Shield tunneling parameter prediction method under composite stratum

Abstract

The invention discloses a method for predicting shield tunneling parameters under a composite stratum, which comprises the following steps of: collecting and sorting shield tunneling historical data under the composite stratum, including geological parameters and construction parameters, and performing abnormal value elimination, missing value filling and data standardization; the geological parameters and the geometric parameters are combined according to the tunneling ring number, and a quantitative geological matrix of each ring is generated; the geometric parameters represent the position of the soil layer relative to the tunnel excavation face by adopting 1-5 codes, and input dimension unification is ensured through zero filling alignment; a CNN-Attention-SVR model is constructed, and the CNN-Attention-SVR model is trained; new geological data are input into the trained model, tunneling parameters are predicted, and construction adjustment is guided; reliable technical support and decision basis are provided for tunnel construction under the composite stratum condition, and therefore the overall safety and economical efficiency of shield construction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of shield tunneling construction, and particularly relates to a method for predicting shield tunneling parameters under a composite stratum. Background Art

[0002] With the acceleration of the urbanization process, shield tunneling construction has been widely used in tunnel construction due to its advantages such as high mechanization, high safety, and high construction efficiency. However, there is a complex non-linear relationship between tunneling parameters and geological environment. Especially under complex geological conditions, it is particularly important to select reasonable tunneling parameters. Although the selection of tunneling parameters mainly relies on manual experience at present, with the rapid development of artificial intelligence technology, intelligent decision-making and unmanned shield tunneling construction have become research hotspots.

[0003] Machine learning algorithms, such as random forest, K-nearest neighbor algorithm, and neural network, have been widely used in the research of tunneling parameter prediction due to their powerful non-linear modeling ability. The existing research mainly falls into two directions: one is to construct an intelligent prediction model through the time series and correlation of tunneling parameters themselves without considering geological parameters; the other is to construct an intelligent prediction model of tunneling parameters considering the characteristic parameters of geological conditions. However, due to the complexity of geological conditions, especially the large differences in different geological characteristic parameters in composite strata, both methods have deficiencies in quantifying complex geological conditions and feature extraction. Summary of the Invention

[0004] Object of the Invention: The object of the present invention is to provide a method for predicting shield tunneling parameters under a composite stratum. By dividing different types of soil layers, each ring forms a matrix, the geological parameters of different strata are respectively counted, and geometric parameters are introduced through unit coding, so as to establish a quantification matrix of geological conditions. The method of combining convolutional neural network with attention mechanism is used to automatically extract the features of the geological quantification matrix, and finally the regression support vector machine is used to realize the intelligent prediction of tunneling parameters.

[0005] Technical Solution: A method for predicting shield tunneling parameters under a composite stratum according to the present invention includes the following steps: (1) Data preprocessing: Collect and sort out the historical data of shield tunneling under a composite stratum, including geological parameters and construction parameters, and perform outlier removal, missing value filling, and data standardization; (2) Construct a quantified geological matrix: Combine geological parameters and geometric parameters according to the tunneling ring number to generate a quantified geological matrix for each ring; The geometric parameters are represented by encoding from 1 to 5 to indicate the position of the soil layer relative to the tunnel excavation surface, and zero padding is used to align to ensure the unity of the input dimension; (3) Construct the CNN-Attention-SVR model: The input layer receives the quantified geological matrix; the convolutional neural network CNN extracts local geological features, including two convolutional layers and a pooling layer, introducing an attention mechanism to dynamically calculate feature weights and highlighting key areas; Support Vector Regression SVR combined with the Radial Basis Function RBF kernel is used to perform non-linear regression on the weighted features and output the predicted values of the tunneling parameters; (4) Model training: Use the Adam optimizer to train the model, prevent overfitting through cross-validation, and the loss function is the mean square error MSE; (5) Parameter prediction: Input the new geological data into the trained model, predict the tunneling parameters and guide the construction adjustment.

