A large-diameter slurry shield safe stability multi-objective online optimization method and system based on GAT-LSTM-MOMPA

Abstract

The application provides a large-diameter slurry shield safety stability multi-objective online optimization method and system based on GAT-LSTM-MOMPA, and belongs to the technical field of shield tunnel construction parameter multi-objective online optimization. The method comprises the following steps: obtaining monitoring data of shield construction parameters and preprocessing, and establishing an original sample set; adopting a method combining knowledge driving and correlation analysis, input parameters and output targets of a prediction model are determined; a space-time deep learning network framework GAT-LSTM is built, the mapping relationship between the input and the output is determined, and accurate prediction of the future time step optimization value is realized; through connection with MOMPA and TOPSIS, the historical value of the model is constantly updated, and accurate and effective real-time online optimization is achieved. The GAT-LSTM-MOMPA multi-objective online optimization model established by the application effectively improves the safety and stability of the shield machine operation.

Description

A Multi-Objective Online Optimization Method and System for Safety and Stability of Large-Diameter Slurry Shield Tunneling Machines Based on GAT-LSTM-MOMPA Technical Field

[0001] This invention relates to the field of multi-objective online optimization technology for shield tunnel construction parameters, and more specifically, to a multi-objective online optimization method and system for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA. Background Technology

[0002] Urban development has led to a continuous increase in the scale of underground transportation, promoting the development of tunnels towards larger diameters and longer distances. Large-diameter slurry shield tunneling machines (TBMs), thanks to their efficient tunneling capabilities, have become the mainstream choice for large-diameter underground tunnel construction. During the excavation of large-diameter tunnels, the safety and stability control of the TBM is more challenging than that of traditional small-diameter TBMs. Overturning moment, roll angle, and pitch angle are key indicators reflecting the overall stability of large-diameter slurry shield tunnels. Due to the complexity of their working environment, many uncontrollable factors exist during the tunneling process. When operating parameters are unstable, major safety accidents such as TBM overturning can easily occur, causing casualties. Currently, TBM operation mainly relies on manual experience for adjustment, resulting in large fluctuations in overturning moment, pitch angle, and roll angle, which easily leads to safety accidents and poses a significant threat to the personal safety of construction workers. Therefore, achieving multi-system intelligent coordinated control of large-diameter slurry shield tunnels is imperative, thereby realizing intelligent and unmanned operation of the TBM, which will be the future development trend of TBMs. Summary of the Invention

[0003] To address the aforementioned shortcomings or improvement needs of existing methods, this invention provides a multi-objective online optimization method and system for the safety and stability of large-diameter slurry shield tunneling machines based on GAT-LSTM-MOMPA (Graph Attention Network-Long Short-Term Memory Network-Multi-Objective Marine Predator Algorithm). First, an index system influencing the overturning moment, pitch angle, and roll angle of the large-diameter slurry shield tunneling machine is constructed based on a dual-drive approach of knowledge data. Then, a high-performance GAT-LSTM model is developed based on this system. Finally, this model is used as the objective function for improving the MOMPA algorithm, and the improved MOMPA online multi-objective optimization model is used to optimize the safety and stability objectives of the large-diameter slurry shield tunneling machine.

[0004] The present invention achieves the above-mentioned technical objectives through the following technical solutions.

[0005] A multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA:

[0006] Acquire and preprocess monitoring data of shield tunneling construction parameters to establish an original sample set;

[0007] In the original sample set, the influencing indicators affecting the safety and reliability of the tunnel boring machine were initially screened. The heat map generated by Pearson correlation coefficient was used to further screen the influencing indicators, resulting in a set of key influencing indicators.

[0008] A spatiotemporal deep learning network framework, GAT-LSTM, was built as the prediction model. The spatial and temporal features of the key influencing indicator set were used as the input of the prediction model, and the overturning moment, pitch angle, and roll angle were used as the output of the prediction model. The mapping relationship between the input and output was determined.

[0009] The mapping relationship serves as the fitness function of the MOMPA optimization algorithm. After setting the optimization objective, determining the decision variables, and setting the constraints, the MOMPA optimization algorithm generates a non-dominated optimal solution set. The TOPSIS algorithm is then used to determine the optimal solution at time t and feeds it back to the physical entity of the tunneling system. The next optimization step is then executed.

