Earthwork excavation volume prediction method for shield tunneling

Abstract

The application provides a shield tunnel earthwork excavation amount prediction method, comprising: obtaining shield tunnel earthwork excavation amount historical data; obtaining shield machine historical operation data, and preprocessing the operation data to generate a shield machine operation data set; determining key shield machine operation parameters affecting the earthwork excavation amount by correlation analysis and importance analysis; constructing a Bayesian optimized XGBoost model, inputting real-time obtained key shield machine operation parameters into the Bayesian optimized XGBoost model, and calculating to obtain the earthwork excavation amount corresponding to the current tunneling section. Before construction, the key shield machine operation parameters can be set according to the calculated earthwork excavation amount, thereby providing a reference for the reasonable setting of the shield machine operation parameters by the driver during construction. The shield tunnel earthwork excavation amount prediction method has good prediction accuracy, and provides important support for the rapid prediction and real-time evaluation of the earthwork excavation amount in the shield tunnel construction process, and the setting and recommendation of the shield machine operation parameters.

Description

Technical Field

[0001] This invention relates to a method for predicting the earthwork excavation volume of a shield tunnel. Specifically, it relates to a method for predicting the earthwork excavation volume of a shield tunnel based on key operating parameters of the shield machine that affect the earthwork excavation volume and a Bayesian-optimized XGBoost model. This invention belongs to the field of shield tunneling technology. Background Technology

[0002] Due to its advantages such as environmental friendliness, safety, and efficiency, the application scale and number of tunnel boring machines (TBMs) in the construction of large-scale water conservancy tunnels, traffic tunnels, and urban subway tunnels have shown a continuous upward trend in recent years. Currently, in the construction process of various tunnels, the construction management model is based on the construction cycle, designing the daily and monthly tunneling distances of the TBMs. However, the calculation of the earthwork excavation volume is usually done after the tunneling process is completed, by summarizing manual paper records, and then judging whether the earthwork excavation volume is reasonable. The biggest drawback of this management model is that it is a typical "hindsight bias." Once the calculation reveals that the earthwork excavation is over-excavated, only further support measures can be taken to prevent damage to nearby buildings or ground subsidence; if under-excavation is found, it is necessary to reduce the soil chamber pressure in time to prevent instability of the strata at the excavation face or surface heave.

[0003] Furthermore, with increasingly in-depth research into factors affecting tunnel construction quality, it has been found that the amount of earthwork excavation in shield tunnels has a significant impact on tunnel quality, surrounding rock strata, surrounding topography, and surface settlement. However, currently, there are very few methods for calculating / predicting the amount of earthwork excavation for shield tunnels. The pre-excavation volume is generally not calculated before shield tunnel construction; it is only estimated after excavation is completed, based on manually tallied quantities of excavated soil. This calculation of earthwork excavation volume is significantly delayed, severely impacting tunnel construction quality. Summary of the Invention

[0004] In view of the above reasons, the purpose of this invention is to provide a method for predicting the earthwork excavation volume of a shield tunnel based on key operating parameters of the shield machine that affect the earthwork excavation volume of the shield tunnel and a Bayesian optimized XGBoost model.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting the earthwork excavation volume of a shield tunnel, comprising the following steps:

[0006] S1: Obtain historical data on earthwork excavation volume of shield tunnel;

[0007] S2: Obtain historical operating data of the tunnel boring machine, preprocess the historical operating data to remove spurious data, and obtain preprocessed historical operating data of the tunnel boring machine;

[0008] S3: Merge the historical data of earthwork excavation in the shield tunnel with the preprocessed historical operation data of the shield machine to form a dataset, perform correlation analysis and importance analysis, and determine the key operating parameters of the shield machine that affect the earthwork excavation in the shield tunnel.

[0009] S4: Based on the key operating parameters of the shield machine that affect the earthwork excavation of the shield tunnel obtained in step S3, construct a Bayesian optimized XGBoost model;

[0010] S4.1: Construct an XGBoost model to predict the earthwork excavation volume of shield tunnels;

[0011] The historical data of earthwork excavation volume of the shield tunnel obtained in step S3 and the preprocessed historical operation data of the shield machine are merged to form a dataset. This dataset is then standardized. 80% of the standardized dataset is used as the training set, and the remaining 20% ​​is used as the test set. The training dataset Data={(x1,y1),(x2,y2),…,(x m ,y m Construct an XGBoost model to predict the earthwork excavation volume of a shield tunnel;

[0012] Step S4.2: Construct the black-box function;

[0013] The parameters contained in the black-box function are the structural parameters of the XGBoost model, including the learning rate, penalty coefficient, number of decision trees, and maximum depth of decision trees;

[0014] Step S4.3: Determine the initial domain space of each structural parameter in the black-box function;

[0015] Step S4.4: Construct the Bayesian optimization objective function for the XGBoost model and determine the number of random search steps and the number of Bayesian optimization iterations;

[0016] The Bayesian optimization objective function is the prediction performance of the XGBoost model on the test set under the current combination of structural parameters.

