Intelligent prediction method and equipment for shield excavation face stability based on graph network learning

Through the graph network learning model, the shield parameter network is constructed using Hilbert-Huang transform and partial correlation coefficient, which solves the problem of judging the stability of excavation surface in shield excavation, realizes intelligent prediction of the stability of the shield excavation surface, and improves construction Efficiency and safety.

CN115422971BActive Publication Date: 2024-11-05HUAZHONG UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211045510.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-30
Publication Date
2024-11-05
Estimated Expiration
2042-08-30

AI Technical Summary

Technical Problem

It is difficult to accurately judge the stability of the excavation surface during the shield excavation process, resulting in high safety risks. Lag-down adjustments by the prior art can easily lead to ground ups and collapse.

Method used

Using a graph network learning method, a shield parameter network diagram is constructed through Hilbert-Huang transformation and partial correlation coefficient, and a graph network learning model is trained to realize intelligent prediction of the stability of the shield excavation surface.

Benefits of technology

The shield excavation construction efficiency and construction quality are improved, and the excavation surface stability can be quickly and accurately judged, and the safety risks are reduced. The prediction results suitable for multi-pressure cabin shield machines are more accurate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115422971B_ABST
    Figure CN115422971B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for intelligent prediction of shield excavation face stability based on graph network learning, which belongs to the field of shield tunneling safety. The method comprises: S1: obtaining pressure data of a shield pressure chamber, and characterizing the real-time shield tunneling stability according to Hilbert-Huang transformation; S2: constructing a shield parameter partial correlation threshold network based on the partial correlation coefficients between various shield parameters to reduce the dimension of shield parameters and visualize the relationship between various shield parameters; S3: constructing a graph network learning model of shield excavation face stability and shield parameter network characteristics and training it using a graph network learning algorithm; S4: predicting shield excavation face stability using a graph network learning model. The present invention can improve the efficiency and construction quality of shield tunneling construction, and improve the stability of the shield tunneling excavation face.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of shield tunneling safety, and more specifically, relates to a method and device for intelligently predicting shield excavation face stability based on graph network learning. Background Art

[0002] During shield tunneling, it is very important to keep the excavation face stable. No matter what type of shield machine is used, the support pressure in the pressure chamber must be balanced with the water and soil pressure in front of the excavation face so that the excavation face can remain stable. During the construction of the tunnel, if the support pressure is insufficient, it may cause the surface to collapse, and if the support pressure is too high, it will cause the surface to rise.

[0003] The current slurry and earth pressure balance shield machines all have pre-set slurry or soil compartment pressure values, which are then adjusted according to the excavation face pressure conditions during excavation to maintain the stability of the excavation face. This delayed adjustment can easily lead to ground uplift or collapse at the machine head.

[0004] Since shield tunneling is carried out deep underground, it is difficult to accurately and comprehensively understand the stability of the excavation face during the excavation process, and the risk of instability cannot be known in the first place, which leads to a higher safety risk in shield tunneling.

[0005] Therefore, there is an urgent need for a method that can quickly and accurately determine the stability of the shield excavation face. Summary of the invention

[0006] In view of the above defects or improvement needs of the prior art, the present invention provides an intelligent prediction method for shield excavation face stability based on graph network learning, which aims to intelligently and quickly predict the excavation face stability according to shield parameters to prevent the excavation face from becoming unstable.

[0007] To achieve the above object, according to one aspect of the present invention, a method for intelligently predicting shield excavation face stability based on graph network learning is provided, comprising:

[0008] Offline training phase:

[0009] S1: Perform Hilbert-Huang transform on the shield pressure signal of the shield pressure chamber and use DS to characterize the real-time shield tunneling stability:

[0010]

[0011] Wherein, DS is the non-stationarity of the shield pressure signal, H(ω,t) is the Hilbert time-frequency spectrum of the shield pressure signal, ω represents the frequency domain, B(ω) is the average marginal spectrum of the shield pressure signal, T represents the time axis of the Hilbert time-frequency spectrum integration, and t represents time;

[0012] S2: Collect the time series data of various shield parameters, obtain the weighted directed connections between the shield parameters based on the partial correlation coefficients between the shield parameters, construct the shield parameter partial correlation threshold network, and obtain the shield parameter network diagram; among them, the method for obtaining the weighted directed connections between the shield parameters based on the partial correlation coefficients between the shield parameters is as follows:

[0013] For any three non-repeated shield parameters X, Y, and Z, the weight of the directed connection Z→X between two shield parameters in the shield parameter network diagram is determined by the following formula:

[0014] d(X,Y:Z)≥<d(X,Y:Z)> Z +kσ Z (d(X,Y:Z))

[0015] The weight is set to the number of Ys that satisfy the above formula, where

[0016] d(X,Y:Z)=ρ(X,Y)-ρ(X,Y:Z)

[0017] <d(X,Y:Z)> Z and σ Z (d(X,Y:Z)) are the mean and standard deviation of d(X,Y:Z) for all shield parameters except X and Y; ρ(X,Y) is the correlation coefficient between X and Y, and ρ(X,Y:Z) is the partial correlation coefficient between X and Y excluding the influence of Z;

[0018] S3: constructing a graph network learning model of shield excavation face stability and shield parameter network characteristics, training the graph network learning model with the shield parameter network graph obtained in step S2 as input and the DS value as output, and obtaining an intelligent prediction model of shield excavation face stability based on graph network learning;

[0019] Online prediction stage:

[0020] S4: Collect the time series data of various shield parameters, process them according to step S2 to obtain the shield parameter network diagram, and then input it into the shield excavation face stability intelligent prediction model based on graph network learning to obtain the DS value to predict the shield excavation face stability.

[0021] Further, for a shield machine having multiple pressure chambers, S1 includes:

[0022] S11: extracting the pressure chamber data on the shield excavation face pressure sensor collected by the shield machine during the excavation process, and obtaining the shield pressure signal corresponding to each pressure chamber;

[0023] S12: using the Hilbert-Huang transform method to process the shield pressure signal to obtain the non-stationarity of each pressure chamber data, i.e., the value of DS;

[0024] S13: weighted sum of the DS values ​​of the shield pressure signals of each pressure chamber to obtain the stability of the shield excavation face.

[0025] Further, the S12 includes:

[0026] The shield pressure signal X(t) obtained in step S11 is subjected to empirical mode decomposition to obtain the basic mode components of X(t):

[0027]

[0028] Among them, c1, c2, …, c i ,…,c M is the fundamental mode component, r M is the residual component, t is the time parameter, and M is the total number of basic mode components.

[0029] Furthermore, the graph network learning model in S3 includes a network graph updating module and a DS conversion module. The network graph updating module is used to iteratively update the input shield parameter network graph according to steps S31-S32. The DS conversion module is used to convert the iterated shield parameter network graph into DS. The loss function of the DS conversion module is calculated according to step S33:

[0030] S31: For node n in the shield parameter network diagram, construct a state function x based on its network adjacency information n , f w is a local transfer function, which represents the dependency of node n on its adjacent nodes, then:

[0031] x n =f w (l n ,l co[n] ,x ne[n] ,l ne[n] )

[0032] Among them, ln, l co[n] 、x ne[n] , l ne[n] They represent node n, i.e., the type of shield parameters, the edge of node n, i.e., the partial correlation between shield parameters, the state of node n, i.e., the time series data of shield parameters, and the adjacent nodes of node n. The adjacent nodes represent nodes connected by edges, i.e., the partial correlation coefficient between nodes is greater than the threshold.

[0033] S32: For the node state function x n, calculate the iteration of the state function:

[0034] x n (m+1)=F w (xn(m),l)

[0035] Among them, x n (m) is x n The mth iteration of F w f w The global transition function extended to all nodes until the state x n(m) Iterate until a fixed point solution x is obtained, and then proceed to step S33;

[0036] S33: Calculate the loss function e w and its gradient

[0037]

[0038] in, is the conversion function that converts the iterated shield parameter network diagram into DS; w is Internal reference; DS i is the shield stability of the i-th sample under the shield tunneling state, which is the value actually calculated in step S1; G i is the adjacency matrix describing the shield parameter network graph structure of the i-th sample, X i It is the characteristic matrix obtained after iteration of the shield parameter network diagram of the i-th sample, which is composed of the node states after iteration.

[0039] In order to achieve the above-mentioned purpose, the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the intelligent prediction method for the stability of the shield excavation face as described in any of the preceding items is implemented.

