A method and system for regulating internal forces of a tunnel support structure

Abstract

The application discloses a tunnel supporting structure internal force regulation method and system, and the method steps are as follows: multiple sub-models based on the same neural network architecture are constructed to respectively perform different functions; a target and an adjuster are installed in the initial support construction process of a construction section; the cross-section deformation of the construction section is input into a first sub-model to obtain a continuous displacement field of the current construction section in the cross-section direction; a second sub-model obtains a continuous internal force field by analyzing the continuous displacement field, and then a third sub-model generates a regulation scheme according to the continuous internal force field and predicts a predetermined position required for installing the adjuster in the next construction section. The system is used for executing the method. The application has the beneficial effects that: through fewer data points and mechanical principles, the application can inversely obtain the full-field response of the initial support structure, and then cooperating with the corresponding hardware equipment, the internal force field of the tunnel surrounding rock deformation can be adaptively compensated.

Description

Technical Field

[0001] This application relates to the field of tunnel construction technology, and in particular to a method and system for regulating the internal forces of tunnel support structures. Background Technology

[0002] Currently, during tunnel construction, the initial support structure is generally formed by steel arches spliced ​​from I-beams or steel grids spliced ​​from steel trusses, combined with shotcrete to bear the pressure of the surrounding rock. The drawback of this current technology is that it can only passively bear the load of the surrounding rock, providing limited adjustment and lacking deformation control. However, in actual construction, due to varying geological conditions, the deformation degree of the surrounding rock differs at different locations. This easily leads to uneven stress on the initial support structure, causing uneven internal forces in the overall structure, resulting in common quality problems such as cracking, spalling, and water leakage in the lining structure. Summary of the Invention

[0003] One objective of this application is to provide a method for regulating the internal forces of tunnel support structures that can solve at least one of the defects in the aforementioned background art.

[0004] Another objective of this application is to provide a tunnel support structure internal force control system that can solve at least one of the defects in the above-mentioned background art.

[0005] To achieve at least one of the above objectives, the technical solution adopted in this application is: a method for regulating the internal forces of a tunnel support structure, comprising the following steps:

[0006] S100: Based on the structural characteristics of the tunnel and the initial support, multiple sub-models based on the same neural network architecture are constructed, including the first sub-model that performs displacement global inversion, the second sub-model that performs internal force analysis, and the third sub-model that performs regulation response prediction.

[0007] S200: The tunnel is spatiotemporally discretized to obtain multiple continuous construction segments; during the initial support construction of each construction segment, the regulators required for the active adjustment network and the targets required for the displacement monitoring network are installed at predetermined positions.

[0008] S300: During the initial support construction process of the entire construction section, the cross-sectional deformation of the current construction section is continuously collected through the displacement monitoring network, and the collected deformation data is input into the first sub-model to obtain the continuous displacement field of the current construction section.

[0009] S400: The second sub-model obtains the continuous internal force field by analyzing the load distribution field obtained by the inversion of the continuous displacement field. Then, the third sub-model generates the control scheme based on the continuous internal force field and predicts the pre-positioning required for the installation of the regulator in the next construction section.

[0010] S500: The active control network drives the corresponding adjustment amount of the regulators installed at different predetermined positions in the current construction section according to the obtained control scheme.

[0011] Preferably, in step S100, each sub-model adopts the PINN architecture, and the specific construction includes the following process: constructing control equations based on the structural parameters of the initial support material and the structural parameters of the tunnel; defining the boundaries of the control equations based on the tunnel construction method; obtaining the PINN basic model based on the combination of the DNN basic network model and the control equations and boundaries; setting multiple loss functions to optimize the PINN basic model, and obtaining the required first sub-model, second sub-model and third sub-model based on the different types and / or weights of the loss functions required for optimization.

[0012] Preferably, the loss function includes physical equation residual loss, boundary condition residual loss, initial condition residual loss, and real data acquisition residual loss; wherein, the PINN basic model is optimized using physical equation residual loss, boundary condition residual loss, and real data acquisition residual loss to obtain the first sub-model; the PINN basic model is optimized using physical equation residual loss and boundary condition residual loss, and the second and third sub-models are obtained according to the different weights of physical equation residual loss and boundary condition residual loss.

[0013] Preferably, the total loss function L corresponding to the first sub-model tol In the middle, the residual loss L of the physical equation phy Residual loss L of actual collected data true The weights are adaptive weights, and the specific acquisition process is as follows: A trainable noise parameter is automatically learned for each loss function using homoscedasticity uncertainty. Then, the L of the total loss function improved by the first sub-model is... tol The expression is:

[0014] ;

[0015] Where, σ phy and σ true L represents the noise parameters corresponding to the residual loss of the physical equation and the residual loss of the actual acquired data, respectively. bc λ represents the boundary condition residual loss. bc The weights represent the residual loss due to boundary conditions.

