Earthwork volume prediction method based on kriging interpolation and classification regression tree for shield construction

Abstract

This invention discloses a method for predicting earthwork volume in shield tunneling construction based on Kriging interpolation and classification regression trees, relating to the field of earthwork volume prediction in shield tunneling construction. This invention solves the prediction accuracy problem of traditional Kriging interpolation in areas of local geological variation by real-time acquisition of construction parameters, geological parameters, and environmental parameters, combined with a dynamic adjustment of the variogram function and a multi-scale interpolation fusion strategy. Simultaneously, it improves the accuracy of spatial prediction by dynamically correcting the stratum interpolation results using a Bayesian update method. Furthermore, by employing a dynamic weight adjustment strategy combined with a Bayesian fusion model, the prediction model weights at different stages can be adjusted according to error feedback during construction, thereby achieving more accurate earthwork volume prediction.

Description

Technical Field

[0001] This invention relates to the field of earthwork volume prediction in shield tunneling construction, specifically a method for predicting earthwork volume in shield tunneling construction based on Kriging interpolation and classification regression trees. Background Technology

[0002] In tunnel boring machine (TBM) construction, the prediction of earthwork volume is crucial for construction progress and cost control. Traditional earthwork volume prediction methods typically rely on historical data and simple mathematical models, but these methods often fail to adequately account for the dynamic changes in construction parameters, geological conditions, and environmental factors. Therefore, many researchers have attempted to combine machine learning algorithms with traditional numerical methods to improve the accuracy of earthwork volume prediction. For example, some methods employ multiple linear regression models to predict earthwork volume, based on statistical analysis of construction parameters and geological data. However, traditional methods often exhibit low prediction accuracy when dealing with complex geological conditions and construction environments, especially in scenarios with drastic geological changes and complex construction conditions.

[0003] Most existing earthwork volume prediction technologies rely on traditional regression analysis methods or simple interpolation algorithms. These methods often struggle to adapt to complex situations when faced with variability in construction parameters and geological environments. While traditional Kriging interpolation can spatially interpolate geological data to some extent, its variogram has limited adjustment capabilities for areas with abrupt changes in strata, resulting in poor prediction accuracy in these regions. Therefore, existing technologies still have significant shortcomings in terms of accuracy and adaptability for multi-parameter, multi-stage earthwork volume prediction. Summary of the Invention

[0004] This invention proposes a method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees. By collecting construction parameters, geological parameters, and environmental parameters in real time, and combining a dynamic adjustment of the variogram function and a multi-scale interpolation fusion strategy, it solves the prediction accuracy problem of traditional Kriging interpolation in areas with local geological changes.

[0005] The method for predicting earthwork volume in tunnel boring machine (TBM) construction based on Kriging interpolation and classification regression trees includes the following steps:

[0006] S1. Real-time acquisition of construction parameters, geological parameters, and environmental parameters generated during the tunnel boring machine construction process, and outlier detection, missing value imputation, and smoothing of the acquired data;

[0007] S2. Establish a spatial data model of the shield tunneling area through geological parameters, and use a combination of local and global interpolation, dynamic adjustment of the variogram function and multi-scale interpolation fusion strategy to spatially predict the strata data. During the shield tunneling process, use the Bayesian update method to dynamically correct the interpolation results.

[0008] S3. Based on construction parameters, geological parameters, and environmental parameters, feature selection and interactive analysis are performed to construct an initial classification and regression tree model. The model is then updated and optimized online using a dynamic pruning algorithm and an incremental learning strategy.

[0009] S4. Based on the trained adaptive classification and regression tree model, predict the changes in earthwork volume in the short, medium and long term respectively, and input the short, medium and long term prediction results into the Bayesian fusion model to calculate the confidence weight of different models. Through a dynamic weight adjustment strategy, adjust the weight of each prediction model according to the actual error in the construction process to generate the final earthwork volume prediction result.

[0010] Furthermore, in step S1, the construction parameters include at least the tunnel boring machine's advance speed, cutterhead rotation speed, soil chamber pressure, and screw conveyor rotation speed; the geological parameters include at least the stratum distribution, moisture content, sand content, and porosity within the tunnel construction area; and the environmental parameters include at least the groundwater level and surface settlement within the tunnel construction area.

