A method for constructing a universal segment point selection model for shield tunneling and its application

Abstract

The present invention belongs to the technical field of shield construction, and specifically discloses a construction method and application of a universal shield segment point selection model, wherein the construction method comprises: constructing a training sample set, training a neural network to obtain a universal shield segment point selection model; wherein the method for obtaining each training sample is: establishing a shield machine model, a segment model and a DTA; based on the DTA, randomly determining the coordinates of the corresponding points of the previous ring of segments to be assembled on the DTA, the posture and position of the previous ring of segments, the coordinates of the corresponding points of the shield machine on the DTA, and the current posture of the shield machine, so as to measure or calculate the construction parameters under the current model state; based on the posture and position of the previous ring of segments, the construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA, the points of the segments to be assembled are determined as labels to obtain a training sample. The present invention solves the problem of insufficient data volume for training the segment point selection model, ensures the rationality of the segment point selection, and improves the efficiency of segment point selection.

Description

Technical Field

[0001] The present invention belongs to the technical field of shield construction, and more specifically, relates to a method for constructing a universal shield segment point selection model and an application thereof. Background Art

[0002] The shield method is a tunnel construction method that keeps the excavation face stable within the steel shell of the shield machine while safely advancing forward, assembling the segments at its tail, and then using jacks to push the assembled segments, using their reaction force to advance the shield. Every time the shield machine advances a working stroke, a segment is assembled under the support of the shield tail, and grouting is promptly injected into the gap between the stratum behind the shield tail and the outer periphery of the segment to prevent the tunnel and the ground from sinking, while continuously discharging earth from the excavation face. A significant feature of the shield method compared to other construction methods is that the segments assembled into a ring directly become the final lining of the tunnel. Therefore, shield construction has extremely high requirements for the assembly quality of the segments, and the assembly quality of the segments directly affects the quality of the tunnel.

[0003] Universal segment refers to the use of only one type of ring with a certain amount of wedge in the entire tunnel construction, and the orderly rotation of the wedge ring is used to fit the tunnel axis of various linear shapes. The selection of universal segment is restricted by many factors and the selection process is complicated. At present, most of the selection is done manually by experienced personnel based on on-site construction data. This selection method is blind and arbitrary, and it is easy to have unreasonable segment selection, making it difficult to ensure the overall quality of the tunnel, and it is inefficient and has extremely high requirements for operators. Summary of the invention

[0004] The present invention provides a method for constructing a universal shield segment point selection model and an application thereof, so as to solve the technical problem of low point selection efficiency in the existing universal shield segment point selection method.

[0005] The technical solution of the present invention to solve the above technical problem is as follows: A method for constructing a universal shield segment point selection model, comprising:

[0006] Construct a training sample set and train the neural network to obtain a universal segment point selection model for shield tunneling;

[0007] The method for obtaining each training sample is as follows:

[0008] Establish shield machine model, segment model and DTA;

[0009] Based on the DTA, randomly determine the coordinates of the corresponding points of the previous ring of segments to be assembled on the DTA, the posture and position of the previous ring of segments, the coordinates of the corresponding points of the shield machine on the DTA, and the current posture of the shield machine;

[0010] According to the posture of the previous ring of segments and the current posture of the shield machine, the construction parameters under the current model state are measured or calculated; and based on the posture and position of the previous ring of segments, the construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA, the positions of the segments to be assembled are determined as labels to obtain a training sample.

[0011] The beneficial effects of the present invention are as follows: in order to solve the problem that the assembly points of shield construction segments are restricted by many factors and difficult to reasonably select, the present invention proposes to obtain a segment point selection model based on reasonable segment point selection data training and optimization of the neural network model. Compared with manual point selection, this method can ensure the rationality of segment point selection. In view of the problem that it is difficult to obtain sufficient historical construction data to train and optimize the automatic segment point selection model due to the lack of historical segment data and incomplete data records, this method can effectively solve this problem by obtaining sample data by randomly determining the posture and other information of the previous ring of segments based on computer simulation technology. The automatic segment point selection method of this method can ensure the rationality of segment point selection, thereby ensuring the quality of the formed tunnel and improving the efficiency of segment point selection. At the same time, it also solves the problem of insufficient data for training the segment point selection model.

[0012] Based on the above technical solution, the present invention can also be improved as follows.

