteristics also show obvious unevenness, such as uneven thickness distribution of clay layers, local intercalation of silt and other weak soil layers; abundant groundwater, and seepage, pipe bursts, quicksand, and sudden surges often occur during construction. These characteristics exacerbate the geological uncertainty in the construction process, making the accurate prediction and dynamic assignment of soil parameters an important requirement for the design and construction of foundation pit excavation projects.

[0003] As a machine learning tool, the random forest algorithm has been widely used in geotechnical engineering in recent years because it can effectively handle multidimensional features and complex nonlinear relationships. It can accurately predict the physical and mechanical parameters of soil by analyzing multiple variables that affect soil properties (such as soil layer type, excavation depth, precipitation, groundwater level, etc.). At the same time, the random forest algorithm can generate soil parameter values ​​for multiple scenarios for numerical simulation and risk assessment, providing a scientific basis for the optimization of foundation pit design and construction plans. Combined with real-time dynamic monitoring and data optimization, the random forest assignment method is gradually becoming an efficient and reliable solution in complex geological engineering.

[0004] Existing soil parameter prediction methods mainly rely on traditional geotechnical engineering empirical formulas or limited numerical analysis methods, which are usually unable to effectively cope with the complex geological conditions in floodplain areas. These methods usually assume that soil properties are uniformly distributed and fail to fully consider randomness and spatial variability, resulting in insufficient prediction accuracy in complex geological environments. In addition, due to the lack of dynamic monitoring and real-time adjustment capabilities, existing technologies are difficult to reflect the dynamic changes of soil parameters during construction, and cannot provide engineers with a basis for adjustment in a timely manner, which may increase construction risks.

[0005] On the other hand, some machine learning methods applied to geotechnical engineering, such as support vector machines or ordinary regression models, have improved the accuracy of soil parameter prediction to a certain extent, but they are usually sensitive to data noise and have difficulty in handling complex nonlinear relationships between multidimensional input features. At the same time, these methods cannot evaluate the contribution of different factors to the prediction results through feature importance analysis, and lack the ability to assign values ​​to the random distribution of prediction parameters in space, which limits their application value in risk assessment and construction optimization. Therefore, current technology urgently needs an efficient method that can take into account high prediction accuracy, dynamic adjustment capabilities, and spatial variability modeling.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a random prediction and assignment method for soil parameters of floodplain foundation pit excavation, which effectively processes the multidimensional geological data of floodplain foundation pit excavation through a random forest model, captures the nonlinear relationship between complex variables, and combines feature weighting and parameter optimization to greatly improve the prediction accuracy of soil parameters and reduce the risk of construction uncertainty. Through dynamic monitoring and real-time prediction functions, it is possible to quickly respond to changes in soil parameters during construction, continuously optimize the model, ensure that the prediction results are accurate and reliable, and significantly improve construction safety. Using a random field assignment method, the spatial variability of soil parameters is modeled through an exponential decay correlation function, supporting multi-scenario construction risk assessment, optimizing design solutions, and improving the scientificity and construction economy of engineering design under complex geological environments, so as to solve the problems in the above-mentioned background technology.

[0008] In order to achieve the above object, the present invention provides the following technical solution: a random prediction and assignment method for soil parameters of floodplain foundation pit excavation, comprising the following steps:

[0009] Collect geological survey data and construction information of the floodplain foundation pit excavation area, filter out the redundant parts of characteristic parameters, retain the variables that have an impact on soil mechanical parameters, and build a complete original database;

[0010] The collected data are divided into a training set and a validation set. The training set is used for preliminary training of the model, and the validation set is used for tuning and verification of the model. Features related to soil parameters are selected as input variables, and the output variables are the physical and mechanical parameters of soil related to construction. The input parameters are standardized and normalized through feature engineering to improve the model calculation efficiency and prediction accuracy.

[0011] Use the training set to build a random forest model, and implement preliminary training by setting model parameters. Use the early stopping method to avoid overfitting during training, determine the relative importance of input features, and perform weighted averaging on the prediction results of each decision tree to generate the initial prediction results of the model.

