Shield tunnel segment dynamic type selection method based on deep learning and multi-objective decision

Abstract

The invention provides a shield tunnel segment dynamic type selection method based on deep learning and multi-objective decision, and the method comprises the steps: building a segment type selection neural network model through a deep learning algorithm based on the actual measurement data of shield construction, and carrying out the training to obtain a nonlinear mapping relation between an assembly point location and a shield body, between the assembly point location and a segment center coordinate, and between the assembly point location and an oil cylinder stroke difference; determining all to-be-selected point position combinations meeting staggered joint assembly requirements of subsequent n rings according to the current ring segment point positions, and solving shield tail gaps, axis deviations and oil cylinder stroke differences of the subsequent n ring segments under different to-be-selected point position combinations on the basis of a neural network model and in combination with geometric relationships among the obtained coordinates; and carrying out adaptive distribution on the weight of each loss subitem according to a segment type selection loss function, traversing all to-be-selected point position combinations, and realizing optimal assembly point position combination selection by taking loss function minimization as a target.

Description

Technical Field

[0001] The present invention belongs to the technical field of shield tunnel construction, and specifically relates to a method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making. Background Art

[0002] The shield method has been widely used in various urban underground engineering constructions due to its advantages such as high degree of automation, good safety, and little impact on the surrounding environment. During the shield construction process, tunnel ring construction is as important as shield excavation, and segment selection is the key guiding basis for tunnel ring construction. The current segment selection methods can be divided into manual selection and automatic selection according to the working mode.

[0003] Manual selection usually relies on the construction experience of on-site operators and the help of professional selection devices to select the segment assembly points. For example, Chinese Patent 202122270973.5 and Patent 202210204858.4 simulate the segment assembly process by designing a wedge-shaped segment staggered selection compass, thereby assisting on-site construction personnel to quickly locate the segment assembly points.

[0004] Automated selection mainly combines simplified calculation models with mature manual experience and embeds them into the core of the selection algorithm to achieve automated selection of assembly points. For example, patent 202310224025.9 derived the quantitative correlation function of shield tail gap, cylinder stroke difference and shield trend between adjacent ring segments, and further combined with manual selection experience to form an intelligent decision-making algorithm to achieve dynamic selection of the current ring assembly point; such as patent 201910387068.2, by designing an intelligent decision-making system for segment selection, the real-time calculated construction parameters such as cylinder stroke difference and shield tail gap are dynamically compared with the manually set thresholds, thereby predicting the optimal selection of the current ring segment.

[0005] With the rapid development of data science, data-driven technology has been widely used in the field of shield construction to explore the intrinsic connections between massive measured data. Among them, deep learning has been applied to the prediction of key construction parameters such as shield posture and shield tail gap due to its powerful nonlinear approximation ability. For example, patent 202210739896.X uses wavelet transform and deep learning algorithm to train the measured engineering data to predict the shield posture parameters at future times. For example, patent 202410266325.8 combines deep learning algorithm with the minimum shield tail gap theory to form a PDNN model, and then accurately predicts the shield tail gap through online learning.

[0006] For segment selection, manual selection has problems such as over-reliance on decision-makers' experience and lack of quantitative calculation support, which can easily lead to unreasonable selection and construction, thus burying huge hidden dangers for shield posture and the quality of formed segments. Although automated selection can solve many of the above problems, its complex calculation model also limits the rapid prediction of multi-ring assembly points. Summary of the invention

[0007] In view of the problems and shortcomings of the prior art, the present invention provides a method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making. The method utilizes the characteristics of deep learning to achieve accurate prediction without relying on cumbersome calculation models. The deep learning algorithm is used to predict the construction parameters that affect the selection of segments, so as to achieve efficient selection of multi-ring assembly points.

[0008] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0009] A dynamic selection method for shield tunnel segments based on deep learning and multi-objective decision-making: Based on the measured data of shield construction, a segment selection neural network model is constructed through a deep learning algorithm, and the nonlinear mapping relationship between the assembly points and the shield body, the segment center coordinates, and the cylinder stroke difference is obtained through training; all the candidate point combinations of the subsequent n rings that meet the staggered assembly requirements are determined according to the current ring segment points, and based on the neural network model and the geometric relationship between the obtained coordinates, the shield tail gap, axis deviation and cylinder stroke difference of the subsequent n ring segments under different candidate point combinations are solved; according to the segment selection loss function and the weights of each loss item are adaptively allocated, all candidate point combinations are traversed, and the optimal assembly point combination is selected with the goal of minimizing the loss function.