[0010] Furthermore, the monitoring data undergoes preprocessing, including shutdown data removal, outlier handling, and missing value handling; the shutdown data removal involves selecting the propulsion speed v, cutterhead rotation speed n, total propulsion force F, and cutterhead torque T as discriminant indicators D. When D≠0, the tunnel boring machine is in working state and the corresponding data is retained. When D=0, the tunnel boring machine is in non-working state and the data of non-working state is directly discarded.

[0011] Furthermore, preliminary screening was conducted on the influencing indicators affecting the safety and reliability of the tunnel boring machine, including propulsion speed, cutterhead rotation speed, cutterhead torque, total propulsion force, cutterhead power, piston rod pressure of cutterhead telescopic cylinder 1, piston rod pressure of cutterhead telescopic cylinder 2, piston rod pressure of cutterhead telescopic cylinder 3, propulsion group A pressure, propulsion group B pressure, propulsion group E pressure, slurry density, slurry discharge density, main slurry discharge flow rate, slurry chamber pressure, central chamber reaction force, left-middle-upper grouting pressure, left-upper EP2 external sealing pressure, and right-upper EP2 external sealing pressure.

[0012] Furthermore, key influencing indicators include: propulsion speed X1, cutterhead torque X3, total propulsion force X4, slurry density X6, slurry discharge density X7, main slurry discharge flow rate X8, mud and water chamber pressure X9, left-middle-upper grouting pressure X11, left-upper EP2 external sealing pressure X12, cutterhead telescopic cylinder 2 piston rod pressure X15, propulsion group A pressure X17, and propulsion group E pressure X19.

[0013] Furthermore, the prediction model is composed of two layers of GAT blocks, three layers of LSTM blocks, and one dense layer stacked together. Each GAT block has 18 neurons, and the LSTM layers have 256, 128, and 64 neurons, respectively.

[0014] Furthermore, determining the decision variables involves optimizing and adjusting the active parameters of the tunnel boring machine (TBM), while keeping the passive parameters of the TBM at their original values. The active parameters are: propulsion speed X1, cutterhead torque X3, total propulsion force X4, grout density X6, main grout discharge flow rate X8, slurry chamber pressure X9, left, middle and upper grouting pressure X11, propulsion group A pressure X17, and propulsion group E pressure X19. The passive parameters are: grout discharge density X7, EP2 external sealing pressure X12, and cutterhead telescopic cylinder 2 piston rod pressure X15.

[0015] Furthermore, the constraint is set based on 40% of the original value of the key impact indicator.

[0016] A multi-objective online optimization system for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA includes:

[0017] The first main module acquires and preprocesses monitoring data of shield tunneling construction parameters to establish an original sample set;

[0018] The second main module uses a combination of knowledge-driven and Pearson correlation analysis to determine the input parameters and output targets of the prediction model.

[0019] The third main module builds the spatiotemporal deep learning network framework GAT-LSTM to extract the spatial and temporal features of key influencing indicators, enabling accurate prediction of overturning moment, pitch angle and roll angle.

[0020] The fourth main module learns the mapping relationship between input parameters and optimization objectives through GAT-LSTM, establishes a fitness function, generates a non-dominated optimal solution set through the MOMPA optimization algorithm, and uses the TOPSIS algorithm to determine the optimal solution at time t.

[0021] An electronic device, comprising:

[0022] At least one processor, at least one memory, and a communication interface; wherein,

[0023] The processor, memory, and communication interface communicate with each other;

[0024] The memory stores program instructions that can be executed by the processor, which calls the program instructions to execute the above method.

[0025] A non-transitory computer-readable storage medium storing computer instructions that cause the computer to perform the above-described method.

[0026] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0027] (1) The present invention provides a multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunneling machines based on GAT-LSTM-MOMPA. It establishes the overall safety and stability optimization objectives and key influencing index set of large-diameter slurry shield tunneling machines, develops a GAT-LSTM prediction model, which can capture the correlation between parameters affecting the overall safety and stability of large-diameter slurry shield tunneling machines, and proposes a real-time parameter optimization model that embeds spatiotemporal deep learning blocks (GAT-LSTM) into the MOMPA algorithm, which improves the ability of the optimization model to adjust in advance and ensures the safety and stability of the shield tunneling process.

[0028] (2) The present invention provides a multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA. Based on dual-drive of knowledge data, it effectively screens the factors affecting the safety and stability of shield tunnels, identifies the optimization objectives and the set of influencing parameter indicators for safety and stability, and provides a reference for the safety and stability research of shield tunnels.

