Shield tunnel irregular load inversion method based on neural network agent model

Abstract

The invention relates to a shield tunnel irregular load inversion method based on a neural network agent model, and belongs to the field of shield tunnel load inversion. According to the method, assumption of a load distribution mode in a traditional lining load inversion method is abandoned, an inversion result can reveal an irregular load distribution mode of a field tunnel, and technical guarantee is provided for safe service of the tunnel; according to the method, the lining full-circumference deformation obtained based on the three-dimensional laser scanning technology is inverted, the field data volume is remarkably increased, the dimension of the deformation vector can be fully increased along with the increase of n, and the defect that an inversion result is not unique easily caused by a traditional inversion method is overcome.

Description

Technical Field

[0001] The invention belongs to the field of shield tunnel load inversion, and in particular relates to a shield tunnel irregular load inversion method based on a neural network proxy model. Background Art

[0002] The load size and distribution of shield tunnels are important factors affecting the structural safety of tunnels during their service life. To obtain the load size and distribution of shield tunnels during their service life, the traditional method requires the deployment of water and soil pressure sensors behind the lining. Although water and soil pressure sensors can directly obtain the load acting on the tunnel lining, the deployment is complex, the sensor survival rate is low, and it is difficult to apply them on a large scale in actual projects due to cost constraints.

[0003] Compared with directly obtaining the lining load, obtaining the internal force and deformation of the tunnel lining is cheaper and the technology is more mature. Based on the internal force and deformation of the lining obtained by detection or monitoring, combined with the mechanical model of the shield tunnel, the size and distribution of the load acting on the lining can be inverted.

[0004] The existing lining load inversion methods are mostly based on the load distribution pattern assumed during tunnel design. Through the mechanical forward model, combined with the intelligent algorithm, global optimization is performed to obtain the load size that meets the on-site internal force and deformation monitoring. However, this method relies on the load distribution pattern in the design period, which is quite different from the actual load distribution pattern of the project. If the assumed load distribution pattern is abandoned for inversion, there may be a dilemma that the inversion load is not unique. At this time, various regularization techniques or prior distributions need to be introduced, which involves human subjectivity. Therefore, for the inversion of lining loads, there is currently no inversion method that is both low-cost, highly feasible on-site, and has a solid theoretical basis.

[0005] Since the load distribution pattern of the actual tunnel project cannot be known with certainty, the assumed load distribution pattern must be abandoned. The resulting problem of non-unique inversion load can be avoided by increasing the field measured data. The 3D laser scanning technology can obtain the full circumferential deformation of the shield tunnel by scanning the tunnel contour, which greatly enriches the measured data of tunnel deformation. The circumferential deformation obtained by 3D laser scanning can improve the inversion accuracy of the tunnel lining load. Summary of the invention

[0006] The purpose of the present invention is to overcome the problems existing in the prior art and provide a method for inversion of irregular loads in a shield tunnel based on a neural network proxy model.

[0007] To achieve the above object, the technical solution of the present invention is: a method for inversion of irregular loads in shield tunnels based on a neural network proxy model, comprising:

[0008] Step S1, establishing a refined three-dimensional finite element model of the shield tunnel lining according to the geometric dimensions, joint structure and material mechanical parameters of the actual tunnel project;

[0009] Step S2: Divide the lining into n equal parts according to the circumference to form n load inversion target points and form an N-dimensional load inversion vector P = (P1, P2, ..., P n ); randomly generate m load vectors P based on the Monte Carlo method;

[0010] Step S3, applying the load determined in step S2 to the lining, inputting it into the finite element model established in step S1, calculating the full-circle deformation of the lining corresponding to each load vector P; dividing the lining deformation into 2n equal parts of the circumference, obtaining the lining deformation of each equally divided point, and forming m 2n-dimensional deformation vectors D;

[0011] Step S4, selecting a neural network proxy model, taking the deformation vector D as the model input, taking the load vector P as the model output, and training and verifying the neural network proxy model with the m samples obtained in step S3;

[0012] Step S5, using three-dimensional laser scanning technology to obtain the full-circle deformation of the actual tunnel; according to the division method in step S3, determine the lining deformation vector d at the engineering site;

[0013] Step S6: using the neural network proxy model trained in step S4, based on the on-site lining deformation vector, calculate the load vector p actually acting on the lining.