[0010] Furthermore, in the process of constructing the segment selection neural network model, the data set is generated in the following manner:

[0011] Collect on-site construction data related to segment selection as the original data set, including at least: stratum information, shield key point coordinates, shield construction parameters and segment assembly points;

[0012] Preprocess the original data set: including zero value processing, outlier processing, noise reduction processing and normalization processing;

[0013] Perform feature screening to select features that are relevant to the output target, including the shield center coordinates, segment center coordinates, and cylinder stroke difference;

[0014] Determine the prediction step size and sliding window of the neural network model, and continuously read the data set to generate a time series data set.

[0015] Furthermore, the construction process of the segment selection loss function is as follows:

[0016] According to the geometric relationship of each component of the shield construction model, the shield tail gap, axis deviation and cylinder stroke difference are solved; the obtained shield tail gap, axis deviation and cylinder stroke difference are weighted and calculated to form the segment selection loss function; each loss sub-item needs to be adaptively weighted according to the empirical threshold, and the sum of the weights must be 1.

[0017] Furthermore, based on various sensors on the shield machine, historical construction data related to segment selection are preliminarily collected as the original data set, including at least: formation information, key point coordinates, construction parameters and segment assembly points; wherein the formation information at least includes: base bed coefficient, base friction coefficient, deformation modulus; the key point coordinates at least include: shield body center coordinates, segment center coordinates; the construction parameters at least include: cylinder thrust, propulsion speed, cylinder stroke difference.

[0018] Furthermore, the original data set is preprocessed:

[0019] Zero value processing uses a binary discriminant function to identify and eliminate data in unstable tunneling states;

[0020] Outlier processing uses the box plot method to detect and remove outliers ring by ring, and linear interpolation is used to fill in missing data;

[0021] Noise reduction processing Savitzky-Golay filtering method is used to reduce the noise of the data set;

[0022] Normalization processing uses min-max normalization to map the original data to the [0,1] interval without changing the distribution of the data.

[0023] As a preferred solution, the feature screening uses empirical screening combined with random forest and Pearson correlation coefficient method to perform feature screening on the data.

[0024] Furthermore, the segment selection neural network model includes three neural network models built based on the LSTM algorithm, each model corresponds to only one output target; the root mean square error, mean absolute error, and determination coefficient R 2 As an evaluation indicator of model training effect;

[0025] The Bayesian optimization algorithm was used, and RMSE was used as the objective function to iteratively optimize the hyperparameter combinations of the three LSTM models, including the hidden layer structure, activation function, and optimizer.

[0026] Three LSTM models are trained according to the optimized hyperparameter combination, and the generalization performance is evaluated on the test set. If the three evaluation indicators meet the accuracy requirements on the test set, the model structure and parameters are saved to form a segment selection neural network model.

[0027] As a preferred solution, according to the current ring segment point position, among all the candidate point combinations of the subsequent n rings that meet the staggered assembly requirements, the relationship between the optional point position K of the i+1th ring and the selected point position M of the ith ring is as follows;

[0028] K=M+2+3N(N=0,1,2,...).

[0029] Furthermore, according to the geometric relationship of each component of the shield construction model, the process of solving the shield tail gap, axis deviation and cylinder stroke difference is as follows:

[0030] Substituting the candidate point combinations and related input parameters of the subsequent n rings into the segment selection neural network model to predict the shield body, segment center coordinates and cylinder stroke difference of the subsequent n rings;

[0031] Among them, the coordinates of the shield center B and the segment center C of the segment ring to be assembled at a certain assembly point are obtained by the neural network model; it is assumed that point A and point C are at the same mileage and located on the design axis; it is assumed that the coordinates of points A, B, and C in the world coordinate system are (x A ,y A ,z A )、(x B ,y B ,z B )、(x C ,y C ,z C ), then the axis deviation of the segment ring to be assembled at the selected point is expressed by the following formula:

[0032]

[0033] The shield tail clearance is calculated by the following formula:

[0034]

[0035] Where: d h ,d v Represent the horizontal and vertical axis deviations respectively; s represents the shield tail gap.

