A real-time prediction method for TBM tunneling parameters in complex strata

Abstract

The present application relates to the technical field of tunneling parameter prediction, and particularly relates to a real-time prediction method for TBM tunneling parameters in complex strata. The technical scheme comprises data collection and preprocessing, model selection and training, model prediction and evaluation, result application and feedback. The data collection and preprocessing comprises real-time collection of operation parameters through various sensors on the TBM equipment, including cutterhead speed, penetration, thrust, torque, total thrust, shoe pressure, advancing speed, current and power. The present application realizes real-time prediction of TBM tunneling parameters in complex strata by fusing machine learning algorithms, can optimize tunneling parameters in real time, improve construction efficiency, stabilize the tunneling process to ensure project quality, reduce tool wear, reduce energy consumption and other construction costs, prevent equipment failure and accidents to ensure construction safety, and can also promote construction technology innovation, realize intelligent construction management, comprehensively improve the comprehensive benefits of TBM construction, and promote the efficient, safe and intelligent development of tunnel (gallery) projects.

Description

Technical Field

[0001] The present invention relates to the technical field of tunneling parameter prediction, and in particular to a real-time prediction method for TBM tunneling parameters in complex strata. Background Art

[0002] As an advanced tunneling and roadway construction equipment, full-face rock tunnel boring machines (TBMs) have been widely used in various tunnel and coal mine roadway projects. However, tunneling operations in complex strata face many severe challenges.

[0003] (1) Uncertainty in tunneling parameter control

[0004] Currently, TBM excavation parameter control relies heavily on the operator's experience. Drivers adjust key excavation parameters such as cutterhead speed, thrust, and torque based on their own experience and subjective judgment of site conditions. This approach has significant limitations, as it is difficult for operators to accurately predict changing geological conditions and precisely grasp the complex relationship between excavation parameters and formation characteristics. When formation conditions suddenly change, such as when transitioning from soft rock to hard rock, or encountering complex situations like rock fracture zones or areas with abundant groundwater, operators often fail to optimize and adjust excavation parameters in a timely and accurate manner. This can lead to low excavation efficiency, increased cutter wear, increased construction costs, and even accidents, impacting construction progress and quality.

[0005] (2) Hysteresis in response to formation changes

[0006] The geological conditions of complex strata are highly heterogeneous and uncertain. During excavation, factors such as the hardness, strength, integrity, and groundwater conditions of the strata may change at any time. Due to the lack of effective real-time monitoring and prediction methods, the TBM system's response to stratum changes is significantly delayed. Traditional construction methods typically rely on post-analysis and adjustments after problems arise during construction. This passive response is not only inefficient but also fails to fundamentally resolve the problem. For example, when a tool is excessively worn due to encountering hard rock, the excavation progress may have already been significantly impacted. At this point, replacing the tool not only increases construction costs but also causes interruptions and prolongs the construction period.

[0007] (3) Limitations of existing technologies

[0008] While some technologies and methods exist for controlling TBM excavation parameters, most of these methods have limitations. While some traditional mathematical models and empirical formulas can, to a certain extent, reflect the relationship between excavation parameters and stratum conditions, the complexity and diversity of strata significantly limit the accuracy and applicability of these models. Furthermore, while existing monitoring technologies can capture some basic construction parameters, their ability to obtain detailed geological information within the strata is limited, making it difficult to accurately predict and adjust excavation parameters in real time. Furthermore, as tunneling projects advance into deeper and more complex strata, the requirements for precision and real-time control of TBM excavation parameters are becoming increasingly stringent, and existing technologies are no longer able to meet the demands of actual projects.

[0009] In summary, this application proposes a real-time prediction method for TBM excavation parameters in complex strata. Summary of the Invention

[0010] The purpose of the present invention is to propose a real-time prediction method for TBM excavation parameters in complex strata to address the problem in the background technology that TBM excavation parameters are still mainly operated by the driver based on experience, and the changing laws of the excavation parameters cannot be accurately known, and it is difficult to make timely optimization decisions on the excavation parameters when the geological conditions of the stratum change.

[0011] The technical solution of the present invention is a method for real-time prediction of TBM excavation parameters in complex strata, comprising the following steps:

[0012] Data collection and preprocessing:

[0013] Various sensors on the TBM equipment collect real-time operating parameters, including cutterhead speed, penetration, thrust, torque, total thrust, gripper pressure, propulsion speed, current, and power. The surrounding rock grade is determined as a geological parameter based on geological survey reports and the judgment of on-site geological engineers.