[0011] Furthermore, step S2 specifically includes the following steps:

[0012] S201. Establish an initial spatial data grid based on the preprocessed geological parameters, and calculate the variogram function of the initial spatial data grid using the traditional Kriging interpolation method;

[0013] S202. For areas with abrupt changes in strata, dynamically adjust the parameters of the local variogram and remodel the local geological features to establish a local interpolation model;

[0014] S203. Based on historical data, calculate the geological trend of the entire shield tunneling area and use global kriging interpolation to fit it and establish a global interpolation model;

[0015] S204. Utilize a multi-scale interpolation fusion strategy to assign weights and fuse local interpolation models with global interpolation models.

[0016] Furthermore, in step S201, the mutation function is specifically expressed as follows:

[0017] ;

[0018] Among them, the Represents the variogram function, the Indicating spatial distance, the Indicates distance as The number of sample point pairs, the Indicates the sample index, the Indicates position The formation parameter values ​​at the location, the Indicates position The stratigraphic parameter values ​​at that location.

[0019] Furthermore, step S3 specifically includes the following steps:

[0020] S301. Construct a candidate feature set based on the preprocessed construction parameters, geological parameters, and environmental parameters;

[0021] S302. Perform information entropy calculation and feature interaction analysis on the constructed candidate feature set, quantitatively evaluate the impact of each feature on the change of earthwork volume, screen key influencing factors that have significant contributions to the prediction results, and remove redundant or low-contribution features.

[0022] S303. Construct an initial classification regression tree model based on the selected key influencing factors, and use the Gini coefficient or information gain criterion to determine the optimal split point at each split node to capture the nonlinear relationship between construction parameters and earthwork volume.

[0023] S304. The initial classification and regression tree is optimized by an adaptive pruning algorithm, and redundant branches are automatically removed by evaluating the error contribution of each branch;

[0024] S305. Through an incremental learning strategy, the classification regression tree model is updated online based on new data during the tunnel boring machine construction process.

[0025] Furthermore, step S302 specifically includes the following sub-steps:

[0026] S3021. Calculate the mutual information values ​​among construction parameters, geological parameters, and environmental parameters, measure the degree of correlation between each variable and earthwork volume, and use the information entropy analysis method to evaluate the contribution of variables to the final prediction accuracy;

[0027] S3022. Using the maximum correlation minimum redundancy algorithm, variables that are highly correlated with other features but have low information gain are eliminated;

[0028] S3023. After feature selection is completed, different construction scenarios are classified, and key influencing factors under each scenario are categorized.

[0029] Furthermore, step S303 specifically includes the following sub-steps:

[0030] S3031. Combining the CART algorithm, using the selected key influencing factors as input variables and the earthwork volume per unit time as the target variable, an initial tree structure is constructed;

[0031] S3032. Calculate the Gini coefficient or information gain of all candidate features at each split node, and select the optimal feature as the basis for splitting;

[0032] S3033. Determine the optimal split point based on data characteristics, and capture the nonlinear relationship between construction parameters and earthwork volume for each split by using the minimum variance decreasing criterion information gain ratio.

[0033] Furthermore, step S4 specifically includes the following sub-steps:

[0034] S401. Generate short-term, medium-term, and long-term prediction results based on the trained adaptive classification and regression tree model;

[0035] S402. Input the short-term, medium-term and long-term prediction results into the constructed Bayesian fusion model, and calculate the confidence weights through the Bayesian fusion model;

[0036] S404. A dynamic weight adjustment strategy is adopted to correct the weights of each prediction model online based on real-time feedback data and error analysis results during the construction process, so that the Bayesian fusion model can be updated adaptively.

[0037] S405. Generate the final comprehensive prediction results.

[0038] Furthermore, step S401 specifically includes the following sub-steps:

[0039] S4011. In the short-term prediction phase, a sliding window with a time step of 5 minutes is used to summarize the real-time collected construction data into a short-term dataset, and this dataset is input into an adaptive classification and regression tree model to output the earthwork volume prediction results for the next 5 minutes.

[0040] S4012. In the intermediate prediction stage, a sliding window with a time step of 30 minutes is used to summarize the construction data collected in the most recent 30 minutes into an intermediate dataset. This intermediate dataset is then input into an adaptive classification and regression tree model to output the earthwork volume prediction results for the next 30 minutes.

[0041] S4013. In the long-term prediction phase, historical data covering the entire construction cycle is used as a long-term dataset, and this long-term dataset is input into an adaptive classification and regression tree model to output a prediction of the earthwork volume change trend throughout the entire construction cycle.