[0036] Furthermore, the axis deviation, shield tail clearance and cylinder stroke difference are weighted and calculated to form the segment selection loss function. The loss sub-items and the total loss function are shown in the following formula; the weight of each loss sub-item needs to be adaptively allocated according to the empirical threshold, and should be amplified in equal proportion to ensure that the sum of the weights is 1;

[0037]

[0038] Where: Loss1, Loss2, and Loss3 are the loss items represented by axis deviation, shield tail clearance, and cylinder stroke difference respectively; Losstotal is the total loss function of segment selection; 1iv , 1ih is the weight corresponding to the vertical and horizontal axis deviation of the i-th ring; 2i is the weight corresponding to the gap at the tail of the i-th ring shield; 3iv , 3ih is the weight corresponding to the stroke difference of the upper and lower and left and right cylinders of the i-th ring; k i is the weight magnification coefficient corresponding to the ith ring, which is obtained by the following formula;

[0039]

[0040] The method for selecting the optimal assembly point combination of subsequent rings is as follows:

[0041] According to the following formula, traverse all candidate point combinations G, and use the loss function Loss total Minimize the target and complete the optimal assembly point combination G of the subsequent n ring segments best Selection of

[0042] G best =argmin Loss total (G).

[0043] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making as described above are implemented.

[0044] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making as described above.

[0045] Compared with the prior art, the present invention and its preferred solution are driven by measured data of shield construction, and use deep learning algorithms to construct a neural network model for segment selection, and explore the nonlinear mapping relationship between assembly points and shield body, segment center coordinates, and cylinder stroke difference; determine all the candidate point combinations of subsequent n rings that meet the requirements of staggered assembly according to the current ring segment points; based on the neural network model and combined with the geometric relationship between the obtained coordinates, solve the shield tail gap, axis deviation and cylinder stroke difference of the subsequent n ring segments under different candidate point combinations; comprehensively consider the above three to design the segment selection loss function and adaptively allocate the weights of each loss item; traverse all candidate point combinations, and minimize the loss function as the goal to achieve the selection of the optimal assembly point combination. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0047] Figure 1 This is a flow chart of the principle of dynamic selection of shield tunnel segments according to an embodiment of the present invention;

[0048] Figure 2 This is a flow chart of constructing a neural network model for selecting a segment according to an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of assembling the pipe segments according to an embodiment of the present invention (taking the yoz plane as an example);

[0050] Figure 4 This is a schematic diagram of the shield tail gap according to an embodiment of the present invention;

[0051] Figure 5 Schematic diagram of the weight function of multi-objective decision making in an embodiment of the present invention: (a) axis deviation (b) shield tail clearance (c) cylinder stroke difference;

[0052] In the figure: 1- assembled segment ring; 2- segment ring to be assembled; 21- axis of segment ring to be assembled; 22- outer edge of segment ring to be assembled; 23- shield tail gap I; 24- shield tail gap II; 25- shield tail gap III; 26- shield tail gap IV; 3- shield shell; 31- shield shell inner wall; 4- shield body; 41- shield body axis; 5- cutter head; 6- thrust cylinder; 61- cylinder stroke difference; 7- designed axis. DETAILED DESCRIPTION

[0053] In order to make the features and advantages of this patent more obvious and easy to understand, the following embodiments are specifically described in detail as follows:

[0054] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0055] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0056] Combine the following Figure 1 to Figure 5 The present invention is further described with specific embodiments:

[0057] like Figure 3 , Figure 4As shown, in this embodiment, the coordinate axes of the world coordinate system are defined as: the x-axis is horizontal to the right, the z-axis is vertically downward, and the y-axis, the x-axis and the z-axis form a right-handed coordinate system. On this basis, Figure 1 As shown, the following steps are performed to select the optimal assembly point combination for the subsequent 5-ring segments.

[0058] Step 1: Build a neural network model for segment selection

[0059] Step 11: Raw data set collection

[0060] like Figure 2 As shown, this embodiment preliminarily collects historical construction data related to segment selection based on various sensors on the shield machine as the original data set, which includes: stratum information, key point coordinates, construction parameters and segment assembly points. Among them, the stratum information includes: bed coefficient, base friction coefficient, deformation modulus, etc.; the key point coordinates include: shield center coordinates (i.e., propulsion cylinder action surface center coordinates), segment center coordinates; construction parameters include: cylinder thrust, propulsion speed, cylinder stroke difference, etc.;

[0061] Step 12: Dataset preprocessing

[0062] like Figure 2 As shown, this step performs preprocessing on the original data set collected in step 11, including zero value processing, outlier processing, noise reduction processing and normalization processing;