[0017] Step S4.5: Run the Bayesian optimization model. When the Bayesian optimization objective function is minimized, the optimal initial threshold space of the XGBoost model is obtained.

[0018] Step S4.6: Based on the optimal initial threshold space obtained in step S4.5, further narrow the threshold space and perform secondary optimization on the XGBoost model. Iterate and update continuously until the accuracy improvement is less than 0.01, determine the optimal parameter combination setting of the XGBoost model, and obtain the Bayesian optimized XGBoost model.

[0019] Step S5: Input the key operating parameters from the real-time operating parameters of the tunnel boring machine into the Bayesian-optimized XGBoost model to calculate the earthwork excavation volume of the tunnel corresponding to the current tunneling section.

[0020] Preferably, the method for determining the key operating parameters of the tunnel boring machine in step S3 is as follows:

[0021] S3.1: Merge the historical data of earthwork excavation volume of the shield tunnel and the historical operation data of the shield machine obtained in steps S1 and S2 into a complete dataset using a Python program;

[0022] S3.2: Solving for the historical operating parameters x of the tunnel boring machine using the Pearson correlation coefficient method. i earthwork excavation volume y i Correlation coefficient P xy Remove duplicate and irrelevant variables;

[0023] Correlation coefficient P xy The calculation formula is as follows:

[0024]

[0025] S3.3: Using the random forest importance analysis method, with the Gini coefficient as the best splitting feature, the importance of the shield machine operation parameters affecting the earthwork excavation volume of the shield tunnel is ranked, and the top 10 important parameters are selected as the key operation parameters of the shield machine.

[0026] Preferably, the key operating parameters of the tunnel boring machine are: E-group propulsion pressure, foam pressure, cutterhead rotation speed, bentonite pressure, F-group propulsion pressure, A-group propulsion pressure, C-group propulsion pressure, center flushing pressure, propulsion speed, and D-group propulsion pressure.

[0027] Preferably, the initial threshold space of the black-box function structure parameters in step S4.3 is as follows: the learning rate is set to 0.001 to 0.1, the penalty coefficient is set to 0 to 10, the number of decision trees is set to 2 to 100, and the maximum depth of the decision trees is set to 2 to 10.

[0028] In step S4.4, when constructing the Bayesian optimization objective function of the XGBoost model, the number of random search steps is 10, and the number of Bayesian optimization iterations is 100.

[0029] Preferably, the method for preprocessing the historical operating data of the tunnel boring machine in step S2 is as follows:

[0030] First, acquire the tunnel boring machine's operating data during the tunneling process and integrate it into a .csv file;

[0031] Then, outliers in the .csv file are filtered out using the 3-Sigma method, and empty characters are filtered out using the Dropna function in Python.

[0032] Furthermore, based on the operating characteristics of the tunnel boring machine, the cutterhead rotation speed is used as the dividing condition, and the minimum and maximum tunneling lengths are used as constraints to filter out the shutdown data in the tunneling data and retain the valid tunneling data;

[0033] Finally, the Savitzky-Golay filtering method was used to filter and reduce noise in the effective tunneling data.

[0034] Preferably, during construction, newly generated shield machine operation data and measured earthwork excavation data can be re-entered into the database at any time to update the database and continuously iterate. The earthwork excavation volume of the shield tunnel in the next tunneling stage is calculated through the Bayesian optimized XGBoost model.

[0035] Compared with the prior art, the present invention has the following beneficial effects: it can predict / calculate the earthwork excavation volume of the shield tunnel in advance, providing a reliable reference for ensuring the quality of tunnel construction; and it provides a reference for the operator to reasonably set the operating parameters of the shield machine during construction. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method for predicting the earthwork excavation volume of a shield tunnel according to the present invention.

[0037] Figure 2 This is a statistical table of slag removal volume for the shield tunnel section between Zhengxingwan Station and Honglian Station of Chengdu Metro Line 19 Phase 2 Tunnel Project.