[0042] Furthermore, in step S402, the specific process for calculating the credibility weight through the Bayesian fusion model is as follows:

[0043] ;

[0044] ;

[0045] ;

[0046] Among them, the The credibility weight is represented by the following. Indicates the current construction data The lower model The posterior probability, the Indicates the current construction data The lower model The posterior probability, the Indicates the type of result, including short-term, medium-term, and long-term forecast results. Representation Model Generate current data The likelihood probability, the Representation Model Generate current data The likelihood probability, the Represents the prediction variance, the This indicates the prediction error.

[0047] The beneficial effects of the invention are:

[0048] This invention improves the accuracy of spatial prediction by dynamically correcting the stratigraphic interpolation results using a Bayesian update method. In constructing the classification and regression tree model, this invention introduces an adaptive pruning algorithm and an incremental learning strategy, enabling online updates and optimization of the model based on real-time construction data. This avoids the overfitting problem in traditional methods and enhances the model's adaptability to new data. Attached Figure Description

[0049] Figure 1 The flowchart illustrates the method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees, as provided in this embodiment of the invention. Detailed Implementation

[0050] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the invention, and not all of them. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0053] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or machine that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or machine. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or machine that includes said element.

[0054] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0055] Example 1

[0056] Among them, such as Figure 1 A method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees includes the following steps:

[0057] S1. Real-time acquisition of construction parameters, geological parameters, and environmental parameters generated during the tunnel boring machine construction process, and outlier detection, missing value imputation, and smoothing of the acquired data;

[0058] S2. Establish a spatial data model of the shield tunneling area through geological parameters, and use a combination of local and global interpolation, dynamic adjustment of the variogram function and multi-scale interpolation fusion strategy to spatially predict the strata data. During the shield tunneling process, use the Bayesian update method to dynamically correct the interpolation results.

[0059] S3. Based on construction parameters, geological parameters, and environmental parameters, feature selection and interactive analysis are performed to construct an initial classification and regression tree model. The model is then updated and optimized online using a dynamic pruning algorithm and an incremental learning strategy.

[0060] S4. Based on the trained adaptive classification and regression tree model, predict the changes in earthwork volume in the short, medium and long term respectively, and input the short, medium and long term prediction results into the Bayesian fusion model to calculate the confidence weight of different models. Through a dynamic weight adjustment strategy, adjust the weight of each prediction model according to the actual error in the construction process to generate the final earthwork volume prediction result.

[0061] Furthermore, in step S1, the construction parameters include at least the tunnel boring machine's advance speed, cutterhead rotation speed, soil chamber pressure, and screw conveyor rotation speed; the geological parameters include at least the stratum distribution, water content, sand content, and porosity within the tunnel construction area; and the environmental parameters include at least the groundwater level and surface settlement within the tunnel construction area. Specifically, the implementation process of step S1 is as follows:

[0062] During the real-time acquisition and preprocessing stage, sensors are installed to collect construction parameters, geological parameters, and environmental parameters of the tunnel boring machine to ensure the real-time nature and stability of the data. Outlier detection is performed on the collected data, and statistical methods are used to identify and remove outliers. At the same time, for missing values, interpolation or machine learning methods are used to fill them in. Then, smoothing is performed using moving average and weighted average methods to reduce noise in the data and ensure data stability.

[0063] Furthermore, step S2 specifically includes the following steps:

[0064] S201. Establish an initial spatial data grid based on the preprocessed geological parameters, and calculate the variogram function of the initial spatial data grid using the traditional Kriging interpolation method;

[0065] S202. For areas with abrupt changes in strata, dynamically adjust the parameters of the local variogram and remodel the local geological features to establish a local interpolation model;

[0066] S203. Based on historical data, calculate the geological trend of the entire shield tunneling area and use global kriging interpolation to fit it and establish a global interpolation model;

[0067] S204. Utilize a multi-scale interpolation fusion strategy to assign weights and fuse local interpolation models with global interpolation models.

[0068] Furthermore, in step S201, the mutation function is specifically expressed as follows:

[0069] ;

[0070] Among them, the Represents the variogram function, the Indicating spatial distance, the Indicates distance as The number of sample point pairs, the Indicates the sample index, the The stratigraphic parameter values ​​at the specified location, Indicates position The stratigraphic parameter values ​​at that location.