[0029] (3) The present invention provides a multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA. By building a GAT-LSTM model that can capture the temporal and spatial characteristics of the dataset, after training, the nonlinear fitting relationship between the input and output is reflected by the trained model parameters, achieving a more accurate fitting and providing a reliable foundation for subsequent multi-objective optimization.

[0030] (4) The present invention provides a multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA. The MOMPA online optimization algorithm can better complete the optimization task of shield tunnel safety and stability, and solves the problems of traditional optimization algorithms being unable to provide dynamic parameter optimization and having small optimization range, which are unreasonable.

[0031] (5) This invention introduces an absolute coefficient (R) 2 The performance of the prediction model was evaluated using three commonly used metrics: root mean square error (RMSE), mean absolute error (MAE), and mean square error (RMSE). The effectiveness and accuracy of the BO-Autoformer model in predicting overturning moment, pitch angle, and roll angle over multiple time steps were analyzed and verified. Attached Figure Description

[0032] Figure 1 is a flowchart of the online multi-objective optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA provided in an embodiment of the present invention.

[0033] Figure 2 is a schematic diagram of the online optimization system for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA provided in an embodiment of the present invention.

[0034] Figure 3 is a schematic diagram of the physical structure of the electronic device provided in an embodiment of the present invention;

[0035] Figure 4 is a schematic diagram of the overturning moment during the tunneling process of the tunnel boring machine provided in the embodiment of the present invention;

[0036] Figure 5 is a heatmap of the Pearson correlation coefficient provided in the embodiment of the present invention;

[0037] Figure 6 is a graph of the loss function during the training process of the overturning moment prediction model provided in an embodiment of the present invention.

[0038] Figure 7 is a diagram of the loss function during the training process of the pitch angle prediction model provided in an embodiment of the present invention.

[0039] Figure 8 is a graph of the loss function during the training process of the roll angle prediction model provided in an embodiment of the present invention;

[0040] Figure 9 is a distribution diagram of the prediction results of the test set of the overturning moment prediction model provided in the embodiment of the present invention;

[0041] Figure 10 is a distribution diagram of the prediction results of the pitch angle prediction model test set provided in the embodiment of the present invention;

[0042] Figure 11 is a distribution diagram of the prediction results of the test set of the roll angle prediction model provided in the embodiment of the present invention;

[0043] Figure 12 is a histogram of the overturning moment optimization effect provided by the embodiment of the present invention;

[0044] Figure 13 is a line graph showing the overturning moment optimization effect provided by the embodiment of the present invention;

[0045] Figure 14 is a histogram of the pitch angle optimization effect provided by the embodiment of the present invention;

[0046] Figure 15 is a line graph showing the pitch angle optimization effect provided by the embodiment of the present invention;

[0047] Figure 16 is a histogram of the roll angle optimization effect provided by the embodiment of the present invention;

[0048] Figure 17 is a line graph showing the roll angle optimization effect provided by the embodiment of the present invention. Detailed Implementation

[0049] 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.

[0050] In a first aspect, as shown in Figure 1, this invention provides a multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA, comprising the following steps:

[0051] Step (1): Obtain monitoring data of shield tunneling construction parameters from the tunneling system database, perform data preprocessing, filter out irrelevant data such as shutdown data, outliers and missing values, and establish the original sample set;

[0052] Modern large-diameter slurry shield tunneling machines are equipped with advanced sensing and Internet of Things (IoT) technologies to collect and record data at a fixed frequency. Data processing was performed on 192,251 samples and 1,250 shield machine parameters collected from study segments 1500-1550. The shield machine shutdown data contained a large number of zero-value data, affecting the accuracy of the results. After shutdown data removal, outlier handling, and missing value processing, a total of 84,396 valid research data samples and 1,250 shield machine parameters were obtained, forming the original sample set.

[0053] (1) Shutdown data filtering

[0054] Four main construction parameters—propulsion speed, cutterhead rotation speed, total propulsion force, and cutterhead torque—are selected as the discrimination indicators. The discrimination function is as follows:

[0055] (1)

[0056] (2)

[0057] In the above formula, F, T, v, and n correspond to the total propulsion force, cutterhead torque, propulsion speed, and cutterhead rotation speed, respectively, and D represents the product of the four parameter values. When the product of the four parameters is not 0, it indicates that the tunnel boring machine is in working condition, and the corresponding data is retained; when the product of the four parameters is 0, that is, the value of any one of the parameters is 0, it is considered that the tunnel boring machine is in non-working condition, and the data for non-working condition is directly discarded.