[0016] Preferably, when installing the target, multiple points are set at equal intervals along the cross-sectional direction of the current construction section of the tunnel as the predetermined installation positions of the target; wherein, the midpoint of the cross-section is used as one of the points, and the remaining points are symmetrically distributed to both sides of the midpoint of the cross-section.

[0017] Preferably, after obtaining the deformation data at predetermined locations collected by the displacement monitoring network, the first sub-model interpolates the deformation at adjacent predetermined locations along the cross-sectional direction of the construction section to obtain the required continuous displacement field. When the construction length of the tunnel is lower than the set value of the total tunnel length, the initial PINN basic model is trained using the measured data from the initial stage of tunnel construction. The trained initial PINN basic model is used as a physical discriminator and connected in parallel with the traditional discriminator to construct a generative adversarial network discriminator. The control equations for constructing each sub-model are embedded into the generative adversarial network discriminator, thereby suppressing overfitting during the construction of the continuous displacement field by the first sub-model through the generative adversarial network discriminator.

[0018] Preferably, multiple installation positions for installing regulators are set at equal intervals on the arch frame used for the initial support of the construction section; based on the continuous internal force field output by the second sub-model, the third sub-model outputs the peak point of the continuous internal force field; the installation position on the arch frame closest to the peak point of the internal force is selected as the predetermined installation position of the regulator for the next construction section.

[0019] A tunnel support structure internal force control system is provided for implementing the aforementioned tunnel support structure internal force control method. The system includes a displacement monitoring network, an active control network, and an edge analysis module. The displacement monitoring network collects the cross-sectional deformation of the current construction section of the tunnel. The active control network compensates for the continuous internal force field of the current construction section based on the received internal force control signal. The edge analysis module is connected to both the displacement monitoring network and the active control network. The edge analysis module constructs multiple sub-models based on the same neural network architecture. The edge analysis module is adapted to receive the collected data from the displacement monitoring network, generate the internal force control signal based on the constructed sub-models, and send it to the active control network.

[0020] Preferably, the displacement monitoring network includes a displacement acquisition component installed on the drainage board trolley and a target installed on the tunnel construction section; the displacement acquisition component is adapted to move the drainage board trolley and then identify the target installed on the current construction section.

[0021] Preferably, the active control network includes a hydraulic station installed on the drainage board trolley and an adjuster installed on the arch frame required for the initial support construction; the adjuster includes a lifting component and a hydraulic loading component; the lifting component is detachably installed on the arch frame, and is adapted to extend and retract within a set adjustment range, and to mechanically lock when the set extension / retraction amount is reached; the hydraulic loading component is detachably installed on the drive end of the lifting component to drive the lifting component to extend; the hydraulic loading component is adapted to separate and disassemble from the lifting component after the lifting component is mechanically locked; the hydraulic station is adapted to receive the internal force control signal from the edge analysis module, and the hydraulic station is adapted to connect with the hydraulic loading component of the adjuster via an oil circuit, thereby controlling the loading amount of the hydraulic loading component according to the internal force control signal.

[0022] Compared with the prior art, the beneficial effects of this application are as follows:

[0023] Compared to traditional methods, this application uses fewer data points combined with mechanical principles to invert the full-field response of the initial support structure. Then, with the corresponding hardware equipment, it can adaptively compensate for the internal force field of the tunnel surrounding rock deformation, thereby effectively saving equipment costs while ensuring the safety of the initial support structure of the surrounding rock; at the same time, it can also achieve the effect of compensation while tunnel construction is underway. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the working steps of the control method in this application.

[0025] Figure 2 This is a schematic diagram of the PINN architecture for each sub-model in this application.

[0026] Figure 3 This is a simplified structural diagram of the tunnel construction in this application.

[0027] Figure 4 This is a simplified structural diagram of the tunnel cross-section in this application.

[0028] Figure 5 This is a schematic diagram of the regulator in this application.

[0029] In the diagram: Surrounding rock 01, Target 011, Arch frame 02, Installation station 021, Drainage board trolley 03, Displacement acquisition component 04, Hydraulic station 05, Edge analysis module 06, Support component 1, Base 11, Outer tube 12, Lifting component 2, Locking component 3, Hydraulic loading component 4, Hydraulic cylinder 41, Connecting component 42. Detailed Implementation

[0030] The present application will now be further described in conjunction with specific embodiments. It should be noted that, in the description of this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicates that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0031] In the description of this application, it should be noted that the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., which indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of this application.

[0032] It should be noted that the terms "first," "second," etc., in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0033] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0034] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0035] The terms “comprising” and “having”, and any variations thereof, in the specification and claims of this application are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0036] One aspect of this application provides a method for regulating the internal forces of a tunnel support structure, such as... Figure 1 and Figure 4 As shown, one preferred embodiment includes the following steps:

[0037] S100: Based on the structural characteristics of the tunnel and the initial support, multiple sub-models based on the same neural network architecture are constructed, including the first sub-model that performs displacement global inversion, the second sub-model that performs internal force analysis, and the third sub-model that performs regulation response prediction.