[0006] Further, in step (1), the geological parameters include: unit weight, internal friction angle, cohesion; the construction parameters include: penetration, tunneling speed, total thrust, cutterhead rotation speed, cutterhead torque, total grouting volume, HBW grease consumption, EP2 grease consumption.

[0007] Further, in step (1), the 3σ standard deviation method is used to remove the data of the non-steady tunneling section, and the moving window smoothing polynomial is used to denoise the tunneling parameters.

[0008] Further, in step (2), the dimension of the quantified geological matrix is 18×5, each row corresponds to a soil layer unit, including 3 geological parameters and 2 geometric parameters, and the input matrix dimensions of all rings are ensured to be consistent by padding with zeros; the geometric parameters are soil layer thickness and position encoding.

[0009] Further, in step (2), the coding of 1-5 is as follows: 1 is the upper part of the horizontal plane above the excavation face, 2 is the lower part of the horizontal plane above the excavation face, 3 is the upper half of the excavation face, 4 is the lower half of the excavation face, and 5 is below the excavation face.

[0010] Further, in step (3), CNN adopts the LeNet architecture, including two convolutional layers with a convolutional kernel size of 3×3 and two max-pooling layers with a pooling window of 2×2; the attention mechanism is realized through self-attention calculation, and the formula is: ; where, is the dimension of the key; Q, K, V are the linear transformations of the input feature matrix.

[0011] Further, in step (3), the loss function of SVR is: ; where, is the weight vector, is the regularization parameter, is the slack variable.

[0012] Further, in step (4), the ratio of the training set, the validation set, and the test set is 8:1:1, and the training termination condition is that the loss value does not decrease significantly for 30 consecutive iterations.

[0013] Further, in step (5), the prediction result is restored to the actual parameter value through denormalization. The denormalization formula is: ; where is the original data; is the normalized data; and are the maximum and minimum values of the corresponding parameters, respectively.

[0014] A shield tunneling parameter prediction system under a composite stratum according to the present invention includes: A data preprocessing module: used to collect and organize the shield tunneling historical data under the composite stratum, including geological parameters and construction parameters, and perform outlier removal, missing value filling, and data standardization; A module for constructing a quantified geological matrix: used to combine geological parameters and geometric parameters according to the tunneling ring number to generate a quantified geological matrix for each ring; the geometric parameters are represented by codes from 1 to 5 to indicate the position of the soil layer relative to the tunnel excavation face, and zero padding is used to align the input dimensions to ensure unity; A CNN-Attention-SVR module: used to receive the quantified geological matrix at the input layer; extract local geological features through a convolutional neural network CNN, including two convolutional layers and a pooling layer, introduce an attention mechanism to dynamically calculate feature weights, and highlight key areas; use support vector regression SVR combined with a radial basis function RBF kernel to perform non-linear regression on the weighted features and output the predicted values of the tunneling parameters; A model training module: used to train the model using an Adam optimizer, prevent overfitting through cross-validation, and the loss function is the mean square error MSE; A parameter prediction module: used to input new geological data into the trained model, predict the tunneling parameters, and guide construction adjustment.

[0015] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: By comprehensively considering the complexity and uncertainty of shield tunneling under composite strata conditions, the present invention constructs a shield tunneling parameter prediction model based on the CNN-Attention-SVR architecture; based on geological survey data and historical construction data, by establishing a quantitative geological matrix, the model fully considers various influencing factors including geological type, formation hardness, water content, etc., combined with key parameters such as thrust force, torque, and tunneling speed recorded during the shield tunneling process, and the model can accurately predict the construction parameter requirements for shield tunneling; through a large number of model trainings and optimizations, the present invention forms a complete set of shield tunneling parameter prediction and guidance solutions, which can not only improve the construction accuracy and efficiency, but also significantly reduce the construction risk, providing reliable technical support and decision-making basis for tunnel construction under composite strata conditions, thereby improving the overall safety and economy of shield construction. Brief Description of the Drawings

[0016] Figure 1 is a flowchart of the present invention; Figure 2 is a schematic diagram of the quantitative geological feature matrix of the present invention; Figure 3 is a schematic diagram of the CNN-Attention-SVR model of the present invention. Detailed Embodiments

[0017] The technical solutions of the present invention will be further described below with reference to the drawings.