[0058] (2) Outlier identification and handling

[0059] Outliers are identified and removed using box plots. For removed outliers, data from the previous or next second is used to fill in the gaps.

[0060] (3) Handling missing values

[0061] By combining direct deletion with constant filling, missing data is processed according to different situations to better ensure data validity.

[0062] Step (2): Based on the knowledge of the tunneling process of large-diameter slurry shield tunneling machine, preliminary screening of the influencing indicators of the safety and reliability of shield tunneling machine was carried out in the original sample set;

[0063] The overturning moment of a large-diameter slurry shield tunneling machine refers to the moment formed by the working load or the load outside the overturning line relative to the overturning line during the operation of the shield tunneling machine. A schematic diagram of the overturning moment is shown in Figure 4. In the figure, Mx and My represent the torques in the x-axis and y-axis directions, respectively, and Fx and Fy represent the forces in the x-axis and y-axis directions, respectively. The x and y axes are a rectangular coordinate system with the cutterhead center as the origin.

[0064] As can be seen from the formation of the cutterhead overturning moment, the working load is the main factor affecting the cutterhead overturning moment. In the shield machine parameter system, the propulsion speed, cutterhead rotation speed, cutterhead torque, total propulsion force, cutterhead power, piston rod pressure of cutterhead telescopic cylinder 1, piston rod pressure of cutterhead telescopic cylinder 2, piston rod pressure of cutterhead telescopic cylinder 3, propulsion group A pressure, propulsion group B pressure, and propulsion group E pressure all affect the working load of the shield machine, and thus affect the overturning moment.

[0065] Based on the working principle of roll angle and pitch angle, the following input parameters are used for preliminary screening: slurry density, slurry discharge density, main slurry discharge flow rate, mud-water chamber pressure, central chamber reaction force, left-middle-upper grouting pressure, EP2 external sealing pressure (upper left), and EP2 external sealing pressure (upper right).

[0066] Based on the analysis of the working principle of large-diameter slurry shield tunneling machines and a large amount of engineering practice experience, a total of 19 influencing indicators were selected for processing. The detailed data distribution is shown in Table 1.

[0067] Table 1 Sample Data

[0068]

[0069] Step (3): Use the heatmap generated by Pearson correlation coefficient to further screen 19 influencing indicators to obtain a set of key influencing indicators;

[0070] Pearson correlation coefficients were calculated for the parameters determined by the above experience, and a correlation coefficient heatmap was generated as shown in Figure 5. The heatmap shows that the correlation coefficients of cutterhead rotation speed (X2), EP2 external seal pressure (X13), cutterhead telescopic cylinder 3 piston rod pressure (X14), cutterhead telescopic cylinder 1 piston rod pressure (X16), and propulsion group B pressure (X18) with the safety target are all less than 0.4. Meanwhile, the cutterhead torque (X3) has a very strong correlation with the cutterhead power (X5), and the slurry chamber pressure (X9) has a strong correlation with the central chamber reaction force (X10). To reduce model redundancy, two correlation coefficients were... Only one of the highly sensitive parameters is retained as the input parameter, and the following 12 input parameters are selected: propulsion speed (X1), cutterhead torque (X3), total propulsion force (X4), slurry density (X6), slurry discharge density (X7), main slurry discharge flow rate (X8), mud and water chamber pressure (X9), left, middle and upper grouting pressure (X11), EP2 external seal pressure - left upper (X12), cutterhead telescopic cylinder 2 piston rod pressure (X15), propulsion group A pressure (X17), and propulsion group E pressure (X19).

[0071] The correlation between data was analyzed using the Pearson correlation coefficient, as follows:

[0072] (3)

[0073] In the formula: For sample size, , For two variable sample points, and The mean of the samples within the two variable sets.

[0074] Step (4): Build the spatiotemporal deep learning network framework GAT-LSTM as the prediction model. The spatial and temporal features of the key influencing indicator set are used as inputs, and the overturning moment, pitch angle and roll angle are used as outputs. Determine the mapping relationship between input and output to achieve accurate prediction of the future time step optimization value.

