Method, system and storage medium for predicting shield tunneling efficiency in water-rich sandy gravel strata

Abstract

The present invention discloses a method, system and storage medium for predicting shield tunneling efficiency in water-rich sand and gravel strata. The prediction method steps are as follows: S1: real-time acquisition of stratum characteristic data, tunneling efficiency data and stratum response data; S2: cleaning, denoising, interpolation preprocessing operations and normalization of the collected raw data; S3: extracting characteristic variables that significantly affect shield tunneling efficiency; S4: based on the extracted characteristic variables, using machine learning to construct a quantitative evaluation model for key construction parameters; S5: using historical data and expert experience to construct a training set and a verification set, and training and verifying the quantitative evaluation model; S6: applying the trained quantitative evaluation model to actual construction, adjusting shield tunneling parameters and optimizing construction technology according to the real-time acquired data and model prediction results. The present invention realizes quantitative evaluation of key construction parameters through intelligent means, optimizes construction technology, and improves construction efficiency and safety.

Description

Technical Field

[0001] The invention relates to the technical field of shield tunneling, and in particular to a method, system and storage medium for predicting shield tunneling efficiency in water-rich sand and gravel strata. Background Art

[0002] When shield tunneling is carried out in water-rich sand and gravel strata, the water-rich sand and gravel strata have poor stratum stability due to their large particles, wide particle size distribution, and strong permeability. They are extremely sensitive to disturbances during tunneling, resulting in great construction difficulty and high risk. At present, the artificial intelligence determination method for shield tunneling efficiency in water-rich sand and gravel strata is still immature, with problems such as scattered information storage, complex data processing, and low model prediction accuracy.

[0003] Traditional shield tunneling methods often have difficulty in accurately predicting and responding to stratum changes, resulting in low tunneling efficiency, difficulty in controlling stratum settlement, and even safety accidents. Therefore, it is urgent to develop a prediction method for shield tunneling efficiency in water-rich sandy gravel strata. Summary of the invention

[0004] Based on the technical problems existing in the background technology, the present invention proposes a method, system and storage medium for predicting the shield tunneling efficiency in water-rich sand and gravel strata, which realizes quantitative evaluation of key construction parameters through intelligent means, optimizes construction technology, and improves construction efficiency and safety.

[0005] The method for predicting shield tunneling efficiency in water-rich sand and gravel strata proposed by the present invention comprises the following steps: S1: Use sensors on the shield machine to collect real-time data on stratum characteristics, excavation efficiency and stratum response during excavation; S2: Clean, denoise, interpolate and preprocess the collected raw data, and normalize them to eliminate dimensional differences; S3: Extract characteristic variables that have a significant impact on shield tunneling efficiency from the normalized data; S4: Based on the extracted characteristic variables, a quantitative evaluation model of key construction parameters is constructed using machine learning; S5: Use historical data and expert experience to build training sets and validation sets to train and validate the quantitative evaluation model; S6: Apply the trained quantitative evaluation model to actual construction, adjust the shield tunneling parameters and optimize the construction process based on the real-time collected data and model prediction results.

[0006] Preferably, the formula for normalization processing in S2 is:

[0007] In the formula, Xrepresents the normalized data. X 0 represents the collected data. X max and X min are the maximum and minimum values ​​of the collected data respectively.

[0008] Preferably, the extraction of characteristic variables in S3 adopts a combination of principal component analysis and entropy weight method. The n-dimensional data involved in all parameters is reduced to k dimensions using principal component analysis. On this basis, the entropy weight method is used to preliminarily determine the weight of each indicator after dimensionality reduction. A judgment value is given based on experience. Indicators with weight values ​​less than the judgment value are considered to be redundant indicators, and finally m characteristic variables are screened out.

[0009] Preferably, the quantitative evaluation model in S4 introduces the attention mechanism on the basis of the BP neural network to construct an AM-BP neural network prediction model, and assigns different weights to each input feature.