[0018] As Figures 1 - 3 shown, an embodiment of the present invention provides a method for predicting shield tunneling parameters under composite strata, including the following steps: Step S1, collect and sort out historical data of shield tunneling under composite strata, and perform data preprocessing; including: Step S11, collect geological data related to composite strata and construction data during shield tunneling from completed shield projects. The geological data includes degree, internal friction angle, and cohesion, and the construction data includes penetration degree, tunneling speed, total thrust force, cutter head rotation speed, cutter head torque, total grouting volume, HBW grease consumption, and EP2 grease consumption; Step S12: Clean and process the collected data, remove outliers, fill in missing data, and ensure the integrity and accuracy of the data. First, integrate the data within the rings into inter-ring data; eliminate those records in non-tunneling states and non-steady states, such as data when the shield machine stops tunneling due to maintenance, segment erection, or faults, and data during the process of the shield from startup to reaching a stable tunneling state. Use the 3σ standard deviation method to distinguish the data in the starting stage and the stable tunneling stage, thereby retaining the data in the stable section. Perform data noise reduction processing on the tunneling parameters, and use the moving window smoothing polynomial method to reduce the impact of noise on data analysis.

[0019] Step S2, as Figure 2 shown, construct the model input parameters, establish a quantified geology matrix by ring, and the matrix includes geological parameters and geometric parameters, including: Step S21: Based on the processed geological data and shield tunneling parameters, group the data by tunneling ring number to generate a quantified geology matrix for each ring; Step S22: Add geometric parameters to the quantified geology matrix, including soil layer thickness and relative position of the face, and these parameters can significantly affect the tunneling state and parameters of the shield. Next, perform parameter setting and initialization work, and the data processing work in this process needs to ensure the accuracy and consistency of the data.

[0020] Step S23: Adjust the dimension of the quantified geology matrix to ensure that all input data is in the same dimension, and further standardize the matrix; the geometric parameters in the quantified geology matrix are represented by a coding method of 1 to 5. 1 and 2 respectively represent above and below the horizontal plane above the tunnel excavation face; 3 and 4 both represent the position at the tunnel excavation face, 3 is the upper half of the excavation face, and 4 is the lower half; the number 5 represents the soil layer below the tunnel excavation face. Arrange the geological parameters and geometric parameters in sequence to form a quantified row vector corresponding to the soil layer, and then combine them from top to bottom in the direction of soil layer burial depth to form a geological feature quantified matrix corresponding to the ring number. In order to ensure that the geological quantification matrix of each ring has the same dimension for input into the prediction model, all matrices are padded with 0 for alignment.

[0021] Step S3, as Figure 3 shown, construct a multi-layer convolutional neural network model, using the quantified geology matrix as the input and outputting the predicted values of the shield tunneling parameters. It includes: Step S31: Design a convolutional neural network model. The input layer receives the quantified geology matrix, and the initial layer includes multiple convolutional kernels for extracting local features of geological characteristics and geometric parameters; the convolutional layer is the core component of the CNN, and its main function is to extract data features from multiple angles. The essence of the convolutional layer is to perform a discrete convolution operation to calculate the result after convolution of each receptive field matrix.

[0022] ; where: W is the convolutional kernel matrix; b is the bias term; is the receptive field matrix.

[0023] Step S32, add a pooling layer after the convolutional layer. By performing downsampling operations, reduce the size of the feature matrix, enhance the generalization ability of the model, and avoid overfitting. The main role of the pooling layer is to compress the data, thereby reducing the number of parameters and the risk of model overfitting. Common pooling operations include max pooling and average pooling. The pooling layer reduces the spatial size of the data, reducing the computational complexity and memory consumption, thus helping to build a more efficient and stable deep learning model. The calculation formula of the pooling layer is: ; where: z is the result after pooling; is the pooling window matrix of the output matrix of the previous layer; l is the window shape.

