A data-driven multi-objective optimization decision-making method for tunnel boring machine parameters

Abstract

This invention provides a data-driven multi-objective optimization decision-making method for tunnel boring machine (TBM) excavation parameters, comprising: constructing a multi-objective evaluation system for the TBM construction process; acquiring TBM operation data and preprocessing the data; dividing the dataset based on the preprocessed data; constructing a mapping relationship between geological parameters, excavation parameters, and excavation performance objectives using the LightGBM model to form an objective function; evaluating and verifying the objective function; performing multi-objective optimization using a non-dominated sorting genetic algorithm within a given range of excavation parameters and under the constraint of balancing hydraulic cylinder thrust and total TBM thrust, generating a Pareto optimal solution set; evaluating individuals in the Pareto optimal population based on the TOPSIS method, and selecting the closest parameter combination based on subjective weighting. This invention is applicable to the field of TBM construction, can significantly improve the excavation efficiency of TBMs, reduce construction risks, and provide effective support for the optimization decision-making of TBM excavation parameters.

Description

Technical Field

[0001] This invention relates to the field of tunnel boring machine (TBM) construction technology, and more specifically to a data-driven multi-objective optimization decision-making method for TBM tunneling parameters. Background Technology

[0002] In shield tunnel construction, optimizing the tunneling parameters of the tunnel boring machine (TBM) is crucial, as it is a core element in ensuring the efficient and safe progress of the project. Traditional methods primarily rely on the operator's experience to adjust parameters. However, due to the high complexity and dynamic changes in geological conditions, as well as the subjectivity of human judgment, existing methods have the following significant limitations:

[0003] Experience-based dependence and uncertainty: Operators typically adjust parameters (such as thrust, cutterhead speed, etc.) based on historical experience or qualitative rules, making it difficult to quantify the complex nonlinear relationship between geological conditions and tunneling performance. When geological conditions change abruptly (such as alternation between soft and hard strata), experience-driven parameter adjustments are prone to lag or inaccuracy, leading to a decrease in tunneling speed or loss of attitude control.

[0004] Limitations of single-objective optimization: Traditional methods often consider a single objective in isolation (such as maximizing tunneling speed), neglecting the conflicts between multiple objectives. For example, increasing tunneling speed may exacerbate deviations in the tunnel boring machine's pitch or roll angles, threatening the accuracy of the tunnel axis and equipment safety; conversely, excessively constraining attitude adjustment may sacrifice efficiency. This trade-off is difficult to achieve a dynamic balance through human experience.

[0005] Insufficient data utilization: During the operation of the tunnel boring machine, a massive amount of sensor data is generated (such as cylinder pressure, cutterhead torque, and ground parameters). However, traditional methods lack systematic data modeling capabilities and fail to fully explore the tunneling patterns hidden in the data, leading to blind parameter decision-making.

[0006] Insufficient real-time performance and adaptability: Real-time changes in geological conditions require rapid parameter response, while manual adjustments are delayed and cannot adapt to complex constraints (such as multi-cylinder thrust balancing and equipment physical limitations). In addition, existing optimization methods are mostly based on static models, making it difficult to dynamically adapt to the time-varying characteristics during the tunneling process.

[0007] To balance the tunneling efficiency and safety of a tunnel boring machine (TBM), tunneling efficiency primarily focuses on tunneling speed, while safety is reflected in minimizing attitude deviations. However, there is often a conflict between the stability of attitude control and efficient tunneling, making it difficult to achieve an optimal balance. Summary of the Invention

[0008] The purpose of this invention is to propose a data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters, in order to solve the aforementioned problems existing in the prior art.

[0009] The technical solution is a data-driven multi-objective optimization decision-making method for tunnel boring machine (TBM) excavation parameters. This method combines machine learning models and multi-objective optimization algorithms to achieve a dual improvement in efficiency and safety during TBM excavation. To achieve a balance between stable attitude control and efficient excavation during TBM excavation, specific combinations of TBM excavation parameters and excavation strategies are provided, including:

[0010] Step S1: Construct a multi-objective evaluation system for the shield tunneling process, acquire shield machine operation data, and preprocess the data;

[0011] Step S2: Based on the preprocessed data, divide the dataset;

[0012] Step S3: Use the LightGBM model to construct the mapping relationship between geological parameters, tunneling parameters and tunneling performance targets, and form the objective function;

[0013] Step S4: Evaluate and verify the objective function obtained in step S3;

[0014] Step S5: Using a non-dominated sorting genetic algorithm, multi-objective optimization is performed within the given range of tunneling parameters and under the constraint of balancing the hydraulic cylinder thrust and the total thrust of the tunnel boring machine, to generate the Pareto optimal solution set.