[0014] In one embodiment of the present invention, in step S1, the characteristics of the three-dimensional finite element numerical model of the tunnel lining are: a geometric model is established according to the diameter, lining thickness and lining segmentation of the actual tunnel project; the joints of different segment segments are connected by bolts; friction contact is adopted between the segments; the constitutive model of the concrete lining adopts a three-fold line elastoplastic model, and the mechanical parameters are selected according to the concrete grade; the joint bolts adopt a double-fold line elastoplastic model, and the mechanical parameters are selected according to the bolt model.

[0015] In one embodiment of the present invention, in step S2, the loads between the inversion target points are obtained using a cubic spline interpolation method.

[0016] In one embodiment of the present invention, in step S2, the load acquisition method between the inversion target points is specifically as follows:

[0017] For the i-th load inversion target point, its polar coordinates are θ i , on which a load P is applied i , forming data points (θ i ,P i ); for adjacent data points, a cubic polynomial y = ax 3+bx 2 +cx+d for interpolation; according to the continuity requirement, each cubic function must satisfy:

[0018]

[0019]

[0020]

[0021]

[0022] …

[0023]

[0024]

[0025] We get 2n equations. According to the requirement of smoothness, we get n equations by taking the first-order derivative of the piecewise function at the connection point:

[0026]

[0027]

[0028] …

[0029]

[0030]

[0031] Then, we can obtain n equations continuously based on the second-order derivative:

[0032] 6a1θ2+2b1=6a2θ2+2b2

[0033] 6a2θ3+2b2=6a3θ3+2b3

[0034] …

[0035] 6a n-1 θ n +2b n-1 =6a n θ n +2b n

[0036] 6a n θ1+2b n =6a1θ1+2b1

[0037] By combining the above 4n equations, we can solve the 4n unknown coefficients a1, b1, c1, d1, ..., a of the n piecewise functions. n , b n , cn , d n .

[0038] In one embodiment of the present invention, in step S3, the load at any position of the lining is calculated according to the cubic spline interpolation result, and the corresponding load is applied in the finite element numerical model, the finite element model is iterated until convergence, and the full-circle deformation of the inner contour of the tunnel lining under a given load is obtained, and the lining ring is divided into 2n equal parts to obtain a 2n-dimensional deformation vector D; the finite element model calculation is repeated m times to obtain m deformation vectors D.

[0039] In one embodiment of the present invention, in step S4, a BP neural network proxy model is constructed with a 2n-dimensional deformation vector D as input and an n-dimensional load vector P as output. The BP neural network proxy model is trained with the m samples obtained in step S3 to obtain a neural network proxy model P=f(D) for predicting the load vector P with the lining deformation vector D.

[0040] In one embodiment of the present invention, in step S5, three-dimensional laser scanning technology is used to obtain the contour point cloud data of the on-site tunnel, and after processing, the inner contour of the lining in the current state is obtained; by comparing with the inner contour data of the tunnel just after construction, the full-circle deformation of the tunnel in the current state is obtained; according to the equally divided positions in step S3, the deformation vector d of the on-site tunnel is obtained.

[0041] In one embodiment of the present invention, in step S6, the neural network proxy model P=f(D) trained in step S4 is input with the on-site deformation vector d to obtain the on-site tunnel load vector p=f(d).