[0040] In order to achieve the above-mentioned purpose, the present invention further provides a device for intelligently predicting the stability of a shield excavation face, comprising a computer-readable storage medium as described above and a processor, wherein the processor is used to call and process a computer program stored in the computer-readable storage medium.

[0041] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0042] 1. The present invention calculates the stability of the shield excavation face by collecting the pressure data of the shield pressure chamber, and uses the shield parameter network to describe the correlation between the shield parameters and visualize them. On this basis, the graph network learning technology is used to realize the intelligent prediction of the stability of the shield excavation face, which can improve the efficiency and construction quality of the shield excavation construction and improve the stability of the shield excavation face.

[0043] 2. For a multi-pressure chamber shield machine, the present invention calculates the DS values ​​separately through the Hilbert-Huang transformation method and performs weighted summation to improve the accuracy of the prediction results.

[0044] 3. First, the shield parameter network diagram is iteratively updated until the fixed point solution is obtained, and then the conversion function between the iteratively updated shield parameter network diagram and DS is trained. The purpose is: based on the fixed point theory, the shield parameter network diagram is iteratively updated so that the information contained in it can be converged during the transmission process, thereby integrating the information of all nodes and comprehensively processing the information in the shield parameter network diagram from a more systematic perspective, ensuring that the DS obtained after pooling and dimensionality reduction through the DS conversion module is more accurate.

[0045] 4. The present invention proposes an intelligent prediction device for shield excavation face stability based on graph network learning. Based on graph network learning technology, it can use the big data recorded and stored by the machine itself to guide the machine driver, thereby improving the shield machine driver and management personnel's correct understanding of the excavation face stability, and has certain guiding significance for the intelligent and unmanned development of shield machines. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of a shield excavation face stability intelligent prediction method provided by the present invention;

[0047] Figure 2 It is a flow chart of shield parameter network construction in the present invention;

[0048] Figure 3 It is the realization of the shield parameter network model in the present invention;

[0049] Figure 4 It is a schematic diagram of the shield data export interface in a specific implementation case of the present invention;

[0050] Figure 5 It is the original signal, the basic mode component and the residual component after EMD decomposition in the specific implementation case of the present invention;

[0051] Figure 6 is the Hilbert spectrum of each IMF component in a specific implementation case of the present invention;

[0052] Figure 7 yes Figure 6The Hilbert spectrum of the shield pressure signal obtained by summing;

[0053] Figure 8 yes Figure 7 The Hilbert marginal spectrum. DETAILED DESCRIPTION

[0054] 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.

[0055] like Figure 1 As shown, the present invention provides an intelligent prediction method for shield excavation face stability based on graph network learning, comprising:

[0056] Offline training phase:

[0057] S1: Perform Hilbert-Huang transform on the shield pressure signal of the shield pressure chamber and use DS to characterize the real-time shield tunneling stability:

[0058]

[0059] Wherein, DS is the non-stationarity of the shield pressure signal, H(ω,t) is the Hilbert time-frequency spectrum of the shield pressure signal, ω represents the frequency domain, B(ω) is the average marginal spectrum of the shield pressure signal, T represents the time axis of the Hilbert time-frequency spectrum integration, and t represents time;

[0060] S2: Collect various shield parameters, obtain the weighted directed connections between shield parameters based on the partial correlation coefficients between the shield parameters, construct the shield parameter partial correlation threshold network, and obtain the shield parameter network diagram; among them, Figure 2 As shown in Figure 2, the method for obtaining the weighted directed connection between shield parameters based on the partial correlation coefficient between shield parameters is as follows:

[0061] For any three non-repeated shield parameters X, Y, and Z, the weight of the directed connection Z→X between two shield parameters in the shield parameter network diagram is determined by the following formula:

[0062] d(X,Y:Z)≥<d(X,Y:Z)> Z +kσ Z (d(X,Y:Z))

[0063] The weight is set to the number of Ys that satisfy the above formula, where

[0064] d(X,Y:Z)=ρ(X,Y)-ρ(X,Y:Z)

[0065] <d(X,Y:Z)> Z and σ Z (d(X,Y:Z)) are the mean and standard deviation of d(X,Y:Z) for all shield parameters except X and Y; ρ(X,Y) is the correlation coefficient between X and Y, and ρ(X,Y:Z) is the partial correlation coefficient between X and Y excluding the influence of Z;

[0066] S3: constructing a graph network learning model of shield excavation face stability and shield parameter network characteristics, training the graph network learning model with the shield parameter network graph obtained in step S2 as input and the DS value as output, and obtaining an intelligent prediction model of shield excavation face stability based on graph network learning;

[0067] Online prediction stage:

[0068] S4: Collect the time series data of various shield parameters, process them according to step S2 to obtain the shield parameter network diagram, and then input it into the shield excavation face stability intelligent prediction model based on graph network learning to obtain the DS value to predict the shield excavation face stability.