[0015] Step S6: Evaluate individuals in the Pareto optimal population based on the TOPSIS method, and select the closest parameter combination based on subjective weighting.

[0016] According to a further improvement of the present invention, the method for preprocessing the tunnel boring machine (TBM) operation data in step S1 is to remove abnormal data, including the TBM's shutdown data during operation, the TBM's data from the start of tunneling until it reaches a steady state, and outliers; the 3σ criterion method is used to remove outliers.

[0017] According to a further improvement of the present invention, the dataset in step S2 includes necessary data and auxiliary data; the necessary data includes the average value of the tunneling speed of each ring, the average value of the pitch angle and the average value of the roll angle at the last five moments; the auxiliary data includes: internal friction angle, cohesion, unit weight, tunnel burial depth, uniaxial compressive strength, total thrust, cutterhead torque, cutterhead rotation speed, thrust of AF group hydraulic cylinder, pressure data of three working chambers, slurry discharge / slurry inlet pump speed, equipment side roll, equipment tilt angle, stroke data of two main drive anti-torsion hydraulic cylinders, and pressure data of two main drive anti-torsion hydraulic cylinders, totaling twenty-five data. Both the necessary data and the auxiliary data are obtained by averaging between rings.

[0018] According to a further improvement of the present invention, the objective function constructed in step S3 is:

[0019] O1=f1(Xi ), i = 1, 2, ..., 25;

[0020] O2=f2(X i ), i = 1, 2, ..., 25;

[0021] O3=f3(X i ), i = 1, 2, ..., 25;

[0022] Where O1, O2, and O3 represent tunneling speed, pitch angle, and roll angle, respectively; X represents the expert features selected for constructing the LGBM model, which includes five passive parameters and twenty active parameters, namely internal friction angle, cohesion, unit weight, tunnel burial depth, uniaxial compressive strength, total thrust, cutterhead torque, cutterhead speed, thrust of AF group hydraulic cylinders, pressure data of three working chambers, slurry discharge / inlet pump speed, equipment side roll, equipment tilt angle, stroke data of two main drive anti-torsion hydraulic cylinders, and pressure data of two main drive anti-torsion hydraulic cylinders; and f represents the LGBM model.

[0023] According to a further improvement of the present invention, the objective function evaluation and verification steps in step S4 are as follows:

[0024] Step S41: Use grid search to determine hyperparameters, including learning rate, maximum tree depth, minimum number of samples per leaf, minimum split gain, and maximum number of leaves.

[0025] Step S42: After determining the hyperparameters of the model, the five-fold cross method is used for training and evaluation. That is, each time 80% of the data is used as the training set and 20% of the data is used as the test set. The dataset is randomly split, and training is performed on the training set and testing is performed on the test set.

[0026] Step S43: Evaluate the prediction accuracy of the model by using root mean square error and goodness-of-fit indices to measure the accuracy of the model.

[0027] According to a further improvement of the present invention, there are two constraints in step 5: one is the parameter range constraint, i.e., the decision variable range constraint, and the other is the balance constraint between the sum of the cylinder thrust and the total thrust, described as follows:

[0028]

[0029] in, This represents the maximum allowable tunneling speed. These represent the maximum permissible deviations of the pitch and roll angles in engineering, respectively. Let be the lower bound of the i-th decision variable. Let P be the upper limit of the i-th decision variable. jIt is the pressure of the j-th cylinder, and A represents the effective area of ​​the piston. Due to force transmission loss, an error band of α is allowed.

[0030] According to a further improvement of the present invention, the specific steps of multi-objective optimization using a non-dominated sorting genetic algorithm in step S5 are as follows:

[0031] Step S51: Initialize the population, initialize the dominance count of each individual to zero, and create an empty set to record the individuals it dominates;

[0032] Step S52: Use the LGBM model prediction function to evaluate the objective function value corresponding to the current decision variable;

[0033] Step S53 involves iterating the population, maintaining population diversity through crowding, and using an elitist strategy to ensure the algorithm maintains a good solution, continuously optimizing to obtain the optimal solution set for the optimization problem.