[0075] This invention establishes three different deep learning models: an overturning moment prediction model, a roll angle prediction model, and a pitch angle prediction model. These models predict and optimize the parameters of the overturning moment, roll angle, and pitch angle, and construct the model fitness function. The models are trained on a computer equipped with an i9-13980HX CPU, an RTX4070 GPU, and 32GB of RAM. During training, 80% of the data is randomly selected as the training set, and 20% as the test set. The established prediction model consists of two GAT layers, three LSTM layers, and one dense layer stacked together. Each GAT layer has 18 neurons, and the LSTM layers have 256, 128, and 64 neurons respectively. Adam is used to dynamically adjust the learning rate, and Dropout is used to prevent overfitting. The specific settings of the prediction model are shown in Tables 2 and 3. A prediction model is built using PyTorch. The model's inputs are 12 key impact indicators selected through knowledge-driven analysis from time step t-2 to t, and 3 safety indicator values ​​from time step t-2 to t-1. The output is the 3 safety indicator values ​​at time step t. This input-output parameter framework facilitates subsequent online model optimization.

[0076] (1) Time series data reconstruction

[0077] The input and output parameter structures are as follows:

[0078] (4)

[0079] (5)

[0080] Among them, X train X test Y represents the input to the training and test sets. train Y test This represents the output of the training and test sets, where t represents the time step and i represents the number of samples.

[0081] (2) Establish and train the prediction model: Randomly divide the data in the original sample set into a training sample set and a test sample set, establish the prediction model based on the PyTorch framework in Python, and train it using the training sample set.

[0082] Table 2 Prediction Model Architecture

[0083]

[0084] Table 3 Hyperparameter settings for the prediction model

[0085]

[0086] Step (5), GAT-LSTM spatiotemporal algorithm prediction;

[0087] A spatiotemporal safety index prediction model is built using GAT-LSTM. The nonlinear temporal and spatial relationship between the input parameters found by GAT-LSTM is used as the fitness function of the optimization algorithm to optimize the objective. The accuracy of the GAT-LSTM spatiotemporal prediction algorithm reflects the fit of the fitness function and plays a decisive role in the final optimization effect. Therefore, the accuracy of the spatiotemporal model prediction is crucial.

[0088] The model was trained and validated based on the constructed GAT-LSTM. The optimal hyperparameters were determined by adjusting the model's hyperparameters. Figures 6, 7, and 8 show the training and validation loss functions for the three spatiotemporal safety index prediction models, respectively. As can be seen from the figures, after 100 epochs, the loss curves of all three models tended to plateau, and the difference between the loss on the test set and the training set was small, indicating no overfitting. This demonstrates that 100 epochs are sufficient to improve model performance. Based on the determined key hyperparameters, R... 2 The model's performance is evaluated using three metrics: fitness function (MAE), fitness rate (RMSE), and fitness rate (RMSE), ensuring a good fit to the fitness function and improving the model's reliability. 2 Used to measure the goodness of fit between predicted and observed values, RMSE and MAE can effectively reflect the deviation between the predicted and observed values. The smaller the RMSE and MAE values, the better the model performance. 2 The closer the value is to 1, the higher the model's prediction accuracy.

[0089] (6)

[0090] (7)

[0091] (8)

[0092] In the formula, The total number of samples, For predicted values, This is the actual value. This is the average of the true values.

[0093] To more intuitively observe the accuracy of the models, Figures 9, 10, and 11 show partial prediction fitting effects of the three spatiotemporal safety index prediction models. Table 4 presents the evaluation metrics for the three spatiotemporal safety index prediction models. The results show that the developed GAT-LSTM spatiotemporal safety index prediction model achieves high-precision predictions for overturning moment, pitch angle, and roll angle when the input parameter is 8, and the data distribution presented also conforms to the trend of the actual collected data. For overturning moment prediction, R... 2The value is 0.919, MAE is 135.204, and RMSE is 106.241; for pitch angle prediction, R... 2 The values ​​are 0.923, MAE is 0.017, and RMSE is 0.014; for roll angle prediction, R... 2 The R-values ​​are 0.976, MAE is 0.055, and RMSE is 0.032. (The remaining text appears to be incomplete and contains several errors. A more accurate translation would require the full context.) 2 All are greater than 0.91.

[0094] In summary, the prediction accuracy of the three models satisfies the condition that the model parameters serve as the fitness fitting function. This means that after training the GAT-LSTM model, the nonlinear fitting relationship between the input and output is reflected by the trained model parameters. This provides a reliable foundation for subsequent multi-objective optimization.