[0038] Figure 3 This is a schematic diagram illustrating the process of extracting earthwork excavation data of shield tunnels using the CNN convolution algorithm in this invention.

[0039] Figure 4A This is a .csv data file generated by preprocessing tunnel boring machine operation data according to the present invention;

[0040] Figure 4B This is a schematic diagram illustrating the process of deleting outliers and null characters from tunnel boring machine operating data according to the present invention.

[0041] Figure 4C This is a schematic diagram illustrating the extraction and processing of valid data from tunnel boring machine operation data according to the present invention;

[0042] Figure 4D This is a schematic diagram illustrating the Savitzky-Golay filtering and noise reduction process applied to tunnel boring machine (TBM) operation data according to the present invention.

[0043] Figure 5This is a correlation diagram between the shield machine operation data and the shield tunnel earthwork excavation data of this invention;

[0044] Figure 6 The ranking of key operating parameters of shield tunneling machines that affect the earthwork excavation volume of shield tunnels, as determined by this invention;

[0045] Figure 7 Flowchart of the Bayesian optimization XGBoost model for predicting earthwork excavation in shield tunnels for this invention;

[0046] Figure 8 This is a diagram illustrating the Bayesian threshold space search process of this invention.

[0047] Figure 9A This is a schematic diagram comparing the measured and predicted values ​​of earthwork excavation volume of shield tunnels in the training set of this invention.

[0048] Figure 9B This is a schematic diagram comparing the measured and predicted values ​​of earthwork excavation volume in a centralized shield tunnel according to an embodiment of the present invention.

[0049] Figure 10 A recommended electronic form for setting tunnel boring machine operating parameters in embodiments of the present invention. Detailed Implementation

[0050] The structure and features of the present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that various modifications can be made to the embodiments disclosed herein; therefore, the embodiments disclosed in this specification should not be considered as limitations on the present invention, but merely as examples to make the features of the present invention readily apparent.

[0051] like Figure 1 As shown, the method disclosed in this invention for predicting the earthwork excavation volume of a shield tunnel based on key operating parameters of the shield machine affecting the earthwork excavation volume and a Bayesian optimized XGBoost model is as follows:

[0052] Step S1: Obtain historical data on earthwork excavation volume of the shield tunnel;

[0053] Step S2: Obtain the historical operation data of the tunnel boring machine, and preprocess the historical operation data to remove the pseudo data and obtain the preprocessed historical operation data of the tunnel boring machine.

[0054] Step S3: Merge the historical data of earthwork excavation volume of the shield tunnel and the historical operation data of the shield machine to form a dataset, and perform correlation analysis and importance analysis to determine the key operating parameters of the shield machine that affect the earthwork excavation volume of the shield tunnel.

[0055] Step S4: Based on the key operating parameters of the tunnel boring machine that affect the earthwork excavation volume of the tunnel boring machine obtained in Step S3, construct a Bayesian optimized XGBoost model.

[0056] Step S5: Input the key operating parameters from the real-time operating parameters of the tunnel boring machine into the Bayesian-optimized XGBoost model to calculate the earthwork excavation volume of the tunnel corresponding to the current tunneling section.

[0057] The following example, using the shield tunneling construction of the Zhengxingwan Station to Honglian Station section of Chengdu Metro Line 19 Phase 2 as an illustration, demonstrates how this invention predicts (i.e., pre-calculates) the earthwork excavation volume of the shield tunnel.

[0058] Step S1: Obtain historical data on earthwork excavation volume of shield tunnel.

[0059] The historical data can be the earthwork excavation data of completed shield tunnels in the same region, or the earthwork excavation data of shield tunnels under construction in the initial stage. In this embodiment of the invention, the applicant obtained the earthwork excavation data of completed shield tunnels during the construction of the Zhengxingwan Station to Honglian Station section of the Chengdu Metro Line 19 Phase 2 Tunnel Project.

[0060] Since most engineering projects currently use paper documents to record the earthwork excavation volume of shield tunnels, it is necessary to first scan the paper documents to generate digital images, and then perform image processing to extract the earthwork excavation volume data. The specific steps are as follows:

[0061] Step S1.1: Obtain digital image files.

[0062] Place the paper document in the center of the scanning device, ensuring the entire area is free of shadows during shooting / scanning to maximize the overall quality of the digital image and prevent artifacts and other noise. The paper document can be converted into the following format after being photographed and scanned: Figure 2 The digital image shown.