[0012] Real-time monitoring of dynamic information during foundation pit excavation is performed, and the random forest model is input with dynamic variables to predict soil physical and mechanical parameters. According to actual construction needs, key soil parameters are collected on site for verification to ensure the reliability of the model prediction results, and the input is dynamically updated according to parameter changes during the construction process;

[0013] The prediction accuracy of the model is evaluated through the validation set, and the mean square error is used to measure the difference between the predicted value and the measured value. If the prediction result error exceeds the set threshold, the model parameters are adjusted until the validation result meets the engineering requirement standard. At the same time, the model structure is dynamically adjusted according to the field data to ensure that it continues to adapt to changes in the construction environment and improve prediction accuracy and stability;

[0014] Based on the optimized random forest model, the soil parameters of the construction area are dynamically predicted, and the random field grid is assigned values ​​by the center point method.

[0015] Preferably, when constructing the original database, low-correlation characteristic variables are eliminated through feature importance analysis. Specifically, the Pearson correlation coefficient or mutual information method is used to calculate the correlation between each variable and the physical and mechanical parameters of the soil, and variables with correlations lower than a preset threshold are eliminated to construct an efficient data set.

[0016] Preferably, during the data grouping process, the ratio of the training set to the validation set is adjusted according to the size of the data. When the data size is large, the training set accounts for 70%-80%, and the validation set accounts for 20%-30%. At the same time, the missing values ​​in the data set are filled by interpolation or random forest embedding to ensure the integrity of the model input data.

[0017] Preferably, the parameter settings of the random forest model include:

[0018] Number of trees: set to 100 to 1000, dynamically adjusted based on model stability and computational cost;

[0019] Maximum depth: no initial limit, the maximum depth is automatically determined based on the training data;

[0020] Minimum number of sample splits: set to 2 to ensure that the split nodes have sufficient data support;

[0021] Feature selection ratio: set to the square root of the total number of features to improve the efficiency of split node selection and the generalization ability of the model.

[0022] Preferably, when training the random forest model, the prediction results of the decision trees are weighted averaged, and the weight value is determined according to the importance of each feature, and the feature importance is calculated by recursive feature elimination (RFE). Each feature weight is dynamically assigned to ensure that the prediction results can better reflect the impact of key features.

[0023] Preferably, the key soil parameters collected during the dynamic monitoring process include pore water pressure, shear strength of soil and deformation data of retaining structures. At the same time, the excavation depth and speed of the foundation pit are recorded in real time and dynamically input into the random forest model to improve the accuracy of the prediction. The time interval for collecting data is set according to the construction conditions, and it is recommended to be 10 minutes to 1 hour.

[0024] Preferably, during the model optimization process, an adaptive threshold is used for the mean square error (MSE) evaluation value between the predicted results and the measured data. The threshold is dynamically adjusted according to the project type, and is set in the range of 0.05-0.1 for conventional projects and in the range of 0.01-0.05 for high-risk projects. When the MSE value exceeds the set threshold, the model parameters are automatically adjusted, including increasing the number of decision trees or adjusting the feature importance weights.

[0025] Preferably, when assigning values ​​to the random field grid by the central point method, an exponential decay correlation function is used to determine the spatial correlation between random variables. The specific formula is:

[0026] Where (x1, y1) and (x2, y2) are the coordinates of any two points in space, ρ(x1-x2, y1-y2) represents the correlation function between two spatial points, which is used to describe the degree of correlation between (x1, y1) and (x2, y2) in space. θ1 and θ2 are the horizontal and vertical correlation distances, respectively. The correlation distance is the integral of the correlation function over the entire space. The calculation expression is: Where τ represents the spatial distance, that is, the distance between two spatial points, which is used to describe the spatial position change of variable correlation, and ρ(τ) represents the correlation function, which describes the correlation strength between two spatial points when the distance is τ.

[0027] Preferably, the center point method is based on spatial correlation, and uses the relative position of the grid center point to calculate the parameter correlation matrix. The spatial distribution characteristics between random variables are determined according to the correlation function, and finally the spatially randomly distributed soil parameter values ​​are generated for construction simulation and multi-scenario risk assessment, so as to realize the dynamic change of soil parameters in different spatial positions.