[0042] The present invention also provides a shield tunnel irregular load inversion system based on a neural network agent model, which is characterized in that it includes a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0043] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0044] Compared with the prior art, the present invention has the following beneficial effects: the present invention abandons the assumption of load distribution pattern in traditional lining load inversion method, and the inversion result can reveal the irregular load distribution pattern of the on-site tunnel, providing technical guarantee for the safe service of the tunnel; the inversion is based on the full-circle deformation of the lining obtained by three-dimensional laser scanning technology, which significantly increases the amount of on-site data. As n increases, the dimension of the deformation vector can be fully increased, avoiding the defect that the traditional inversion method easily leads to non-unique inversion results. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The flowchart of the lining load inversion algorithm of the present invention is shown;

[0046] Figure 2 This is a schematic diagram of the cross section of a subway shield tunnel;

[0047] Figure 3 Refine the three-dimensional finite element numerical model for the tunnel lining;

[0048] Figure 4 is the constitutive relationship between the segments and bolts in the finite element model;

[0049] Figure 5 For training, testing and verification of BP neural network;

[0050] Figure 6 It is the deformation monitoring and load inversion results of on-site shield tunnel. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0052] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0053] 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.

[0054] like Figure 1 As shown, the present invention provides a shield tunnel irregular load inversion method based on a neural network proxy model, comprising:

[0055] Step S1, establishing a refined three-dimensional finite element model of the shield tunnel lining according to the geometric dimensions, joint structure and material mechanical parameters of the actual tunnel project;

[0056] Step S2: Divide the lining into n equal parts according to the circumference to form n load inversion target points and form an N-dimensional load inversion vector P = (P1, P2, ..., P n ); randomly generate m load vectors P based on the Monte Carlo method;

[0057] Step S3, applying the load determined in step S2 to the lining, inputting it into the finite element model established in step S1, calculating the full-circle deformation of the lining corresponding to each load vector P; dividing the lining deformation into 2n equal parts of the circumference, obtaining the lining deformation of each equally divided point, and forming m 2n-dimensional deformation vectors D;

[0058] Step S4, selecting a neural network proxy model, taking the deformation vector D as the model input, taking the load vector P as the model output, and training and verifying the neural network proxy model with the m samples obtained in step S3;

[0059] Step S5, using three-dimensional laser scanning technology to obtain the full-circle deformation of the actual tunnel; according to the division method in step S3, determine the lining deformation vector d at the engineering site;

[0060] Step S6: using the neural network proxy model trained in step S4, based on the on-site lining deformation vector, calculate the load vector p actually acting on the lining.

[0061] Taking the current subway shield tunnel as an example, the relevant geometric dimensions are as follows: Figure 2 As shown. The tunnel outer diameter is 3.1m, the lining thickness is 0.35m, and the ring width is 1.2m. The lining ring is assembled from 6 segments, including 1 capping block (F block), 2 adjacent blocks (L1 block and L2 block), 2 standard blocks (B1 block and B2 block) and 1 bottom block (D block). The segments are connected by straight bolts.

[0062] Step S1: According to the geometric dimensions provided by the on-site tunnel, the following is established in the finite element analysis software ABAQU: Figure 3 The detailed numerical model shown in the figure simulates the joint bolts and bolt hand holes. In this three-dimensional finite element model, the lining is made of concrete and Figure 4 The three-fold strain-strain relationship shown in (a) is that the relevant mechanical parameters are the elastic modulus E1 of concrete, the yield stress σ b , plastic modulus E r , ultimate stress σ c and the ultimate strain ε d The stress-strain relationship of the bolt is as follows: Figure 4 (b) The related mechanical parameters are elastic modulus E0, yield stress σ s , plastic modulus E r , ultimate strain ε c (or ultimate stress σ c Penalty functions are used to simulate friction contact between segments and between bolt holes and bolts. The friction coefficient between segments is 0.4, and the friction coefficient between bolt holes and bolts is 0.1. C3D8 solid elements are used for both segments and bolts.