[0063] Step S1.2: Extract historical data on earthwork excavation volume of shield tunnel using CNN convolutional recognition algorithm.

[0064] The digital image obtained in step S1.1 is processed line by line using CNN convolutional recognition and extraction. The designed CNN convolutional layer includes one input layer, one convolutional layer, one pooling layer, one fully connected layer, and one output layer. Figure 3As shown, in the convolutional layer, the convolutional kernel slides across the digital image output from the previous layer, summing the products of each element within the corresponding region of the kernel. This summation is then performed after bias correction and activation by the activation function. By adjusting the kernel size, local feature information of the earthwork excavation data in the digital image is obtained. Next, in the pooling layer, max pooling is used to reduce the resolution of the feature surface, thus obtaining the invariant properties of the earthwork excavation data within the region. In the fully connected layer, a convolutional kernel of the same size as the feature map output from the pooling layer is used to convolve with all outputs from the previous layer, thereby summarizing the entire digital image of the earthwork excavation volume. Furthermore, before the digital image is input into the CNN model, image enhancement methods such as random cropping, flipping, and color dithering are used to improve the digital image recognition capability.

[0065] Step S1.3: Merge the identified earthwork excavation data and save it as a .csv file.

[0066] Step S2: Obtain historical operating data of the tunnel boring machine and preprocess the historical operating data to remove spurious data and obtain high-quality historical operating data of the tunnel boring machine.

[0067] In tunnel boring machine (TBM) operation data, most of the tunneling data is valid construction data. However, due to various reasons such as electromagnetic interference, sensor malfunction, human damage, and extreme geological conditions, the operation data may contain various "dirty data," including "too short," "fluctuating," "discrepancy," "steps," "spurs," and "missing" data. Simultaneously, there may be tunneling segment data with multiple anomaly types coexisting. If this data is not cleaned or repaired in a timely manner, it will affect the establishment of subsequent data models, operational efficiency, and prediction results. Therefore, this invention preprocesses the obtained historical TBM operation data, such as... Figures 4A-4D As shown.

[0068] First, the tunnel boring machine (TBM) operational data during the tunneling process is acquired. Since the TBM operates 24 / 7, generating 86,400 data points per day, this invention merges all the TBM's operational data into a single, complete .csv file, organized by day. (See attached image.) Figure 4A ;

[0069] Then, outliers in the .csv file are filtered out using the 3-Sigma method, and obvious anomalous data such as empty characters are filtered out using the Dropna function in Python. Failure to clean these data will interfere with the subsequent modeling process. (See...) Figure 4B ;

[0070] Furthermore, based on the operating characteristics of the tunnel boring machine (TBM), using cutterhead rotation speed as the segmentation condition and minimum and maximum tunneling length as constraints, the shutdown data in the tunneling data was filtered out, retaining the valid tunneling data. The extraction results are shown below. Figure 4C ;

[0071] Next, the Savitzky-Golay filtering method was used to filter and reduce noise in the effective tunneling data. While filtering and reducing noise, the shape and width of the signal were kept unchanged. The purpose was to counteract the interference of strong electromagnetic fields in the tunnel boring machine's operating environment on the data acquisition and storage process. The noise reduction results are shown in [Figure number missing]. Figure 4D ;

[0072] Finally, the pre-processed tunnel boring machine operation data is saved as a .csv file.

[0073] Through the above preprocessing steps, operational data containing spurious data can be removed, resulting in high-quality historical operational data of the tunnel boring machine. The above preprocessing steps have a strict sequential order: ... Figure 4A If the process is delayed, there is a risk of losing valid data; Figure 4B If the process is delayed, it will have an impact. Figure 4C The accuracy of effective tunneling data extraction; Figure 4D If done in advance, it is easier to filter and reduce noise from pseudo-data into valid data.

[0074] Step S3: Merge the historical data of earthwork excavation volume of the shield tunnel and the historical operation data of the shield machine to form a dataset, and perform correlation analysis and importance analysis to determine the key operating parameters of the shield machine that affect the earthwork excavation volume of the shield tunnel.

[0075] Tunnel boring machines (TBMs) are large and complex pieces of equipment, with a total of 530 operating parameters. Including all these parameters in the model would reduce the model's generalization ability and increase the training time. Therefore, this invention uses the Pearson correlation coefficient method and random forest importance analysis to conduct correlation and importance analysis on the TBM operating parameters that affect the earthwork excavation volume of the tunnel, and to determine the key operating parameters of the TBM that affect the earthwork excavation volume of the tunnel.