[0028] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0029] By processing the multidimensional data of the floodplain foundation pit excavation area through the random forest model, the present invention can capture the nonlinear relationship between complex variables, overcome the limitations of traditional empirical formulas and linear models in complex geological environments, and combine the random forest model with feature weighting and parameter optimization to ensure that the prediction results are closer to the actual soil mechanical properties, especially in the prediction of key parameters such as soil strength, internal friction angle and moisture content. It shows high accuracy. During the dynamic adjustment process, the model is verified and optimized in real time to keep it adapted to the ever-changing construction environment. Compared with traditional methods, the present invention significantly improves the prediction accuracy, provides more reliable soil parameter input for engineering design, and reduces the risk of uncertainty in construction.

[0030] By combining real-time dynamic monitoring with the random forest model, the present invention can quickly respond to changes in soil parameters during construction, timely adjust the model prediction results, and provide accurate guidance for construction operations. The dynamic monitoring and real-time prediction functions of the present invention reduce the delayed response problem in construction and improve the real-time safety management and control capabilities of the project. In addition, by continuously inputting on-site collection data and error assessment optimization models, it can better adapt to construction needs under complex conditions and significantly reduce the possibility of construction accidents.

[0031] Through the random field assignment method, the present invention solves the problem that traditional models cannot effectively reflect the spatial variability of soil parameters. The exponential decay correlation function is used to calculate the spatial correlation of soil parameters, and the generated random field can not only reflect the natural fluctuations of parameters at different locations, but also support multi-scenario construction risk assessment. For example, by simulating the soil strength distribution at multiple locations, possible weak soil areas can be identified in advance, thereby optimizing support design or adjusting construction plans. The present invention provides accurate and flexible input data for numerical simulations in complex geological environments, improves the scientificity and reliability of engineering design, and provides guarantees for safety and economy during construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present application 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 recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0033] Figure 1 The present invention is a method flow chart of the random prediction and assignment method of soil parameters for floodplain foundation pit excavation.

[0034] Figure 2 Schematic diagram of the random forest weighted prediction process of the present invention.

[0035] Figure 3It is a schematic diagram of the grid of spatial variability of soil parameters based on random fields in the present invention.

[0036] Figure 4 This is a flow chart for predicting soil parameters for floodplain foundation pit excavation according to the present invention. DETAILED DESCRIPTION

[0037] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0038] The present invention provides Figures 1 to 4 The random prediction and assignment method of soil parameters for floodplain foundation pit excavation shown in the figure includes the following steps:

[0039] Collect geological survey data and construction information of the floodplain excavation area, including but not limited to cohesion, internal friction angle, shear strength, soil thickness, density, water content, soil type, soil particle characteristics, geological age, hydrological characteristics, excavation depth, excavation speed, retaining structure type and other parameters. These data are obtained through geological exploration, field tests and laboratory tests, and formatted according to data characteristics to provide standardized input for subsequent modeling. At the same time, screen the redundant parts of the characteristic parameters, retain the variables that have a significant impact on the mechanical parameters of the soil, and build a complete original database.

[0040] When constructing the original database, low-correlation characteristic variables are eliminated through feature importance analysis. Specifically, the Pearson correlation coefficient or mutual information method is used to calculate the correlation between each variable and the physical and mechanical parameters of the soil, and the variables with correlations lower than the preset threshold are eliminated to construct an efficient data set.

[0041] The collected data is divided into a training set and a validation set. The training set is used for the initial training of the model, and the validation set is used for model tuning and verification. The main features related to soil parameters are further selected as input variables, such as soil layer type, geological age, foundation pit depth, etc. The output variables are the physical and mechanical parameters of soil related to construction (including but not limited to shear strength, internal friction angle and cohesion). The input parameters are standardized and normalized through feature engineering to improve the model calculation efficiency and prediction accuracy.

[0042] During the data grouping process, the ratio of the training set to the validation set is adjusted according to the size of the data. When the data volume is large, the training set accounts for 70%-80%, and the validation set accounts for 20%-30%. At the same time, the missing values ​​in the data set are filled by interpolation or random forest embedding method to ensure the integrity of the model input data.