[0013] Furthermore, the method for determining the coordinates of the corresponding point of the shield machine on the DTA is:

[0014] Based on the coordinates of the corresponding points of the previous ring of segments on the DTA and the geometric dimensions of the shield machine, the coordinates of the corresponding points of the current shield machine on the DTA are determined.

[0015] Furthermore, the coordinates of the corresponding points of the previous ring segment on the DTA, the posture and point positions of the previous ring segment are randomly determined as follows:

[0016] The Monte Carlo method is used to collect historical data of the segment position during the actual construction process, and the data is analyzed and processed to obtain the first probability distribution of the deviation between the actual segment position and DTA;

[0017] According to the first probability distribution, the segment posture deviation is randomly sampled, and combined with the coordinates of the corresponding points of the previous ring of segments on the DTA determined randomly, the posture and point positions of the previous ring of segments are calculated.

[0018] Furthermore, the coordinates of the corresponding points of the shield machine on the DTA and the current posture of the shield machine are randomly determined as follows:

[0019] Monte Carlo method is used to collect the shield machine posture data in the formed tunnel, and the second probability distribution of the deviation between the shield machine posture and DTA is obtained by analyzing and processing the data;

[0020] According to the second probability distribution, the shield machine posture deviation is randomly sampled, and combined with the coordinates of the corresponding points of the shield machine on the DTA, the shield machine posture is calculated.

[0021] A further beneficial effect of the present invention is that, based on the Monte Carlo method, random sampling can be performed on the basis of actual historical data to ensure the reliability and diversity of training data.

[0022] Furthermore, the coordinates of the corresponding points of the segments to be assembled on the DTA are determined as follows:

[0023] According to the posture of the previous ring of segments, the geometric parameters of the segments and the DTA, the coordinates of the corresponding points of the segments to be assembled on the DTA are determined.

[0024] Furthermore, the construction parameters include: the thrust cylinder stroke difference of the shield machine in the horizontal direction, the thrust cylinder stroke difference of the shield machine in the vertical direction, and the shield tail gap between the shield machine and the segment in four directions of up, down, left and right.

[0025] Furthermore, in the construction of the training sample set, the line types of the DTA include straight lines, circular curves and transition curves.

[0026] A further beneficial effect of the present invention is that in order to make the simulation data represent the correspondence between the actual construction parameters and the points, it is necessary to ensure that the collected shield machine posture data includes the posture of the shield machine under various DTA line shapes, and the data must be comprehensive and sufficient to improve the reliability of model training.

[0027] Further, the method of training the neural network is:

[0028] S1. Build a neural network model and initialize the weights and biases of the neural network model;

[0029] S2, input the training sample into the input layer of the neural network in the form of a vector, the input layer inputs the data into the first hidden layer, thereby outputting a feature value consistent with the number of neurons in the hidden layer; in this way, sequentially inputting into the remaining multiple hidden layers of the neural network to obtain corresponding feature values; finally, inputting into the output layer of the neural network and obtaining multiple feature values ​​consistent with the number of segment points, thereby obtaining the predicted point position;

[0030] S3, comparing the predicted point with the actual point of the training sample, and generating a loss function, if the loss function is greater than a preset value, updating the weight and bias of the neural network and repeatedly executing S2 until the error values ​​between each predicted point and the actual point in the corresponding training sample group are within the preset range;

[0031] S4. Save the weights and biases, and generate a universal segment point selection model.

[0032] The present invention also provides a universal shield segment point selection method, comprising:

[0033] The posture and position of the previous ring of segments to be assembled, the current construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA are collected during the real-time construction process, and the point selection model constructed by the construction method of a universal shield segment point selection model as described above is input to obtain the position of the segments to be assembled, thereby completing the point selection of the universal segment.

[0034] The beneficial effect of the present invention is that the point selection model constructed by the above-mentioned method for constructing a universal shield segment point selection model can ensure the rationality of segment point selection, thereby ensuring the quality of the formed tunnel and improving the efficiency of segment point selection.