[0014] The collected data were cleaned, checked for completeness, and missing data were supplemented using linear interpolation. Outliers were identified and removed, and the data were mapped to the [0,1] interval using the maximum-minimum normalization method.

[0015] Model selection and training:

[0016] The SVR and AdaBoost fusion algorithms are used. SVR determines the hyperplane and derives the decision function by solving a specific optimization problem. AdaBoost trains weak learners by adjusting sample weights and combines them into strong learners.

[0017] First, use SVR for preliminary predictions. Then, use the SVR prediction results and the original input features as AdaBoost inputs. The data is divided into training and test sets at 90% and 10% respectively. Five-fold cross-validation is used to optimize the SVR hyperparameters and train the model. The type and number of AdaBoost weak learners are determined and trained, and finally the prediction model is obtained by fusion.

[0018] Model prediction and evaluation:

[0019] During the tunneling process, parameters are acquired in real time and normalized before being input into the model to obtain real-time prediction values. The model is dynamically updated using online learning.

[0020] Use the coefficient of determination, mean absolute error, and mean square error to evaluate the prediction effect and draw a comparison curve;

[0021] Application of results and feedback:

[0022] Provide optimization suggestions to drivers based on prediction results, establish a decision support system to comprehensively analyze multiple factors to generate optimization plans and monitor early warnings;

[0023] Collect actual data and compare and analyze the predicted results, optimize and adjust the model, including improving data collection accuracy, adjusting model parameter structure, introducing new variables, and regularly updating and paying attention to new technologies.

[0024] Optionally, during data cleaning, for missing thrust data, use the formula Make an estimate supplement, where F t is the thrust value at the missing moment, F t-1 and F t+1 is the thrust value at adjacent moments, and t is the time variable.

[0025] Optionally, in data normalization, the normalization formula for a parameter is x min and x max are the minimum and maximum values ​​of the parameter in the data set, respectively.

[0026] Optionally, the optimization problem solved by SVR is The constraints are

[0027]

[0028] Where w is the normal vector of the hyperplane, b is the intercept, The input data x i The function mapped to the high-dimensional feature space, C is the penalty parameter used to control the complexity of the model and the tolerance to error, ∈ is the parameter of the insensitive loss function, ξ i and is a slack variable.

[0029] Optionally, the decision function of SVR is

[0030] Among them, α i and is the Lagrange multiplier, K(x i ,x) is the kernel function. Commonly used kernel functions include linear kernel functions Polynomial kernel function Where c and d are constants, Gaussian kernel function Where σ is the kernel function parameter.

[0031] Optionally, the sample weight update formula in AdaBoost is in, e t is the error rate of the weak learner on the training set, and are the weights of sample i at the tth and t+1th iterations, y i is the actual value of the sample, h t (x i ) is the predicted value of the tth weak learner for sample i.

[0032] Optionally, in the result application, the tunneling parameter decision support system generates an optimization plan based on the formation conditions and construction requirements, using preset rules and algorithms, combined with the model prediction results, construction technology and equipment performance, and issues an early warning when the deviation between the actual parameters and the predicted parameters exceeds the threshold.

[0033] Optionally, in model feedback improvements, new construction data and geological information can be incorporated into the model when regularly updating and maintaining it, and attention can be paid to the development of new machine learning technologies. The performance of the prediction method can be improved by increasing the accuracy and frequency of data collection, adjusting model hyperparameters or structure, and introducing new feature variables.

[0034] Optionally, during the data acquisition process, high-precision sensors are used to monitor the current and power parameters in real time to ensure the accuracy and stability of the data, and the sensors are calibrated regularly.

[0035] Optionally, during the model selection and training process, when SVR uses a Gaussian kernel function, the optimal combination of the Gaussian kernel function parameter σ and the penalty parameter C is determined by cross-validation combined with a grid search method, where the parameter range of the grid search is pre-set based on historical data and experience.

[0036] Compared with the prior art, this application has at least one of the following beneficial technical effects:

[0037] 1. This invention accurately predicts tunneling parameters, enabling the TBM to adjust promptly to changes in the stratum. If rock hardness changes, the cutterhead speed and thrust are automatically adjusted, reducing downtime for adjustments. It provides real-time optimization suggestions to the operator, speeding up operational decisions and increasing the equipment's effective operating time. Predicting stratum changes and adjusting parameters in advance reduces cutter wear and equipment failure, minimizing construction interruptions. For example, if hard strata are predicted, advance adjustments can be made to avoid cutter damage. Promptly detecting and addressing anomalies allows for early warning and action, ensuring construction continuity.