[0071] Specifically, in the spatial modeling process of geological data, the first step is to establish a preliminary spatial data grid based on the processed geological data. The construction of the spatial data grid takes into account the geological characteristics of the region and ensures that it can reflect various geological changes within the region. At this point, the Kriging interpolation method is used. Its basic principle is to quantify the degree of geological change at different locations in space by calculating the variogram based on the distance relationship between geological data points. When performing spatial interpolation, considering the existence of abrupt changes in stratigraphy, a strategy of dynamically adjusting the variogram is adopted. This means that in abrupt change areas, the parameters of the interpolation method are automatically adjusted according to local geological characteristics to enhance the accuracy of the interpolation. For non-abrupt change areas, global Kriging interpolation is used to fit the overall stratigraphic trend within the region. The advantage of Kriging interpolation is that it can use spatial autocorrelation to infer geological parameters at unknown locations and has a certain robustness to data noise. By fusing local and global interpolation methods, it is ensured that the model can not only accurately describe the characteristics of abrupt change areas but also reasonably reflect the overall stratigraphic changes.

[0072] Furthermore, step S3 specifically includes the following steps:

[0073] S301. Construct a candidate feature set based on the preprocessed construction parameters, geological parameters, and environmental parameters;

[0074] S302. Perform information entropy calculation and feature interaction analysis on the constructed candidate feature set, quantitatively evaluate the impact of each feature on the change of earthwork volume, screen key influencing factors that have significant contributions to the prediction results, and remove redundant or low-contribution features.

[0075] S303. Construct an initial classification regression tree model based on the selected key influencing factors, and use the Gini coefficient or information gain criterion to determine the optimal split point at each split node to capture the nonlinear relationship between construction parameters and earthwork volume.

[0076] S304. The initial classification and regression tree is optimized by an adaptive pruning algorithm, and redundant branches are automatically removed by evaluating the error contribution of each branch;

[0077] S305. Through an incremental learning strategy, the classification regression tree model is updated online based on new data during the tunnel boring machine construction process.

[0078] Specifically, the Classification and Regression Tree (CART) model is the core of predicting construction earthwork volume. Through recursive splitting of a tree structure, the CART model can capture the complex nonlinear relationships between different construction parameters and earthwork volume. When constructing a CART model, feature selection and interaction analysis are first performed on construction parameters, geological parameters, and environmental parameters. The core purpose of this step is to eliminate redundant information and identify the features most relevant to changes in earthwork volume. During feature selection, the contribution of each feature to the final prediction result can be evaluated by calculating the information entropy value. Furthermore, mutual information analysis is used to quantify the interrelationships between different parameters and earthwork volume. After feature selection, a decision tree is constructed based on the CART algorithm. The optimal split point is determined by calculating the Gini coefficient or information gain of each node, thereby optimizing the tree structure. The advantage of the CART model is its ability to handle large-scale high-dimensional data, and through the partitioning of nodes, the model can capture the impact of key parameters during construction on earthwork volume. To improve the model's generalization ability, an adaptive pruning method is used to dynamically prune redundant branches during the tree's growth to avoid overfitting. Furthermore, to further improve the accuracy of the prediction model, an incremental learning strategy is employed to dynamically update the classification and regression tree. The incremental learning method continuously updates the model based on new construction data during the tunnel boring machine (TBM) construction process, making the predictions more closely reflect the gradually changing environment and conditions during construction. With each update, the model adjusts the tree structure and weights based on the newly added data, making it more adaptable to the new construction environment.

[0079] Furthermore, step S302 specifically includes the following sub-steps:

[0080] S3021. Calculate the mutual information values ​​among construction parameters, geological parameters, and environmental parameters, measure the degree of correlation between each variable and earthwork volume, and use the information entropy analysis method to evaluate the contribution of variables to the final prediction accuracy;

[0081] S3022. Using the maximum correlation minimum redundancy algorithm, variables that are highly correlated with other features but have low information gain are eliminated;

[0082] S3023. After feature selection is completed, different construction scenarios are classified, and key influencing factors under each scenario are categorized.