[0010] Preferably, for the quantitative evaluation model, the m feature variables are divided into p Input layer variables, q Output layer variables; The number of hidden layer neuron nodes is given by the empirical formula Decide, among others, is the number of neurons in the hidden layer, is a regulation constant between 1 and 10, thus determining the network structure as ; Input layer data Input model, where The data is preprocessed and normalized; Calculate the output layer:

[0011]

[0012] In the formula, is the output layer output, is the output of the input layer node, is the weight of the neuron node connection, is the bias term, is the activation function of the hidden layer nodes, The activation function of the output layer nodes, is the effect coefficient of the attention mechanism.

[0013] Preferably, the attention mechanism's effect coefficient The calculation formula is:

[0014]

[0015] In the formula, For the dataset The input information at a certain time slice; q is the query vector, which is represented by the output of the network hidden layer node. ; and is the similarity score between the output feature and the query vector (i.e., attention score); is the attention scoring function.

[0016] Preferably, the model training method in S5 is: Divide the collected historical data into training set, validation set and test set in proportion; Input the training set data into the model and define the loss function of the model as:

[0017] In the formula, is the output value of model training, is the actual value; is the error between the output value and the actual value; Minimize the loss function using gradient descent , and continuously improve the fitting effect of the model.

[0018] Preferably, it also includes information storage, using distributed database and cloud storage technology to uniformly store and manage collected data, normalized data, characteristic variables, quantitative evaluation models, and prediction and evaluation result information.

[0019] The shield tunneling efficiency prediction system for water-rich sand and gravel strata proposed by the present invention comprises: The data acquisition module is used to collect the stratum characteristic data, excavation efficiency data and stratum response data in real time during the excavation process using the sensors on the shield machine; The data processing module is used to clean, denoise, interpolate and preprocess the collected raw data, and perform normalization to eliminate dimensional differences; and extract characteristic variables that have a significant impact on shield tunneling efficiency from the normalized data; The model prediction module is used to build a quantitative evaluation model of key construction parameters based on the extracted characteristic variables using machine learning; and to build training sets and validation sets using historical data and expert experience to train and validate the quantitative evaluation model; The control execution module is used to apply the trained quantitative evaluation model to actual construction, adjust the shield tunneling parameters and optimize the construction process based on the real-time collected data and model prediction results.

[0020] The storage module uses distributed database and cloud storage technology to store and manage collected data, normalized data, feature variables, quantitative evaluation models, and prediction and evaluation result information.

[0021] The computer-readable medium proposed in the present invention stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for predicting the shield tunneling efficiency in water-rich sand and gravel strata is implemented.

[0022] Beneficial technical effects of the present invention: (1) The present invention combines PCA and entropy weight method to extract features, which can reduce the limitations of a single method, more comprehensively consider the intrinsic structure and information content of the data, and optimize the quality of feature extraction. This improves model performance and stability, prevents overfitting, improves generalization ability, and enhances the model's adaptability to data changes.

[0023] (2) The present invention introduces an attention mechanism into the neural network model, which enables the model to focus on key information and improve learning efficiency. In the BP neural network, the input data often contains a large number of features. The attention mechanism can automatically identify and focus on the features most relevant to the task, accelerate the convergence speed of the model, and improve the prediction accuracy and generalization ability of the model.

[0024] (3) The present invention realizes quantitative evaluation of key construction parameters through intelligent means, realizes intelligent integration of data perception, data prediction, data feedback, and data adjustment, optimizes construction technology, and improves construction efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flowchart of the shield tunneling efficiency prediction method for water-rich sand and gravel strata proposed by the present invention; Figure 2 It is a flowchart of the data acquisition module proposed by the present invention; Figure 3 A flowchart of the data processing module proposed by the present invention; Figure 4 A schematic diagram of the feature extraction algorithm proposed in the present invention; Figure 5 A flowchart for constructing the quantitative evaluation model proposed in the present invention; Figure 6 This is a schematic diagram of the model training and verification process proposed by the present invention; Figure 7 The intelligent decision-making and process optimization flow chart proposed by the present invention; Figure 8 A structural block diagram of the storage unit proposed by the present invention; Fig. 9 This is a schematic diagram of a shield tunneling project according to Application Example 1 proposed by the present invention; Fig.10 This is a comparison chart before and after the shield tunneling efficiency improvement of Application Example 2 proposed by the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further explained below in conjunction with specific embodiments.