[0043] The random forest model is constructed using the training set, and preliminary training is performed by setting model parameters (such as the number of trees, maximum depth, minimum number of sample splits, and number of leaf node samples). The early stopping method is used during training to avoid overfitting, determine the relative importance of input features, and perform weighted averaging on the prediction results of each decision tree to generate the initial prediction results of the model. The basis for feature weighting is the importance of the feature, and the stability and high prediction accuracy of the model are achieved by optimizing the feature selection ratio and tree depth.

[0044] The parameter settings of the random forest model include:

[0045] Number of trees: set to 100 to 1000, dynamically adjusted based on model stability and computational cost;

[0046] Maximum depth: no initial limit, the maximum depth is automatically determined based on the training data;

[0047] Minimum number of sample splits: set to 2 to ensure that the split nodes have sufficient data support;

[0048] Feature selection ratio: set to the square root of the total number of features to improve the efficiency of split node selection and the generalization ability of the model.

[0049] When training a random forest model, the prediction results of the decision trees are weighted averaged, and the weight value is determined by the importance of each feature, which is calculated by recursive feature elimination (RFE). Each feature weight is dynamically assigned to ensure that the prediction results better reflect the impact of key features.

[0050] The dynamic information during the foundation pit excavation process is monitored in real time. The data includes dynamic variables such as construction progress, excavation depth, soil layer type, and retaining structure construction parameters. These dynamic data are combined to input the random forest model to predict the physical and mechanical parameters of the soil. At the same time, according to the actual construction needs, key soil parameters (such as shear strength, pore water pressure, etc.) are collected on site for verification to ensure the reliability of the model prediction results, and the input is dynamically updated according to the parameter changes during the construction process.

[0051] The key soil parameters collected during the dynamic monitoring process include pore water pressure, shear strength of soil and deformation data of retaining structures. At the same time, the excavation depth and speed of the foundation pit are recorded in real time and dynamically input into the random forest model to improve the accuracy of the prediction. The time interval for collecting data is set according to the construction conditions, and it is recommended to be 10 minutes to 1 hour.

[0052] The prediction accuracy of the model is evaluated through the validation set, and the mean square error (MSE) is used to measure the difference between the predicted value and the measured value. If the prediction result error exceeds the set threshold, the model parameters are adjusted, such as increasing the number of decision trees, optimizing the feature selection ratio, or adjusting the model depth, until the verification result meets the engineering requirements. At the same time, the model structure is dynamically adjusted according to the field data to ensure that it continues to adapt to changes in the construction environment, thereby improving the prediction accuracy and stability.

[0053] During the model optimization process, an adaptive threshold is used for the mean square error (MSE) evaluation value between the predicted results and the measured data. The threshold is dynamically adjusted according to the project type, and is set in the range of 0.05-0.1 for conventional projects and in the range of 0.01-0.05 for high-risk projects. When the MSE value exceeds the set threshold, the model parameters are automatically adjusted, including increasing the number of decision trees or adjusting the feature importance weights.

[0054] When assigning values ​​to the random field grid using the central point method, the exponential decay correlation function is used to determine the spatial correlation between random variables. The specific formula is:

[0055] Where (x1, y1) and (x2, y2) are the coordinates of any two points in space, ρ(x1-x2, y1-y2) represents the correlation function between two spatial points, which is used to describe the degree of correlation between (x1, y1) and (x2, y2) in space. θ1 and θ2 are the horizontal and vertical correlation distances, respectively. The correlation distance is the integral of the correlation function over the entire space. The calculation expression is: Where τ represents the spatial distance, that is, the distance between two spatial points, which is used to describe the spatial position change of variable correlation, and ρ(τ) represents the correlation function, which describes the correlation strength between two spatial points when the distance is τ.

[0056] Based on the optimized random forest model, the soil parameters of the construction area are dynamically predicted, and the random field grid is assigned values ​​through the center point method. The center point method is based on spatial correlation, and the parameter correlation matrix is ​​calculated using the relative position of the grid center point. The spatial distribution characteristics between random variables are determined according to the correlation function. Finally, the spatially randomly distributed soil parameter values ​​are generated for construction simulation and multi-scenario risk assessment, realizing the dynamic changes of soil parameters in different spatial positions.