[0083] Furthermore, step S303 specifically includes the following sub-steps:

[0084] S3031. Combining the CART algorithm, using the selected key influencing factors as input variables and the earthwork volume per unit time as the target variable, an initial tree structure is constructed;

[0085] S3032. Calculate the Gini coefficient or information gain of all candidate features at each split node, and select the optimal feature as the basis for splitting;

[0086] S3033. Determine the optimal split point based on data characteristics, and capture the nonlinear relationship between construction parameters and earthwork volume for each split by using the minimum variance decreasing criterion information gain ratio.

[0087] Furthermore, step S4 specifically includes the following sub-steps:

[0088] S401. Generate short-term, medium-term, and long-term prediction results based on the trained adaptive classification and regression tree model;

[0089] S402. Input the short-term, medium-term and long-term prediction results into the constructed Bayesian fusion model, and calculate the confidence weights through the Bayesian fusion model;

[0090] S404. A dynamic weight adjustment strategy is adopted to correct the weights of each prediction model online based on real-time feedback data and error analysis results during the construction process, so that the Bayesian fusion model can be updated adaptively.

[0091] S405. Generate the final comprehensive prediction results.

[0092] Specifically, in the prediction result fusion stage, the Bayesian fusion model combines short-term, medium-term, and long-term prediction results to calculate the credibility weight of each model. The Bayesian inference principle plays an important role here, calculating the posterior probability of each prediction result based on current construction data and historical data, and assigning weights according to the credibility of different prediction models. Furthermore, the Bayesian model dynamically adjusts the weights of each model by calculating the likelihood and posterior probability of each model's prediction result, enabling the Bayesian fusion model to automatically adapt to changes in different construction stages and prediction periods. As real-time data flows in during the construction process, the Bayesian model will correct the prediction results based on actual errors, ensuring the continuous accuracy of the model.

[0093] In this process, the dynamic weight adjustment strategy plays a key role. When new construction data emerges during construction, the Bayesian model adjusts the weights of each prediction model based on this real-time data. This ensures that the contribution of each model matches the current construction environment and progress in different prediction stages, including the short, medium and long term. The adaptive adjustment strategy effectively addresses the uncertainties and complexities in the construction process and guarantees the accuracy of the final prediction results.

[0094] Furthermore, step S401 specifically includes the following sub-steps:

[0095] S4011. In the short-term prediction phase, a sliding window with a time step of 5 minutes is used to summarize the real-time collected construction data into a short-term dataset, and this dataset is input into an adaptive classification and regression tree model to output the earthwork volume prediction results for the next 5 minutes.

[0096] S4012. In the intermediate prediction stage, a sliding window with a time step of 30 minutes is used to summarize the construction data collected in the most recent 30 minutes into an intermediate dataset. This intermediate dataset is then input into an adaptive classification and regression tree model to output the earthwork volume prediction results for the next 30 minutes.

[0097] S4013. In the long-term prediction phase, historical data covering the entire construction cycle is used as a long-term dataset, and this long-term dataset is input into an adaptive classification and regression tree model to output a prediction of the earthwork volume change trend throughout the entire construction cycle.

[0098] Furthermore, in step S402, the specific process for calculating the credibility weight through the Bayesian fusion model is as follows:

[0099] ;

[0100] ;

[0101] ;

[0102] Among them, the The credibility weight is represented by the following. Indicates the current construction data The lower model The posterior probability, the Indicates the current construction data The lower model The posterior probability, the Indicates the type of result, including short-term, medium-term, and long-term forecast results. Representation Model Generate current data The likelihood probability, the Representation Model Generate current data The likelihood probability, the Represents the prediction variance, the This indicates the prediction error.

[0103] Furthermore, the Bayesian fusion model is constructed before step S401, that is, it also includes step S400: establishing an initialized Bayesian model and assigning initial weights to each prediction model. The initial weights are determined based on the model performance during cross-validation in training in S3. For example, short-term models with smaller errors on recent data are assigned higher weights. Further, short-term, medium-term, and long-term prediction models calculate earthwork volume predictions respectively, and the Bayesian model calculates the overall prediction value based on these prediction results. As the tunnel boring machine (TBM) construction progresses, new data is continuously input, and the Bayesian model continuously updates its weights based on the actual errors. Each time a new prediction is calculated, the system analyzes the error and reallocates the weights of each model, reducing the weight of models with larger errors and increasing the weight of models with smaller errors. This dynamic adjustment mechanism ensures that the prediction model can adapt to changes in the construction environment and improve prediction accuracy.