[0076] While the Pearson correlation coefficient method can be used to screen model input features, it can only discover linear relationships between independent and dependent variables, and is poor at describing nonlinear relationships. Random forest importance analysis, on the other hand, by integrating multiple decision trees, can be used to discover relationships between higher-dimensional data features and screen out important features. Therefore, this invention integrates these two methods to determine the key operating parameters of the tunnel boring machine (TBM) that affect the earthwork excavation volume of a shield tunnel. By integrating the two methods—first correlation analysis and then importance analysis—the relationship between earthwork excavation volume and TBM operating parameters can be quickly understood, thereby rapidly removing irrelevant or redundantly correlated variables, reducing computation time for subsequent importance analysis, and improving the accuracy of the importance analysis.

[0077] First, the historical data of earthwork excavation volume of the shield tunnel and the historical operation data of the shield machine obtained in steps S1 and S2 are merged into a complete dataset using a Python program.

[0078] Then, the Pearson correlation coefficient method was used to solve for the historical operating parameters x of the tunnel boring machine. i earthwork excavation volume y i Correlation coefficient P xy This removes duplicate and irrelevant variables.

[0079] Correlation coefficient P xy The calculation formula is as follows:

[0080]

[0081] from Figure 5 The shield tunneling machine operation data x shown i earthwork excavation volume y i The correlation analysis revealed that most data characteristics of the tunnel boring machine showed poor linear correlation with the earthwork excavation volume, and there were a large number of linearly irrelevant variables.

[0082] Finally, based on the previous step, the importance analysis method of random forest was used, with the Gini coefficient as the best dividing feature, to obtain the ranking of the importance of the shield machine operating parameters that affect the earthwork excavation volume of the shield tunnel. The top 10 parameters in terms of importance were selected as the key operating parameters of the shield machine.

[0083] like Figure 6 As shown, the key operating parameters of the tunnel boring machine that ultimately affect the earthwork excavation volume of the shield tunnel are: E group propulsion pressure, foam pressure, cutterhead speed, bentonite pressure, F group propulsion pressure, A group propulsion pressure, C group propulsion pressure, center flushing pressure, propulsion speed, and D group propulsion pressure.

[0084] Step S4: Based on the key operating parameters of the tunnel boring machine that affect the earthwork excavation volume of the shield tunnel obtained in Step S3, construct a Bayesian optimized XGBoost model.

[0085] Compared to grid search and random search methods, Bayesian optimization offers better performance and reduces computation time, such as... Figure 7 As shown, the method for constructing a Bayesian-optimized XGBoost model is as follows:

[0086] Step S4.1: Construct an XGBoost model to predict the earthwork excavation volume of a shield tunnel.

[0087] XGBoost, proposed by Chen and Guestrin in 2016, is a novel gradient-enhanced ensemble learning method that has been widely applied in knowledge mining, data science competitions, and industry. The core idea of ​​this method is to construct a new decision tree, fit the residual between the final predicted value and the actual value through continuous feature segmentation, and then combine the results of all predictors into the final prediction.

[0088] First, the historical data of earthwork excavation volume of the shield tunnel and the historical operation data of the shield machine obtained in step S3 are merged and standardized. Then, 80% of the standardized dataset is used as the training set and the remaining 20% ​​is used as the test set. The training dataset Data={(x1,y1),(x2,y2),…,(x m ,y m Construct an XGBoost model to predict the earthwork excavation volume of a shield tunnel. The specific method is as follows:

[0089] Given a loss function Regularization term Ω(f) k If the earthwork excavation volume is predicted, then the objective function for predicting the earthwork excavation volume is:

[0090]

[0091] In the formula, W(φ) is the representation in linear space; i is the i-th sample; k is the k-th decision tree; y i This is the measured value of the earthwork excavation volume;

[0092] This is the predicted earthwork excavation volume. A gradient boosting tree approach is used to... Represented as Then W(φ) can be converted to:

[0093]

[0094] Next, the objective function is simplified by second-order Taylor expansion, regularization expansion, and combining coefficients of terms of the same degree.

[0095] (i) Second-order Taylor expansion to optimize the loss function term.