[0069] Preferably, for a shield machine having multiple pressure chambers, S1 includes:

[0070] S11: extracting the pressure chamber data on the shield excavation face pressure sensor collected by the shield machine during the excavation process, and obtaining the shield pressure signal corresponding to each pressure chamber;

[0071] S12: using the Hilbert-Huang transform method to process the shield pressure signal to obtain the non-stationarity of each pressure chamber data, i.e., the value of DS;

[0072] S13: weighted sum of the DS values ​​of the shield pressure signals of each pressure chamber to obtain the stability of the shield excavation face.

[0073] Preferably, the S12 includes:

[0074] The shield pressure signal X(t) obtained in step S11 is subjected to empirical mode decomposition to obtain the basic mode components of X(t):

[0075]

[0076] Among them, c1, c2, …, c i ,…,c M is the fundamental mode component, r M is the residual component, t is the time parameter, and M is the total number of basic mode components.

[0077] Preferably, the graph network learning model in S3 includes a network graph updating module and a DS conversion module, wherein the network graph updating module is used to iteratively update the input shield parameter network graph according to steps S31-S32, and the DS conversion module is used to convert the iterated shield parameter network graph into DS, and the loss function of the DS conversion module is calculated according to step S33:

[0078] S31: For node n in the shield parameter network diagram, construct a state function x based on its network adjacency information n , f w is a local transfer function, which represents the dependency of node n on its adjacent nodes, then:

[0079] x n =f w (l n ,l co[n] ,x ne[n] ,l ne[n] )

[0080] Among them, ln, l co[n] 、x ne[n] , l ne[n] They represent node n, i.e., the type of shield parameters, the edge of node n, i.e., the partial correlation between shield parameters, the state of node n, i.e., the time series data of shield parameters, and the adjacent nodes of node n. The adjacent nodes represent nodes connected by edges, i.e., the partial correlation coefficient between nodes is greater than the threshold.

[0081] S32: For the node state function x n , calculate the iteration of the state function:

[0082] x n(m+1) =F w (x n(m) ,l)

[0083] Among them, x n(m) is x n The mth iteration of F w f w The global transition function extended to all nodes until the state x n(m) Iterate until a fixed point solution x is obtained, and then proceed to step S33;

[0084] S33: Calculate the loss function e w and its gradient

[0085]

[0086] in, is the conversion function that converts the iterated shield parameter network diagram into DS; w is Internal reference; DS i is the shield stability of the i-th sample under the shield tunneling state, which is the value actually calculated in step S1; G i is the adjacency matrix describing the shield parameter network graph structure of the i-th sample, X i It is the characteristic matrix obtained after iteration of the shield parameter network diagram of the i-th sample, which is composed of the node states after iteration.

[0087] Below, the present invention is described with reference to corresponding data in a specific case implementation process as follows:

[0088] First, the shield parameters are extracted: In order to record and analyze various shield parameters during the excavation process, a real-time shield information monitoring system is used in this case, which can not only display the values ​​of various shield parameters in real time, but also export them for analysis. During shield excavation, the shield real-time information monitoring system records parameter information every 10 seconds, and records as many as 977 shield parameters, covering all aspects of shield excavation. Due to the particularity of the super-large diameter shield, the propulsion cylinder has 6 partitions and the excavation chamber pressure has 5 compartments, which are recorded separately.

[0089] Selection of specific shield parameters: In the cutterhead drive system, the cutterhead torque, cutterhead speed, etc. reflect the actual excavation of the cutterhead, the cutterhead current, the total current of the cutterhead drive and the total torque of the cutterhead drive reflect the shield machine's control over the cutterhead drive, the cutterhead telescopic cylinder stroke is a reflection of the hydraulic system, the left support moment, the torque support left and right pressures are parameters that support the rotation of the cutterhead, these parameters comprehensively reflect the various properties of the cutterhead and its drive, and should be included in the analysis. The main parameters involved in the mud and water circulation, such as mud inlet pressure, mud inlet and mud discharge ratio, working chamber volume flow detection, mud and water circulation pipeline flow detection, mud discharge and mud inlet volume flow detection, etc., all need to be included in the study. In addition, the total excavation volume and the actual excavation volume are a reflection of the excavated soil, and this embodiment is also included in the research parameters. In the shield real-time monitoring system, the permeability of the excavated soil is also recorded, and this parameter is also taken as a representation of the mud and water circulation system and included in the study.