[0104] Example 2

[0105] Furthermore, as a preferred embodiment of the above embodiments, it is proposed to construct the classification regression tree by calculating the Gini index for each feature. For the sample set in node t, the Gini index is: ; wherein, the Representing a number system, the aforementioned Indicates that the sample belongs to the category The probability, the This represents the number of categories; the splitting at each node selects the feature with the smallest Gini index and the corresponding threshold: ; wherein, the The characteristic that represents the minimum Gini index is... This indicates that it corresponds to the feature. The threshold, the Indicates feature-based and threshold The split subset, the Representing a subset The Gini index.

[0106] Example 3

[0107] Furthermore, as an exemplary technical solution of the above embodiments, an application scenario is proposed. This scenario applies the Kriging interpolation method, assuming the geological location is a two-dimensional coordinate system (x, y), and infers the soil type at an unknown location using data points from known locations. Specifically:

[0108] Given soil type and depth data, calculate the variogram (e.g., semivariogram) between every two locations, and calculate the spatial variability based on the geological characteristics (soil type, depth) of different locations; set the earthwork volume for the location (6.5, 12.5) to be predicted, and calculate it using Kriging interpolation based on samples from known locations, i.e.: ; wherein, the The earthwork volume representing the predicted target spatial location, the The weighting coefficients are represented by the variogram calculation. Indicates a known location The data values ​​are used. The interpolated geological prediction results are fused with real-time collected construction and environmental parameters. A classification and regression tree (CART) model is used to explore the nonlinear relationships affecting earthwork volume changes, and an online update mechanism is used to achieve adaptive optimization of the model. Based on the features in the data (such as humidity, thrust, etc.), the classification and regression tree selects the optimal features and split points based on minimizing the Gini index. The shield machine thrust is selected as the split feature, and the split effect is calculated using the Gini index. Each time new data arrives (e.g., the soil humidity of (10,18,21) is 40%), incremental learning adapts to the new data by updating the model weights. Three models are used: model M1 based on soil humidity, model M2 based on shield machine thrust, and model M3 based on depth. The posterior probability of each model is calculated and weighted and fused according to Bayes' theorem. Each time new data arrives, the Bayesian posterior probability is updated based on the real-time data. Assuming that new data points include shield machine thrust data and its posterior probability is updated, the posterior probabilities of other models are adjusted accordingly. The final prediction results are output.

[0109] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for predicting earthwork volume in shield tunneling construction based on Kriging interpolation and classification regression trees, characterized in that, Includes the following steps: S1. Real-time acquisition of construction parameters, geological parameters, and environmental parameters generated during the tunnel boring machine construction process, and outlier detection, missing value imputation, and smoothing of the acquired data; S2. Establish a spatial data model of the shield tunneling area through geological parameters, and use a combination of local and global interpolation, dynamic adjustment of the variogram function and multi-scale interpolation fusion strategy to spatially predict the strata data. During the shield tunneling process, use the Bayesian update method to dynamically correct the interpolation results. S3. Based on construction parameters, geological parameters, and environmental parameters, feature selection and interactive analysis are performed to construct an initial classification and regression tree model. The model is then updated and optimized online using a dynamic pruning algorithm and an incremental learning strategy. S4. Based on the trained adaptive classification and regression tree model, predict the changes in earthwork volume in the short, medium and long term respectively, and input the short, medium and long term prediction results into the Bayesian fusion model to calculate the confidence weight of different models. Through a dynamic weight adjustment strategy, adjust the weight of each prediction model according to the actual error in the construction process to generate the final earthwork volume prediction result. Step S2 specifically includes the following steps: S201. Establish an initial spatial data grid based on the preprocessed geological parameters, and calculate the variogram function of the initial spatial data grid using the traditional Kriging interpolation method; S202. For areas with abrupt changes in strata, dynamically adjust the parameters of the local variogram and remodel the local geological features to establish a local interpolation model; S203. Based on historical data, calculate the geological trend of the entire shield tunneling area and use global kriging interpolation to fit it and establish a global interpolation model; S204. Utilize a multi-scale interpolation fusion strategy to perform weight allocation and fusion of local and global interpolation models; In step S201, the mutation function is specifically expressed as follows: ； Among them, the Represents the variogram function, the Indicating spatial distance, the Indicates distance as The number of sample point pairs, the Indicates the sample index, the Indicates position The formation parameter values ​​at the location, the Indicates position The stratigraphic parameter values ​​at that location.

2. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 1, characterized in that, In step S1, the construction parameters include at least the tunnel boring machine's advance speed, cutterhead rotation speed, soil chamber pressure, and screw conveyor rotation speed; the geological parameters include at least the stratum distribution, water content, sand content, and porosity within the tunnel construction area; and the environmental parameters include at least the groundwater level and surface settlement within the tunnel construction area.

3. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 1, characterized in that, Step S3 specifically includes the following steps: S301. Construct a candidate feature set based on the preprocessed construction parameters, geological parameters, and environmental parameters; S302. Perform information entropy calculation and feature interaction analysis on the constructed candidate feature set, quantitatively evaluate the impact of each feature on the change of earthwork volume, screen key influencing factors that have significant contributions to the prediction results, and remove redundant or low-contribution features. S303. Construct an initial classification regression tree model based on the selected key influencing factors, and use the Gini coefficient or information gain criterion to determine the optimal split point at each split node to capture the nonlinear relationship between construction parameters and earthwork volume. S304. The initial classification and regression tree is optimized by an adaptive pruning algorithm, and redundant branches are automatically removed by evaluating the error contribution of each branch; S305. Through an incremental learning strategy, the classification regression tree model is updated online based on new data during the tunnel boring machine construction process.

4. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 3, characterized in that, Step S302 specifically includes the following sub-steps: S3021. Calculate the mutual information values ​​among construction parameters, geological parameters, and environmental parameters, measure the degree of correlation between each variable and earthwork volume, and use the information entropy analysis method to evaluate the contribution of variables to the final prediction accuracy; S3022. Using the maximum correlation minimum redundancy algorithm, variables that are highly correlated with other features but have low information gain are eliminated; S3023. After feature selection is completed, different construction scenarios are classified, and key influencing factors under each scenario are categorized.

5. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 4, characterized in that, Step S303 specifically includes the following sub-steps: S3031. Combining the CART algorithm, using the selected key influencing factors as input variables and the earthwork volume per unit time as the target variable, an initial tree structure is constructed; S3032. Calculate the Gini coefficient or information gain of all candidate features at each split node, and select the optimal feature as the basis for splitting; S3033. Determine the optimal split point based on data characteristics, and capture the nonlinear relationship between construction parameters and earthwork volume for each split by using the minimum variance decreasing criterion information gain ratio.

6. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 1, characterized in that, Step S4 specifically includes the following sub-steps: S401. Generate short-term, medium-term, and long-term prediction results based on the trained adaptive classification and regression tree model; S402. Input the short-term, medium-term and long-term prediction results into the constructed Bayesian fusion model, and calculate the confidence weights through the Bayesian fusion model; S404. A dynamic weight adjustment strategy is adopted to correct the weights of each prediction model online based on real-time feedback data and error analysis results during the construction process, so that the Bayesian fusion model can be updated adaptively. S405. Generate the final comprehensive prediction results.

7. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 6, characterized in that, Step S401 specifically includes the following sub-steps: S4011. In the short-term prediction phase, a sliding window with a time step of 5 minutes is used to summarize the real-time collected construction data into a short-term dataset, and this dataset is input into an adaptive classification and regression tree model to output the earthwork volume prediction results for the next 5 minutes. S4012. In the intermediate prediction stage, a sliding window with a time step of 30 minutes is used to summarize the construction data collected in the most recent 30 minutes into an intermediate dataset. This intermediate dataset is then input into an adaptive classification and regression tree model to output the earthwork volume prediction results for the next 30 minutes. S4013. In the long-term prediction phase, historical data covering the entire construction cycle is used as a long-term dataset, and this long-term dataset is input into an adaptive classification and regression tree model to output a prediction of the earthwork volume change trend throughout the entire construction cycle.

8. The method for predicting earthwork volume in shield tunneling based on Kriging interpolation and classification regression trees as described in claim 6, characterized in that, In step S402, the specific process for calculating the credibility weights using the Bayesian fusion model is as follows: ； ； ； Among them, the The credibility weight is represented by the following. Indicates the current construction data The lower model The posterior probability, the Indicates the current construction data The lower model The posterior probability, the Indicates the type of result, including short-term, medium-term, and long-term forecast results. Representation Model Generate current data The likelihood probability, the Representation Model Generate current data The likelihood probability, the Represents the prediction variance, the This indicates the prediction error.

Images