[0027] Example 1 Reference Figure 1 The present invention proposes a method for predicting shield tunneling efficiency in water-rich sand and gravel strata, and the method steps are as follows: (1) Data collection: Using sensors installed on the shield machine, including vibration sensors installed on the cutterhead, stress sensors and displacement sensors installed on the shield body of the shield machine, and geological radar, real-time data on formation characteristics, excavation efficiency data, and formation response data during the excavation process are collected. Figure 2 As shown, starting from the start of the sensor, after signal conversion, analog-to-digital conversion, data packaging and other links, the collected raw data is finally transmitted to the data processing center; (2) Data preprocessing: Clean, denoise, and interpolate the collected raw data to identify outliers, duplicate values, and missing values ​​in the data, and complete the data based on statistical methods; in the presence of noise, use filtering technology to smooth the data and normalize it, mapping the raw data to the [0, 1] interval to eliminate dimensional differences, as follows: Figure 3 As shown in the figure, the specific steps of data preprocessing include data cleaning (removing noise, outliers, etc.), data compression (reducing storage space), data standardization (unifying dimensions) and data segmentation (facilitating subsequent processing), etc.; the following formula is used for normalization:

[0028] In the formula, X represents the normalized data. X 0 represents the collected data. X max and X min are the maximum and minimum values ​​of the collected data respectively.

[0029] (3) Feature extraction: Statistical analysis methods are used to extract characteristic variables that have a significant impact on shield tunneling efficiency from the preprocessed data, including but not limited to relevant parameters such as stratum strength index, water content change rate, tunneling speed fluctuation rate, torque peak value, etc., so as to reveal the distribution characteristics, change rules and correlation of the data, clarify the relationship and change trend between stratum characteristics and tunneling efficiency, and provide a basis for subsequent model construction; (4) Construction of quantitative evaluation model: Based on the extracted characteristic variables, machine learning is used to construct a quantitative evaluation model of key construction parameters, which can predict the changing trend of shield tunneling efficiency and evaluate the effects under different construction parameter combinations; (5) Model training and validation: Use historical data and expert experience to construct training sets and validation sets, and train and validate the quantitative evaluation model to improve the model's prediction accuracy and generalization ability; (6) Intelligent decision-making and process optimization: Apply the trained quantitative evaluation model to actual construction. Based on the real-time collected data and model prediction results, the prediction results are evaluated and verified in combination with historical data and expert experience to improve the accuracy and reliability of the prediction results. Based on the prediction results and preset decision-making rules, the excavation parameters are intelligently adjusted to optimize the construction process.

[0030] In addition, it also includes an information storage step, which uses distributed database and cloud storage technology to uniformly store and manage raw data, preprocessed data, feature variables, quantitative evaluation models, and prediction and evaluation result information.

[0031] Reference Figure 4 , which is the algorithm flow of feature extraction proposed by the present invention, including key steps such as signal decomposition (such as wavelet transform), feature selection (based on statistics or machine learning algorithm) and feature construction (synthesizing new features). Feature extraction adopts the combination of principal component analysis (PCA) and entropy weight method. PCA is used to reduce the n-dimensional data involved in all parameters to k-dimensional data. On this basis, the entropy weight method is used to preliminarily determine the weight of each indicator after dimensionality reduction. A judgment value is given according to experience. The indicators with weight values ​​less than the judgment value are considered to be redundant indicators. Finally, m feature variables are screened.

[0032] Reference Figure 5 , which is a flowchart of the construction of the quantitative evaluation model proposed in the present invention, including data partitioning (training set, validation set, test set), model selection (such as machine learning model, deep learning model, etc.), parameter tuning (through cross-validation and other methods) and model evaluation (evaluation indicators such as accuracy, recall, etc.). The quantitative evaluation model specifically introduces AM (attention mechanism) on the basis of BP neural network to construct an AM-BP neural network prediction model, eliminates the unreasonable influence of input data on output data, and assigns different weights to each input feature, so that the model can focus on more important features, improve the influence of key input data, and thus improve the accuracy of prediction.