[0038] S200: The tunnel is spatiotemporally discretized to obtain multiple continuous construction segments; during the initial support construction of each construction segment, the regulators required for the active adjustment network and the target 011 required for the displacement monitoring network are installed at the predetermined positions.

[0039] S300: During the initial support construction of the entire construction section, the cross-sectional deformation of the current construction section is continuously collected through the displacement monitoring network, and the collected deformation data is input into the first sub-model to obtain the continuous displacement field of the current construction section.

[0040] S400: The second sub-model obtains the continuous internal force field by analyzing the load distribution field obtained by the inversion of the continuous displacement field. Then, the third sub-model generates the control scheme based on the continuous internal force field and predicts the pre-positioning required for the installation of the regulator in the next construction section.

[0041] S500: The active control network drives the corresponding adjustment amount of the regulators installed at different predetermined positions in the current construction section according to the obtained control scheme.

[0042] Understandably, traditional methods for analyzing the deformation of tunnel surrounding rock 01 primarily employ either purely mechanical or purely data-driven approaches. The purely mechanical approach mainly involves numerical analysis of tunnel internal forces using finite element method (FEM) software; however, this method requires manual mesh generation and boundary setting, making it difficult to guide timely adjustments of internal forces during the demanding tunnel construction process. The purely data-driven approach uses deep learning to predict tunnel internal forces based on the required data points; however, this method requires a large number of data points, sometimes necessitating in-depth measurements within the surrounding rock 01. Furthermore, the response of the surrounding rock 01 generally lags behind the initial support construction process, thus failing to guide the adjustment of internal forces in the initial support structure.

[0043] In the technical solution of this application, a data-driven model based on physical equations, namely the first to third sub-models, can be constructed based on the structural characteristics of the tunnel and the initial support. Then, during the initial stage of the initial support construction in the construction section, a target 011 is installed at a predetermined position on the tunnel cross-section to collect deformation data at certain locations of the tunnel cross-section. Substituting the collected data points into the constructed sub-model, and using these data points in conjunction with mechanical principles, the full-field response of the initial support structure can be obtained through inversion. This, combined with appropriate hardware equipment, allows for adaptive compensation of the internal force field of the tunnel surrounding rock 01 deformation. Furthermore, based on the current full-field response of the initial support structure, the number and location of hardware equipment to be installed in the next construction section can be predicted in advance, thereby effectively saving equipment costs while ensuring the safety of the initial support structure of the surrounding rock 01. Simultaneously, it can achieve the effect of compensation during tunnel construction. For ease of understanding, the various steps of the technical solution of this application will be explained in detail below.

[0044] In this embodiment, various types of neural network architectures are used to implement the corresponding functions of each sub-model. The preferred architecture is the PINN architecture, or Physical Information Neural Network architecture, which is an extension of Deep Neural Networks (DNNs). This means that by adding physical rules or constraints to the basic DNN network model, the required PINN basic model can be obtained. The construction of each sub-model in this embodiment includes the following process: Constructing control equations based on the structural parameters of the initial support material and the tunnel's structural parameters. Defining the boundaries of the control equations based on the tunnel's construction method. Obtaining the PINN basic model by combining the basic DNN network model with the control equations and boundaries. Optimizing the PINN basic model by setting various loss functions. Based on the different types and / or weights of the loss functions required for optimization, the required first, second, and third sub-models are obtained.

[0045] Understandably, since tunnel construction is carried out in segments and the physical parameters between different segments may differ, the construction of the PINN basic model can be continuously based on the tunnel construction progress. That is, the PINN basic model is built using the structural parameters of the initial construction segment of the tunnel, and as tunnel construction continues, the PINN basic model is continuously updated using the structural parameters of subsequent construction segments to ensure that the PINN basic model always corresponds to the parameters of the current construction segment.

[0046] Therefore, before constructing the PINN basic model, it is necessary to discretize the tunnel in the spatiotemporal domain based on the tunnel construction requirements. Specifically, appropriate coordinates are selected, and uniform sampling is performed within the tunnel's length domain to achieve spatial domain discretization. Then, each sampling point is discretized in the time domain according to the on-site construction process to obtain the spatiotemporal coordinates of each sampling point. It should be noted that the tunnel segment corresponding to adjacent sampling points is a single construction segment; the length of a single construction segment can be selected according to the actual needs of those skilled in the art. Based on industry experience, the length of a single construction segment can be 10m.

[0047] In this embodiment, for the construction of the sub-model, taking any construction segment as an example, the required parameters first need to be obtained, including the equivalent elastic modulus E of the required initial support material, the cross-sectional thickness A of the initial support, the moment of inertia I of the cross-section, the calculated width B of the initial support, the foundation modulus K of the tunnel, the cross-sectional radius ρ of the tunnel, the central angle θ corresponding to the tunnel cross-section, and the tangential uniformly distributed load distribution function q of the surrounding rock. ρ and q θ Based on the parameters mentioned above, the governing equations are expressed as follows:

[0048] ; .