[0035] The present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute the method for constructing a universal shield segment point selection model as described above and / or the method for selecting universal shield segment points as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flowchart of a method for obtaining training samples in a method for constructing a universal shield segment point selection model provided in an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a pipe segment provided in an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of a pipe segment through-seam connection and a pipe segment staggered seam connection provided in an embodiment of the present invention;

[0039] Figure 4 A schematic diagram of the relationship between the shield machine, the segment and the DTA provided in an embodiment of the present invention;

[0040] Figure 5 A schematic diagram of determining the coordinates of corresponding points of the segments to be assembled on the DTA provided in an embodiment of the present invention;

[0041] Figure 6 A schematic diagram of a three-layer BP neural network provided by an embodiment of the present invention;

[0042] Figure 7 A flowchart of a method for selecting points for a universal shield segment provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0044] Embodiment 1

[0045] A method for constructing a universal segment point selection model for shield tunneling 100, such as Figure 1 As shown, including:

[0046] Construct a training sample set and train the neural network to obtain a universal segment point selection model for shield tunneling;

[0047] The method for obtaining each training sample is as follows:

[0048] S110, establish shield machine model, segment model and DTA;

[0049] S120, based on the DTA, randomly determine the coordinates of the corresponding points of the previous ring of segments of the current segment to be assembled on the DTA, the posture and position of the previous ring of segments, the coordinates of the corresponding points of the current shield machine on the DTA, and the current posture of the shield machine;

[0050] S130. According to the posture of the previous ring of segments and the current posture of the shield machine, the construction parameters under the current model state are measured or calculated; and based on the posture and position of the previous ring of segments, the construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA, the positions of the segments to be assembled are determined as labels to obtain a training sample.

[0051] The pipe segment refers to a circular cylindrical structure with a certain amount of wedge formed by several assembled blocks. The assembled blocks are connected by circumferential bolt holes, and the pipe segments are connected by axial bolt holes. The installation of the pipe segment needs to consider various influencing factors, and reasonable points are selected on the end face of the previous ring of pipe segments for assembly. By selecting different points, the pipe segment can obtain different wedge amounts in the axial and radial directions, thereby fitting different tunnel axes. For example, Figure 2 As shown, it is a universal pipe segment composed of 6 assembly blocks with 10 points (as shown by the 10 circles in the left figure) and a wedge amount of 2S, where S represents the single-sided wedge amount.

[0052] The reference data in S130 includes the points of the previous ring of segments, mainly to avoid the occurrence of through-seam assembly. Figure 3As shown in the left figure, through-seam assembly means that the longitudinal seams between the segments are aligned, which is easy to cause uneven annular surface and increase the length of the annular seam, which will reduce the compression rate of the elastic sealing gasket and is not conducive to waterproofing. During the construction process, it is generally necessary to ensure that the segments are assembled with staggered seams, such as Figure 3 As shown in the right figure, staggered assembly requires that the longitudinal seams of the front and rear segments be staggered. This assembly method has a relatively flat ring surface, and the circular ring can be approximately considered as a homogeneous stiffness. The length of the ring seam is short, and sealing and waterproofing are easier to handle. The reference data in S130 contains the pose of the previous ring of segments in order to calculate the coordinates of the segments to be assembled at all possible points, and combined with the coordinates of the corresponding points of the segments to be assembled on the DTA, the deviation between the actual axis of the segment and the axis of the DTA is calculated, that is, the axis fitting deviation. Since the collected sample data is used to train the segment selection model based on the neural network in the subsequent steps, it is necessary to ensure that the acquired data is comprehensive and sufficient to represent the correspondence between the actual construction parameters and the points.

[0053] The collected sample data are trained based on a neural network, with the aim of obtaining the influence of the construction parameters and the position of the previous ring of segments on the selection of the current ring of segments, and finally generating an automatic segment selection model.

[0054] It should be noted that in S130, the positions of the segments to be assembled can be determined by manual point selection based on the posture and position of the previous ring of segments, construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA.

[0055] The shield machine model, segment model and DTA are established in the computer simulation environment; the established DTA needs to include all line types, that is, it needs to include straight lines, circular curves and transition curves.

[0056] Regarding the acquisition of construction parameters in S120, such as Figure 4 As shown in the figure, once the positions of the segments and the shield machine are determined in the simulation environment, various construction parameters can be obtained by solving the spatial distance method.