[0057] Specific implementation method 1: In the floodplain foundation pit excavation project, first collect relevant data of the area through geological survey and construction records. These data include cohesion, internal friction angle, shear strength, soil thickness, density, moisture content, soil type, soil particle characteristics, geological age and hydrological characteristics. In order to ensure data quality, feature importance analysis is used to eliminate redundant variables. Common methods include Pearson correlation coefficient and mutual information analysis. For example, if certain soil particle characteristics have little effect on cohesion, they can be removed from the input variables to reduce the model calculation burden and improve accuracy. The data is normalized, for example, the shear strength value is normalized to the interval [0, 1] to avoid the impact of dimensional differences between eigenvalues ​​on model performance.

[0058] Next, the processed data is divided into a training set and a validation set. The training set usually accounts for 70%-80%, and the validation set accounts for 20%-30%. The proportion is adjusted according to the amount of data to ensure that the model training has sufficient data and the verification results are representative. In the training stage, the random forest algorithm is used to build the model, and preliminary training is achieved by adjusting parameters such as the number of trees, maximum depth, minimum number of split samples, etc. For example, the number of trees is set to 500 and the maximum depth is infinite. The model can automatically determine the growth depth of the tree to capture the nonlinear relationship between the input variables. In order to prevent overfitting, the early stopping method is used to stop training when the model performance is no longer significantly improved. After the model training is completed, weights are assigned according to the importance of the features to determine the impact of different features in the final prediction. For example, the internal friction angle may be given a higher weight because it contributes significantly to the prediction of soil strength.

[0059] The prediction results are generated by weighted averaging. For example, assuming that the input variables include shear strength and moisture content, with importance weights of 1.5 and 1.0 respectively, the prediction results of the decision tree of the random forest are weighted according to these weights, and the final output is the accurate physical and mechanical parameter values ​​of the soil.

[0060] This implementation method improves the accuracy of soil parameter prediction through the nonlinear fitting ability of the random forest model. Feature importance analysis and normalization processing effectively reduce the interference of redundant data on the model, thereby optimizing model performance. Compared with traditional linear models or empirical formulas, this method can better capture the nonlinear relationship of multidimensional variables under complex geological conditions and provide a scientific basis for construction.

[0061] Specific implementation method 2: During the excavation of the floodplain foundation pit, the construction site collects dynamic data in real time, including excavation depth, deformation of the retaining structure, pore water pressure, soil type and construction speed. In order to achieve dynamic monitoring, a sensor network is arranged in the construction area, and the time interval of the sensor can be adjusted according to the project requirements (such as 10 minutes to 1 hour) to ensure that the data update frequency meets the real-time requirements. For example, deformation monitors are set up at key locations to collect displacement data of the retaining structure; pore water pressure sensors are set up to record the changes in groundwater in real time during the foundation pit excavation process.

[0062] The collected dynamic data is directly input into the trained random forest model for prediction. For example, when the excavation depth increases, the friction angle in the soil may decrease. At this time, the model can adjust the value of the predicted soil parameters by inputting dynamic variables. In addition, key physical parameters (such as shear strength) need to be collected during the construction process, and these actual data are used to verify the accuracy of the model prediction results. The difference between the model prediction value and the measured value is evaluated by the mean square error (MSE). For example, when the MSE is 0.01, it means that the model has a high prediction accuracy. If the MSE exceeds the set threshold (such as 0.01-0.05 in high-risk projects), the model will automatically adjust, such as increasing the number of decision trees, adjusting feature weights, or optimizing input parameters.

[0063] The dynamic prediction results are fed back to engineers through the construction visualization platform. For example, if the model indicates that the current soil strength is lower than the design value, engineers can immediately take reinforcement measures, such as reducing the excavation depth or installing support structures. This construction guidance method that combines dynamic monitoring with real-time prediction can quickly respond to changes in soil conditions during construction.

[0064] The combination of dynamic monitoring and real-time prediction enables real-time control of construction risks. Compared with traditional static prediction methods, this method can dynamically adjust input variables to reflect geological changes during foundation pit excavation, thereby improving construction safety and reliability. At the same time, combined with data collected by sensors, the optimized model can continuously adapt to site conditions and provide timely and scientific guidance for construction.