[0049] Where Q represents the shear force, obtained by differentiating the bending moment M, the bending moment... k represents curvature; N represents axial force. ε represents tangential strain.

[0050] ; .

[0051] Where u represents tangential displacement and w represents radial displacement.

[0052] It is understandable that after obtaining the governing equations, the boundary governing equations can be determined according to the actual working conditions. For ease of understanding, the following will take the construction stage of the drill-and-blast tunnel as an example. Without considering the optimal anchor bolts or the shear stiffness of the lining-surrounding rock 01 interface, the boundary conditions are: u=0, M=0, Q=0. If the tunnel cross section is a semicircle, the boundary range of the central angle θ is [0, π].

[0053] Based on the above governing equations and boundary conditions, we can construct, as follows: Figure 2 The diagram shows a basic DNN network model. This model consists of an input layer x, hidden layers, and an output layer y; where the input layer x represents the coordinates of the sampling points for the deformation of the construction section, x = [ρ, θ]. T The output layer y represents the bidirectional displacement of all sampling points, y = [w, u]. T By combining the basic DNN network model with the governing equations and boundary conditions, a PINN foundation model for an elastic foundation beam can be formed. It's important to know that the loss function corresponding to the PINN foundation model includes the physical equation residual loss L. phy Boundary condition residual loss L bc Initial condition residual loss L ini and the residual loss L of the actual collected data true The specific expressions for each loss function are well-known to those skilled in the art, and therefore will not be elaborated upon here. Based on the four types of loss functions and the weight combinations given, different total loss functions L can be derived. tol The resulting total loss function L tol Backpropagation to the basic DNN network model yields the required first sub-model D-PINN, second sub-model F-PINN, and third sub-model A-PINN.

[0054] Specifically, the PINN base model is optimized using residual losses from physical equations, boundary conditions, and actual acquired data to obtain the first sub-model D-PINN for performing global displacement inversion. That is, the total loss function corresponding to the first sub-model D-PINN. ; where λ phy , λ bc , λ true L represents the residual loss of the physical equations, respectively. phy Boundary condition residual loss L bc and the residual loss L of the actual collected data true The corresponding weights.

[0055] The PINN basic model is optimized using physical equation residual loss and boundary condition residual loss. Based on the different weights of these two residual losses, a second sub-model F-PINN performing internal force analysis and a third sub-model A-PINN performing regulation response prediction are obtained. That is, the total loss function L corresponds to the second sub-model F-PINN and the third sub-model A-PINN. tol The expressions are all: The weights λ are only different for different sub-models. phy and λ bc The values ​​can be different.

[0056] To facilitate understanding, the following is a detailed description of the specific working process of the first sub-model D-PINN, the second sub-model F-PINN, and the third sub-model A-PINN after their construction is completed.

[0057] like Figure 3 As shown, the construction section of the tunnel is represented by D. As the tunnel construction progresses, the construction section continues from D1 to D2. i That is, the current construction section is D. i At this point, the input parameter for the first sub-model D-PINN is the equivalent elastic modulus E of the initial support material. i initial support section thickness A i Moment of inertia I of cross section i Initial support calculation width B i (Generally taken as 1m), foundation modulus K i Monitoring point coordinates x ture and the corresponding displacement data w ture u ture The output of the first sub-model D-PINN is the construction section D. i The uniformly distributed load distribution function q of the cross section ρ (θ ij ), q θ (θ ij ), displacement distribution function w(θ) ij ), u(θ) ij ), θ ij Indicates construction section D i The central angles on the cross section corresponding to different target 011 installation positions; where, the uniformly distributed load distribution function q on the cross section. ρ (θ ij ), q θ (θ ij The displacement distribution function w(θ) ij ), u(θ) ij The result is obtained through inversion.

[0058] The input parameter for the second sub-model F-PINN is the equivalent elastic modulus E of the initial support material. iinitial support section thickness A i Moment of inertia I of cross section i Initial support calculation width B i (Generally taken as 1m), foundation modulus K i Construction section D i The uniformly distributed load distribution function q of the cross section ρ (θ ij ), q θ (θ ij The output of the second sub-model F-PINN is construction section D. i The internal force distribution function N(θ) of the cross section ij ), M(θ) ij ), Q(θ) ij ).

[0059] The input parameter for the third sub-model A-PINN is the equivalent elastic modulus E of the initial support material. i initial support section thickness A i Moment of inertia I of cross section i Initial support calculation width B i (Generally taken as 1m), foundation modulus K i Construction section D i The internal force distribution function N(θ) of the cross section ij ), M(θ) ij ), Q(θ) ij ), and the displacement distribution function w(θ) ij ), u(θ) ij The output of the third sub-model A-PINN is the magnitude of the regulating load of the regulators installed at each predetermined location, and the predicted value of the next construction segment D. i+1 The corresponding installation location of the regulator.

