Shield tunnel construction stability online multi-objective optimization method and system based on BO-autoformer-shap-nswoa

Abstract

The application discloses a shield tunnel construction stability online multi-objective optimization method and system based on BO-Autoformer-SHAP-NSWOA, which comprises the following steps: selecting important slurry shield construction parameters, geological condition parameters and tunnel burial depth as input variables, taking slurry pressure, pitch angle and roll angle as output variables, and establishing an original sample set; adopting BO optimization Autoformer regression model hyperparameters to construct a BO-Autoformer prediction model, predicting each output variable, and obtaining a corresponding regression prediction function; introducing a SHapley method to determine key regulation and control parameters affecting slurry shield tunnel construction stability; taking the prediction function as a fitness function, combining construction parameter constraint conditions and attitude control constraint conditions to construct a multi-objective optimization function; using an NSWOA algorithm to perform global optimization on each output variable of the optimization model, and determining a unique optimal slurry shield tunnel construction parameter combination through a TOPSIS method; and realizing high-precision prediction and collaborative optimization control of slurry pressure, pitch angle and roll angle.

Description

Technical Field

[0001] This invention belongs to the field of shield tunnel construction parameter optimization technology, and more specifically, relates to an online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA. Background Technology

[0002] With the rapid advancement of urbanization, the shortage of urban space resources and traffic congestion have become increasingly prominent problems, making underground space development a core means to alleviate these contradictions. In major projects such as urban tunnels and cross-river / sea tunnels, large-diameter slurry shield tunneling machines have become core equipment for large-section tunnel construction due to their advantages such as uniform slurry pressure transmission, high precision in excavation face balance control, low cutterhead wear, and high construction efficiency. However, large-diameter slurry shield tunneling faces severe challenges brought about by complex geological conditions: on the one hand, as the tunnel diameter increases, the probability of the excavation face encountering complex geological environments such as layered strata and alternating soft and hard strata increases significantly, leading to abrupt changes in the mechanical properties of the excavation face and greatly increasing the difficulty of tunnel stability control; on the other hand, the current control of key parameters in shield tunneling still relies on the experience and judgment of operators, resulting in problems such as response lag and insufficient control precision, which can easily lead to a series of engineering risks.

[0003] Specifically, slurry pressure, as a core parameter for maintaining the stability of the excavation face, directly determines the equilibrium state of the tunnel face with its control precision. Excessive or insufficient pressure can lead to accidents such as tunnel face collapse and surface subsidence. Simultaneously, the uneven soil forces acting on the tunnel boring machine (TBM) under complex geological conditions can cause deviations in pitch and roll angles. This not only alters the contact state between the TBM and the soil and reduces the soil's bearing capacity, but also leads to uneven segment settlement and bolt stress imbalance, ultimately affecting the overall stability of the tunnel structure. In existing technologies, traditional manual control methods are ill-suited to the dynamic changes under complex geological conditions, while existing prediction models often suffer from insufficient nonlinear fitting accuracy and poor model interpretability. Optimization algorithms are prone to getting trapped in local optima during multi-objective collaborative optimization, failing to achieve precise coordinated control of slurry pressure, pitch angle, and roll angle. Therefore, there is an urgent need to develop an online multi-objective optimization method that combines high-precision prediction capabilities, strong interpretability, and efficient optimization performance to ensure the safety and stability of large-diameter slurry shield tunnel construction. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA. It proposes a collaborative technical solution integrating the BO-Autoformer prediction model, SHAP interpretability analysis, and the NSWOA multi-objective optimization algorithm. By using Bayesian optimization to accurately optimize the hyperparameters of the Autoformer algorithm, a high-precision regression prediction model for slurry pressure, pitch angle, and roll angle is constructed. Simultaneously, the SHAP method is introduced to quantify the contribution and positive / negative nature of each construction parameter to the prediction results, thus solving the problem of... Traditional models suffer from the "black box" problem. This paper uses the regression function of the prediction model as the fitness function, sets constraints based on engineering specifications and actual needs, and uses the NSWOA algorithm to achieve global multi-objective optimization and obtain the Pareto optimal solution set. The TOPSIS method is used to objectively select the unique optimal combination of construction parameters for large-diameter slurry shield tunnels. Furthermore, an online update mechanism is adopted to replace historical actual values ​​with the current time step optimization results, enabling dynamic optimization and adjustment of the construction process. This accurately maps the complex nonlinear relationship between construction parameters and tunnel stability control objectives, significantly improving the accuracy, efficiency, and engineering applicability of stability control in large-diameter slurry shield tunnel construction.

[0005] To achieve the above objectives, one aspect of the present invention provides an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, comprising the following steps: S1: Select important slurry shield tunneling construction parameters, geological condition parameters, and tunnel burial depth as input variables, collect data on stability control targets and regulation parameters of slurry shield tunnels as output variables, and establish an original sample set; S2: Bayesian optimization was used to optimize the hyperparameters of the Autoformer regression model. The optimal parameter combination was determined by combining 5-fold cross-validation. A BO-Autoformer prediction model was constructed to predict slurry pressure, pitch angle, and roll angle, and three regression prediction functions were obtained. The Shapley additive interpretation method was introduced to calculate the Shapley value of each input feature variable for each regression prediction result. The contribution of each input feature variable to slurry pressure, pitch angle, and roll angle and the positive or negative contribution were analyzed to identify the key control parameters affecting the stability of slurry shield tunnel construction. S3: The three regression prediction functions obtained in step S2 are used as fitness functions, and a multi-objective optimization function is constructed by combining construction parameter constraints and attitude control constraints; the multi-objective whale optimization algorithm based on non-dominated sorting, NSWOA, is used to globally optimize the slurry pressure, pitch angle and roll angle of the multi-objective optimization model, and the unique optimal combination of slurry shield tunneling construction parameters is determined from the obtained Pareto solution set by the TOPSIS method. S4: Using an online update mechanism, the optimization result of the current time step is used to replace the historical actual value of the corresponding output variable as the input data for the next time step optimization. Steps S2 to S3 are repeated to achieve dynamic online optimization of tunnel construction stability.