[0065] Specific implementation method 3: Random field assignment to achieve spatial variability modeling

[0066] In the foundation pit construction area, the soil is divided into multiple random field grid units, each unit represents the soil characteristics of a specific spatial position. The division density of the random field grid unit is dynamically adjusted according to the construction accuracy requirements. For example, in high-precision projects, the unit grid side length is 5 meters, and in ordinary projects it is 10 meters. The soil parameter value is calculated at the center point of each grid unit. The center point method is based on the prediction results of the random forest model and is assigned in combination with the spatial correlation formula.

[0067] The spatial correlation is represented by an exponential decay function, as follows:

[0068] Among them, θ1 and θ2 represent the horizontal and vertical correlation distances respectively. The specific values ​​are determined according to the regional geological conditions and monitoring data. For example, a smaller value may be taken in sandy soil and a larger value may be taken in clay soil. The predicted value of the soil parameters of each grid unit is calculated through the correlation function, and a random field is generated. The random field can not only reflect the spatial variability of soil parameters, but also provide input data for construction simulation of multiple scenarios.

[0069] The generated random fields are used for construction risk assessment. For example, if the parametric simulation predicts that there may be a risk of low soil strength at a certain location, engineers can design reinforcement measures in advance to avoid unexpected problems during construction.

[0070] The random field assignment method effectively reflects the randomness and spatial variability of soil parameters through the spatial correlation formula. Compared with the traditional method of uniform assumption, this implementation method significantly improves the simulation accuracy and provides a scientific basis for construction optimization in complex geological environments. At the same time, the random field can generate multi-scenario simulation data, providing more possibilities for design and risk control.

[0071] By processing the multidimensional data of the floodplain foundation pit excavation area through the random forest model, the present invention can capture the nonlinear relationship between complex variables and overcome the limitations of traditional empirical formulas and linear models in complex geological environments. The random forest model combines feature weighting and parameter optimization to ensure that the prediction results are closer to the actual soil mechanical properties, especially in the prediction of key parameters such as soil strength, internal friction angle and moisture content. In the dynamic adjustment process, the model is verified and optimized in real time so that it can always adapt to the ever-changing construction environment. Compared with traditional methods, the present invention significantly improves the prediction accuracy, provides more reliable soil parameter input for engineering design, and reduces the risk of uncertainty in construction.

[0072] By combining real-time dynamic monitoring with a random forest model, the present invention can quickly respond to changes in soil parameters during construction, adjust the model prediction results in a timely manner, and provide accurate guidance for construction operations. For example, if the soil strength is monitored to be weakened during construction, the model can immediately feedback risk warnings to help engineers develop response plans, such as adjusting the excavation speed or adding support structures. The dynamic monitoring and real-time prediction functions of the present invention reduce the problem of delayed response during construction and improve the real-time safety management and control capabilities of the project. In addition, by continuously inputting on-site collection data and error assessment optimization models, it can better adapt to construction needs under complex conditions and significantly reduce the possibility of construction accidents.

[0073] Through the random field assignment method, the present invention solves the problem that traditional models cannot effectively reflect the spatial variability of soil parameters. The exponential decay correlation function is used to calculate the spatial correlation of soil parameters, and the generated random field can not only reflect the natural fluctuations of parameters at different locations, but also support multi-scenario construction risk assessment. For example, by simulating the soil strength distribution at multiple locations, possible weak soil areas can be identified in advance, thereby optimizing support design or adjusting construction plans. The present invention provides accurate and flexible input data for numerical simulations in complex geological environments, improves the scientificity and reliability of engineering design, and provides guarantees for safety and economy during construction.