[0063] Step S2: Taking the tunnel crown as θ=0 (0°), counterclockwise rotation as positive, and the tunnel base as θ=π (180°), the lining is divided into 36 equal parts to form a 36-dimensional load inversion vector P=(P1,P1,…,P 36 ) and the corresponding position coordinate vector θ=(0,18 / π,9 / π,…,π,…,35 / 18π). For adjacent load inversion target points, the form y=ax 3 +bx 2 +cx+d cubic spline interpolation, according to the continuity and smoothness requirements of the spline function, the difference coefficients of each piecewise function are a1, b1, c1, d1, ..., a 36 , b 36 , c 36 , d 36 The following 144 equations must be satisfied:

[0064]

[0065]

[0066]

[0067]

[0068] Solving the above equations yields 144 spline interpolation coefficients.

[0069] Step S3: Randomly generate 500 36-dimensional load vectors P = (P1, P1, ..., P 36 ), each load component P i It obeys the uniform distribution of (0kN, 100kN). The load at any position of the lining is calculated according to the cubic spline interpolation results, and the above load is applied in the finite element numerical model. The finite element model is iterated until convergence to obtain the full-circle deformation of the inner contour of the tunnel lining under the given load. The lining ring is divided into 360 equal parts to obtain a 360-dimensional deformation vector D. The finite element model is calculated 500 times to obtain 500 deformation vectors D. The horizontal displacement of the inner surface of the lining is selected here.

[0070] Step S4, using the 360-dimensional deformation vector D as input and the 36-dimensional load vector P as output, construct a BP neural network proxy model, and train the BP neural network with the 500 samples obtained in step S3 to obtain a proxy model P=f(D) for predicting the load vector P using the lining deformation vector D.

[0071] In the BP neural network in step S4 above, the number of hidden layer neurons is set to 20, the activation function is the hyperbolic tangent function (Tanh function), the output layer activation function is the linear function (purelin function), the mean square error (MSE) between the predicted load vector and the sample load vector is used as the loss function, and the gradient descent algorithm is used for training. After 178 iterations, the training target is reached. The training, testing and verification of the BP neural network are as follows: Figure 5 shown.

[0072] In step S5, the contour point cloud data of the on-site tunnel is obtained by using 3D laser scanning technology, and the inner contour of the lining in the current state is obtained after processing. By comparing with the inner contour data of the tunnel just after construction, the full-circle deformation of the tunnel in the current state is obtained, and the horizontal displacement distribution of the lining is as follows: Figure 6 (a) As shown in FIG. 1 , according to the equally divided positions in step S3, the horizontal displacement vector d of the on-site tunnel is obtained.

[0073] In step S6, according to the proxy model P=f(D) trained in step S4, the on-site deformation vector d is input to obtain the load vector p=f(d) of the on-site tunnel. The tunnel load distribution obtained by inversion is as follows: Figure 6 (b) as shown.

[0074] The present invention also provides a shield tunnel irregular load inversion system based on a neural network agent model, which is characterized in that it includes a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0075] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0076] 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.

[0077] 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.

[0078] 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 comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0079] 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 The steps for the functions specified in one or more boxes.

[0080] 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.

Claims

1. A method for inversion of irregular loads in shield tunnels based on a neural network proxy model, characterized in that: include: Step S1, establishing a refined three-dimensional finite element model of the shield tunnel lining according to the geometric dimensions, joint structure and material mechanical parameters of the actual tunnel project; Step S2: Divide the lining into n equal parts according to the circumference to form n load inversion target points and form an N-dimensional load inversion vector P = (P1, P2, ..., P n ); randomly generate m load vectors P based on the Monte Carlo method; Step S3, applying the load determined in step S2 to the lining, inputting it into the finite element model established in step S1, calculating the full-circle deformation of the lining corresponding to each load vector P; dividing the lining deformation into 2n equal parts of the circumference, obtaining the lining deformation of each equally divided point, and forming m 2n-dimensional deformation vectors D; Step S4, selecting a neural network proxy model, taking the deformation vector D as the model input, taking the load vector P as the model output, and training and verifying the neural network proxy model with the m samples obtained in step S3; Step S5, using three-dimensional laser scanning technology to obtain the full-circle deformation of the actual tunnel; according to the division method in step S3, determine the lining deformation vector d at the engineering site; Step S6: using the neural network proxy model trained in step S4, based on the on-site lining deformation vector, calculate the load vector p actually acting on the lining.