[0006] Furthermore, the important slurry shield tunneling construction parameters, geological condition parameters, and tunnel burial depth parameters mentioned in step S1 are used as input variables, including slurry flow rate, slurry density, discharge flow rate, discharge density, slurry pressure difference between inlet and outlet, air cushion chamber pressure, slurry chamber level, grouting pressure, grouting volume, cutterhead torque, cutterhead rotation speed, cutterhead extrusion pressure, propulsion speed, penetration depth, total propulsion force, burial depth, and groundwater level; the output variables include slurry pressure, pitch angle, and roll angle.

[0007] Further, step S2 includes: The input feature variables and output variables in the original sample set are normalized to map all variables to the interval [-1, 1]. The Bayesian optimization algorithm was used to optimize the four hyperparameters of the Autoformer regression model: number of encoding layers, number of decoding layers, learning rate, and batch size. The parameter combination with the highest accuracy was determined as the optimal parameter for the Autoformer prediction model. The preprocessed sample set is divided into a training set and a test set. The hyperparameter optimization results are input into the model to build and train the BO-Autoformer prediction model. The accuracy of the BO-Autoformer prediction model is evaluated using the coefficient of determination, root mean square error, and mean absolute error. The Shapley additive interpretation method is introduced to calculate the Shapley value of each input feature variable on the prediction result of the output variable, quantify the relative contribution of each input feature variable to slurry pressure, pitch angle and roll angle, and thus determine the key influencing parameters.

[0008] Furthermore, the initial ranges of the four parameters—number of encoding layers, number of decoding layers, learning rate, and batch size—are [0,5], [0,5], [1×10⁻⁶], and [1×10⁻⁶], respectively. -5 1×10 -1 ] and [16,64].

[0009] Further, step S3 includes: Based on the three prediction functions of slurry pressure, pitch angle and roll angle output by the BO-Autoformer prediction model, a multi-objective optimization objective function with tunnel stability as the core is constructed. Based on engineering specifications and actual construction needs, the two types of constraint boundaries, namely construction parameter constraints and attitude control constraints, are clearly defined. Determine the target number, population size, crossover and mutation operator values, and optimization stopping criteria for the genetic algorithm; By simulating three predation behaviors of whales—surrounding prey, spiraling approach, and random search—and combining non-dominated sorting and crowding calculation, multi-objective collaborative optimization is achieved. The TOPSIS method is used to select a unique optimal solution from the Pareto optimal solution set as the optimal combination of slurry shield tunneling parameters.

[0010] Furthermore, the objective functions for the slurry pressure, pitch angle, and roll angle determined by the BO-Autoformer prediction regression equation in step S3 are: (6) In the formula: , and Let be the objective functions for slurry pressure, pitch angle, and roll angle, respectively. The values ​​represent the predicted mud pressure, and x1 to x15 are 15 adjustable construction parameters. The pitch angle, The roll angle, and These are the upper and lower limits of the slurry pressure, respectively, which can be calculated using empirical formulas (7) and (8): (7) (8) In the formula: For groundwater pressure, For static earth pressure, For variable earth pressure, For active earth pressure, The coefficient of earth pressure at rest. The active earth pressure coefficient, The weight of soil. The density of water, For cohesion, The depth of the soil at the top of the tunnel. The depth of the tunnel roof below the water level.

[0011] Furthermore, when controlling the attitude of the tunnel boring machine during tunnel boring construction, the allowable deviation limit for pitch angle and roll angle is ±30mm / m; 80% of the allowable deviation limit is selected as the warning value at the construction site, and correction adjustment is required when the deviation exceeds the warning value.

[0012] Furthermore, the method of simulating whale predation behavior and combining non-dominated ranking with crowding calculation to achieve multi-objective collaborative optimization includes the following steps: Several sets of construction parameter combinations that meet the constraints are randomly generated to form the initial population; Input each set of construction parameters into the three objective functions in step S3, and calculate the fitness values ​​corresponding to the slurry pressure deviation, pitch angle, and roll angle. The parameter combinations in the population are sorted according to the multi-objective fitness of mud pressure, pitch angle, and roll angle, and the "non-dominated solutions" are selected. Calculate the crowding distance of individuals within each frontier layer to measure the distribution density of non-dominated solutions in the solution space, preserve relatively sparse solutions, and ensure the diversity of multi-objective optimization. The NSWOA algorithm iteratively updates the position of the parameter combination to be optimized by simulating three predation behaviors of whales: surrounding prey, spiraling approach, and random search.

[0013] Furthermore, the mathematical model expression for surrounding the prey is: (10) (11) In the formula: This represents the current iteration number. Used to measure the distance between a whale and its prey. , All are whale position update coefficients. This represents the position vector of the whale with the best position in the current iteration. It is the current position vector of the whale; The mathematical model for spiral approximation is expressed as: (12) in, Indicates the distance between the whale and its prey. It is a constant used to determine the shape of the spiral. , All are random numbers with values ​​in the range [-1, 1]. The mathematical model for random search is as follows: (13) (14) In the formula: This represents the location of a whale in a randomly selected population.

[0014] The second aspect of this invention provides an online multi-objective optimization system for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, used to implement the aforementioned online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, comprising: Data acquisition module: used to select important slurry shield tunneling construction parameters, geological condition parameters and tunnel burial depth as input variables, collect stability control targets and regulation parameter data of slurry shield tunnels as output variables, and establish an original sample set; The module for predictive model construction and key control parameter determination is used to optimize the hyperparameters of the Autoformer algorithm using Bayesian optimization, determine the optimal parameter combination by combining 5-fold cross-validation, construct the BO-Autoformer prediction model, predict slurry pressure, pitch angle, and roll angle, and obtain three regression prediction functions. The Shapley additive interpretation method is introduced to calculate the Shapley value of each input feature variable for each regression prediction result, analyze the degree of contribution and positive or negative contribution of each input feature variable to slurry pressure, pitch angle, and roll angle, and identify the key control parameters affecting the construction stability of slurry shield tunnels. The NSWOA-based multi-objective optimization module is used to construct a multi-objective optimization function by using three regression prediction functions obtained from the prediction model construction and key control parameter determination module as fitness functions, and combining construction parameter constraints and attitude control constraints. The multi-objective whale optimization algorithm based on non-dominated sorting is used to globally optimize the slurry pressure, pitch angle and roll angle of the multi-objective optimization model. The unique optimal combination of slurry shield tunneling construction parameters is determined from the obtained Pareto solution set by the TOPSIS method. The online update module is used to replace the historical actual values ​​of the corresponding output variables with the optimization results of the current time step as the input data for the optimization of the next time step. This repeats the operations of the prediction model construction and key control parameter determination module and the NSWOA-based multi-objective optimization module to achieve dynamic online optimization of tunnel construction stability.