[0063] Zero value processing: In the actual excavation process, the shield machine is in an unstable excavation state for a considerable part of the time, and the monitoring data in these time periods are meaningless for model construction. Therefore, the present invention uses a binary discriminant function to identify and eliminate the data in the unstable excavation state;

[0064] Outlier processing: Outliers in the original data set will not only affect the results of subsequent noise reduction and normalization, but also cause the model to overfit and reduce generalization performance. Therefore, the present invention uses the box plot method to detect and remove outliers ring by ring, and uses linear interpolation to fill in all missing data (zero value + outlier);

[0065] Noise reduction: The noise contained in the original data set may affect the subsequent model training effect. Therefore, the present invention uses the Savitzky-Golay filtering method to perform noise reduction on the data set. The advantage of this method is that the shape of the signal can be retained while filtering out the noise.

[0066] Normalization: The dimensions of different types of parameters in the original data set are different, and the range of differences is huge, which can easily lead to model convergence difficulties and long training time. Therefore, the present invention uses min-max normalization to map the original data to the [0,1] interval without changing the distribution of the data;

[0067] Step 13: Feature Screening

[0068] like Figure 2 As shown, the present invention uses empirical screening combined with random forest (RF) and Pearson correlation coefficient to perform feature screening on the data processed in step 12, and selects parameters (i.e., features) that are strongly correlated with the output targets (shield center coordinates, segment center coordinates, cylinder stroke difference) to reduce the dimension of the data set, thereby achieving the purpose of streamlining the model and improving generalization performance.

[0069] Step 14: Multi-step forecasting strategy design

[0070] like Figure 1 , Figure 2 As shown, this embodiment aims to predict the optimal assembly point combination of the next 5 rings of segments. Therefore, before the model is built, the prediction step is set to 5, and the sliding window is set to 10 according to experience, that is, the input parameters are the features corresponding to the first 10 rings, and the output targets are the shield body, segment center coordinates and cylinder stroke difference of the next 5 rings; on this basis, the data processed in steps 11 to 13 are continuously read to generate a time series data set;

[0071] Step 15: Model structure design and evaluation index selection

[0072] like Figure 2 As shown, the time series data set obtained in step 14 is randomly divided into a training set, a validation set, and a test set in a ratio of 6:2:2, where the training set is used to determine the trainable parameters of the model, the validation set is used to preliminarily evaluate the model performance and adjust the hyperparameters, and the test set is used to finally evaluate the generalization performance of the model.

[0073] In order to simplify the model structure and save computational overhead, the present invention constructs three neural network models based on the LSTM algorithm, and each model corresponds to only one output target; the root mean square error (RMSE), mean absolute error (MAE), and determination coefficient R 2 As an evaluation index of the model training effect, as shown in formulas (1) to (3);

[0074]

[0075] Where: n is the test sample size; is the i-th predicted value, is the average value of the measured values, y i is the i-th measured value.

[0076] Step 16: Hyperparameter Optimization

[0077] like Figure 2As shown in the figure, the Bayesian optimization (BO) algorithm is used, and RMSE is used as the objective function to iteratively optimize the hyperparameter combinations (hidden layer structure, activation function, and optimizer) in the three LSTM models.

[0078] Step 17: Model Training

[0079] According to the hyperparameter combination optimized in step 16, three LSTM models are trained respectively, and their generalization performance is evaluated on the test set. 2 ) meet the accuracy requirements on the test set, the trained model can be considered to have certain prediction performance and generalization ability, and then its model structure and parameters are saved to form a segment selection neural network model.

[0080] Step 2: Determine possible assembly point combinations for subsequent rings

[0081] According to the current ring segment assembly point, all the candidate point combinations G that meet the staggered assembly requirements of the next 5 rings are determined in sequence; the relationship between the optional point K of the i+1th ring and the selected point M of the ith ring (i<5) is shown in formula (4);

[0082] K=M+2+3N(N=0,1,2,...) (4)

[0083] Step 3: Calculate the shield tail clearance, axis deviation, and cylinder stroke difference

[0084] Substitute the candidate point combination and related input parameters of the subsequent 5 ring segments in step 2 into the segment selection neural network model obtained in step 1 to predict the shield body 4, the segment center coordinates and the cylinder stroke difference 61 of the subsequent 5 rings;

[0085] Further, if Figure 3 , Figure 4 As shown, the basic structures involved in shield construction include: assembled segment rings 1, segment rings 2 to be assembled, shield shell 3 (shield shell inner wall 31), shield body 4 (shield body axis 41), cutter head 5, and thrust cylinder 6.