[0060] In this embodiment, as can be seen from the aforementioned working process of each sub-model, the first sub-model D-PINN is the first link in the entire working process. The accuracy of the output results of the first sub-model D-PINN directly affects the accuracy of the results of the subsequent second sub-model F-PINN and third sub-model A-PINN. Since the tunnel is still under construction during the data acquisition process of the first sub-model D-PINN, data jumps may occur during construction. For example, when blasting is used, the vibration generated by the blasting will cause a sudden change in the displacement data of the cross-sectional monitoring points of the current construction section. This sudden change will lead to a loss of L in the actual acquired data residual. true The corresponding weight λ true The contamination forces the first sub-model D-PINN to fit incorrect data, resulting in reduced output accuracy. Therefore, in constructing the total loss function L corresponding to the first sub-model D-PINN... tol At that time, the residual loss of the physical equation Lphy Residual loss L of actual collected data true The weights are adaptive, which can adaptively change the weight coefficients when there are sudden changes in the monitoring point data to ensure the accuracy of the output results.

[0061] Specifically, for the residual loss L of the physical equation phy Residual loss L of actual collected data true The process of obtaining the adaptive weights is as follows: using homoscedastic uncertainty to adjust the residual loss L of the physical equations respectively. phy Residual loss L of actual collected data true Automatically learn a trainable noise parameter σ phy and σ true This causes each residual loss term to automatically reduce its confidence level for high-noise data, thus the L of the total loss function improved by the first sub-model D-PINN is... tol The expression is:

[0062] .

[0063] It is understandable that the noise parameter σ phy and σ true It can be updated via backpropagation along with other weight coefficients, eliminating the need for manual parameter tuning. Based on the improved total loss function of the first sub-model D-PINN, when a sudden change occurs in the actual collected data corresponding to a certain batch of monitoring points, the corresponding noise parameter σ... true This will increase, thus leading to a loss of L in the actual collected data residuals. true The corresponding weighting coefficients are reduced, automatically minimizing the impact of observations on the output accuracy of the first sub-model, D-PINN. If the residuals of the physical equations continue to increase, the corresponding noise parameter σ... phy It will also continue to increase, causing the residual loss L of the physical equations to rise. phy The corresponding weighting coefficients decrease.

[0064] In this embodiment, the monitoring points required by the first sub-model D-PINN are the locations where the target 011 is installed on the surrounding rock 01 during the initial support construction of the entire construction section. Therefore, the location of the monitoring point is the displacement of the target 011, and also the deformation of the surrounding rock 01 corresponding to the target 011 installation point. Thus, the target 011 needs to be installed on the surrounding rock 01 at the initial stage of the initial support construction to ensure that the total deformation of the surrounding rock 01 during the entire initial support construction process can be monitored. Figure 4 As shown, when installing target 011, multiple points are set at equal intervals along the cross-sectional direction of the current construction section of the tunnel as the predetermined installation positions of target 011; among them, the midpoint of the cross-section is used as one of the points, and the remaining points are symmetrically distributed to both sides of the midpoint of the cross-section.

[0065] It is important to understand that the specific number and location of target 011 can be set according to the actual needs of those skilled in the art. The more target 011s installed, the more real data is acquired from the corresponding monitoring points, resulting in more accurate output from the first sub-model D-PINN. However, excessive target 011 installations can lead to data redundancy. Therefore, in this embodiment, five target 011s are preferably installed. One installation point corresponds to the highest point of the construction section's cross-section, i.e., the midpoint. The remaining four points are grouped into two sets, located on either side of the highest point of the cross-section. Extending outwards from the midpoint of the construction section's cross-section, the central angle between adjacent target 011 installation points is 20°~30°.

[0066] Understandably, after the first sub-model D-PINN obtains the deformation data at predetermined locations collected by the displacement monitoring network, it interpolates the deformation at adjacent predetermined locations along the cross-sectional direction of the construction section to obtain the required continuous displacement field, i.e., the displacement distribution function w(θ). ij ), u(θ) ij However, in the early stages of tunnel construction, especially during the initial support erection of the first construction segment, the first sub-model D-PINN acquires the displacement data of target 011 for the first time. Interpolating the continuous displacement field based on this data can easily lead to overfitting. As tunnel construction progresses and the number of construction segments increases, the first sub-model D-PINN can iteratively update the fitting process of the continuous displacement field based on the continuously acquired displacement data of target 011 from different construction segments, ensuring that the fitted continuous displacement field meets the accuracy requirements. Therefore, in the early stages of tunnel construction, when the tunnel construction length is less than the set value of the total tunnel length (e.g., less than 20% of the total tunnel length), it is necessary to correct the fitting process of the continuous displacement field acquired by the first sub-model D-PINN to ensure the output accuracy of the first sub-model D-PINN.