[0015] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) The online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA of the present invention has high prediction accuracy and strong model reliability: Bayesian optimization of BO is used to optimize the hyperparameters of the Autoformer algorithm, including the number of encoding layers, the number of decoding layers, the learning rate and the batch size. The BO-Autoformer prediction model is constructed by combining the 5-fold cross-validation method. The high-precision prediction of slurry pressure, pitch angle and roll angle is achieved by verifying the triple index of coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE). The R² of the training set and the test set are close to 0.92 and above, and the RMSE and MAE values ​​are significantly reduced, which can ensure the accuracy and reliability of the prediction results and provide solid data support for subsequent optimization.

[0016] (2) The online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA of the present invention has strong model interpretability and clear decision basis: By introducing the Shapley additive interpretation method to perform interpretive analysis on the BO-Autoformer prediction model, by calculating the Shapley value of each input feature variable, the contribution and positive or negative of 17 input parameters to the prediction results of slurry pressure, pitch angle and roll angle are quantified, and the key construction parameters affecting the stability of tunnel construction are clarified. It can solve the "black box" problem of traditional deep learning models and provide a clear decision basis for the precise control of shield construction parameters.

[0017] (3) The online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA of the present invention has high optimization efficiency and excellent global optimization capability: it adopts the multi-objective whale optimization algorithm (NSWOA) based on non-dominated sorting for multi-objective optimization, and determines the optimal solution that makes the multi-objective function optimal from the Pareto front solution set obtained by the NSWOA algorithm. It comprehensively considers the state of the three objective functions, the acquisition method is simple, and the results are relatively objective; the algorithm integrates non-dominated sorting and congestion calculation, and has the advantages of high optimization accuracy, strong robustness and fast convergence speed. It can effectively jump out of local optima and quickly search for the Pareto optimal solution set; combined with the TOPSIS method, it objectively selects the unique optimal combination of slurry shield tunnel construction parameters from the Pareto solution set, which can avoid the interference of human subjective preferences and ensure the scientific nature and engineering applicability of the optimization results.

[0018] (4) The online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA of the present invention replaces the historical actual value of the corresponding output variable with the optimization result of the current time step as the input data for the optimization of the next time step through the online update mechanism. This realizes the dynamic adjustment of construction parameters and can respond in real time to changes in geological conditions and dynamic fluctuations during construction. It solves the defect that traditional optimization methods cannot adapt to complex working conditions in real time and significantly improves the stability and safety of large-diameter slurry shield tunnel construction.

[0019] (5) The online multi-objective optimization method and system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA of the present invention uses the BO-Autoformer regression prediction function as the fitness function of the NSWOA algorithm, which accurately maps the complex nonlinear relationship between shield construction parameters and slurry pressure, pitch angle and roll angle, replacing the limitations of traditional mathematical formulas; at the same time, based on the constraints set by GB50446-2017 Code for Construction and Acceptance of Shield Tunnel, the optimization results meet the actual requirements of the project, effectively reducing the risks of face collapse, segment settlement and uneven bolt stress, and ensuring the long-term stability of the tunnel structure. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 2 This is a search result diagram of the slurry pressure hyperparameter for an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 3 This is a graph showing the pitch angle hyperparameter search results for an online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 4 This is a search result diagram of the roll angle hyperparameter of an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the slurry pressure prediction results of an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the pitch angle prediction results of an online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA according to an embodiment of the present invention. Figure 7 This is a schematic diagram of the roll angle prediction results of an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA according to an embodiment of the present invention. Figure 8This is a schematic diagram illustrating the explanatory analysis of the slurry pressure prediction model of an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 9 This is a schematic diagram illustrating the explanatory analysis of the pitch angle prediction model of an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 10 This is a schematic diagram illustrating the interpretive analysis of the roll angle prediction model of an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Figure 11 This is a schematic diagram of the Pareto front and optimal solution obtained by NSWOA optimization in an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA according to an embodiment of the present invention; Figure 12 This is a schematic diagram of an online multi-objective optimization system for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] like Figure 1 As shown, one aspect of the present invention provides an online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, comprising the following steps: S1: Select important slurry shield tunneling construction parameters, geological condition parameters, and tunnel burial depth as input variables, collect data on stability control targets and regulation parameters of slurry shield tunnels as output variables, and establish an original sample set; S2: Bayesian optimization (BO) is used to optimize the hyperparameters of the Autoformer algorithm. The optimal parameter combination is determined by combining 5-fold cross-validation. The BO-Autoformer prediction model is constructed to predict slurry pressure, pitch angle, and roll angle, and three regression prediction functions are obtained. The Shapley Additive Explanation (SHAP) method is introduced to calculate the Shapley value of each input feature variable for each regression prediction result. The contribution of each input feature variable to slurry pressure, pitch angle, and roll angle and the positive or negative contribution are analyzed to identify the key control parameters affecting the stability of slurry shield tunnel construction. S3: Using the three regression prediction functions obtained in step S2 as fitness functions, a multi-objective optimization function is constructed by combining construction parameter constraints and attitude control constraints; the non-dominated sorting whale optimization algorithm (NSWOA) is used to globally optimize the slurry pressure, pitch angle, and roll angle of the multi-objective optimization model, and the unique optimal combination of slurry shield tunneling construction parameters is determined from the obtained Pareto solution set using the TOPSIS method; S4: Using an online update mechanism, the optimization result of the current time step is used to replace the historical actual value of the corresponding output variable as the input data for the next time step optimization. Steps S2 to S3 are repeated to achieve dynamic online optimization of tunnel construction stability.