[0095] Table 4. Model Prediction Evaluation Indicators Results

[0096]

[0097] Step (6) learns the mapping relationship between input parameters and optimization objective through GAT-LSTM, establishes a high-precision mapping relationship as the fitness function of MOMPA optimization algorithm, generates non-dominated optimal solution set through optimization algorithm, and uses the ideal solution similarity ranking priority technique (TOPSIS) to provide recommendations for the optimal value of the model, continuously updates the historical value of the model, and achieves accurate and effective real-time online optimization.

[0098] Based on the general principles of multi-objective optimization methods, the implementation of the MOMPA online optimization algorithm mainly consists of a fitness function, optimization objective setting, decision variable determination, constraint setting, TOPSIS decision-making, and real-time optimization. The fitness function is the nonlinear mapping relationship between the input and output sought by the spatiotemporal safety index prediction model. The optimization objective is the overturning moment, pitch angle, and roll angle of a large-diameter slurry shield tunneling machine. Decision variables are determined based on the selected input parameters and actual shield machine control. Constraints should control the decision parameters within a reasonable range to prevent the optimization result from being unattainable or causing excessive damage to the large-diameter slurry shield tunneling machine. TOPSIS selects the optimal decision parameters from the Pareto set solution of MOMPA optimization based on the set evaluation criteria. Since the model output is the target prediction value for the next time step, the optimization result contains information about the output target for the next time step. To achieve real-time optimization, the historical values ​​of the optimization objective need to be replaced with the results of the optimization algorithm. Specifically, the historical values ​​Y at time steps t-2 and t-1... i,t-2 Y i,t-1 It should be replaced with the optimized value Y at time steps t-1 and t. i,optThen, the optimization for the next time step is performed accordingly. This process is repeated each time step is moved forward.

[0099] (1) Establishment of fitness function

[0100] (9)

[0101] (2) Determination of decision variables

[0102] Of the 12 key influencing indicators identified through a dual-drive approach of knowledge and data, the following are active parameters of the tunnel boring machine (TBM): propulsion speed (X1), cutterhead torque (X3), total propulsion force (X4), grout density (X6), main grout discharge flow rate (X8), slurry chamber pressure (X9), left-middle-upper grouting pressure (X11), propulsion group A pressure (X17), and propulsion group E pressure (X19). The following are passive parameters, reflecting more of the working state of the cutterhead telescopic cylinder; their pressure values ​​are typically determined by factors such as cylinder design, manufacturing, and wear during use. During TBM tunneling, the active parameters are adjusted to achieve safe and stable tunneling, while the passive parameters mainly reflect local geological conditions and states. Therefore, active parameters can be optimized and adjusted, while passive parameters should be maintained at their original values.

[0103] (3) Setting constraints

[0104] Setting constraints on key influencing indicators plays a crucial role in the reliability of optimization results and the safety of tunnel boring machines. Methods for adjusting constraints can be divided into adjustments based on the original values ​​and adjustments based on upper and lower limits of the data. To enhance the adaptability of the control parameters and make the constraints more aligned with actual conditions and needs, this invention selects 40% of the original values ​​as a constraint for multi-objective optimization, allowing them to be adjusted within the range of [-40%, +40%].

[0105] (4) Basic ideas and steps of the TOPSIS algorithm

[0106] Step 1: Forwarding and standardizing the original data matrix;

[0107] Step 2: Calculate the distance between the optimal and worst solutions:

[0108] (10)

[0109] (11)

[0110] Step 3: Unnormalized score of the i-th evaluation object:

[0111] (12)

[0112] in, Let i be the score for the i-th decision option. It is the distance between the i-th evaluation object and the minimum value. is the distance between the i-th evaluation object and the maximum value, and m represents the total number of objects. , These are the most ideal and most negative conditions among all solutions, respectively. , .

[0113] Step (7), optimize the results analysis;

[0114] With the objective function, decision variables, and constraints set, the model parameters are initialized, the population size is set to 300, the maximum number of iterations is set to 100, and the population positions are initialized. The developed GAT-LSTM-MOMPA ensemble algorithm is applied to determine the Pareto optimal solution set after iteration termination. Then, TOPSIS analysis is applied to these Pareto front solutions to determine the final ideal solution. For ease of observation, a validation set of 2000 time steps is selected to demonstrate the model performance.