[0057] Since the historical data of segment selection is obtained during the actual tunnel construction process, a set of historical data can only be obtained by assembling a ring of segments. Due to the complexity of segment assembly construction, data generation is slow and historical data is small. Moreover, since a complete set of historical data needs to contain many construction parameters, point information, etc., it is very easy to produce incomplete data records. Due to the small amount of historical data and incomplete data records, it is difficult to classify the segments by combining machine learning methods. This method is designed to solve the problem that the assembly points of shield construction segments are restricted by many factors and difficult to select reasonably. It is proposed to train and optimize the neural network model based on reasonable segment selection data to obtain the segment selection model. Compared with manual selection, this method can ensure the rationality of segment selection. In view of the problem that it is difficult to obtain sufficient historical construction data to train and optimize the automatic segment selection model due to the small amount of historical data and incomplete data records, this method obtains sample data by randomly determining the posture and other information of the previous ring of segments based on computer simulation technology, which can effectively solve this problem. The automatic segment point selection method of this method can ensure the rationality of segment point selection, thereby ensuring the quality of the formed tunnel, improving the efficiency of segment point selection, and also solving the problem of insufficient data for training the segment point selection model.

[0058] Preferably, the method for determining the coordinates of the corresponding point of the shield machine on the DTA is:

[0059] Based on the coordinates of the corresponding points of the previous ring of segments on the DTA and the geometric dimensions of the shield machine, the coordinates of the corresponding points of the current shield machine on the DTA are determined.

[0060] Preferably, the coordinates of the corresponding points of the previous ring segment on the DTA, the posture and point positions of the previous ring segment are randomly determined in the following manner:

[0061] The Monte Carlo method is used to collect historical data of the segment position during the actual construction process, and the data is analyzed and processed to obtain the first probability distribution of the deviation between the actual segment position and DTA;

[0062] According to the first probability distribution, the segment posture deviation is randomly sampled, and combined with the coordinates of the corresponding points of the previous ring of segments on the DTA determined randomly, the posture and point positions of the previous ring of segments are calculated.

[0063] Preferably, the coordinates of the corresponding points of the shield machine on the DTA and the current posture of the shield machine are randomly determined as follows:

[0064] Monte Carlo method is used to collect the shield machine posture data in the formed tunnel, and the second probability distribution of the deviation between the shield machine posture and DTA is obtained by analyzing and processing the data;

[0065] According to the second probability distribution, the shield machine posture deviation is randomly sampled, and combined with the coordinates of the corresponding points of the shield machine on the DTA, the shield machine posture is calculated.

[0066] Since the shield machine aims to tunnel along the DTA, but the actual tunneling axis of the shield machine cannot completely coincide with the DTA, a deviation value can be added to the DTA to simulate the position of the shield machine. The position relationship between the shield machine and the DTA is as follows: Figure 4 To sample the deviation value, it is necessary to collect historical shield machine posture data and DTA data, and analyze and calculate the probability distribution of the deviation.

[0067] Taking the deviation obeying normal distribution as an example, the probability density function of the deviation between the actual position of the shield machine and the tunnel design axis is:

[0068]

[0069] Where x is the deviation between the actual position of the shield machine and the corresponding point on the tunnel design axis, μ and σ are the expectation and variance of the deviation, respectively.

[0070] In order to make the simulation data represent the correspondence between the actual construction parameters and the points, it is necessary to ensure that the collected shield machine posture data includes the posture of the shield machine under various DTA line shapes, and the data must be comprehensive and sufficient; the historical shield machine excavation posture data collected under conventional working conditions and special working conditions should be processed separately, and the probability density functions of the deviations under conventional working conditions and the probability density functions of the deviations under special working conditions should be calculated respectively.

[0071] Preferably, the coordinates of the corresponding points of the above-mentioned segments to be assembled on the DTA are determined in the following manner:

[0072] According to the position and posture of the previous ring of segments, the segment geometric parameters and DTA, the coordinates of the corresponding points on the DTA of the segments to be assembled are determined.

[0073] like Figure 5 As shown in the figure, the center point of the end face of the segment to be assembled at all possible points is on a plane Ψ. When the geometric dimensions of the segment are known, the position of the previous ring of segments is determined, and the equation of the plane Ψ can be obtained by calculation. Then, the intersection of the plane Ψ and the DTA is calculated, which is the coordinate of the corresponding point of the segment on the DTA. It should be noted that the segment coordinates refer to the coordinates of the center point of the end face of the segment.

[0074] Preferably, the construction parameters include: the thrust cylinder stroke difference of the shield machine in the horizontal direction, the thrust cylinder stroke difference of the shield machine in the vertical direction, and the shield tail gap between the shield machine and the segment in four directions of up, down, left and right.