[0023] Furthermore, the important slurry shield tunneling construction parameters, geological condition parameters, and tunnel burial depth parameters mentioned in step S1 include 17 factors: slurry flow rate, slurry density, discharge flow rate, discharge density, slurry pressure difference between inlet and outlet, air cushion chamber pressure, slurry chamber level, grouting pressure, grouting volume, cutterhead torque, cutterhead rotation speed, cutterhead extrusion pressure, propulsion speed, penetration depth, total propulsion force, burial depth, and groundwater level. These 17 factors are used as input characteristic variables, and slurry pressure, pitch angle, and roll angle are used as output variables.

[0024] Further, step S2 includes: (1) Data preprocessing: In a specific embodiment of the present invention, 39,762 sets of sample data from the construction site were collected, as shown in Table 1. Due to the different dimensions of the selected input variables, in order to prevent the data from being overwhelmed or failing to converge due to excessively large or small data, it is necessary to preprocess the sample data. In the embodiment of the present invention, the input and output variables are normalized to the interval [-1, 1]; 80% of the samples were randomly selected from all samples to form the training set for training the model. To test the generalization performance of the model, the remaining 20% ​​of samples were used as the test set to verify the model's effectiveness.

[0025] Table 1 Sample Data

[0026] (2) Optimization of hyperparameters of Autoformer prediction model: In order to achieve better prediction results for Autoformer regression model, this invention uses BO to optimize the four parameters of the Autoformer regression model: number of encoding layers, number of decoding layers, learning rate and batch size. The model accuracy is verified by combining the 5-fold cross-validation method, so as to determine the parameter combination with the highest accuracy as the optimal parameters of Autoformer prediction model. In a specific embodiment of the present invention, before optimizing the parameters, an initial range for hyperparameter optimization is given. The initial ranges for the four parameters—number of encoding layers, number of decoding layers, learning rate, and batch size—are [0,5], [0,5], [1×10], respectively. -5 1×10 -1 [16,64]; the hyperparameter search results for slurry pressure, pitch angle, and roll angle are as follows: Figure 2 , Figure 3 and Figure 4 As shown; from Figure 2 As can be seen from the data, the Autoformer prediction model for slurry pressure performs best when the number of encoding layers is 2, the number of decoding layers is 4, the learning rate is 0.036, and the batch size is 32. Figure 3 As can be seen from the data, the Autoformer prediction model for pitch angle performs best when the number of encoding layers is 4, the number of decoding layers is 2, the learning rate is 0.044, and the batch size is 32. Figure 4 As can be seen from the data, the Autoformer prediction model for the rolling angle performs best when the number of encoding layers is 2, the number of decoding layers is 3, the learning rate is 0.067, and the batch size is 16. (3) Model training and validation: The preprocessed sample set is divided into a training set and a test set in a ratio (e.g., 4:1). The training set is used to train the model and test the generalization performance of the model. The test set is used to verify the model effect. The BO-Autoformer prediction model is built and trained based on the Python environment. In a specific embodiment of the present invention, learning simulation is performed using a training set based on the parameter optimization results, and verification is performed using a test set. Autoformer prediction models for slurry pressure, pitch angle, and roll angle are established respectively, and the prediction results are as follows: Figure 5 , Figure 6 and Figure 7 As shown; This invention introduces the absolute coefficient (R) 2 The accuracy of the BO-Autoformer prediction model is evaluated using three commonly used metrics: root mean square error (RMSE), and mean absolute error (MAE). 2 Used to measure the goodness of fit between predicted and observed values, RMSE and MAE can effectively reflect the deviation between predicted and observed values. 2 The closer the value is to 1, the higher the model's prediction accuracy; the smaller the RMSE and MAE values, the better the model's performance. (1) (2) (3) In the formula, For the sample size, For predicted values, These are actual observed values. This is the average of the actual observed values; Depend on Figure 5 It can be seen that the BO-Autoformer model can predict changes in slurry pressure very well. The coefficient of determination R0 for the BO-Autoformer model on the training set for predicting slurry pressure is... 2 The root mean square error (RMSE) was 0.942, the mean absolute error (MAE) was 0.034, and the coefficient of determination (R²) for the test set was 0.029. 2 The mean square error (RMSE) was 0.928, the root mean square error (RMSE) was 0.008, and the mean absolute error (MAE) was 0.006, indicating that the model has fully learned the relationship between the input and output variables of the sample set and has a good fitting effect. from Figure 6 It can be observed that the BO-Autoformer model can predict pitch angle changes very well. The coefficient of determination R0 for pitch angle prediction on the training set based on the BO-Autoformer model is... 2 The root mean square error (RMSE) is 0.935, the mean absolute error (MAE) is 0.731, and the coefficient of determination (R²) for the test set is 0.633. 2 The mean square error (RMSE) was 0.918, the root mean square error (RMSE) was 0.216, and the mean absolute error (MAE) was 0.186, indicating that the model has fully learned the relationship between the input and output variables of the sample set and has a good fitting effect. from Figure 7 It can be observed that the BO-Autoformer model can predict pitch angle changes very well. The coefficient of determination R0 for roll angle prediction on the training set based on the BO-Autoformer model is also shown. 2The root mean square error (RMSE) was 0.939, the mean absolute error (MAE) was 0.109, and the coefficient of determination (R²) for the test set was 0.094. 