[0086] Among them, points B and C represent the shield center and segment center of the segment ring 2 to be assembled at a certain assembly point, respectively, and their coordinates are obtained by the neural network model; point A and point C are at the same mileage and located on the design axis 7. Assume that the coordinates of points A, B, and C in the world coordinate system are (x A ,y A ,z A )、(x B ,y B ,z B )、(x C ,y C ,z C), then the axis deviation of the segment ring axis 21 to be assembled at the selected point can be expressed by equation (5). Figure 4 It can be seen that the relative distance s between the segment center C and the shield center B is an important indicator reflecting the gaps Ⅰ23, Ⅱ24, Ⅲ25 and Ⅳ26 between the outer edge 22 of the segment ring to be assembled and the shield tail. Therefore, the shield tail gap is defined as shown in formula (6). The cylinder stroke difference 61 includes the up and down and left and right stroke differences (Δl v , Δl h ), can be directly obtained by the neural network model without further calculation.

[0087]

[0088] Where: d h ,d v Represent the horizontal and vertical axis deviations respectively; s represents the shield tail clearance;

[0089] Step 4: Loss function design

[0090] The axis deviation, shield tail clearance and cylinder stroke difference obtained in step 3 are weighted and calculated to form the segment selection loss function. The loss sub-items and the total loss function are shown in formula (7). The weight of each loss sub-item needs to be adaptively allocated according to the empirical threshold, and should be amplified in equal proportion to ensure that the sum of the weights is 1.

[0091]

[0092] Where: Loss1, Loss 2、 Loss3 is the loss item represented by axis deviation, shield tail clearance, and cylinder stroke difference; Loss total is the total loss function of segment selection; 1iv , 1ih is the weight corresponding to the vertical and horizontal axis deviation of the i-th ring; 2i is the weight corresponding to the gap at the tail of the i-th ring shield; 3iv , 3ih is the weight corresponding to the stroke difference of the upper and lower and left and right cylinders of the i-th ring; k i is the weight magnification coefficient corresponding to the ith ring, obtained by formula (8).

[0093]

[0094] like Figure 5 As shown in FIG. 1 , in this embodiment, the empirical thresholds of axis deviation, shield tail clearance and cylinder stroke difference are set to 50 mm, 30 mm and 100 mm, and the initial weights of the three are set to 0.1, 0.6 and 0.1, then λ 1iv , 1ih ,2i , 3iv , 3ih The expressions of are shown in (9) to (11).

[0095]

[0096]

[0097] Step 5: Select the optimal assembly point combination for subsequent rings

[0098] According to formula (12), traverse all candidate point combinations G and use the loss function Loss in step 4 total Minimize the target and complete the optimal assembly point combination G of the subsequent 5-ring segment best Selection of

[0099] G best =argmin Loss total (G) (12).

[0100] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0101] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0104] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0105] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0106] This patent is not limited to the above-mentioned optimal implementation mode. Anyone can derive other forms of a dynamic selection method for shield tunnel segments based on deep learning and multi-objective decision-making under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of this invention should be covered by this patent.

Claims

1. A dynamic selection method for shield tunnel segments based on deep learning and multi-objective decision making. Features: Based on the measured data of shield construction, a neural network model for segment selection is constructed through a deep learning algorithm, and the nonlinear mapping relationship between the assembly points and the shield body, segment center coordinates, and cylinder stroke difference is obtained through training; all the candidate point combinations of the subsequent n rings that meet the requirements of staggered assembly are determined according to the current ring segment points, and the shield tail gap, axis deviation and cylinder stroke difference of the subsequent n ring segments under different candidate point combinations are solved based on the neural network model and combined with the geometric relationship between the obtained coordinates; according to the segment selection loss function and the weights of each loss item are adaptively allocated, all candidate point combinations are traversed, and the optimal assembly point combination is selected with the goal of minimizing the loss function.

2. According to claim 1, the dynamic selection method for shield tunnel segments based on deep learning and multi-objective decision-making, Features: During the construction of the segment selection neural network model, the data set is generated as follows: Collect on-site construction data related to segment selection as the original data set, including at least: stratum information, shield key point coordinates, shield construction parameters and segment assembly points; Preprocess the original data set: including zero value processing, outlier processing, noise reduction processing and normalization processing; Perform feature screening to select features that are relevant to the output target, including the shield center coordinates, segment center coordinates, and cylinder stroke difference; Determine the prediction step size and sliding window of the neural network model, and continuously read the data set to generate a time series data set.