[0067] In this embodiment, the modification method for fitting the continuous displacement field obtained by the first sub-model D-PINN in the initial stage of tunnel construction is as follows: An initial PINN basic model is trained using measured data from the initial stage of tunnel construction; the trained initial PINN basic model is then used as a physical discriminator and connected in parallel with a traditional discriminator to construct a generative adversarial network (GAN) discriminator. The control equations for constructing each sub-model are embedded into the GAN discriminator, thereby suppressing overfitting during the construction of the continuous displacement field in the first sub-model D-PINN. For ease of understanding, detailed explanations using specific parameters will follow.

[0068] Specifically, the initial PINN basic model was trained using measured data from the first construction segment, i.e., 10m of tunnel construction. This initial PINN basic model served as the physical discriminator, which was then connected in parallel with a traditional discriminator to obtain a generative adversarial network (GAN) discriminator. For every 5m of tunnel construction progress, the GAN discriminator generated multiple sets of pseudo-displacement sample data to fit the displacement continuity field. Simultaneously, the obtained pseudo-displacement sample data was mixed 1:1 with the collected real data to form a training set, which was used to train and update the GAN discriminator.

[0069] In this embodiment, as Figure 4 As shown, multiple installation positions 021 for installing regulators are set at equal intervals on the arch frame 02 used for the initial support of the construction section; according to the continuous internal force field output by the second sub-model F-PINN, the third sub-model A-PINN outputs the peak points of the continuous internal force field; the installation position 021 on the arch frame 02 closest to the peak points of the internal force is selected as the predetermined installation position of the regulator for the next construction section.

[0070] Understandably, during the initial support erection process, arch frame 02 supports the entire surrounding rock 01. Since arch frame 02 and surrounding rock 01 are closely fitted and supported by adjusters, the gap between them is relatively small. Therefore, the adjusters are often installed before arch frame 02 is erected; that is, the adjusters are first installed on arch frame 02, and then the arch frame 02 with the adjusters installed is used for internal force field compensation of the surrounding rock 01 corresponding to the current construction section. To improve tunnel construction efficiency, the adjusters are installed during the initial support erection process. The first sub-model D-PINN outputs the continuous displacement field just as the initial support erection is completed; that is, the adjuster installation precedes the output of the first sub-model D-PINN. Therefore, the installation position of the adjusters needs to be predicted in advance. During the entire tunnel construction process, the deformation trend of the surrounding rock 01 along the tunnel construction direction changes gradually, with virtually no abrupt changes. Therefore, the installation position of the adjusters for the next construction section can be predicted using the third sub-model A-PINN based on the monitoring data of the current construction section.

[0071] It's important to understand that the arch frame 02 has multiple installation positions 021 along the tunnel cross-section. When installing the regulator, it's not necessary to install it at every installation position 021. For the deformation of the surrounding rock 01, placing the regulator precisely at the location of maximum deformation in the surrounding rock 01 yields the best deformation compensation effect. Therefore, the peak internal force point of the surrounding rock 01 can be obtained in the internal force field output by the second sub-model F-PINN. Since the installation position of the arch frame 02 and the location of the installation positions 021 are fixed, the installation positions 021 on the arch frame 02 may not correspond to the peak internal force point of the surrounding rock 01. Therefore, the regulator can only be installed at the installation position 021 closest to the peak internal force point to ensure that deformation compensation of the surrounding rock 01 is achieved with the minimum number of regulators.

[0072] Another aspect of this application provides a tunnel support structure internal force control system for implementing the aforementioned tunnel support structure internal force control method, such as... Figure 4 As shown, one preferred embodiment includes a displacement monitoring network, an active control network, and an edge analysis module 06. The displacement monitoring network is used to collect the cross-sectional deformation of the current construction section of the tunnel; the active control network compensates for the continuous internal force field of the current construction section based on the received internal force control signals; the edge analysis module 06 is connected to both the displacement monitoring network and the active control network, and is used to construct multiple sub-models based on the same neural network architecture. The edge analysis module 06 can receive the data collected by the displacement monitoring network, and then generate internal force control signals based on the constructed sub-models and send them to the active control network.

[0073] Specifically, such as Figure 4 As shown, the displacement monitoring network includes a displacement acquisition component 04 installed on the drainage board trolley 03 and a target 011 installed on the tunnel construction section; the displacement acquisition component 04 can move the drainage board trolley 03 to identify the target 011 installed on the current construction section.

[0074] Understandably, based on the number of monitoring points, the number of targets 011 can be set to five, spaced out along the midpoint of the cross-section and both sides. The displacement acquisition component 04 uses a camera or radar; the specific number of components 04 is determined according to the tunnel construction conditions. After the tunnel construction is completed and the initial support is erected, targets 011 are installed at predetermined positions in the surrounding rock 01, and the displacement changes of targets 011 reflect the local deformation of the surrounding rock 01. After data acquisition for each construction segment is completed, targets 011 can be removed for installation in the next segment.