2 The mean square error (RMSE) was 0.920, the root mean square error (RMSE) was 0.117, and the mean absolute error (MAE) was 0.099, indicating that the model has fully learned the relationship between the input and output variables of the sample set and has a good fitting effect. (4) Parameter Importance Analysis: This invention introduces the Shapley Additive Explanation (SHAP) method to interpret and analyze the prediction models for slurry pressure, pitch angle, and roll angle based on Autoformer: The Shapley value of each input feature variable for each regression prediction result is calculated, and the contribution degree and positive / negative nature of each input feature variable to slurry pressure, pitch angle, and roll angle are analyzed to clarify the key control parameters affecting the stability of slurry shield tunnel construction; the interpretation function of the SHAP method is expressed as: (4) In the formula, This represents the interpretation function. Indicates the number of input features. This indicates whether each feature can be observed. The Shapely value represents the virtual model without features. Indicates the first The Shapely value of each feature; the Shapely value contributed by each feature is calculated using the following formula: (5) In the formula, Represents the eigenvector. Indicates that features are not included. All feature subsets, for The number of non-zero elements in the neutron. It is the set of all features in the model; This invention utilizes the SHAP method to perform an overall analysis of the regression prediction results for each input feature variable, interpreting the degree of contribution and the positive or negative nature of each construction parameter to the existing tunnel deformation prediction results, thereby identifying the important construction parameters affecting the control objectives and providing a decision-making basis for shield tunneling construction management. In a specific embodiment of the present invention, the SHAP method is used to interpret and analyze the prediction models of BO-Autoformer for slurry pressure, pitch angle, and roll angle, and the effects of the input parameters on the prediction results of slurry pressure, pitch angle, and roll angle are obtained as follows: Figure 8 , Figure 9 and Figure 10 As shown; Depend on Figure 8 It can be seen that the order of influence of the 17 input parameters on the slurry pressure is as follows: air cushion chamber pressure, slurry chamber level, slurry inlet flow rate, slurry inlet density, slurry outlet flow rate, slurry outlet density, pressure difference between inlet and outlet slurry, total propulsion force, propulsion speed, cutterhead rotation speed, cutterhead extrusion pressure, penetration depth, cutterhead torque, grouting pressure, grouting volume, burial depth, and groundwater level. Depend on Figure 9 It can be seen that the order of influence of the 17 input parameters on the pitch angle is as follows: total thrust, air cushion chamber pressure, mud chamber liquid level, slurry density, slurry discharge density, slurry flow rate, slurry discharge flow rate, cutterhead extrusion pressure, propulsion speed, cutterhead torque, cutterhead rotation speed, grouting pressure, slurry pressure difference between inlet and outlet, penetration, grouting volume, burial depth, and groundwater level. Depend on Figure 10 It can be seen that the order of influence of the 17 input parameters on the roll angle is as follows: total propulsion force, air cushion chamber pressure, slurry chamber level, slurry density, slurry flow rate, discharge density, discharge flow rate, propulsion speed, cutterhead torque, cutterhead extrusion pressure, cutterhead rotation speed, grouting pressure, penetration, pressure difference between slurry inlet and outlet, grouting volume, burial depth, and groundwater level. Further, step S3 includes: (1) Determine the objective function: According to the SHAP analysis results, the shield tunneling parameters have a greater impact on slurry pressure, pitch angle, and roll angle than on burial depth and groundwater level, and these two parameters cannot be adjusted. Therefore, this invention chooses to continuously adjust the shield tunneling parameters x1 to x15 to achieve stability control of the shield tunnel. Specifically, based on the interpretability analysis of SHAP, the parameters that need to be adjusted are determined, including slurry flow rate x1, slurry density x2, slurry discharge flow rate x3, slurry discharge density x4, slurry pressure difference between slurry inlet and outlet x5, air cushion chamber pressure x6, slurry chamber level x7, grouting pressure x8, grouting volume x9, cutterhead torque x10, cutterhead rotation speed x11, cutterhead extrusion pressure x12, propulsion speed x13, penetration depth x14, and total propulsion force x15. This invention optimizes the design using slurry pressure, pitch angle, and roll angle as decision objectives. BO-Autoformer regression prediction can fit the complex nonlinear relationship between shield tunneling parameters and slurry pressure, pitch angle, and roll angle. A trained regression function is introduced as the fitness function for optimization, replacing the traditional mathematical relationship as the fitness function in the multi-objective genetic algorithm. Excessive or insufficient slurry pressure can lead to tunnel face collapse; therefore, the slurry pressure needs to be controlled between upper and lower limits. The pitch angle and roll angle are minimized. This invention constructs a multi-objective optimization objective function with tunnel stability as the core, based on the three regression prediction functions of slurry pressure, pitch angle, and roll angle output by the BO-Autoformer prediction model, clarifying the optimization direction of each objective. The objective functions for slurry pressure, pitch angle, and roll angle determined by the BO-Autoformer prediction regression equation are as follows: (6) In the formula: , and Let be the objective functions for slurry pressure, pitch angle, and roll angle, respectively. The values ​​represent the predicted mud pressure, and x1 to x15 are 15 adjustable construction parameters. The pitch angle, The roll angle, and These are the upper and lower limits of the slurry pressure, respectively, which can be calculated using empirical formulas (7) and (8): (7) (8) In the formula: For groundwater pressure, For static earth pressure, For variable earth pressure, For active earth pressure, The coefficient of earth pressure at rest. The active earth pressure coefficient, The weight of soil. The density of water, For cohesion, The depth of the soil at the top of the tunnel. The depth of the tunnel roof below the water level; (2) Determine the range of constraints: By combining engineering specifications and actual construction needs, two types of constraint boundaries—construction parameter constraints and attitude control constraints—are clearly defined to ensure the feasibility and safety of the optimization results. Construction parameter constraints: To ensure the practical significance of optimized construction parameter design and to guarantee that the resulting parameter settings are more reasonable, feasible, and free from safety hazards, constraints need to be set on the values ​​of decision parameters based on project requirements and existing project data. The range of values ​​for each adjustable parameter x1 to x15 is limited to the minimum and maximum values ​​of the measured data. These constraints are generally expressed as follows: (9) In the formula, and These represent the lower and upper limits of the values ​​that the operation parameter can take, respectively. In the embodiments of the present invention, the constraint range of each operating parameter is set according to the actual engineering situation. Since the project does not have specific requirements for the values ​​of the operating parameters, the present invention sets the constraint range of the parameters based on the existing data of the project, and uses the maximum and minimum values ​​of the measured data as the upper and lower limits of the values ​​of each construction parameter, respectively. The specific constraint conditions of each construction parameter are shown in Table 2. Table 2 Input Parameter Range Table