[0074] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A random prediction and assignment method for soil parameters of floodplain foundation pit excavation, characterized in that: The following steps are involved: Collect geological survey data and construction information of the floodplain foundation pit excavation area, filter out the redundant parts of characteristic parameters, retain the variables that have an impact on soil mechanical parameters, and build a complete original database; The collected data are divided into a training set and a validation set. The training set is used for preliminary training of the model, and the validation set is used for tuning and verification of the model. Features related to soil parameters are selected as input variables, and the output variables are the physical and mechanical parameters of soil related to construction. The input parameters are standardized and normalized through feature engineering to improve the model calculation efficiency and prediction accuracy. Use the training set to build a random forest model, and implement preliminary training by setting model parameters. Use the early stopping method to avoid overfitting during training, determine the relative importance of input features, and perform weighted averaging on the prediction results of each decision tree to generate the initial prediction results of the model. Real-time monitoring of dynamic information during foundation pit excavation is performed, and the random forest model is input with dynamic variables to predict soil physical and mechanical parameters. According to actual construction needs, key soil parameters are collected on site for verification to ensure the reliability of the model prediction results, and the input is dynamically updated according to parameter changes during the construction process; The prediction accuracy of the model is evaluated through the validation set, and the mean square error is used to measure the difference between the predicted value and the measured value. If the prediction result error exceeds the set threshold, the model parameters are adjusted until the validation result meets the engineering requirement standard. At the same time, the model structure is dynamically adjusted according to the field data to ensure that the model continues to adapt to changes in the construction environment and improve the prediction accuracy and stability; Based on the optimized random forest model, the soil parameters of the construction area are dynamically predicted, and the random field grid is assigned values ​​by the center point method.

2. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: When constructing the original database, low-correlation characteristic variables are eliminated through feature importance analysis. Specifically, the Pearson correlation coefficient or mutual information method is used to calculate the correlation between each variable and the physical and mechanical parameters of the soil, and the variables with correlations lower than the preset threshold are eliminated to construct an efficient data set.

3. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: During the data grouping process, the ratio of the training set to the validation set is adjusted according to the size of the data. When the data volume is large, the training set accounts for 70%-80%, and the validation set accounts for 20%-30%. At the same time, the missing values ​​in the data set are filled by interpolation or random forest embedding method to ensure the integrity of the model input data.

4. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: The parameter settings of the random forest model include: Number of trees: set to 100 to 1000, dynamically adjusted based on model stability and computational cost; Maximum depth: no initial limit, the maximum depth is automatically determined based on the training data; Minimum number of sample splits: set to 2 to ensure that the split nodes have data support; Feature selection ratio: set to the square root of the total number of features to improve the efficiency of split node selection and the generalization ability of the model.

5. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: When training the random forest model, the prediction results of the decision trees are weighted averaged. The weight value is determined based on the importance of each feature. The feature importance is calculated by recursive feature elimination. The weight of each feature is dynamically assigned to ensure that the prediction results reflect the impact of key features.

6. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: The key soil parameters collected during the dynamic monitoring process include pore water pressure, shear strength of the soil and deformation data of the retaining structure. At the same time, the excavation depth and speed of the foundation pit are recorded in real time and dynamically input into the random forest model to improve the accuracy of the prediction.

7. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: During the model optimization process, an adaptive threshold is used for the mean square error evaluation value between the predicted results and the measured data. The threshold is dynamically adjusted according to the project type, and is set in the range of 0.05-0.1 for conventional projects and in the range of 0.01-0.05 for high-risk projects. When the mean square error exceeds the set threshold, the model parameters are automatically adjusted, including increasing the number of decision trees or adjusting the feature importance weights.

8. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: When assigning values ​​to the random field grid using the central point method, the exponential decay correlation function is used to determine the spatial correlation between random variables. The specific formula is: Where (x1, y1) and (x2, y2) are the coordinates of any two points in space, ρ(x1-x2, y1-y2) represents the correlation function between two spatial points, which is used to describe the degree of correlation between (x1, y1) and (x2, y2) in space. θ1 and θ2 are the horizontal and vertical correlation distances, respectively. The correlation distance is the integral of the correlation function over the entire space. The calculation expression is: Where τ represents the spatial distance, that is, the distance between two spatial points, which is used to describe the spatial position change of variable correlation, and ρ(τ) represents the correlation function, which describes the correlation strength between two spatial points when the distance is τ.

9. The random prediction and assignment method for soil parameters of floodplain foundation pit excavation according to claim 1 is characterized in that: The center point method is based on spatial correlation. It uses the relative position of the grid center point to calculate the parameter correlation matrix, determines the spatial distribution characteristics between random variables according to the correlation function, and finally generates spatially randomly distributed soil parameter values ​​for construction simulation and multi-scenario risk assessment, realizing the dynamic changes of soil parameters in different spatial positions.

Images