[0038] 2. Real-time prediction and parameter adjustment ensure a stable excavation process, reduce overbreak and underbreak, minimize ground disturbance, ensure uniform tunnel (tunnel) cross-sections and stable surrounding rock mass, provide a good foundation for support linings, and improve tunnel (tunnel) formation quality and overall structural safety. Accurate prediction ensures that tunnel (tunnel) excavation profiles conform to the design, optimizes lining construction, improves lining durability and waterproofing, and reduces quality issues such as lining cracks and leakage.

[0039] 3. Rational prediction of parameters reduces tool wear, extends tool life, reduces tool replacement frequency and procurement costs, and avoids the time and labor consuming of frequent tool changes. Optimizing parameters improves TBM operation efficiency, reduces energy consumption, and saves energy costs such as electricity, achieving significant benefits in long-term, large-scale projects. Accurate prediction and monitoring reduce the probability of construction accidents, minimize casualties, equipment damage, and downtime and repair costs, thus ensuring economic benefits.

[0040] 4. Real-time monitoring and prediction of parameter anomalies provide early warnings to prevent safety incidents caused by equipment overload and damage, ensuring safe equipment operation. Stable excavation and appropriate parameter control reduce the risk of geological disasters such as collapse and water inrush, providing accurate information to support safety management and control, ensuring the safety of construction personnel.

[0041] 5. Employ advanced machine learning algorithms to provide the industry with new technologies, promote progress and innovation in construction technology, and enhance my country's engineering construction technology level and competitiveness. Establish a decision support system to comprehensively analyze multiple factors for intelligent management, automatically generate plan warnings, reduce manual intervention, and promote the development of intelligent and automated tunnel (lane) construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a real-time prediction method for TBM excavation parameters in complex strata. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0044] Example 1

[0045] like Figure 1As shown in the figure, the present invention proposes a real-time prediction method for TBM excavation parameters in complex strata, which includes data collection and preprocessing, model selection and training, model prediction and evaluation, and result application and feedback. The steps are described in detail below.

[0046] 1. Data Collection and Preprocessing

[0047] Data collection

[0048] The following operating parameters are collected in real time by various sensors installed on the TBM equipment: cutterhead speed (n, unit: rpm), penetration (p, unit: mm / rev), thrust (F, unit: kN), torque (T, unit: kN·m), total thrust (F total , unit: kN), gripper pressure (P shoe , unit: MPa), propulsion speed (v, unit: m / min), current (I, unit: ampere), power (P, unit: kilowatt).

[0049] At the same time, combined with the geological survey report and the professional judgment of the on-site geological engineer, the surrounding rock grade (G, discrete values ​​can be used to represent different grades) is determined as a geological parameter.

[0050] Data preprocessing

[0051] Data cleaning:

[0052] Check the integrity of the data to ensure that all parameters have corresponding values. For missing data points, if the adjacent data points in the time series change relatively smoothly, use linear interpolation to supplement them. t , it is known that the thrust data at adjacent moments are F t-1 and F t+1 , then by the formula Perform estimation supplementation, where t is the time variable.

[0053] Identify and eliminate outliers. Set a reasonable threshold range and treat data outside this range as outliers. For thrust F, according to the performance parameters of the equipment and previous construction experience, set the normal thrust range to [F min ,F max ], if the thrust value F collected at a certain moment i Satisfy F i <F min or F i >F max , then the data point will be removed.

[0054] Data normalization:

[0055] The maximum-minimum normalization method is used to map each parameter data to the [0,1] interval to eliminate the influence of different parameter dimensions and facilitate subsequent model training and calculation. For a certain parameter x, cutterhead speed n, the normalization formula is:

[0056]

[0057] Among them, x min and x max are the minimum and maximum values ​​of the parameter in the data set, respectively.

[0058] 2. Model Selection and Training

[0059] Model selection

[0060] This scheme adopts the algorithm of SVR (support vector machine) and AdaBoost fusion to predict tunneling parameters.