[0075] Specifically, such as Figure 4 and Figure 5As shown, the active control network includes a hydraulic station 05 installed on the drainage board trolley 03 and an adjuster installed on the arch frame 02 required for initial support construction. The adjuster includes a support component 1, a lifting component 2, a locking component 3, and a hydraulic loading assembly 4. The support component 1 and the lifting component 2 can cooperate to form a lifting assembly, which is detachably installed on the arch frame 02. The lifting assembly can extend and retract within a set adjustment range and is mechanically locked by the locking component 3 when the set extension / retraction amount is reached. The hydraulic loading assembly 4 is detachably installed on the drive end of the lifting assembly to drive the lifting assembly to extend. The hydraulic loading assembly 4 can be separated and disassembled from the lifting assembly after the lifting assembly is mechanically locked. The hydraulic station 05 can receive the internal force control signal from the edge analysis module 06. The hydraulic station 05 can be connected to the hydraulic loading assembly 4 of the adjuster via an oil circuit, thereby controlling the loading amount of the hydraulic loading assembly 4 according to the internal force control signal. For the specific structure of the regulator, please refer to the technical solution of Chinese invention patent application with publication number CN120004171A and title: "A composite lifting device based on hydraulic loading and mechanical locking and its usage method".

[0076] It is understood that the support component 1 includes a fixedly connected base 11 and an outer sleeve 12; the base 11 can be fixedly installed on the arch frame 02 by bolts or welding. The lifting component 2 is slidably installed inside the outer sleeve 12, and the lifting component 2 is mainly used to lift and support the surrounding rock 01 to achieve deformation compensation. The hydraulic loading assembly 4 includes a hydraulic cylinder 41 and a connecting assembly 42; the hydraulic cylinder 41 is detachably installed on the base 11 of the support component 1 through the connecting assembly 42 and can drive the lifting component 2 so that when lifting is required, the hydraulic cylinder 41 drives the lifting component 2 to slide along the support component 1 until it abuts against the surrounding rock 01 through the oil output by the hydraulic station 05, and then applies a set lifting force to the surrounding rock 01 according to the control scheme. The locking component 3 can be installed on the base 11 of the support component 1 after the lifting component 2 has completed lifting and lock the position of the lifting component 2. The hydraulic loading assembly 4 can be removed from the support component 1 after the locking component 3 has been locked.

[0077] Specifically, the edge analysis module 06 utilizes pre-deployed sub-models to calculate the displacement changes at each monitoring point on the tunnel section at a fixed frequency (the frequency is determined based on site conditions). This displacement is then converted into a global continuous displacement field using the first sub-model D-PINN, while the internal force changes are analyzed using the second sub-model F-PINN. After displacement acquisition, the latest global continuous displacement field and internal force field are read from the database. The third sub-model A-PINN, which performs regulation response prediction, is then invoked to design the regulation load. The regulation loads required for each preset regulation position are converted into hydraulic pressure signals, which are then connected to the hydraulic loading component 4 of the regulator at the hydraulic station 05. The corresponding lifting force can then be applied to the lifting component 2 of the regulator via the corresponding hydraulic pressure signals.

[0078] The basic principles, main features, and advantages of this application have been described above. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely the principles of this application. Various changes and modifications can be made to this application without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. A method for regulating internal forces of a tunnel support structure, characterized by, The method comprises the following steps: S100: According to the structural characteristics of the tunnel and the primary support, a plurality of sub-models based on the same neural network architecture are constructed, including a first sub-model for performing global inversion of displacement, a second sub-model for performing internal force analysis, and a third sub-model for performing adjustment response prediction; S200: The tunnel is discretized in space and time to obtain a plurality of continuous construction sections; during the construction of the primary support in each construction section, the adjustors required by the active adjustment network and the targets required by the displacement monitoring network are installed at predetermined positions; S300: During the construction of the primary support in the entire construction section, the cross-sectional deformation of the current construction section is continuously collected through the displacement monitoring network, and the collected deformation data is input into the first sub-model to obtain the continuous displacement field of the current construction section; S400: The second sub-model analyzes the load distribution field obtained by the continuous displacement field inversion to obtain a continuous internal force field, and then the third sub-model generates a control scheme according to the continuous internal force field and predicts the predetermined position required for the installation of the adjustor in the next construction section; S500: The active adjustment network drives the adjustors installed at different predetermined positions in the current construction section according to the obtained control scheme; After the first sub-model obtains the deformation data of the predetermined positions collected by the displacement monitoring network, the deformation interpolation of adjacent predetermined positions along the cross-sectional direction of the construction section is performed to obtain the required continuous displacement field; When the construction length of the tunnel is less than a set value of the total length of the tunnel, an initial PINN base model is trained based on the measured data in the initial stage of tunnel construction; the trained initial PINN base model is connected in parallel with the traditional discriminator to construct a generative adversarial network discriminator; The control equation for constructing each sub-model is embedded into the generative adversarial network discriminator, and then in the process of constructing the continuous displacement field by the first sub-model, the generative adversarial network discriminator is used for overfitting suppression.