[0075] Preferably, the above-mentioned method of training the neural network is:

[0076] S1. Build a neural network model and initialize the weights and biases of the neural network model;

[0077] S2, input the training sample into the input layer of the neural network in the form of a vector, the input layer inputs the data into the first hidden layer, thereby outputting a feature value consistent with the number of neurons in the hidden layer; in this way, sequentially inputting into the remaining multiple hidden layers of the neural network to obtain corresponding feature values; finally, inputting into the output layer of the neural network and obtaining multiple feature values ​​consistent with the number of segment points, thereby obtaining the predicted point position;

[0078] S3, comparing the predicted point with the actual point of the training sample, and generating a loss function, if the loss function is greater than a preset value, updating the weight and bias of the neural network and repeatedly executing S2 until the error values ​​between each predicted point and the actual point in the corresponding training sample group are within the preset range;

[0079] S4. Save the weights and biases, and generate a universal segment point selection model.

[0080] like Figure 6 As shown, taking a three-layer BP neural network model as an example (where "three layers" include an input layer, a hidden layer, and an output layer), training the neural network includes:

[0081] S20, initializing the weights W, V and biases b1, b2 of the BP neural network;

[0082] S21, input the sample data into the hidden layer of the BP neural network model, and the output of the hidden layer is:

[0083] z=g1(W*x+b1);

[0084] Among them, z is the output of the hidden layer, and its dimension is consistent with the number of neurons in this layer, which is an m-dimensional column vector; W is the weight from the input layer to the hidden layer, which is an m×n matrix; x is the input parameter, which is an n-dimensional column vector, including the shield tail gap, cylinder stroke difference, coordinates of the corresponding points on the DTA segment, and the position and posture of the selected segments in the previous ring; b1 is the bias of the input layer, which is an m-dimensional column vector; g1 is the activation function of the hidden layer.

[0085] The output of the hidden layer is input to the output layer, and the output of the output layer is:

[0086] y=g2(V*z+b2),

[0087] Among them, y is the target point. For a segment with p points, y is a p-dimensional column vector; V is the weight from the hidden layer to the output layer, which is a p×m matrix; b2 is the bias of the hidden layer, which is a p-dimensional column vector; g2 is the activation function of the output layer.

[0088] S22, calculate the error between the predicted point outputted from sub-step S21 and the actual point of the sample data, and generate a loss function:

[0089]

[0090] Where y(x) is the point calculated by the neural network model, a is the actual point, and N is the number of training samples. If the loss function is greater than the preset value, the weight and bias of the BP neural network are updated and recalculated until the error values ​​between each target point and the point in the corresponding data sample group are within the preset range;

[0091] S23, saving the weight W of the BP neural network when the error value in step S22 is within the preset range * 、V * and deviation b1 * 、b2 * , and generate the automatic point selection model for the segment accordingly.

[0092] On this basis, as a preferred approach, before each data sample group is trained, the method further includes:

[0093] Each of the data sample groups is normalized to ensure that the data accuracy in each data sample is consistent, thereby ensuring the overall accuracy of the finally generated segment selection model when it is used.

[0094] As a preferred method, the weight parameters, bias values ​​and retraining are adjusted as follows:

[0095] The gradient descent method based on back propagation adjusts the weight parameters and bias values ​​and retrains. The gradient descent method calculates the current gradient of the parameters each time, and then moves the parameters in the opposite direction of the gradient for a certain distance, repeating this process until the loss function approaches zero. The back propagation algorithm uses the structure of the neural network to perform calculations, which can greatly reduce the amount of calculations in the gradient descent method.

[0096] Embodiment 2

[0097] A universal shield segment point selection method, comprising:

[0098] The posture and position of the previous ring of segments to be assembled, the current construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA are collected during the real-time construction process, and the point selection model constructed by the method for constructing a universal shield segment point selection model as described in the first embodiment is input to obtain the position of the segments to be assembled, thereby completing the point selection of the universal segments.

[0099] like Figure 7As shown, firstly, a segment automatic point selection model is pre-trained, and on this basis, real-time reference data is received, wherein the reference data is the posture and position of the previous ring of segments to be assembled, the current construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA. The received real-time reference data is generated in real time during the segment installation process of the current shield construction, and this reference data is input as an input parameter to the pre-trained segment automatic point selection model. The segment point selection model generates a relatively reasonable selection result according to the target construction parameters, and the operator assembles the current ring of segments on the previous ring of segments at the corresponding positions according to the selection result.

[0100] The point selection model constructed by the above-mentioned method for constructing a universal shield segment point selection model can ensure the rationality of segment point selection, thereby ensuring the quality of the formed tunnel and improving the efficiency of segment point selection.