[0090] Table 1- Shield tunneling parameter data

[0091]

[0092] Calculation of DS:

[0093] Based on the shield data collected previously, as described in step S1, the shield signal stability calculation can be performed based on the Hilbert-Huang transform. First, the stability calculation of one ring and one shield pressure signal is performed. Here, the example is the 784th ring of a shield tunneling left line, and the pressure signal is the tunneling chamber pressure 4. First, the tunneling chamber pressure 4 is extracted into a column of data, and the EMD decomposition is performed on it to obtain the curves of the original signal, the 7 basic mode components IMF 1-7 and the residual component (such as Figure 5 On this basis, the 8 components are transformed by Hilbert-Huang transform to obtain 8 Hilbert spectra (as shown in Figure 6 By summing up the Hilbert spectra of all components, the Hilbert spectrum of the shield pressure signal can be obtained (as shown in Figure 7 As shown), the Hilbert spectrum is integrated over time to obtain the Hilbert marginal spectrum (as shown Figure 8 As shown). Dividing the Hilbert marginal spectrum by time gives the average marginal spectrum, and the stability of the pressure signal DS = 1.510 can be finally obtained. Similarly, the stability of the other five shield pressure data in this ring can be calculated, and the result is:

[0094] DS(6)=[0.566,0.500,0.369,1.510,0.121,0.088]

[0095] To obtain the stability value of the multi-dimensional shield pressure signal of this ring, it is necessary to perform weighted summation on this 6-dimensional array. In this study, it is considered that these 6 shield pressures are equally important. Therefore, the stability of this ring excavation can be directly obtained by the mean value method:

[0096] DS = mean (DS (6)) = 0.526

[0097] Construction of graph network:

[0098] In a complex system, the shield parameters are interdependent. Although their direct correlation coefficients are small, their partial correlation coefficients can be calculated separately to represent the interaction. As described in the method in S2, a weighted directed complex network can be constructed with the shield parameters (such as propulsion speed, total thrust, etc.) of a ring of shield excavation data as nodes and the partial correlation relationship between shield parameters as edges, such as Figure 3 shown.

[0099] Training of graph neural network model:

[0100] As described in the method of step S3, the generated Figure 3That is, the shield parameter network diagram is used as the input of the graph neural network model, the overall excavation stability calculated by Hilbert-Huang in S1 and the instability value of the i-th sample actually calculated are used as the training benchmark, and DS is used as the output of the graph neural network model. The graph neural network is trained by deep learning to obtain an end-to-end graph neural network model, and an intelligent prediction model for shield excavation face stability based on graph network learning is obtained.

[0101] In the online prediction stage, as described in step S4, the shield parameter time series data is collected and the shield parameter network diagram is generated according to the method in step S2, and input into the shield excavation face stability intelligent prediction model trained offline in step S3, and the predicted shield excavation stability, i.e., DS value, can be output. This method can make timely predictions on the shield excavation face stability without a large amount of manual judgment and calculation, thus improving prediction efficiency and construction quality.