[0061] Support Vector Machine (SVR) Principle:

[0062] The basic idea of ​​SVR is to find an optimal hyperplane in the feature space so that the distance between the sample point and the hyperplane is maximized. For a given training data set {(x i ,y i )},i=1,2,…,n, where x i is the input feature vector, which is a vector composed of operating parameters and geological parameters in this embodiment, i is the corresponding target output, which includes tunneling parameters such as thrust and cutterhead speed. SVR determines the hyperplane by solving the following optimization problem:

[0063]

[0064] subject to

[0065]

[0066] Where w is the normal vector of the hyperplane, b is the intercept, The input data x i The function mapped to the high-dimensional feature space, C is the penalty parameter used to control the complexity of the model and the tolerance to error, ∈ is the parameter of the insensitive loss function, ξ i and is a slack variable used to handle noise and outliers in the data.

[0067] After solving the above optimization problem, the decision function is obtained:

[0068]

[0069] Among them, α i and is the Lagrange multiplier, K(x i ,x) is the kernel function. Commonly used kernel functions include linear kernel functions Polynomial kernel function Where c and d are constants, Gaussian kernel function Where σ is a kernel function parameter. In this embodiment, a suitable kernel function is selected according to the actual data situation.

[0070] AdaBoost principle:

[0071] AdaBoost is an iterative ensemble learning algorithm that trains a series of weak learners (usually simple classifiers or regressors) by continuously adjusting the weights of training samples and combines these weak learners into a strong learner. i ,y i )}, i=1,2,…,n, initially assigning the same weight to each sample In each iteration, a weak learner h is trained t , and then according to the error rate e of the weak learner on the training set t To update the sample weights:

[0072]

[0073] Where I(.) is an indicator function, which is 1 when the condition is met and 0 otherwise.

[0074] The sample weight update formula is:

[0075]

[0076] in, is the weight coefficient of the weak learner.

[0077] Finally, multiple weak learners are combined into a strong learner:

[0078]

[0079] Among them, T is the number of iterations, that is, the number of weak learners.

[0080] Fusion strategy:

[0081] First, use SVR to make a preliminary prediction of the tunneling parameters and obtain the prediction results Then, Together with the original input feature vector x, it is used as the input feature of AdaBoost. AdaBoost further optimizes and adjusts the prediction results of SVR based on multiple weak learners to obtain the final prediction results.

[0082] Dataset partitioning

[0083] The preprocessed data is divided into a training set and a test set at a ratio of 90% and 10% respectively. The training set is used for model training and parameter adjustment, and the test set is used to evaluate the model's generalization ability and prediction effect.

[0084] Training process

[0085] SVR Training:

[0086] Select a suitable kernel function (such as a Gaussian kernel function) and determine the initial value range of the kernel function parameter σ and the penalty parameter C.

[0087] We used five-fold cross-validation to optimize hyperparameters. The training set was divided into five parts, with four parts selected as training subsets and one as validation subset. The SVR model was trained on the training subsets and evaluated on the validation subsets. The kernel function parameter σ and the penalty parameter C were continuously adjusted to achieve optimal performance on the validation set. The mean squared error (MSE) or mean absolute error (MAE) was used as the evaluation metric.

[0088] For a selected set of hyperparameter values, the entire training set is used for model training to obtain a trained SVR model.

[0089] AdaBoost training:

[0090] Determine the type of weak learner (such as decision tree) and the initial value range of the number T of weak learners.

[0091] The prediction results of SVR And the original input feature vector x is used as input, and the sample weights are initialized as n is the number of training samples.

[0092] Perform iterative training:

[0093] In each iteration, a weak learner h is trained according to the current sample weight t .

[0094] Calculate the error rate e of the weak learner on the training set t , and calculate the weight coefficient α of the weak learner according to the error rate t .

[0095] Using the formula Update sample weights to prepare for the next iteration.

[0096] After T iterations, the trained AdaBoost model is obtained.

[0097] SVR and AdaBoost fusion:

[0098] The trained SVR model and AdaBoost model are combined to obtain the final prediction model. When predicting, the real-time collected operating parameters and geological parameters are first input into the SVR model to obtain the preliminary prediction results. Then Input into the AdaBoost model together with the original input features to obtain the final tunneling parameter prediction results

[0099] 3. Model Prediction and Evaluation

[0100] Real-time prediction

[0101] During TBM excavation, current operating parameters and geological parameters are acquired in real time. After being normalized using data preprocessing methods, they are input into a trained prediction model to obtain real-time predictions of the main TBM excavation parameters, including cutterhead speed and thrust.