[0027] Attitude control constraints: According to GB50446-2017 Code for Construction and Acceptance of Shield Tunneling, when controlling the attitude of the shield machine during shield construction, the allowable deviation limit for pitch angle and roll angle is ±30mm / m; 80% of the allowable deviation limit is selected as the warning value at the construction site, that is, the pitch angle and roll angle need to be controlled within ±25mm / m, and correction adjustment is required when the deviation exceeds the warning value; (3) Initialize NSWOA algorithm parameters Before performing NSWOA multi-objective optimization, it is necessary to first determine the number of objectives, population size, crossover and mutation operator values, and optimization stopping criteria for the genetic algorithm. Considering that an appropriate population size and number of iterations can promote the convergence of multi-objective optimization, in a specific embodiment of this invention, the optimization efficiency and global search capability are ensured by configuring the core parameters of the algorithm. Population size: 100 (to balance optimization diversity and computational efficiency). Iteration parameters: maximum number of generations = 60, stopping number of generations = 60 (to avoid over-iteration or premature convergence). Operation operators: crossover operator = 0.7, mutation operator = 0.01 (which can maintain population diversity and reduce the risk of local optima traps). Algorithm coefficients: Initialize whale position update coefficients A and B; A takes values ​​in the range [-2, 2], B is a random number in the range [0, 1], and the spiral shape constant c = 1 (used to control the spiral approximation trajectory characteristics). (4) Perform global multi-objective optimization based on NSWOA This invention achieves multi-objective collaborative optimization by simulating whale hunting behavior (surrounding prey, spiral approach, random search) and combining non-dominated sorting with crowding calculation. Specifically, it includes the following steps: Population initialization: 100 sets of construction parameter combinations that meet the constraints (i.e., whale individual position vectors) are randomly generated to form the initial population; Fitness calculation: Input each set of construction parameters into the three objective functions in step S3, and calculate the fitness values ​​corresponding to the slurry pressure deviation, pitch angle, and roll angle; Non-dominated sorting: The parameter combinations in the population are sorted according to the multi-objective fitness of mud pressure, pitch angle, and roll angle, and the "non-dominated solutions" are selected. Crowding degree calculation: Calculate the crowding degree distance of individuals in each front layer, measure the distribution density of non-dominated solutions in the solution space, retain relatively sparse solutions, and ensure the diversity of multi-objective optimization; Position Update: The NSWOA algorithm iteratively updates the position of the parameter combination to be optimized by simulating three predatory behaviors of whales: surrounding prey, spiraling approach, and random search. The logic for surrounding prey is as follows: the optimal solution in the current iteration is considered the "prey," and the whale approaches the prey's position. The whale updates its own position based on the following mathematical model: (10) (11) In the formula: This represents the current iteration number. Used to measure the distance between a whale and its prey. , All are whale position update coefficients. This represents the position vector of the whale with the best position in the current iteration (i.e., the optimal parameter combination). It is the current position vector of the whale (i.e., the current parameter combination); The logic of spiral approach is as follows: as the whale swims, it gradually approaches its prey in a spiral trajectory, while releasing bubble nets to drive the prey away; the mathematical model of this behavior in the algorithm is expressed as follows: (12) in, Indicates the distance between the whale and its prey. It is a constant used to determine the shape of the spiral. , All values ​​are random numbers in the range [-1, 1]. In actual hunting scenarios, whales surround and spiral towards their prey simultaneously. Usually, the probability of each of these two behaviors is set to 0.5, that is, one of the behaviors is randomly selected to update the position. The logic of random search is as follows: when it is difficult to find a better solution in the current area (the whale is finding it difficult to find prey in its current area), a random search mechanism is activated to update the area (i.e., to try to find other potential prey locations) to avoid getting trapped in local optima; in the algorithm, the mathematical model corresponding to this behavior is as follows: (13) (14) In the formula: This represents the position of a whale in a randomly selected population (i.e., a random combination of parameters). Through this random search method, the algorithm can effectively escape the limitations of local optima, expand the search range, and increase the possibility of finding the global optimum. The NSWOA algorithm of this invention, based on the above-mentioned single-objective whale algorithm, further incorporates two key steps: non-dominated sorting and crowding calculation, thereby greatly improving the performance of the algorithm in handling multi-objective optimization problems. (5) Obtain the optimal solution This invention takes minimizing the difference between slurry pressure and its upper and lower limit average values, pitch angle, and roll angle as optimization objectives. It uses the Non-Dominated Sorting Whale Optimization Algorithm (NSWOA) based on non-dominated sorting to solve the intelligent optimization problem of shield tunneling operation parameters, and then determines the optimal solution of the combination of operation parameters. After setting the parameters, run the NSWOA algorithm in Python to obtain the Pareto solution set as follows: Figure 11 As shown in the figure, there are 73 Pareto optimal solutions. Based on the multi-objective optimization results, multiple shield tunneling parameter optimization schemes can be obtained using the NSWOA algorithm. However, in actual engineering, only one optimization scheme is needed to guide on-site construction, i.e., a unique optimal solution needs to be obtained. In order to achieve the best optimization effect, an optimal scheme needs to be selected from a large number of Pareto solutions. This invention uses the TOPSIS method to select the optimal solution from the Pareto solution set. The core is to score each scheme based on the distance between the ideal solution and the negative ideal solution. The score of each scheme is calculated using formula (15): (15) In the formula, For the first The fractions corresponding to each Pareto solution. For the first The distance from a Pareto solution to a negative ideal solution. For the first The distance from each Pareto solution to the ideal solution; and The calculation formulas are shown in (15) and (16): (16) (17) In the formula, To optimize the number of targets, In the first Under the first goal A Pareto solution, For the first The average value under each objective and In the first Maximum and minimum values ​​under each objective; and Calculate using formulas such as (17) and (18): (17) (18); Furthermore, step S4 includes: online update Because the developed deep learning model requires spatial and temporal features as input, the control results will affect the prediction of the shield tunnel stability at the next time step. Therefore, online updates are needed to simulate this situation. In particular, historical values ​​of slurry pressure, pitch angle, and roll angle should be replaced with optimized results, not the original values. That is, after obtaining the time step... After obtaining the optimization results, before optimizing the next time step, it is necessary to set the respective objectives in... The actual value at each point is replaced with the optimized result. This process is repeated for each step forward; specifically: Set the current time step The optimized results for the corresponding slurry pressure, pitch angle, and roll angle replace the previous time step. The corresponding actual values ​​of slurry pressure, pitch angle, and roll angle are used as the next time step. The optimized initial state enables dynamic online optimization of tunnel construction stability.