[0024] The present invention adopts the classical LeNet architecture, mainly including two convolutional layers, two pooling layers and two fully connected layers. This structure effectively controls the complexity of the model while ensuring the training effect of the model, thereby reducing the risk of overfitting.

[0025] Step S33, set the attention mechanism, train the attention weights through the model, and highlight the feature regions that are more important for the prediction of tunneling parameters. Then, use these weighted feature maps for the regression task of tunneling parameters. Specifically, the self-attention mechanism is adopted to dynamically calculate the weights of the features. Given the input feature matrix, generate a query matrix , key matrix and value matrix through linear transformation. The formulas are as follows: ; where: is the learnable weight matrix; Then, use the softmax activation function to calculate the similarity between the query and the key to obtain the attention weights. The specific calculation formula is as follows: ; where, is the dimension of the key; Q, K, V are the linear transformations of the input feature matrix.

[0026] For the regression task using SVR, the features extracted through the convolutional layer and the attention mechanism are input into the SVR regression model. SVR will use these feature mappings to predict tunneling parameters, including 8 key tunneling parameters: penetration, total thrust, cutterhead torque, tunneling speed, cutterhead rotation speed, total grouting volume, HBW grease consumption, and EP2 grease consumption. SVR uses the radial basis function (RBF) kernel to handle non-linear relationships, mapping the input feature space to a high-dimensional feature space to make the data linearly separable. Such locally sensitive samples have a greater impact on decision-making, thus better adapting to the local structure of the data and having strong generalization ability. The specific loss function is as follows: ; where, is the weight vector, is the regularization parameter, are the slack variables.

[0027] Step S4: Train the convolutional neural network model with historical data to optimize the model parameters, including: Step S41: Divide the historical dataset into a training set and a validation set, and use the training set to initially train the model. Adjust the network weights by minimizing the prediction error; for the training set, validation set, and test set, the data volume ratios are 0.8, 0.1, and 0.1 respectively. The optimizer minimizes the loss function (MSE) through the backpropagation algorithm to adjust the model parameters. The formula is as follows:

[0028] ; In the formula: is the true value; is the predicted value; n is the number of samples.

[0029] Step S42: Use the cross-validation method during training to evaluate the performance of the model on different datasets and avoid overfitting of the model on the training set.

[0030] Step S43: Use an optimization algorithm (Adam) to further adjust the learning rate and parameters of the model to ensure that the model quickly converges to the global optimum during training. The initial learning rate is set to 0.001, and the model is trained for a maximum of 100 epochs. At the same time, to prevent overfitting of the model, when the loss value does not improve significantly within 30 iterations, the training will be terminated early. Before training, the input and output data are normalized to the range of 0 to 1, and the maximum and minimum values of the output parameters are recorded for denormalization processing after the prediction results. The formula is as follows: ; where, is the original data; is the normalized data; , are the maximum and minimum values of the corresponding parameters respectively.

[0031] Step S5: Use the trained model for prediction and output the parameter values of shield tunneling under the composite stratum. The said step S5 includes: Step S51: Input the new geological data into the already trained CNN-Attention-SVR model to predict the key parameter values during the tunneling process; Step S52: Analyze and verify the prediction results, and evaluate the prediction accuracy of the model by comparing with the parameters in the actual construction; Step S53: Adjust the shield tunneling construction parameters according to the prediction results, optimize the construction plan, and improve the tunneling efficiency and safety.

[0032] The method of the present invention aims to comprehensively consider the geological conditions and construction efficiency, optimize the prediction of shield tunneling parameters through the method of artificial intelligence deep learning, and ensure the stable and efficient tunneling process under the composite stratum conditions. The present invention can improve the construction accuracy, reduce the construction risk, and provide technical guarantee for tunnel construction under complex geological conditions.