[0096] Expanding the loss function terms using a second-order Taylor expansion and removing the constant term, we denote the first derivative as... The second derivative is make Since the structure of the t-1 trees is already determined, the constant term is removed. The objective function then becomes:

[0097]

[0098] (ii) Expand and optimize the regularization terms. The regularization terms are split to obtain... Since the structure of the t-1 trees is already determined, the constant term is removed. The objective function is then further optimized:

[0099]

[0100] (iii) Combine the coefficients of terms of the same degree to obtain the final objective function. Before merging, redefine the decision tree. Let the weight vector of the leaf node be w, and the mapping relationship between the leaf nodes be q. Then a tree can be represented as f t (x)=w q(x) ,w∈R T ,q:R d →{1,2,...,T}, define the complexity Ω of a tree, which is affected by the number of leaf nodes and the l2 norm of the leaf node weight vector, and can be expressed as: In the formula, γ is the penalty coefficient, controlling the number of leaf nodes; λ is the norm coefficient, ensuring that the scores of the leaf nodes are not too large. Then, by grouping all training samples according to leaf nodes, the simplified objective function can be obtained:

[0101]

[0102] In the formula, It is the sum of the first-order partial derivatives contained in the leaf node j; Let be the sum of the second-order partial derivatives contained in leaf node j; both are constants. When each leaf node reaches its maximum / minimum point, the objective function also achieves its optimal value, resulting in the completed XGBoost model for predicting the earthwork excavation volume of a shield tunnel.

[0103] Step S4.2: Construct a black-box function.

[0104] When predicting the earthwork excavation volume of a shield tunnel, the XGBoost model needs to determine the corresponding structural parameters, namely the learning rate, penalty coefficient, number of decision trees, and maximum depth of decision trees. These structural parameters together form the black-box function of Bayesian optimization.

[0105] Step S4.3: Determine the initial domain space of each structural parameter in the black-box function.

[0106] After constructing the black-box function, it is necessary to determine the value range of the structural parameters in the black-box function, i.e., the initial threshold space. In a preferred embodiment of the present invention, the learning rate is set to 0.001 to 0.1, the penalty coefficient is set to 0 to 10, the number of decision trees is set to 2 to 100, and the maximum depth of the decision trees is set to 2 to 10.

[0107] Step S4.4: Construct the Bayesian optimization objective function for the XGBoost model and determine the number of random search steps and the number of Bayesian optimization iterations.

[0108] The Bayesian optimization objective function is the prediction performance of the XGBoost model on the test set under the current combination of structural parameters. In a preferred embodiment of the present invention, when constructing the Bayesian optimization objective function of the XGBoost model, the number of random search steps is 10, and the number of Bayesian optimization iterations is 100.

[0109] Step S4.5: Run the Bayesian optimization model. When the Bayesian optimization objective function reaches its minimum value, the optimal initial threshold space of the XGBoost model is obtained.

[0110] Step S4.6: Based on the optimal initial threshold space obtained in step S4.5, iteratively update and narrow the threshold space to determine the optimal parameter combination settings for the XGBoost model, thus obtaining the Bayesian optimized XGBoost model.

[0111] The termination condition for iterative updates in the Bayesian optimization process is:

[0112]

[0113] Where 'a' is a very small constant representing the accuracy threshold, which is set to 0.01 in this step, and 'k' is the number of iterations, which is set to 10 in this step. The optimal parameter combination settings for the XGBoost model, determined through Bayesian optimization, are shown in Table 1.

[0114] Table 1 Optimal parameter combinations for the XGBoost model

[0115]

[0116] In the initial threshold space setting stage, a relatively large threshold search range is usually set. At the same time, to ensure search efficiency, the number of search steps is generally taken to be small. This means that the initially selected threshold space may not be the optimal combination of structure parameters. For example... Figure 8 As shown, this invention continuously iterates and updates the domain space of structural parameters until the improvement in the prediction accuracy of the model is minimal or nonexistent.

[0117] Step S5: Input the key operating parameters from the real-time operating parameters of the tunnel boring machine into the Bayesian-optimized XGBoost model to calculate the earthwork excavation volume of the tunnel corresponding to the current tunneling section.

[0118] The key operating parameters of the tunnel boring machine (TBM) determined in step S3, such as E-group propulsion pressure, foam pressure, cutterhead speed, bentonite pressure, F-group propulsion pressure, A-group propulsion pressure, C-group propulsion pressure, center scour pressure, propulsion speed, and D-group propulsion pressure, are input as independent variables into the Bayesian-optimized XGBoost model constructed in step S4 to calculate the earthwork excavation volume of the tunnel.