[0028] This invention utilizes the BO-Autoformer-NSWOA method to effectively optimize the difference between slurry pressure and its upper and lower bound averages, pitch angle, and roll angle. The proposed method effectively constructs the Pareto front for all cases and generates the optimal solution. Compared to the average absolute values ​​of the original data, the difference between slurry pressure and its upper and lower bound averages, pitch angle, and roll angle are reduced by an average of 22.72%, 20.77%, and 18.41%, respectively, with an overall average improvement of 35.87%. The optimal solution reduces the difference between slurry pressure and its upper and lower bound averages, pitch angle, and roll angle by an average of 26.90%, 24.45%, and 21.22%, respectively, with an overall average improvement of 42.09%. Case studies verify the applicability of the multi-objective optimization results. The results show that the solution based on the BO-Autoformer-NSWOA framework can simultaneously achieve the triple objective optimization of slurry pressure, pitch angle, and roll angle.

[0029] like Figure 12 As shown, a second aspect of the present invention provides an online multi-objective optimization system for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, used to implement the above-mentioned design method, comprising: Data acquisition module: used to select important slurry shield tunneling construction parameters, geological condition parameters and tunnel burial depth as input variables, collect stability control targets and regulation parameter data of slurry shield tunnels as output variables, and establish an original sample set; The module for predictive model construction and key control parameter determination is used to optimize the hyperparameters of the Autoformer algorithm using Bayesian optimization, determine the optimal parameter combination by combining 5-fold cross-validation, construct the BO-Autoformer prediction model, predict slurry pressure, pitch angle, and roll angle, and obtain three regression prediction functions. The Shapley additive interpretation method is introduced to calculate the Shapley value of each input feature variable for each regression prediction result, analyze the degree of contribution and positive or negative contribution of each input feature variable to slurry pressure, pitch angle, and roll angle, and identify the key control parameters affecting the construction stability of slurry shield tunnels. The NSWOA-based multi-objective optimization module is used to construct a multi-objective optimization function by using three regression prediction functions obtained from the prediction model construction and key control parameter determination module as fitness functions, and combining construction parameter constraints and attitude control constraints. The multi-objective whale optimization algorithm based on non-dominated sorting is used to globally optimize the slurry pressure, pitch angle and roll angle of the multi-objective optimization model. The unique optimal combination of slurry shield tunneling construction parameters is determined from the obtained Pareto solution set by the TOPSIS method. The online update module is used to replace the historical actual values ​​of the corresponding output variables with the optimization results of the current time step as the input data for the optimization of the next time step. This repeats the operations of the prediction model construction and key control parameter determination module and the NSWOA-based multi-objective optimization module to achieve dynamic online optimization of tunnel construction stability.

[0030] It should be noted that the online multi-objective optimization system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA provided in this embodiment can be a computer program (including program code) running on a computer device. For example, the online multi-objective optimization system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA is an application software. The online multi-objective optimization system for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA can be used to execute the corresponding steps in the above-described method provided in the embodiments of this application.

[0031] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement... Figure 1 The methods provided in each step are detailed in the implementation methods provided in the above steps, and will not be repeated here.

[0032] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for online multi-objective optimization of shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA, characterized in that, Includes the following steps: S1: Select important slurry shield tunneling construction parameters, geological condition parameters, and tunnel burial depth as input variables, collect data on stability control targets and regulation parameters of slurry shield tunnels as output variables, and establish an original sample set; S2: Bayesian optimization is used to optimize the hyperparameters of the Autoformer regression model. The optimal parameter combination is determined by combining 5-fold cross-validation. A BO-Autoformer prediction model is constructed to predict slurry pressure, pitch angle and roll angle, and three regression prediction functions are obtained. The SHapley additive interpretation method is introduced to clarify the key control parameters that affect the stability of slurry shield tunnel construction. S3: The three regression prediction functions obtained in step S2 are used as fitness functions, and a multi-objective optimization function is constructed by combining construction parameter constraints and attitude control constraints; the multi-objective whale optimization algorithm based on non-dominated sorting, NSWOA, is used to globally optimize the slurry pressure, pitch angle and roll angle of the multi-objective optimization model, and the unique optimal combination of slurry shield tunneling construction parameters is determined from the obtained Pareto solution set by the TOPSIS method. S4: Using an online update mechanism, the optimization result of the current time step is used to replace the historical actual value of the corresponding output variable as the input data for the next time step optimization. Steps S2 to S3 are repeated to achieve dynamic online optimization of tunnel construction stability.

2. The online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in claim 1, characterized in that: The important slurry shield tunneling parameters, geological condition parameters, and tunnel burial depth parameters mentioned in step S1 are used as input variables, including slurry flow rate, slurry density, discharge flow rate, discharge density, slurry pressure difference between inlet and outlet, air cushion chamber pressure, slurry chamber level, grouting pressure, grouting volume, cutterhead torque, cutterhead rotation speed, cutterhead extrusion pressure, propulsion speed, penetration depth, total propulsion force, burial depth, and groundwater level; the output variables include slurry pressure, pitch angle, and roll angle.

3. The online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in claim 2, characterized in that: Step S2 includes: The input feature variables and output variables in the original sample set are normalized to map all variables to the interval [-1, 1]. The Bayesian optimization algorithm was used to optimize the four hyperparameters of the Autoformer regression model: number of encoding layers, number of decoding layers, learning rate, and batch size. The parameter combination with the highest accuracy was determined as the optimal parameter for the Autoformer prediction model. The preprocessed sample set is divided into a training set and a test set. The hyperparameter optimization results are input into the model to build and train the BO-Autoformer prediction model. The accuracy of the BO-Autoformer prediction model is evaluated using the coefficient of determination, root mean square error, and mean absolute error. The Shapley additive interpretation method is introduced to calculate the Shapley value of each input feature variable on the prediction result of the output variable, quantify the relative contribution of each input feature variable to slurry pressure, pitch angle and roll angle, and thus determine the key influencing parameters.