Claims

1. A method for predicting shield tunneling parameters under composite strata, characterized in that: The following steps are involved: (1) Data preprocessing: Collect and organize historical data of shield tunneling under composite strata, including geological parameters and construction parameters, remove outliers, fill in missing values, and standardize data; (2) Constructing a quantitative geological matrix: The geological parameters are combined with the geometric parameters according to the tunneling ring number to generate a quantitative geological matrix for each ring. The geometric parameters are coded from 1 to 5 to represent the position of the soil layer relative to the tunnel excavation surface, and zero padding is used to align the input dimensions to ensure uniformity. (3) Constructing a CNN-Attention-SVR model: The input layer receives the quantized geological matrix; extracting local geological features through a convolutional neural network (CNN), including two convolutional layers and a pooling layer to introduce an attention mechanism to dynamically calculate feature weights and highlight key areas; using support vector regression (SVR) combined with a radial basis function (RBF) kernel, performing nonlinear regression on the weighted features and outputting predicted values ​​of excavation parameters; (4) Model training: The Adam optimizer is used to train the model, and cross-validation is used to prevent overfitting. The loss function is the mean square error (MSE). (5) Parameter prediction: New geological data is input into the trained model to predict excavation parameters and guide construction adjustments.

2. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (1), the geological parameters include: gravity, internal friction angle, and cohesion; the construction parameters include: penetration, excavation speed, total thrust, cutter head speed, cutter head torque, total grouting volume, HBW grease dosage, and EP2 grease dosage.

3. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (1), the 3σ standard deviation method is used to eliminate the non-stationary tunneling section data, and the moving window smoothing polynomial is used to reduce the noise of the tunneling parameters.

4. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (2), the dimension of the quantitative geological matrix is ​​18×5, each row corresponds to a soil layer unit, and contains 3 geological parameters and 2 geometric parameters. Zero padding is used to align the input matrices of all rings to ensure that the dimensions are consistent; the geometric parameters are soil layer thickness and position code.

5. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (2), the codes of 1-5 are as follows: 1 is the upper part of the horizontal plane above the excavation surface, 2 is the lower part of the horizontal plane above the excavation surface, 3 is the upper part of the excavation surface, 4 is the lower part of the excavation surface, and 5 is the lower part of the excavation surface.

6. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (3), CNN adopts the LeNet architecture, including two convolutional layers with a convolution kernel size of 3×3 and two maximum pooling layers with a pooling window of 2×2; the attention mechanism is implemented through self-attention calculation, and the formula is: ； in, is the dimension of the key; Q, K, V are the linear transformations of the input feature matrix.

7. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (3), the loss function of SVR is: ； in, is the weight vector, is the regularization parameter, is a slack variable.

8. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (4), the ratio of training set, validation set, and test set is 8:1:1, and the training termination condition is that the loss value does not decrease significantly after 30 consecutive iterations.

9. The method for predicting shield tunneling parameters in composite strata according to claim 1, characterized in that: In step (5), the prediction results are restored to the actual parameter values ​​through denormalization. The denormalization formula is: ； in, is the original data; is the normalized data; , are the maximum and minimum values ​​of the corresponding parameters respectively.

10. A shield tunneling parameter prediction system under composite strata, characterized in that: include: Data preprocessing module: used to collect and organize historical data of shield tunneling under composite strata, including geological parameters and construction parameters, to remove outliers, fill in missing values ​​and standardize data; Constructing a quantitative geological matrix module: It is used to combine geological parameters with geometric parameters according to the excavation ring number to generate a quantitative geological matrix for each ring; the geometric parameters are coded 1-5 to represent the position of the soil layer relative to the tunnel excavation surface and zero-filled to ensure the uniformity of the input dimension; CNN-Attention-SVR module: used for receiving the quantized geological matrix at the input layer; extracting local geological features through convolutional neural network CNN, including two convolutional layers and pooling layers to introduce attention mechanism to dynamically calculate feature weights and highlight key areas; using support vector regression SVR combined with radial basis function RBF kernel to perform nonlinear regression on weighted features and output predicted values ​​of excavation parameters; Model training module: used to train the model using the Adam optimizer, and to prevent overfitting through cross-validation. The loss function is the mean square error (MSE). Parameter prediction module: used to input new geological data into the trained model, predict excavation parameters and guide construction adjustments.

Images