[0119] It should be noted that during construction, newly generated shield machine operation data and measured earthwork excavation data can be re-entered into the database at any time to update the database and continuously iterate. The earthwork excavation volume of the shield tunnel in the next tunneling stage is calculated through the Bayesian-optimized XGBoost model.

[0120] Of course, before each day's construction, the operating parameters of the tunnel boring machine (TBM), especially the key operating parameters that affect the earthwork excavation volume of the tunnel, can be input into the Bayesian-optimized XGBoost model to predict the current earthwork excavation volume and provide a reference for the TBM operator.

[0121] To verify the accuracy of the shield tunnel excavation volume prediction method disclosed in this invention, this invention compares the prediction results obtained by the prediction model with the real-time excavation volume obtained at the construction site, using the root mean square error (RMSE) and mean absolute error (MAE) as evaluation functions for the comparison results:

[0122]

[0123]

[0124] Figure 9A To compare the measured and predicted values ​​of earthwork excavation volume for training purposes, the RMSE and MAE were 1.227 and 0.990, respectively. Figure 9B To compare the measured and predicted values ​​of the earthwork excavation volume, the RMSE and MAE were 1.986 and 1.595, respectively. Figure 9A , Figure 9B As can be seen, the present invention has high prediction accuracy and can be fully used to predict the earthwork excavation volume of shield tunnels.

[0125] The significance of this invention lies in two aspects: firstly, it allows for the scientific calculation of the earthwork excavation volume of the shield tunnel before construction, providing a reliable reference for the designer and ensuring project quality; secondly, it provides a reference for the construction party to scientifically set the operating parameters of the shield machine during construction.

[0126] Table 2 shows the predicted earthwork excavation volume for the shield tunnel section between Zhengxingwan Station and Honglian Station of Chengdu Metro Line 19 Phase 2 Project, based on the earthwork excavation volume prediction method disclosed in this invention.

[0127] Before construction, 10 different sets of shield machine operating parameter values ​​were preset to obtain 10 sets of earthwork excavation volume data under the preset parameters. Then, the mean absolute error (MAE) obtained from the training set was added or subtracted to obtain the range of earthwork excavation volume data under the preset parameters, as shown in Table 2.

[0128] Table 2. Earthwork excavation volume data range under preset parameters.

[0129]

[0130]

[0131] After obtaining the earthwork excavation volume data range under preset parameters, it is possible to generate data such as... Figure 10 The electronic form shown provides a reference for operators to set tunnel boring machine (TBM) operating parameters during the construction phase. The electronic form mainly includes different combinations of operating parameters, 10 key operating parameters for the TBM, and the expected range of earthwork excavation volume under these parameter settings.

[0132] Advantages of this invention:

[0133] 1. This invention can accurately calculate the earthwork excavation volume of a shield tunnel in advance, providing a reference for the shield machine operator to set reasonable operating parameters for the shield machine.

[0134] 2. This invention integrates the Pearson correlation coefficient algorithm and the random forest importance analysis method. It sorts the features according to their average contribution and selects the key operating parameters of the shield machine that affect the earthwork excavation of the shield tunnel. It then establishes a mathematical model to predict the earthwork excavation, which reduces the training time of the model and improves the generalization ability of the model.

[0135] 3. This invention constructs a black-box function and an initial threshold space, and continuously updates and iterates to reduce the threshold space to obtain the optimal combination of XGBoost model structural parameters, making the earthwork excavation volume of the shield tunnel predicted by the Bayesian-optimized XGBoost model more accurate.