2. The method of internal force regulation of a tunnel support structure according to claim 1, characterized in that, In step S100, each sub-model adopts a PINN architecture, and the specific construction includes the following process: Constructing a control equation according to the structural parameters of the primary support material and the structural parameters of the tunnel; Giving the boundary of the control equation according to the construction method of the tunnel; Based on the combination of the DNN basic network model and the control equation and the boundary, a PINN base model is obtained; Setting multiple loss functions to optimize the PINN base model, based on the different types of loss functions required for optimization, the required first sub-model, second sub-model and third sub-model are obtained.

3. The method of internal force regulation of a tunnel support structure according to claim 2, wherein, The loss function includes physical equation residual loss, boundary condition residual loss, initial condition residual loss and real collected data residual loss; The PINN base model is optimized by the physical equation residual loss, the boundary condition residual loss and the real collected data residual loss to obtain the first sub-model; The PINN base model is optimized by the physical equation residual loss and the boundary condition residual loss, and according to the different weights of the physical equation residual loss and the boundary condition residual loss, the second sub-model and the third sub-model are obtained respectively.

4. The method of internal force regulation of a tunnel support structure according to claim 3, wherein, In the total loss function L corresponding to the first sub-model tol , the weights of the physical equation residual loss L phy and the real acquisition data residual loss L true are adaptive weights, and the specific acquisition process is as follows: With homoscedastic uncertainty, a trainable noise parameter is automatically learned for each loss function, and the L tol The expression is: ； where σ phy and σ true respectively represent the noise parameters corresponding to the physical equation residual loss and the real acquisition data residual loss, L bc represents the boundary condition residual loss, and λ bc represents the weight of the boundary condition residual loss.

5. The method of internal force regulation of a tunnel support structure according to claim 1, characterized in that, When installing the target, a plurality of point positions are arranged as the installation predetermined positions of the target at equal intervals in the cross section direction of the current construction section of the tunnel; wherein the midpoint position of the cross section is one of the point positions, and the remaining point positions are symmetrically distributed to both sides of the midpoint of the cross section.

6. The method of internal force regulation of a tunnel support structure according to claim 1, wherein, A plurality of installation stations for installing the adjuster are arranged at equal intervals on the arch frame used for building the primary support of the construction section. According to the continuous internal force field output by the second sub-model, the third sub-model outputs the internal force peak point position of the continuous internal force field; and the installation station closest to the internal force peak point position on the arch frame is selected as the installation predetermined position of the adjuster of the next construction section.

7. A system for regulating internal forces of a tunnel support structure, for implementing the method for regulating internal forces of a tunnel support structure according to any one of claims 1 to 6, characterized in that, It comprises: a displacement monitoring network; the displacement monitoring network is used to collect the cross section deformation of the current construction section of the tunnel; an active regulation network; the active regulation network compensates the continuous internal force field of the current construction section according to the received internal force regulation signal; and an edge analysis module; the edge analysis module is respectively connected with the displacement monitoring network and the active regulation network, and the edge analysis module is used to construct a plurality of sub-models based on the same neural network architecture; the edge analysis module is adapted to receive the collected data of the displacement monitoring network, and then generate the internal force regulation signal according to the constructed sub-models and send it to the active regulation network.

8. The system for internal force regulation in tunnel support structures according to claim 7, characterized in that, The displacement monitoring network comprises a displacement collection component installed on the drain plate trolley and a target installed on the construction section of the tunnel; the displacement collection component is adapted to move the drain plate trolley, and then identify the target installed on the current construction section.

9. The system for internal force regulation in tunnel support structures according to claim 7, characterized in that, The active regulation network comprises a hydraulic station installed on the drain plate trolley and an adjuster installed on the arch frame required for building the primary support; The adjuster comprises: a jacking assembly; the jacking assembly is detachably installed on the arch frame, and the jacking assembly is adapted to stretch within a set adjustment range and mechanically locked when reaching a set stretching amount; a hydraulic loading assembly; the hydraulic loading assembly is detachably installed on the driving end of the jacking assembly to drive the jacking assembly to stretch out; the hydraulic loading assembly is adapted to be separated and detached from the jacking assembly after the jacking assembly is mechanically locked; The hydraulic station is adapted to receive the internal force regulation signal of the edge analysis module, and the hydraulic station is adapted to be connected with the hydraulic loading assembly of the adjuster through an oil circuit, and then control the loading amount of the hydraulic loading assembly according to the internal force regulation signal.

Images