[0033] According to the model structure, the m feature variables are divided into p input layer variables and q output layer variables according to their characteristics; the number of hidden layer neuron nodes is determined by the empirical formula Decide, among others, is the number of neurons in the hidden layer, for The adjustment constant between , thus determining the network structure as ; Input layer data Input model, where The data is preprocessed and normalized; the output layer is calculated using the following formula:

[0034]

[0035] In the formula, is the output layer output, is the output of the input layer node, is the weight of the neuron node connection, is the bias term, is the activation function of the hidden layer nodes, The activation function of the output layer nodes, is the effect coefficient of the attention mechanism.

[0036] The effect coefficient of the attention mechanism The calculation formula is:

[0037]

[0038] In the formula, For the dataset The input information at a certain time slice; q is the query vector, which is represented by the output of the network hidden layer node. ; and is the similarity score between the output feature and the query vector (i.e., attention score); is the attention scoring function.

[0039] Reference Figure 6 , model training and verification include training data input, model training (iterative optimization parameters), verification data input, verification result output and model adjustment. The specific method of model training is: Divide the collected historical data into training set, validation set and test set in proportion; Input the training set data into the model and define the loss function of the model as:

[0040] In the formula, is the output value of model training, is the actual value; is the error between the output value and the actual value; Minimize the loss function using gradient descent , and continuously improve the fitting effect of the model.

[0041] The validation set is used to adjust the model parameters and structure during the training process; the test set is used to evaluate the final performance of the model.

[0042] Reference Figure 7 The intelligent decision-making and process optimization process based on the quantitative evaluation model includes real-time monitoring data input, model prediction, decision rule matching, excavation parameter adjustment and feedback mechanism.

[0043] Reference Figure 8 The storage unit includes the data storage layer (such as distributed database, cloud storage, etc.), the data processing layer (responsible for data cleaning, conversion, etc.), the data management layer (providing data query, analysis and other functions) and the data security layer (ensuring data security and privacy protection). Distributed database and cloud storage technology are used to uniformly store and manage raw data, preprocessed data, feature variables, quantitative evaluation models, and prediction and evaluation result information.

[0044] Example 2 The shield tunneling efficiency prediction system for water-rich sand and gravel strata proposed by the present invention comprises: The data acquisition module is used to collect the stratum characteristic data, excavation efficiency data and stratum response data in real time during the excavation process using the sensors on the shield machine; The data processing module is used to clean, denoise, interpolate and preprocess the collected raw data, and perform normalization to eliminate dimensional differences; and extract characteristic variables that have a significant impact on shield tunneling efficiency from the normalized data; The model prediction module is used to build a quantitative evaluation model of key construction parameters based on the extracted characteristic variables using machine learning; and to build training sets and validation sets using historical data and expert experience to train and validate the quantitative evaluation model; The control execution module is used to apply the trained quantitative evaluation model to actual construction, adjust the shield tunneling parameters and optimize the construction process based on the real-time collected data and model prediction results.

[0045] The storage module uses distributed database and cloud storage technology to store and manage collected data, normalized data, feature variables, quantitative evaluation models, and prediction and evaluation result information.

[0046] Example 3 The computer-readable medium proposed by the present invention stores a computer program, and when the computer program is executed by a processor, the method for predicting the shield tunneling efficiency in water-rich sand and gravel formations recorded in Example 1 is implemented.

[0047] Application Example 1 Reference Fig. 9 A water diversion project includes gravel, soft soil, bedrock, etc. The geological conditions are changeable and the construction is difficult. The shield machine in a certain section needs to pass through a mixed stratum about 2 kilometers long, of which the gravel layer accounts for as much as 60%, accompanied by a large number of boulders and cracks, which poses a severe challenge to the tunneling performance and stability of the shield machine.

[0048] By installing multiple sets of intelligent monitoring devices on the shield machine, including high-precision vibration sensors, stress sensors, displacement sensors and geological radars, these sensors can monitor the tunneling parameters of the shield machine (such as thrust, torque, tunneling speed) and changes in the surrounding environment (such as stratum deformation, groundwater level fluctuations) in real time. The data is uploaded to the data processing center in real time through wireless transmission to ensure the timeliness and accuracy of the information.