4. The online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in claim 3, characterized in that, The initial ranges for the four parameters—number of encoding layers, number of decoding layers, learning rate, and batch size—are [0,5], [0,5], [1×10⁻⁶], and [1×10⁻⁶], respectively. -5 1×10 -1 [16,64].

5. A method for online multi-objective optimization of shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA according to any one of claims 1-4, characterized in that, Step S3 includes: Based on the three prediction functions of slurry pressure, pitch angle and roll angle output by the BO-Autoformer prediction model, a multi-objective optimization objective function with tunnel stability as the core is constructed. Based on engineering specifications and actual construction needs, the two types of constraint boundaries, namely construction parameter constraints and attitude control constraints, are clearly defined. Determine the target number, population size, crossover and mutation operator values, and optimization stopping criteria for the genetic algorithm; By simulating three predation behaviors of whales—surrounding prey, spiraling approach, and random search—and combining non-dominated sorting and crowding calculation, multi-objective collaborative optimization is achieved. The TOPSIS method is used to select a unique optimal solution from the Pareto optimal solution set as the optimal combination of slurry shield tunneling parameters.

6. The online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in claim 5, characterized in that, In step S3, the objective functions for slurry pressure, pitch angle, and roll angle determined by the BO-Autoformer prediction regression equation are: （6） In the formula: , and Let be the objective functions for slurry pressure, pitch angle, and roll angle, respectively. The values ​​represent the predicted mud pressure, and x1 to x15 are 15 adjustable construction parameters. The pitch angle, The roll angle, and These are the upper and lower limits of the slurry pressure, respectively, which can be calculated using empirical formulas (7) and (8): （7） （8） In the formula: For groundwater pressure, For static earth pressure, For variable earth pressure, For active earth pressure, The coefficient of earth pressure at rest. The active earth pressure coefficient, The weight of soil. The density of water, For cohesion, The depth of the soil at the top of the tunnel. The depth of the tunnel roof below the water level.

7. The online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in claim 6, characterized in that, When controlling the attitude of the tunnel boring machine (TBM) during tunnel boring machine (TBM) construction, the allowable deviation limit for pitch angle and roll angle is ±30 mm / m. 80% of the allowable deviation limit is selected as the warning value at the construction site. If the deviation exceeds the warning value, correction adjustment is required.

8. The online multi-objective optimization method for shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in claim 5, characterized in that, The method of simulating whale foraging behavior and combining non-dominated ranking with crowding calculation to achieve multi-objective collaborative optimization includes the following steps: Several sets of construction parameter combinations that meet the constraints are randomly generated to form the initial population; Input each set of construction parameters into the three objective functions in step S3, and calculate the fitness values ​​corresponding to the slurry pressure deviation, pitch angle, and roll angle. The parameter combinations in the population are sorted according to the multi-objective fitness of mud pressure, pitch angle, and roll angle, and the "non-dominated solutions" are selected. Calculate the crowding distance of individuals within each frontier layer to measure the distribution density of non-dominated solutions in the solution space, preserve relatively sparse solutions, and ensure the diversity of multi-objective optimization. The NSWOA algorithm iteratively updates the position of the parameter combination to be optimized by simulating three predation behaviors of whales: surrounding prey, spiraling approach, and random search.

9. The online multi-objective optimization method for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA as described in claim 8, characterized in that, The mathematical model expression for surrounding the prey is: （10） （11） In the formula: This represents the current iteration number. Used to measure the distance between a whale and its prey. , All are whale position update coefficients. This represents the position vector of the whale with the optimal position in the current iteration. It is the current position vector of the whale; The mathematical model for spiral approximation is expressed as: （12） in, Indicates the distance between the whale and its prey. It is a constant used to determine the shape of the spiral. , All are random numbers with values ​​in the range [-1, 1]. The mathematical model for random search is as follows: （13） （14） In the formula: This represents the location of a whale in a randomly selected population.

10. An online multi-objective optimization system for the stability of shield tunnel construction based on BO-Autoformer-SHAP-NSWOA, characterized in that, The method for implementing the online multi-objective optimization of shield tunnel construction stability based on BO-Autoformer-SHAP-NSWOA as described in any one of claims 1-9 includes: Data acquisition module: used to select important slurry shield tunneling construction parameters, geological condition parameters and tunnel burial depth as input variables, collect stability control targets and regulation parameter data of slurry shield tunnels as output variables, and establish an original sample set; The module for predictive model construction and key control parameter determination is used to optimize the hyperparameters of the Autoformer algorithm using Bayesian optimization, determine the optimal parameter combination by combining 5-fold cross-validation, construct the BO-Autoformer prediction model, predict slurry pressure, pitch angle, and roll angle, and obtain three regression prediction functions. The Shapley additive interpretation method is introduced to calculate the Shapley value of each input feature variable for each regression prediction result, analyze the degree of contribution and positive or negative contribution of each input feature variable to slurry pressure, pitch angle, and roll angle, and identify the key control parameters affecting the construction stability of slurry shield tunnels. The NSWOA-based multi-objective optimization module is used to construct a multi-objective optimization function by using three regression prediction functions obtained from the prediction model construction and key control parameter determination module as fitness functions, and combining construction parameter constraints and attitude control constraints. The multi-objective whale optimization algorithm based on non-dominated sorting is used to globally optimize the slurry pressure, pitch angle and roll angle of the multi-objective optimization model. The unique optimal combination of slurry shield tunneling construction parameters is determined from the obtained Pareto solution set by the TOPSIS method. The online update module is used to replace the historical actual values ​​of the corresponding output variables with the optimization results of the current time step as the input data for the optimization of the next time step. This repeats the operations of the prediction model construction and key control parameter determination module and the NSWOA-based multi-objective optimization module to achieve dynamic online optimization of tunnel construction stability.

Images