[0102] To adapt to changes in ground conditions and equipment operating status during tunneling, the model is dynamically updated using online learning. As new data is generated, it is added to the training set. The model is then retrained and adjusted at regular intervals or as data accumulation increases to ensure real-time and accuracy.

[0103] Prediction effect evaluation

[0104] The following indicators are used to evaluate the prediction effect of the model:

[0105] Coefficient of determination (R 2 ):

[0106]

[0107] Among them, y i is the actual value, is the predicted value, is the average of the actual values. 2 The value range of is between [0,1]. The closer it is to 1, the better the model fits the data and the stronger the predictive ability.

[0108] Mean Absolute Error (MAE):

[0109]

[0110] MAE represents the average absolute error between the predicted value and the actual value. The smaller the value, the more accurate the prediction.

[0111] Mean Squared Error (MSE):

[0112]

[0113] MSE reflects the mean square of the error between the predicted value and the actual value. Similarly, the smaller the value, the better the model performance.

[0114] Evaluate the model on the test set and calculate the values ​​of the aforementioned evaluation metrics. Also, plot a comparison curve between the predicted and actual values ​​to visually demonstrate the model's prediction performance at different data points. By analyzing the evaluation metrics and comparison curves, you can comprehensively assess the model's accuracy, stability, and generalization ability.

[0115] IV. Application of Results and Feedback

[0116] Optimization decision of tunneling parameters

[0117] Based on the model's predicted excavation parameters, combined with actual project conditions and construction experience, the system provides optimized excavation parameter recommendations to the TBM operator. When ground hardness increases, the system automatically adjusts thrust and cutterhead speed to appropriate ranges based on equipment performance and historical construction data to ensure excavation efficiency and tool life. This functionality is achieved by incorporating appropriate parameter warning and automatic adjustment mechanisms into the TBM control system.

[0118] A tunneling parameter decision support system was established. This system comprehensively analyzes model predictions with factors such as construction technology and equipment performance. Based on different ground conditions and construction requirements, the system uses pre-defined rules and algorithms to generate optimized tunneling parameter plans. Furthermore, the system monitors parameter changes during tunneling in real time. When the deviation between actual and predicted parameters exceeds a certain threshold, it issues a timely adjustment warning, providing construction managers with a basis for decision-making and enabling them to take timely corrective measures.

[0119] Feedback and Improvement

[0120] During the actual construction process, we continuously collected data on actual TBM excavation parameters and construction results, comparing and analyzing them with the model's predictions. Based on this feedback, we conducted an in-depth analysis of the sources and distribution of model prediction errors, identifying potential issues, including data collection errors, inappropriate model structure, and inadequate consideration of the complexity of strata.

[0121] Based on the feedback analysis results, the model is further optimized and adjusted. The accuracy and frequency of data collection are increased, model hyperparameters or structure are adjusted, and new characteristic variables are introduced to better reflect the actual conditions of the formation and equipment. Simultaneously, the model is regularly updated and maintained, incorporating new construction data and geological information into model training to adapt to the changing conditions of different project sites and formations. Furthermore, new algorithms and technologies in the field of machine learning are closely monitored, and advanced methods and technologies are promptly introduced into TBM excavation parameter prediction to continuously improve the performance and accuracy of the prediction method, achieving continuous improvement and innovation.

[0122] The present invention enables the TBM to always maintain the best working state under different formation conditions by accurately predicting the excavation parameters, including the cutterhead speed and thrust. When encountering complex formation changes, the system can adjust the parameters in time to avoid low excavation efficiency caused by unreasonable parameters. When the rock hardness increases, the thrust is automatically increased and the cutterhead speed is appropriately reduced to ensure smooth excavation, reduce unnecessary downtime and adjustment time, and thus improve the overall excavation efficiency. It provides real-time optimization suggestions for TBM drivers, allowing them to quickly make correct operational decisions without having to rely on experience to grope and try, which greatly shortens the time for parameter adjustment, increases the effective working time of the equipment, and thus improves construction efficiency.

[0123] Accurate parameter prediction can effectively reduce construction interruptions caused by excessive tool wear and equipment failure. By predicting ground changes and adjusting parameters in advance, the wear rate of key components such as tools can be reduced, extending their service life. When hard formations are predicted, thrust and speed can be adjusted in advance to reduce impact loads on the tools, avoid sudden tool damage and interruptions, ensure construction continuity, and improve efficiency.