[0115] The optimization results are shown in Table 5. The developed GAT-LSTM-MOMPA-TOPSIS integrated hybrid algorithm effectively optimized the three objectives online, significantly reducing the values ​​of the tunnel boring machine's overturning moment, pitch angle, and roll angle, thus further improving the safety and stability of the tunnel boring machine during the tunneling process. For the studied construction section, the overturning moment decreased from the average value of 3766.99 kN.m to 3037.94 kN.m, an optimization improvement of 19.35 percentage points. The pitch angle decreased from the average value of 22.31° to 20.54°, an optimization improvement of 7.93 percentage points. The roll angle increased from the average value of -0.79 to -0.58, an optimization improvement of 26.58 percentage points. The overall optimization of the three objectives was 17.95 percentage points. Although the performance of the tunnel boring machine can be further improved by increasing the allowable limits of the operating parameters, other constraints and safety issues exist to prevent the parameters from increasing indefinitely.

[0116] Figures 12, 13, 14, 15, 16, and 17 respectively show the line graphs and histograms comparing the overturning moment, pitch angle, and roll angle before and after optimization when the constraint condition is 40%. As can be seen from the figures, the developed GAT-LSTM-MOMPA can reduce the values ​​of safety and stability performance indicators during tunnel boring machine excavation. The optimized overturning moment, pitch angle, and roll angle are more stable, clearly demonstrating the optimization and improvement capabilities of the developed model. Similarly, the histogram before optimization has a significantly larger variance and a larger mean than the histogram after optimization, indicating a significant and more stable optimization effect.

[0117] Table 5. Multi-objective optimization improvement effect

[0118]

[0119] Secondly, as shown in Figure 2, an embodiment of the present invention provides a multi-objective online optimization system for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA, comprising: a first main module: acquiring monitoring data of shield tunneling construction parameters, performing data preprocessing, filtering irrelevant data such as shutdown data, outliers, and missing values, and establishing an original sample set; a second main module: using a combination of knowledge-driven and Pearson correlation analysis to determine the input parameters and output objectives of the study; a third main module: building a spatiotemporal deep learning network framework GAT-LSTM, extracting the spatial and temporal features of the index set, and achieving accurate prediction of overturning moment, pitch angle, and roll angle; a fourth main module: learning the mapping relationship between input parameters and optimization objectives through GAT-LSTM, establishing a high-precision fitness function, generating a non-dominated optimal solution set through optimization algorithms; and using the ideal solution similarity ranking priority technique (TOPSIS) to provide recommendations for the optimal values ​​of the model, continuously updating the historical values ​​of the model, and achieving accurate and effective real-time online optimization.

[0120] As shown in Figure 3, in a third aspect, embodiments of the present invention provide an electronic device, including:

[0121] At least one processor; and at least one memory communicatively connected to the processor, wherein:

[0122] The memory stores program instructions that can be executed by the processor. The processor can call the program instructions to execute the online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA, which is provided by any of the various implementation methods of the first aspect.

[0123] Fourthly, embodiments of the present invention provide a non-transitory computer-readable storage medium storing computer instructions that cause a computer to execute a multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunneling machines based on GAT-LSTM-MOMPA, provided by any of the various implementations of the first aspect.

[0124] This invention proposes an integrated model, GAT-LSTM-MOMPA, which combines Graph Attention Network (GAT), Long Short-Term Memory (LSTM) network, and Multi-Objective Marine Predator Algorithm (MOMPA). This model dynamically mines the mechanical coupling relationships among the components of the tunnel boring machine (TBM) using GAT, enhancing its ability to express spatial topological features. Combined with LSTM's precise modeling of the temporal evolution of tunneling parameters, it achieves high-precision prediction of indicators such as overturning moment. While MOMPA can ultimately provide the optimal solution, it cannot provide continuous control over optimal operating parameters. Deep learning algorithms that consider both temporal and spatial information can provide predicted values ​​for future time steps, which can then be integrated with optimization algorithms to form a closed loop for future time-step optimization. By considering the operating parameters of a large-diameter slurry TBM as controllable input features, the model parameters of the spatiotemporal deep learning algorithm as the fitness function, and safety and reliability indicators as the optimization targets, a multi-objective predictive optimization model for optimizing the next time step can be established. This enables real-time, continuous online multi-objective optimization, contributing to the safety and stability of the tunneling process.