2. According to claim 1, a method for inversion of irregular loads in shield tunnels based on a neural network proxy model is characterized in that: In step S1, the characteristics of the three-dimensional finite element numerical model of the tunnel lining are as follows: a geometric model is established according to the diameter, lining thickness and lining segmentation of the actual tunnel project; the joints of different segment segments are connected by bolts; friction contact is adopted between the segments; the constitutive model of the concrete lining adopts a three-fold line elastic-plastic model, and the mechanical parameters are selected according to the concrete grade; the joint bolts adopt a double-fold line elastic-plastic model, and the mechanical parameters are selected according to the bolt model.

3. According to claim 1, a method for inversion of irregular loads in shield tunnels based on a neural network proxy model is characterized in that: In step S2, the loads between the inversion target points are obtained using the cubic spline interpolation method.

4. According to claim 3, a method for inversion of irregular loads in shield tunnels based on a neural network proxy model is characterized in that: In step S2, the load acquisition method between the inversion target points is as follows: For the i-th load inversion target point, its polar coordinates are θ i , on which a load P is applied i , forming data points (θ i ,P i ); for adjacent data points, a cubic polynomial y = ax 3 +bx 2 +cx+d for interpolation; according to the continuity requirement, each cubic function must satisfy: … We get 2n equations. According to the requirement of smoothness, we get n equations by taking the first-order derivative of the piecewise function at the connection point: … Then, we can obtain n equations continuously based on the second-order derivative: 6a1θ2+2b1=6a2θ2+2b2 6a2θ3+2b2=6a3θ3+2b3 … 6a n-1 i n +2b n-1 =6a n i n +2b n 6a n θ1+2b n =6a1θ1+2b1 By combining the above 4n equations, we can solve the 4n unknown coefficients a1, b1, c1, d1, ..., a of the n piecewise functions. n , b n , c n , d n .

5. The method for inversion of irregular loads in shield tunnels based on a neural network proxy model according to claim 3 is characterized in that: In step S3, the load at any position of the lining is calculated according to the cubic spline interpolation result, and the corresponding load is applied in the finite element numerical model. The finite element model is iterated until convergence to obtain the full circumferential deformation of the inner contour of the tunnel lining under a given load, and the lining ring is divided into 2n equal parts to obtain a 2n-dimensional deformation vector D; Repeat the finite element model calculation m times to obtain m deformation vectors D.

6. The method for inversion of irregular loads in shield tunnels based on a neural network proxy model according to claim 1 is characterized in that: In step S4, a BP neural network proxy model is constructed with the 2n-dimensional deformation vector D as input and the n-dimensional load vector P as output. The BP neural network proxy model is trained with the m samples obtained in step S3 to obtain a neural network proxy model P=f(D) that predicts the load vector P with the lining deformation vector D.

7. The method for inversion of irregular loads in shield tunnels based on a neural network proxy model according to claim 1 is characterized in that: In step S5, the contour point cloud data of the on-site tunnel is obtained by using three-dimensional laser scanning technology. After processing, the inner contour of the lining in the current state is obtained; by comparing with the inner contour data of the tunnel just after construction, the full-circle deformation of the tunnel in the current state is obtained; according to the equally divided positions in step S3, the deformation vector d of the on-site tunnel is obtained.

8. The method for inversion of irregular loads in shield tunnels based on a neural network proxy model according to claim 1 is characterized in that: In step S6, according to the neural network proxy model P=f(D) trained in step S4, the on-site deformation vector d is input to obtain the load vector p=f(d) of the on-site tunnel.

9. A shield tunnel irregular load inversion system based on a neural network proxy model, characterized in that: The method comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps as claimed in any one of claims 1 to 8 can be implemented.

10. A computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor, and when the processor executes the computer program instructions, the method steps according to any one of claims 1 to 8 can be implemented.

Images