3. According to claim 1, the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision making, Features: The construction process of the segment selection loss function is as follows: According to the geometric relationship of each component of the shield construction model, the shield tail gap, axis deviation and cylinder stroke difference are solved; the shield tail gap, axis deviation and cylinder stroke difference are weighted and calculated to form the segment selection loss function; each loss sub-item needs to be adaptively weighted according to the empirical threshold, and the sum of the weights must be 1.

4. According to claim 2, the dynamic selection method for shield tunnel segments based on deep learning and multi-objective decision-making, Features: Based on various sensors on the shield machine, historical construction data related to segment selection are preliminarily collected as the original data set, including at least: formation information, key point coordinates, construction parameters and segment assembly points; among which, the formation information at least includes: base bed coefficient, base friction coefficient, deformation modulus; the key point coordinates at least include: shield center coordinates, segment center coordinates; the construction parameters at least include: cylinder thrust, propulsion speed, cylinder stroke difference.

5. According to claim 2, the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making, Features: Preprocessing the original data set: Zero value processing uses a binary discriminant function to identify and eliminate data in unstable tunneling states; Outlier processing uses the box plot method to detect and remove outliers ring by ring, and linear interpolation is used to fill in missing data; Noise reduction processing Savitzky-Golay filtering method is used to reduce the noise of the data set; Normalization processing uses min-max normalization to map the original data to the [0,1] interval without changing the distribution of the data.

6. According to claim 1, the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision making, Features: The segment selection neural network model includes three neural network models built based on the LSTM algorithm, each model corresponds to only one output target; the root mean square error, mean absolute error, and determination coefficient R are used. 2 As an evaluation indicator of model training effect; The Bayesian optimization algorithm was used, and RMSE was used as the objective function to iteratively optimize the hyperparameter combinations of the three LSTM models, including the hidden layer structure, activation function, and optimizer. Three LSTM models are trained according to the optimized hyperparameter combination, and the generalization performance is evaluated on the test set. If the three evaluation indicators meet the accuracy requirements on the test set, the model structure and parameters are saved to form a segment selection neural network model.

7. The method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making according to claim 3, Features: According to the geometric relationship of each component of the shield construction model, the process of solving the shield tail gap, axis deviation and cylinder stroke difference is as follows: Substituting the candidate point combinations and related input parameters of the subsequent n rings into the segment selection neural network model to predict the shield body, segment center coordinates and cylinder stroke difference of the subsequent n rings; Among them, the coordinates of the shield center B and the segment center C of the segment ring to be assembled at a certain assembly point are obtained by the neural network model; it is assumed that point A and point C are at the same mileage and located on the design axis; it is assumed that the coordinates of points A, B, and C in the world coordinate system are (x A ,y A ,z A )、(x B ,y B ,z B )、(x C ,y C ,z C ), then the axis deviation of the segment ring to be assembled at the selected point is expressed by the following formula: The shield tail clearance is calculated by the following formula: Where: d h d v Represent the horizontal and vertical axis deviations respectively; s represents the shield tail gap.

8. According to claim 7, the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision making, Features: The axis deviation, shield tail clearance and cylinder stroke difference are weighted and calculated to form the segment selection loss function. The loss sub-items and the total loss function are shown in the following formula; the weight of each loss sub-item needs to be adaptively allocated according to the empirical threshold, and should be amplified in equal proportion to ensure that the sum of the weights is 1; Where: Loss 1 、Loss 2 、Loss 3 They are the loss items represented by axis deviation, shield tail clearance, and cylinder stroke difference; total is the total loss function of segment selection; 1iv , 1ih is the weight corresponding to the vertical and horizontal axis deviations of the i-th ring; λ 2i is the weight corresponding to the gap at the tail of the i-th ring shield; 3iv , 3ih is the weight corresponding to the stroke difference of the upper and lower and left and right cylinders of the i-th ring; k i is the weight magnification coefficient corresponding to the ith ring, which is obtained by the following formula; The method for selecting the optimal assembly point combination of subsequent rings is as follows: According to the following formula, traverse all candidate point combinations G, and use the loss function Loss total Minimize the target and complete the optimal assembly point combination G of the subsequent n ring segments best Selection of G best =arg min Loss total (G)。 9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the steps of the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making as described in any one of claims 1 to 8 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method for dynamic selection of shield tunnel segments based on deep learning and multi-objective decision-making as described in any one of claims 1 to 8 are implemented.

Images