[0136] Although specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. It will be understood by those skilled in the art that various changes, modifications, substitutions, and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the earthwork excavation volume of a shield tunnel, characterized in that: It includes the following steps: S1: Obtain historical data on earthwork excavation volume of shield tunnel; S2: Obtain historical operating data of the tunnel boring machine, preprocess the historical operating data to remove spurious data, and obtain preprocessed historical operating data of the tunnel boring machine; S3: Merge the historical data of earthwork excavation in the shield tunnel with the preprocessed historical operation data of the shield machine to form a dataset, perform correlation analysis and importance analysis, and determine the key operating parameters of the shield machine that affect the earthwork excavation in the shield tunnel. S4: Based on the key operating parameters of the shield machine that affect the earthwork excavation of the shield tunnel obtained in step S3, construct a Bayesian optimized XGBoost model; S4.1: Construct an XGBoost model to predict the earthwork excavation volume of shield tunnels; The historical data of earthwork excavation volume of the shield tunnel obtained in step S3 and the preprocessed historical operation data of the shield machine are merged to form a dataset. This dataset is then standardized. 80% of the standardized dataset is used as the training set, and the remaining 20% ​​is used as the test set. The training dataset Data = {(x1, y1), (x2, y2), ..., (x...} m , y m Construct an XGBoost model to predict the earthwork excavation volume of a shield tunnel, where x represents historical operating parameters and y represents the earthwork excavation volume; Step S4.2: Construct the black-box function; The parameters contained in the black-box function are the structural parameters of the XGBoost model, including the learning rate, penalty coefficient, number of decision trees, and maximum depth of decision trees; Step S4.3: Determine the initial domain space of each structural parameter in the black-box function; Step S4.4: Construct the Bayesian optimization objective function for the XGBoost model and determine the number of random search steps and the number of Bayesian optimization iterations; The Bayesian optimization objective function is the prediction performance of the XGBoost model on the test set under the current combination of structural parameters. Step S4.5: Run the Bayesian optimization model. When the Bayesian optimization objective function is minimized, the optimal initial threshold space of the XGBoost model is obtained. Step S4.6: Based on the optimal initial threshold space obtained in step S4.5, further narrow the threshold space and perform secondary optimization on the XGBoost model. Iterate and update continuously until the accuracy improvement is less than 0.01, determine the optimal parameter combination setting of the XGBoost model, and obtain the Bayesian optimized XGBoost model. Step S5: Input the key operating parameters from the real-time operating parameters of the tunnel boring machine into the Bayesian-optimized XGBoost model to calculate the earthwork excavation volume of the tunnel corresponding to the current tunneling section.

2. The method for predicting the earthwork excavation volume of a shield tunnel according to claim 1, characterized in that: The method for determining the key operating parameters of the tunnel boring machine in step S3 is as follows: S3.1: Merge the historical data of earthwork excavation volume of the shield tunnel and the historical operation data of the shield machine obtained in steps S1 and S2 into a complete dataset using a Python program; S3.2: Solving for the historical operating parameters x of the tunnel boring machine using the Pearson correlation coefficient method. i earthwork excavation volume y i Correlation coefficient P xy Remove duplicate and irrelevant variables; Correlation coefficient P xy The calculation formula is as follows: S3.3: Using the random forest importance analysis method, with the Gini coefficient as the best splitting feature, the importance of the shield machine operation parameters affecting the earthwork excavation volume of the shield tunnel is ranked, and the top 10 important parameters are selected as the key operation parameters of the shield machine.

3. The method for predicting the earthwork excavation volume of a shield tunnel according to claim 2, characterized in that: The key operating parameters of the tunnel boring machine are: E-group propulsion pressure, foam pressure, cutterhead speed, bentonite pressure, F-group propulsion pressure, A-group propulsion pressure, C-group propulsion pressure, center flushing pressure, propulsion speed, and D-group propulsion pressure.

4. The method for predicting the earthwork excavation volume of a shield tunnel according to claim 3, characterized in that: The initial threshold space of the black-box function structure parameters in step S4.3 is as follows: the learning rate is set to 0.001~0.1, the penalty coefficient is set to 0~10, the number of decision trees is set to 2~100, and the maximum depth of the decision trees is set to 2~10. In step S4.4, when constructing the Bayesian optimization objective function of the XGBoost model, the number of random search steps is 10, and the number of Bayesian optimization iterations is 100.

5. The method for predicting the earthwork excavation volume of a shield tunnel according to claim 4, characterized in that: The method for preprocessing the historical operating data of the tunnel boring machine in step S2 is as follows: First, acquire the tunnel boring machine's operating data during the tunneling process and integrate it into a .csv file; Then, outliers in the .csv file are filtered out using the 3-Sigma method, and empty characters are filtered out using the Dropna function in Python. Furthermore, based on the operating characteristics of the tunnel boring machine, the cutterhead rotation speed is used as the dividing condition, and the minimum and maximum tunneling lengths are used as constraints to filter out the shutdown data in the tunneling data and retain the valid tunneling data; Finally, the Savitzky-Golay filtering method was used to filter and reduce noise in the effective tunneling data.

6. The method for predicting the earthwork excavation volume of a shield tunnel according to any one of claims 1-5, characterized in that: During construction, newly generated shield machine operation data and measured earthwork excavation data can be re-entered into the database at any time to update the database and continuously iterate. The earthwork excavation volume of the shield tunnel in the next tunneling stage is calculated through the Bayesian optimized XGBoost model.

Images