[0124] In addition, it can promptly detect abnormal conditions in equipment operation, such as excessive thrust and abnormal torque, and notify construction personnel in advance through the early warning system to handle them, avoiding long-term shutdown and maintenance caused by the deterioration of the problem, and reducing the impact of construction interruptions on the construction period.

[0125] In this embodiment, real-time prediction and adjustment of tunneling parameters help maintain the stability of the tunneling process. Stable cutterhead speed, thrust and torque parameters can make the excavation of the tunnel (tunnel) section more uniform and reduce the occurrence of over-excavation and under-excavation. In soft rock formations, by precisely controlling the thrust and speed, excessive disturbance of the rock mass around the tunnel (tunnel) due to excessive thrust can be avoided, thereby ensuring the forming quality of the tunnel (tunnel). A stable tunneling process can also reduce disturbance to the stratum, reduce the risk of stratum deformation and collapse, provide good basic conditions for subsequent support and lining work, and ensure the quality of the entire tunnel (tunnel) project. Accurate prediction of tunneling parameters can make the excavation profile of the tunnel (tunnel) more in line with the design requirements, thereby improving the construction quality of the tunnel (tunnel) lining. When the cross-sectional dimensions of the tunnel (tunnel) are accurate, the amount of lining material can be more reasonable, the force on the lining structure is more uniform, and the durability and waterproof performance of the lining can be effectively improved. Reduce quality problems such as tunnel (tunnel) lining cracks and leakage caused by unstable excavation process, improve the overall waterproofness and structural safety of tunnels (tunnels), and ensure the long-term stability of project quality.

[0126] It is worth noting that this embodiment significantly reduces cutter wear by reasonably predicting excavation parameters and optimizing cutter usage based on stratum conditions. Thrust and rotation speed are adjusted at the appropriate time to prevent excessive cutter wear in hard rock and extend the tool life. This reduces the number of tool changes and lowers the cost of purchasing and replacing tools. Tool replacement not only requires significant time and manpower, but also comes at a high cost. Accurate parameter prediction can effectively reduce these costs and improve the economic efficiency of construction. Optimized excavation parameters make TBM equipment more efficient and reduce energy waste. By precisely controlling the cutterhead rotation speed and thrust, the equipment can operate at the lowest energy consumption while meeting excavation requirements. Reducing energy costs, such as electricity consumption, during construction can, in the long term, represent a significant cost savings for large-scale tunnel (lane) projects, helping to reduce overall construction costs. Accurate parameter prediction and real-time monitoring can effectively reduce the probability of construction accidents. Construction accidents not only result in casualties and equipment damage, but also lead to project downtime and additional repair costs. Through the application of this method, accidents can be prevented and avoided in advance, the huge cost losses caused by accidents can be reduced, and the smooth progress of construction and economic benefits can be guaranteed.

[0127] Furthermore, real-time monitoring and prediction of TBM equipment operating parameters, such as torque, thrust, and current, allows for timely detection and early warning of any abnormal changes. A sudden increase in torque, which could indicate tool sticking or unusually hard ground, triggers a timely warning, allowing construction personnel to take appropriate measures to prevent damage from overload and thus prevent accidents caused by equipment failure. Accurate prediction and control of tunneling parameters ensures TBM equipment operates within a safe operating range, minimizing the threat to construction personnel from equipment failures and accidents, and improving overall construction site safety.

[0128] This method utilizes an advanced machine learning algorithm, integrating SVR with AdaBoost, to provide a new technical approach for TBM excavation parameter prediction. This technological innovation is not only applicable to TBM construction but also provides a reference for other similar projects, driving technological progress across the entire construction industry. It encourages the continuous exploration and application of new technologies and methods during the construction process, promotes continuous innovation and development in construction technology, and enhances my country's technological level and international competitiveness in infrastructure construction, such as tunneling and roadway projects. By establishing a tunneling parameter decision support system, the model's prediction results are integrated with multiple factors, including construction technology and equipment performance, enabling intelligent management of TBM construction. The system automatically generates optimization plans and adjusts warnings based on real-time data, reducing the uncertainty of manual intervention and improving the scientific nature and accuracy of construction management. This lays the foundation for intelligent control of the entire tunneling and roadway construction process, helping to improve construction efficiency, quality, and safety, and promoting the development of tunneling and roadway construction towards intelligent and automated processes, in line with future engineering development trends.