[0101] The relevant technical solution is the same as that in Embodiment 1 and will not be described in detail here.

[0102] Embodiment 3

[0103] A computer-readable storage medium, the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute a method for constructing a universal shield segment point selection model as described in the first embodiment and / or a method for selecting universal shield segment points as described in the second embodiment. The relevant technical solutions are the same as those in the first and second embodiments, and will not be repeated here.

[0104] It will be easily understood by those skilled in the art that the above description is only 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 in the protection scope of the present invention.

Claims

1. A method for constructing a universal shield segment point selection model, characterized in that: include: Construct a training sample set and train the neural network to obtain a universal segment point selection model for shield tunneling; The method for obtaining each training sample is as follows: Establish shield machine model, segment model and DTA; Based on the DTA, the coordinates of the corresponding points of the previous ring of segments to be assembled on the DTA, the posture and position of the previous ring of segments, the coordinates of the corresponding points of the shield machine on the DTA, and the current posture of the shield machine are randomly determined; wherein, based on the coordinates of the corresponding points of the previous ring of segments on the DTA and the geometric dimensions of the shield machine, the coordinates of the corresponding points of the shield machine on the DTA are determined; the Monte Carlo method is used to collect historical data on the posture of the segments during the actual construction process, and the data is analyzed and processed to obtain the deviation between the actual posture of the segments and the DTA A first probability distribution is obtained, and according to the first probability distribution, the posture deviation of the segment is randomly sampled, and the posture and point positions of the previous ring of segments are calculated in combination with the coordinates of the corresponding points of the previous ring of segments on the DTA determined randomly; the posture data of the shield machine in the formed tunnel is collected by the Monte Carlo method, and the second probability distribution of the deviation between the posture of the shield machine and the DTA is obtained by analyzing and processing the data; according to the second probability distribution, the posture deviation of the shield machine is randomly sampled, and the current posture of the shield machine is calculated in combination with the coordinates of the corresponding points of the shield machine on the DTA; Determine the coordinates of the corresponding points of the segments to be assembled on the DTA according to the posture of the segments of the previous ring, the geometric parameters of the segments and the DTA; According to the posture of the previous ring of segments and the current posture of the shield machine, the construction parameters under the current model state are measured or calculated; and based on the posture and position of the previous ring of segments, the construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA, the positions of the segments to be assembled are determined as labels to obtain a training sample.

2. The method for constructing a universal shield segment point selection model according to claim 1, characterized in that: The construction parameters include: the thrust cylinder stroke difference of the shield machine in the horizontal direction, the thrust cylinder stroke difference of the shield machine in the vertical direction, and the shield tail gap between the shield machine and the segment in four directions of up, down, left and right.

3. The method for constructing a universal shield segment point selection model according to claim 1 or 2, characterized in that: In the construction of the training sample set, the line type of the DTA is a straight line, a circular curve or a transition curve.

4. The method for constructing a universal shield segment point selection model according to claim 1 or 2, characterized in that: The method of training the neural network is: S1. Build a neural network model and initialize the weights and biases of the neural network model; S2, inputting the training sample into the input layer of the neural network in the form of a vector, and the input layer inputting the data into the first hidden layer, thereby outputting a feature value consistent with the number of neurons in the hidden layer; In this way, the remaining multiple hidden layers of the neural network are sequentially input to obtain corresponding eigenvalues; finally, the output layer of the neural network is input to obtain multiple eigenvalues ​​consistent with the number of segment points, and the predicted point is obtained; S3, comparing the predicted point with the actual point of the training sample, and generating a loss function, if the loss function is greater than a preset value, updating the weight and bias of the neural network and repeatedly executing S2 until the error value between each predicted point and the actual point in the corresponding training sample set is within a preset range; S4. Save the weights and biases, and generate a universal segment point selection model.

5. A method for selecting points for universal shield segments, characterized in that: include: The posture and position of the previous ring of segments to be assembled, the current construction parameters and the coordinates of the corresponding points of the segments to be assembled on the DTA are collected during the real-time construction process, and the point selection model constructed by the method for constructing a universal shield segment point selection model as described in any one of claims 1 to 4 is input to obtain the position of the segments to be assembled, thereby completing the point selection of the universal segments.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed by the processor, the device where the storage medium is located is controlled to execute a method for constructing a universal shield segment point selection model as described in any one of claims 1 to 4 and / or a universal shield segment point selection method as described in claim 5.

Images