[0049] After receiving the raw data, the data processing center first performs preprocessing operations, including data cleaning, denoising and standardization. Subsequently, advanced feature extraction algorithms are used to extract characteristic variables useful for tunneling efficiency evaluation from the preprocessed data. These characteristic variables include fluctuations in tunneling parameters, formation response characteristics, etc. Then, a quantitative evaluation model is constructed to conduct a comprehensive analysis of the characteristic variables and predict the risks and obstacles that may be encountered during the tunneling process.

[0050] Based on the prediction results of the quantitative evaluation model, the intelligent decision-making system can automatically adjust the excavation parameters, such as reducing the thrust, adjusting the rotation speed or changing the grouting pressure, to adapt to the complex and changing formation conditions. At the same time, the system also has a learning function, which can continuously optimize the decision-making rules based on historical data and real-time feedback to improve the accuracy and timeliness of decision-making. In addition, the system also has an early warning and alarm mechanism, which immediately notifies the operator to take corresponding measures once an abnormal situation is found to ensure construction safety.

[0051] Through the above method, the shield tunneling efficiency of the project has been significantly improved. According to statistics, the tunneling speed has increased by about 20% on average, while reducing the number of downtimes and maintenance costs caused by stratum changes. In addition, the system has successfully warned of multiple potential stratum collapse and boulder impact risks, effectively ensuring construction safety and project quality.

[0052] Application Example 2 A certain construction section is an important part of the city subway line and needs to cross a 3-kilometer-long gravel stratum. Due to the large particles, strong permeability and easy collapse of the stratum, traditional excavation methods are difficult to ensure excavation efficiency and safety.

[0053] In view of the characteristics of the gravel strata in this section, highly sensitive vibration sensors and displacement sensors were installed on the shield machine to monitor small changes during the excavation process. At the same time, geological radar was used to scan the strata ahead in real time to detect potential obstacles and unfavorable geological bodies in advance.

[0054] The data processing center uses advanced data mining technology to conduct in-depth analysis of massive monitoring data and dig out the laws and information hidden in the changes in the formation. Based on these data, the project team continuously optimizes the quantitative evaluation model to improve the prediction accuracy and generalization ability of the model. At the same time, by introducing machine learning algorithms to iteratively optimize the model, the model can better adapt to complex and changing formation conditions.

[0055] The intelligent decision-making system automatically adjusts the excavation process parameters, such as excavation speed, cutter head speed, grouting pressure, etc., based on the prediction results of the quantitative evaluation model. The system also has an intelligent recognition function that can automatically identify the characteristic signals of different strata types and adopt corresponding excavation strategies. In addition, the system is also equipped with remote monitoring and fault diagnosis functions, so that operators can grasp the excavation status in real time and respond quickly to abnormal situations.

[0056] Reference Fig.10 After comprehensive optimization of the intelligent monitoring and decision-making system, the efficiency of shield tunneling in the Shupujiang section has been significantly improved. The tunneling speed has increased by about 30% on average, while reducing downtime and maintenance costs caused by stratum collapse and tool wear. In addition, the system's early warning and alarm mechanism effectively reduced construction risks and ensured the safety and health of construction workers. Through comparative analysis, it was found that after the application of the intelligent monitoring and decision-making system, the tunneling cost of this section was reduced by about 15%, with significant economic benefits.

Claims

1. A method for predicting shield tunneling efficiency in water-rich sandy gravel strata, characterized in that: The steps are as follows: S1: Use sensors on the shield machine to collect real-time data on stratum characteristics, excavation efficiency and stratum response during excavation; S2: Clean, denoise, interpolate and preprocess the collected raw data, and normalize them to eliminate dimensional differences; S3: Extract characteristic variables that have a significant impact on shield tunneling efficiency from the normalized data; S4: Based on the extracted characteristic variables, a quantitative evaluation model of key construction parameters is constructed using machine learning; S5: Use historical data and expert experience to build training sets and validation sets to train and validate the quantitative evaluation model; S6: Apply the trained quantitative evaluation model to actual construction, adjust the shield tunneling parameters and optimize the construction process based on the real-time collected data and model prediction results.