[0129] The above specific embodiments are merely several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art may make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A real-time prediction method for TBM excavation parameters in complex strata, characterized by: The following steps are involved: Data collection and preprocessing: Various sensors on the TBM equipment collect real-time operating parameters, including cutterhead speed, penetration, thrust, torque, total thrust, gripper pressure, propulsion speed, current, and power. The surrounding rock grade is determined as a geological parameter based on geological survey reports and the judgment of on-site geological engineers. The collected data were cleaned, checked for completeness, and missing data were supplemented using linear interpolation. Outliers were identified and removed, and the data were mapped to the [0,1] interval using the maximum-minimum normalization method. Model selection and training: The SVR and AdaBoost fusion algorithms are used. SVR determines the hyperplane and derives the decision function by solving a specific optimization problem. AdaBoost trains weak learners by adjusting sample weights and combines them into strong learners. First, use SVR for preliminary predictions. Then, use the SVR prediction results and the original input features as AdaBoost inputs. The data is divided into training and test sets at 90% and 10% respectively. Five-fold cross-validation is used to optimize the SVR hyperparameters and train the model. The type and number of AdaBoost weak learners are determined and trained, and finally the prediction model is obtained by fusion. Model prediction and evaluation: During the tunneling process, parameters are acquired in real time and normalized before being input into the model to obtain real-time prediction values. The model is dynamically updated using online learning. Use the coefficient of determination, mean absolute error, and mean square error to evaluate the prediction effect and draw a comparison curve; Application of results and feedback: Provide optimization suggestions to drivers based on prediction results, establish a decision support system to comprehensively analyze multiple factors to generate optimization plans and monitor early warnings; Collect actual data and compare and analyze the predicted results, optimize and adjust the model, including improving data collection accuracy, adjusting model parameter structure, introducing new variables, and regularly updating and paying attention to new technologies.

2. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: The data is being cleaned. During data cleaning, for the missing thrust data, the formula Make an estimate supplement, where F t is the thrust value at the missing moment, F t-1 and F t+1 is the thrust value at adjacent moments, and t is the time variable.

3. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1 is characterized in that: In data normalization, the normalization formula for a parameter is x min and x max are the minimum and maximum values ​​of the parameter in the data set, respectively.

4. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: The optimization problem solved by SVR is The constraints are Among them, w is the normal vector of the hyperplane, b is the intercept, The input data x i The function mapped to the high-dimensional feature space, C is the penalty parameter used to control the complexity of the model and the tolerance to error, ∈ is the parameter of the insensitive loss function, ξ i and is a slack variable.

5. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1 is characterized in that: The decision function of SVR is Among them, α i and is the Lagrange multiplier, K(x i ,x) is the kernel function. Commonly used kernel functions include linear kernel functions Polynomial kernel function Where c and d are constants, Gaussian kernel function Where σ is the kernel function parameter.

6. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: The sample weight update formula in AdaBoost is: in, e t is the error rate of the weak learner on the training set, and are the weights of sample i at the tth and t+1th iterations, y i is the actual value of the sample, h t (x i ) is the predicted value of the tth weak learner for sample i.

7. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: In the application of the results, the tunneling parameter decision support system uses preset rules and algorithms based on the geological conditions and construction requirements, combined with model prediction results, construction technology and equipment performance to generate an optimization plan, and issues an early warning when the deviation between the actual parameters and the predicted parameters exceeds the threshold.

8. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: In model feedback improvement, new construction data and geological information are incorporated into the regular update and maintenance of the model, attention is paid to the development of new machine learning technologies, and the performance of the prediction method is improved by increasing the accuracy and frequency of data collection, adjusting model hyperparameters or structure, and introducing new feature variables.

9. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: During the data collection process, high-precision sensors are used to monitor the current and power parameters in real time to ensure the accuracy and stability of the data, and the sensors are calibrated regularly.

10. The method for real-time prediction of TBM excavation parameters in complex strata according to claim 1, characterized in that: During the model selection and training process, when SVR uses a Gaussian kernel function, the optimal combination of the Gaussian kernel function parameter σ and the penalty parameter C is determined by cross-validation combined with a grid search method, wherein the parameter range of the grid search is pre-set based on historical data and experience.

Images