[0102] 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. An intelligent prediction method for shield excavation face stability based on graph network learning, characterized in that: include: Offline training phase: S1: Perform Hilbert-Huang transform on the shield pressure signal of the shield pressure chamber and use DS to characterize the real-time shield tunneling stability: Wherein, DS is the non-stationarity of the shield pressure signal, H(ω,t) is the Hilbert time-frequency spectrum of the shield pressure signal, ω represents the frequency domain, B(ω) is the average marginal spectrum of the shield pressure signal, T represents the time axis of the Hilbert time-frequency spectrum integration, and t represents time; S2: Collect the time series data of various shield parameters, obtain the weighted directed connections between the shield parameters based on the partial correlation coefficients between the shield parameters, construct the shield parameter partial correlation threshold network, and obtain the shield parameter network diagram; among them, the method for obtaining the weighted directed connections between the shield parameters based on the partial correlation coefficients between the shield parameters is as follows: For any three non-repeated shield parameters X, Y, and Z, the weight of the directed connection Z→X between two shield parameters in the shield parameter network diagram is determined by the following formula: d(X,Y:Z)≥<d(X,Y:Z)> Z +kσ Z (d(X,Y:Z)) The weight is set to the number of Ys that satisfy the above formula, where d(X,Y:Z)=ρ(X,Y)-ρ(X,Y:Z) <d(X,Y:Z)> Z and σ Z (d(X,Y:Z)) are the mean and standard deviation of d(X,Y:Z) for all shield parameters except X and Y; ρ(X,Y) is the correlation coefficient between X and Y, and ρ(X,Y:Z) is the partial correlation coefficient between X and Y excluding the influence of Z; S3: constructing a graph network learning model of shield excavation face stability and shield parameter network characteristics, training the graph network learning model with the shield parameter network graph obtained in step S2 as input and the DS value as output, and obtaining an intelligent prediction model of shield excavation face stability based on graph network learning; Online prediction stage: S4: Collect the time series data of various shield parameters, process them according to step S2 to obtain the shield parameter network diagram, and then input it into the shield excavation face stability intelligent prediction model based on graph network learning to obtain the DS value to predict the shield excavation face stability.

2. The intelligent prediction method for shield excavation face stability according to claim 1, characterized in that: For a shield machine with multiple pressure chambers, S1 includes: S11: extracting the pressure chamber data on the shield excavation face pressure sensor collected by the shield machine during the excavation process, and obtaining the shield pressure signal corresponding to each pressure chamber; S12: using the Hilbert-Huang transform method to process the shield pressure signal to obtain the non-stationarity of each pressure chamber data, i.e., the value of DS; S13: weighted sum of the DS values ​​of the shield pressure signals of each pressure chamber to obtain the stability of the shield excavation face.

3. The intelligent prediction method for shield excavation face stability according to claim 2, characterized in that: The S12 includes: The shield pressure signal X(t) obtained in step S11 is subjected to empirical mode decomposition to obtain the basic mode components of X(t): Among them, c1, c2, …, c i ,…,c M is the fundamental mode component, r M is the residual component, t is the time parameter, and M is the total number of basic mode components.

4. The intelligent prediction method for shield excavation face stability according to claim 1, characterized in that: The graph network learning model in S3 includes a network graph updating module and a DS conversion module. The network graph updating module is used to iteratively update the input shield parameter network graph according to steps S31-S32. The DS conversion module is used to convert the iterated shield parameter network graph into DS. The loss function of the DS conversion module is calculated according to step S33: S31: For node n in the shield parameter network diagram, construct a state function x based on its network adjacency information n , f w is a local transfer function, which represents the dependency of node n on its adjacent nodes, then: x n =f w (l n ,l co [n],x ne [n],l ne [n]) Among them, ln, l co [n], x ne [n], l ne [n] represents node n, i.e., the type of shield parameters, the edge of node n, i.e., the partial correlation between shield parameters, the state of node n, i.e., the time series data of shield parameters, and the adjacent nodes of node n. The adjacent nodes represent nodes connected by edges, i.e., the partial correlation coefficient between nodes is greater than the threshold; S32: For the node state function x n , calculate the iteration of the state function: x n (m+1)=F w (x n (m),l) Among them, x n (m) is x n The mth iteration of F w f w The global transition function extended to all nodes until the state x n (m) Iterate until a fixed point solution x is obtained, and then proceed to step S33; S33: Calculate the loss function e w and its gradient in, is the conversion function that converts the iterated shield parameter network diagram into DS; w is Internal reference; DS i is the shield stability of the i-th sample under the shield tunneling state, which is the value actually calculated in step S1; G i is the adjacency matrix describing the shield parameter network graph structure of the i-th sample, X i It is the characteristic matrix obtained after iteration of the shield parameter network diagram of the i-th sample, which is composed of the node states after iteration.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for intelligently predicting the stability of a shield excavation face as claimed in any one of claims 1 to 4 is implemented.

6. A device for intelligent prediction of shield excavation face stability, characterized in that: The method comprises the computer-readable storage medium as claimed in claim 5 and a processor, wherein the processor is used to call and process the computer program stored in the computer-readable storage medium.