2. The method for predicting shield tunneling efficiency in water-rich sand and gravel strata according to claim 1 is characterized in that: The formula for normalization in S2 is: ； In the formula, X represents the normalized data. X 0 represents the collected data. X max and X min are the maximum and minimum values ​​of the collected data respectively.

3. The method for predicting shield tunneling efficiency in water-rich sand and gravel strata according to claim 1 is characterized in that: The extraction of characteristic variables in S3 adopts the combination of principal component analysis and entropy weight method. The n-dimensional data involved in all parameters is reduced to k dimensions using principal component analysis. On this basis, the entropy weight method is used to preliminarily determine the weight of each indicator after dimensionality reduction. A judgment value is given according to experience. Indicators with weight values ​​less than the judgment value are considered to be redundant indicators. Finally, m characteristic variables are screened out.

4. The method for predicting shield tunneling efficiency in water-rich sand and gravel strata according to claim 3 is characterized in that: The quantitative evaluation model in S4 introduces the attention mechanism on the basis of the BP neural network to construct the AM-BP neural network prediction model, and assigns different weights to each input feature.

5. The method for predicting shield tunneling efficiency in water-rich sand and gravel strata according to claim 4 is characterized in that: For the quantitative evaluation model, m feature variables are divided into p Input layer variables, q Output layer variables; The number of hidden layer neuron nodes is given by the empirical formula Decide, among others, is the number of neurons in the hidden layer, is a regulation constant between 1 and 10, thus determining the network structure as ; Input layer data Input model, where is the data after normalization; Calculate the output layer: ； ； In the formula, is the output layer output, is the output of the input layer node, is the weight of the neuron node connection, is the bias term, is the activation function of the hidden layer nodes, The activation function of the output layer nodes, is the effect coefficient of the attention mechanism.

6. The method for predicting shield tunneling efficiency in water-rich sandy gravel strata according to claim 5, characterized in that: The effect coefficient of the attention mechanism The calculation formula is: ； ； In the formula, For the dataset The input information at a certain time slice; q is the query vector, which is represented by the output of the network hidden layer node. ; and is the similarity score between the output feature and the query vector (i.e., attention score); is the attention scoring function.

7. The method for predicting shield tunneling efficiency in water-rich sand and gravel strata according to claim 6, characterized in that: The method for model training in S5 is: Divide the collected historical data into training set, validation set and test set in proportion; Input the training set data into the model and define the loss function of the model as: ； In the formula, is the output value of model training, is the actual value; is the error between the output value and the actual value; Minimize the loss function using gradient descent , and continuously improve the fitting effect of the model.

8. The method for predicting shield tunneling efficiency in water-rich sandy gravel strata according to claim 1, characterized in that: It also includes information storage, which uses distributed database and cloud storage technology to uniformly store and manage collected data, normalized data, characteristic variables, quantitative evaluation models, and prediction and evaluation result information.

9. The shield tunneling efficiency prediction system for water-rich sand and gravel strata is characterized by: include: The data acquisition module is used to collect the stratum characteristic data, excavation efficiency data and stratum response data in real time during the excavation process using the sensors on the shield machine; The data processing module is used to clean, denoise, interpolate and preprocess the collected raw data, and perform normalization to eliminate dimensional differences; and extract characteristic variables that have a significant impact on shield tunneling efficiency from the normalized data; The model prediction module is used to build a quantitative evaluation model of key construction parameters based on the extracted characteristic variables using machine learning; and to build training sets and validation sets using historical data and expert experience to train and validate the quantitative evaluation model; The control execution module is used to apply the trained quantitative evaluation model to actual construction, adjust the shield tunneling parameters and optimize the construction process based on the real-time collected data and model prediction results; The storage module uses distributed database and cloud storage technology to store and manage collected data, normalized data, feature variables, quantitative evaluation models, and prediction and evaluation result information.

10. A computer readable medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting shield tunneling efficiency in water-rich sand and gravel strata as described in any one of claims 1 to 8 is implemented.

Images