[0034] According to a further improvement of the present invention, step 6 uses the TOPSIS method to evaluate the population in the Pareto solution, and the specific steps include:

[0035] Step S61: Use a comprehensive analysis method, combined with expert experience, to determine the target weights;

[0036] Step S62: Through weighted normalization, a scheme decision is made based on the distance from the ideal solution, and a scheme for adjusting and setting tunneling parameters is given.

[0037] Step S63, the evaluation method is as follows:

[0038]

[0039] in, C represents the best ideal solution and the negative ideal solution within a set of optimized solutions with specific weights. i Let represent the performance score of the i-th individual in the population under subjective weighting. The value is between 0 and 1. The closer the performance score is to 1, the better the performance. Select the individual with the best performance score from the population as the result of this optimization.

[0040] Beneficial effects: By optimizing key performance indicators such as tunneling speed, pitch angle, and roll angle of the tunnel boring machine (TBM), the tunneling performance of the TBM can be effectively improved, and scientific decision support can be provided for parameter optimization during the TBM construction process. By integrating machine learning models and multi-objective optimization algorithms, a complete framework from data preprocessing and mapping modeling to parameter optimization is constructed, which solves the bottlenecks of traditional methods such as experience dependence, multi-objective conflict, and insufficient real-time performance, and significantly improves the level of intelligence in TBM construction. The invention has also been experimentally verified to prove the effectiveness of the method in practical engineering applications, especially in complex geological conditions, which can ensure the stable and efficient operation of the TBM. Attached Figure Description

[0041] Figure 1 This is an overall flowchart of the present invention.

[0042] Figure 2 This is a flowchart of the optimization solution for the balance between tunneling speed and attitude angle in the optimization model of this invention.

[0043] Figure 3 This is a schematic diagram illustrating the changes in target values ​​among the three objectives during the optimization iteration process of this invention.

[0044] Figure 4 This is a graph showing the change in the target value before and after optimization in this invention. Detailed Implementation

[0045] The technical solution of the present invention will be further described in detail below through embodiments and with reference to the accompanying drawings.

[0046] To make the technical problems of the present invention clearer, embodiments of the present invention will be provided below, along with a more comprehensive description. However, the specific embodiments given herein are only for explaining the present invention and are not intended to limit the scope or application of the invention.

[0047] like Figure 1 As shown, the present invention provides a data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters, which can be divided into an offline modeling part and an online optimization part, specifically including the following steps:

[0048] Acquire tunnel boring machine (TBM) operation data under normal operating conditions, preprocess the TBM operation data, extract key variables from the TBM operation data, and generate the mean value of key variables for each ring.

[0049] The preprocessing methods include: collecting data from the tunnel boring machine during excavation and processing the data, including removing data from shutdown periods and data from the ascent phase, and using the 3σ criterion to remove outliers. Key variables include: internal friction angle, cohesion, unit weight, tunnel depth, uniaxial compressive strength, total thrust, cutterhead torque, cutterhead speed, AF group cylinder thrust, working chamber pressure (three groups), slurry discharge / inlet pump speed, equipment roll, equipment tilt angle, main drive anti-torsion cylinder stroke (two groups), and main drive anti-torsion cylinder pressure (two groups), totaling twenty-five.

[0050] The model is then trained to obtain the objective function. First, a grid search method is used to determine the hyperparameters, including the learning rate, maximum tree depth, minimum number of samples per leaf, minimum split gain, and maximum number of leaves. After determining the hyperparameters, the model needs to be trained and evaluated. This approach uses a five-fold cross-validation method for training and evaluation, where 80% of the data is used as the training set and 20% as the test set each time. The dataset is randomly split, and training is performed on the training set, while testing is performed on the test set. The model's prediction accuracy is evaluated using the root mean square error (RMSE) and goodness of fit (R²). 2 The accuracy of the model is measured by indicators such as ( ).

[0051] The above descriptions all refer to the objective function models constructed during the offline phase.

[0052] During the online phase, the tunnel boring machine's real-time monitoring system continuously acquires and preprocesses operational data. This data serves as input to the model, used to predict the tunneling performance targets of the tunnel boring machine in real time, including tunneling speed, pitch angle, and roll angle.

[0053] O1=f1(X i ), i = 1, 2, ..., 25;

[0054] O2=f2(X i ), i = 1, 2, ..., 25;

[0055] O3=f3(X i ), i = 1, 2, ..., 25;

[0056] Where O1, O2, and O3 are the tunneling speed, pitch angle, and roll angle, respectively, and X is the expert feature selected for constructing the LGBM model.