[0125] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels based on GAT-LSTM-MOMPA, characterized in that: Monitoring data of shield tunneling construction parameters were acquired and preprocessed to establish an initial sample set. Within this initial sample set, indicators affecting the safety and reliability of the shield machine were initially screened. A heatmap generated using the Pearson correlation coefficient was used for further screening of these indicators, resulting in a set of key influencing indicators. These indicators included: propulsion speed, cutterhead rotation speed, cutterhead torque, total propulsion force, cutterhead power, piston rod pressure of cutterhead telescopic cylinder 1, piston rod pressure of cutterhead telescopic cylinder 2, piston rod pressure of cutterhead telescopic cylinder 3, pressure of propulsion group A, pressure of propulsion group B, pressure of propulsion group E, grout density, grout discharge density, main grout discharge flow rate, slurry chamber pressure, reaction force of the central chamber, grouting pressure of the upper left and middle sections, external sealing pressure of the upper left EP2, and external sealing pressure of the upper right EP2. A spatiotemporal deep learning network framework, GAT-LSTM, was constructed as the prediction model. The spatial and temporal characteristics of the key influencing indicator set were used as the input to the prediction model, while overturning moment, pitch angle, and roll angle were used as the output. The mapping relationship between the input and output was determined. Key influencing indicators include: propulsion speed X1, cutterhead torque X3, total propulsion force X4, slurry inlet density X6, slurry outlet density X7, main slurry outlet flow rate X8, mud-water chamber pressure X9, left-middle-upper grouting pressure X11, left-upper EP2 external seal pressure X12, cutterhead telescopic cylinder 2 piston rod pressure X15, propulsion group A pressure X17, and propulsion group E pressure X19. The mapping relationship serves as the fitness function of the MOMPA optimization algorithm. After setting the optimization objective, determining the decision variables, and setting constraints, the MOMPA optimization algorithm generates a non-dominated optimal solution set, and the TOPSIS algorithm is used for decision-making. The optimal solution at time t is obtained and fed back to the physical entity of the tunneling system. Then, the optimization of the next time step is performed. The determination of decision variables is to optimize and adjust the active parameters of the tunnel boring machine, while the passive parameters of the tunnel boring machine remain at their original values. The propulsion speed X1, cutterhead torque X3, total propulsion force X4, slurry density X6, main slurry discharge flow rate X8, slurry chamber pressure X9, left, middle and upper grouting pressure X11, propulsion group A pressure X17, and propulsion group E pressure X19 are active parameters. The slurry discharge density X7, left upper EP2 external sealing pressure X12, and cutterhead telescopic cylinder 2 piston rod pressure X15 are passive parameters.

2. The multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels according to claim 1, characterized in that, The monitoring data undergoes preprocessing, including shutdown data removal, outlier handling, and missing value handling. The shutdown data removal is as follows: the propulsion speed v, cutterhead rotation speed n, total propulsion force F, and cutterhead torque T are selected as the discrimination index D, where D = F·T·v·n. When D≠0, the tunnel boring machine is in working condition, and the corresponding data is retained. When D = 0, the tunnel boring machine is in non-working condition, and the non-working condition data is directly removed.

3. The multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels according to claim 1, characterized in that, The prediction model consists of two layers of GAT blocks, three layers of LSTM blocks, and one dense layer stacked together. Each GAT block has 18 neurons, and the LSTM layers have 256, 128, and 64 neurons, respectively.

4. The multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels according to claim 1, characterized in that, The constraint is set based on 40% of the original value of the key impact indicator.

5. A system for implementing the multi-objective online optimization method for the safety and stability of large-diameter slurry shield tunnels as described in any one of claims 1-4, characterized in that, include: The first main module acquires and preprocesses monitoring data of shield tunneling construction parameters to establish an original sample set. The second main module uses a combination of knowledge-driven and Pearson correlation analysis to determine the input parameters and output targets of the prediction model. The third main module builds a spatiotemporal deep learning network framework, GAT-LSTM, to extract the spatial and temporal features of key influencing indicators, enabling accurate prediction of overturning moment, pitch angle, and roll angle. The fourth main module learns the mapping relationship between input parameters and optimization targets through GAT-LSTM, establishes a fitness function, generates a non-dominated optimal solution set through the MOMPA optimization algorithm, and uses the TOPSIS algorithm to determine the optimal solution at time t.

6. An electronic device, characterized in that, include: The system includes at least one processor, at least one memory, and a communication interface; wherein the processor, memory, and communication interface communicate with each other. The memory stores program instructions that can be executed by the processor, which invokes the program instructions to perform the method described in any one of claims 1-4.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method described in any one of claims 1-4.

Images