[0057] Secondly, the LGBM model constructed in the offline phase is used as the objective function in the online phase.

[0058] Then, the decision variables and constraints are determined, including parameter range constraints and thrust balance constraints.

[0059] Using the twenty tunneling parameters of the tunnel boring machine in the above model as decision variables, and the range of tunneling parameters and the thrust balance between the total thrust and the propulsion cylinder during the tunneling process as constraints, a multi-objective optimization algorithm is constructed, which can be explained by the following formula:

[0060]

[0061] in, This represents the maximum allowable tunneling speed. These represent the maximum permissible deviations of the pitch and roll angles in engineering, respectively. Let be the lower bound of the i-th decision variable. Let P be the upper limit of the i-th decision variable. j Let be the pressure of the j-th cylinder, and A represent the effective area of ​​the piston. Due to force transmission losses, an error band of α is allowed.

[0062] The objective function value under the current decision variables is predicted using the LGBM model, and then optimized using the Non-Dominated Sorting Genetic Algorithm (NSGA-II) under given constraints. The optimization process is as follows: Figure 2 As shown.

[0063] A second-generation non-dominated genetic algorithm is used to solve the optimization problem, employing population iteration to find the optimal solution. The optimization problem is solved using a second-generation non-dominated sorting genetic algorithm to find a solution that balances excavation efficiency and safety in complex geological environments, including:

[0064] Initialize the population, initialize the dominance count of each individual to zero, and create an empty set to record the individuals it dominates;

[0065] The predictive capabilities of the LGBM model are used to evaluate the objective function values ​​corresponding to the current decision variables.

[0066] Population iteration is performed, population diversity is maintained through crowding, and an elitist strategy is used to ensure that the algorithm can maintain good solutions, and the optimal solution set of the optimization problem is obtained through continuous optimization.

[0067] In the second-generation non-dominated genetic algorithm employed, the algorithm traverses the entire population, identifying individuals not dominated by any other individual and classifying them into the first-level non-dominated layer. For each individual in this layer, the algorithm identifies all individuals it dominates and updates the dominance counts of these individuals. If an individual is dominated by only one individual in the first-level non-dominated layer, its dominance count is set to one. These individuals are then assigned to the next non-dominated layer. This process is repeated until all individuals have been classified into their respective non-dominated layers.

[0068] Individual solutions in the Pareto optimal solution set are evaluated. Based on expert experience and pre-defined reasonable weights, the TOPSIS method is used for weighting and normalization. The final decision is made based on the distance from the ideal solution, including:

[0069] The target weights were determined using a comprehensive analysis method combined with expert experience.

[0070] By weighting and normalizing, the scheme decision is made based on the distance from the ideal solution, and the tunneling parameter adjustment and setting scheme is given.

[0071] The evaluation method is as follows:

[0072]

[0073] in, C represents the best ideal solution and the negative ideal solution within a set of optimized solutions with specific weights. i Let represent the performance score of the i-th individual in the population under subjective weighting. The value is between 0 and 1. The closer the performance score is to 1, the better the performance. Select the individual with the best performance score from the population as the result of this optimization.

[0074] The optimal solution is selected, and an adjustment and setting scheme for the tunnel boring machine's (TBM) tunneling parameters is generated to ensure an optimal balance between efficiency and safety. Finally, the parameter setting scheme derived from the optimal solution is sent to the TBM to achieve online control and optimization.

[0075] This method has been validated on the Haizhu Bay Tunnel project. The method was validated using actual engineering data from rings 390 to 690. The curves for each objective before and after optimization are shown below. Figure 4 As shown.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. That is, all simple and equivalent changes and modifications made based on the claims and description of this invention fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.

Claims

1. A data-driven multi-objective optimization decision-making method for tunneling parameters of a tunnel boring machine, characterized in that, include: Step S1: Construct a multi-objective evaluation system for the shield tunneling process, acquire shield machine operation data, and preprocess the data; Step S2: Based on the preprocessed data, divide the dataset; the dataset includes necessary data and auxiliary data; the necessary data includes the average tunneling speed of each ring, the average pitch angle, and the average roll angle at the last five moments; the auxiliary data includes: internal friction angle, cohesion, unit weight, tunnel depth, uniaxial compressive strength, total thrust, cutterhead torque, cutterhead speed, AF group cylinder thrust, three sets of working chamber pressure data, slurry discharge / inlet pump speed, equipment side roll, equipment tilt angle, two sets of main drive anti-torsion cylinder stroke data, and two sets of main drive anti-torsion cylinder pressure data, totaling twenty-five; both the necessary data and auxiliary data are obtained by averaging between rings; Step S3: Use the LightGBM model to construct the mapping relationship between geological parameters, tunneling parameters and tunneling performance targets, and form the objective function; Step S4: Evaluate and verify the objective function obtained in step S3; Step S5: Using a non-dominated sorting genetic algorithm, multi-objective optimization is performed within the given range of tunneling parameters and under the constraint of balancing the hydraulic cylinder thrust and the total thrust of the tunnel boring machine, generating a Pareto optimal solution set. There are two constraints: one is the parameter range constraint, representing the range constraint of the decision variables; the other is the balance constraint between the sum of the hydraulic cylinder thrust and the total thrust, described as follows: ;in, This represents the maximum allowable tunneling speed. These represent the maximum permissible deviations of the pitch and roll angles in engineering, respectively. For the first The lower bound of each decision variable. For the first The upper limit of a decision variable, It is the first The pressure of each hydraulic cylinder This represents the effective area of ​​the piston. Due to force transmission losses, an error band of α is allowed. Step S6: Evaluate individuals in the Pareto optimal population based on the TOPSIS method, and select the closest parameter combination based on subjective weighting.

2. The data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters according to claim 1, characterized in that, The method for preprocessing the tunnel boring machine (TBM) operating data in step S1 is to remove abnormal data. Abnormal data includes downtime data during TBM operation, data from the start of tunneling until reaching a steady state, and outliers. A 3D model is used. Criterion-based methods are used to eliminate outliers.

3. The data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters according to claim 1, characterized in that, The objective function constructed in step S3 is: ； ； ； in, , , These are tunneling speed, pitch angle, and roll angle, respectively. The expert features selected for constructing the LGBM model include five passive parameters and twenty active parameters: internal friction angle, cohesion, unit weight, tunnel burial depth, uniaxial compressive strength, total thrust, cutterhead torque, cutterhead speed, AF group cylinder thrust, three sets of working chamber pressure data, slurry discharge / inlet pump speed, equipment side roll, equipment tilt angle, two sets of main drive anti-torsion cylinder stroke data, and two sets of main drive anti-torsion cylinder pressure data. This represents the LGBM model.

4. The data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters according to claim 1, characterized in that, The objective function evaluation and verification steps in step S4 are as follows: Step S41: Use grid search to determine hyperparameters, including learning rate, maximum tree depth, minimum number of samples per leaf, minimum split gain, and maximum number of leaves. Step S42: After determining the hyperparameters of the model, the five-fold cross method is used for training and evaluation. Each time, 80% of the data is used as the training set and 20% of the data is used as the test set. The dataset is randomly split, and the training is performed on the training set and the test is performed on the test set. Step S43: Evaluate the prediction accuracy of the model by using root mean square error and goodness-of-fit indices to measure the accuracy of the model.

5. The data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters according to claim 1, characterized in that, The specific steps for multi-objective optimization using the non-dominated sorting genetic algorithm in step S5 are as follows: Step S51: Initialize the population, initialize the dominance count of each individual to zero, and create an empty set to record the individuals it dominates; Step S52: Use the LGBM model prediction function to evaluate the objective function value corresponding to the current decision variable; Step S53 involves iterating the population, maintaining population diversity through crowding, and using an elitist strategy to ensure the algorithm maintains a good solution, continuously optimizing to obtain the optimal solution set for the optimization problem.

6. The data-driven multi-objective optimization decision-making method for tunnel boring machine excavation parameters according to claim 1, characterized in that, Step 6 uses the TOPSIS method to evaluate the population in the Pareto solution. The specific steps include: Step S61: Use a comprehensive analysis method, combined with expert experience, to determine the target weights; Step S62: Through weighted normalization, a scheme decision is made based on the distance from the ideal solution, and a scheme for adjusting and setting tunneling parameters is given. Step S63, the evaluation method is as follows: ； in, , For the best ideal solution and the negative ideal solution in the optimal solution set with specific weights and optimization, The first in the population The performance degree of each individual under subjective weighting is between 0 and 1. The closer the performance degree is to 1, the better the performance. The individual with the best performance degree is selected from the population as the result of this optimization.

Images