Shield tunnel risk early warning method and system based on digital twinborn and federated learning

Abstract

The invention discloses a shield tunnel risk early warning method and system based on digital twinning and federated learning, and the method comprises the steps: a shield tunnel construction side client serves as an edge calculation terminal, and obtains basic parameters; a first shield digital twinning system, a second shield digital twinning system and a third shield digital twinning system are correspondingly constructed by the shield machine producer client side, the shield tunnel construction side client side and the third-party client side according to the basic parameters; a shield construction risk early warning model is arranged in each shield digital twin system; the shield machine production side client, the shield tunnel construction side client and the third side client are all participants, and each participant sends model parameters of the respective shield construction risk early warning model to the central server; the central server is used for averaging the parameters of the three models and issuing the average value of the parameters to each participant; and each participant updates the model of the participant according to the average value of the parameters, and realizes shield construction risk early warning based on the updated model.

Description

Technical Field

[0001] The present invention relates to the technical field of shield tunnel construction, and in particular to a shield tunnel risk early warning method and system based on digital twin and federated learning. Background Art

[0002] In the process of tunnel construction, the shield method has been increasingly widely used due to its advantages of high automation, high construction efficiency, and low impact on the surrounding environment. Shield tunnels are developing towards longer distances, larger cross-sections, greater burial depths, and more complex geological conditions, which puts forward higher requirements and greater challenges for shield tunnel construction.

[0003] Traditional shield construction risk warning faces network disconnection, power outages, network speed fluctuations and frequent data transmission interruptions in terms of data transmission; real-time shield monitoring data flows into the central server, causing network congestion, data packet loss and untimely warnings.

[0004] Traditional shield construction only analyzes single data at the construction site, without considering transferring multi-source data from the construction site to a unified digital model, and lacks the integrated analysis and utilization of geological data and construction data.

[0005] Traditional shield construction relies on manual experience, and the regulation of key shield excavation parameters is determined by operators, making shield construction decisions highly subjective and random. If the parameters are not adjusted properly, it can affect the construction progress at best and cause serious construction accidents at worst. Summary of the invention

[0006] In order to solve the deficiencies of the prior art, the present invention provides a shield tunnel risk warning method and system based on digital twin and federated learning;

[0007] On the one hand, a risk warning method for shield tunnels based on digital twins and federated learning is provided, including:

[0008] The shield tunnel construction client, as an edge computing terminal, obtains geological parameters and construction parameters; the shield tunnel construction client transmits the obtained data to the shield machine production client and third-party clients after homomorphic encryption; the shield machine production client transmits the shield equipment parameters to the shield tunnel construction client and third-party clients after homomorphic encryption; geological parameters, construction parameters and shield equipment parameters are collectively referred to as basic parameters;

[0009] The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively based on basic parameters; each shield digital twin system is equipped with a shield construction risk warning model;

[0010] The shield machine manufacturer client, shield tunnel construction client, third-party client and central server together constitute a federated learning system; the shield machine manufacturer client, shield tunnel construction client and third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server;

[0011] The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant; each participant updates its own model according to the average value of the parameters and implements shield construction risk warning based on the updated model.

[0012] On the other hand, a shield tunnel risk early warning system based on digital twins and federated learning is provided, including: shield tunnel construction client, shield machine manufacturer client, third-party client and central server;

[0013] The shield tunnel construction client, as an edge computing terminal, obtains geological parameters and construction parameters; the shield tunnel construction client transmits the obtained data to the shield machine production client and third-party clients after homomorphic encryption; the shield machine production client transmits the shield equipment parameters to the shield tunnel construction client and third-party clients after homomorphic encryption; geological parameters, construction parameters and shield equipment parameters are collectively referred to as basic parameters;

[0014] The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively based on basic parameters; each shield digital twin system is equipped with a shield construction risk warning model;

[0015] The shield machine manufacturer client, shield tunnel construction client, third-party client and central server together constitute a federated learning system; the shield machine manufacturer client, shield tunnel construction client and third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server;

[0016] The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant; each participant updates its own model according to the average value of the parameters and implements shield construction risk warning based on the updated model.

[0017] The above technical solution has the following advantages or beneficial effects:

[0018] (1) The edge computing system in the present invention is deployed at the edge of the network close to the data source, and performs data processing directly near the data source, including deletion of non-excavation state data of the shield machine, correction of abnormal data, and noise processing, and transmits necessary data to the data cloud platform, thereby reducing bandwidth requirements and transmission costs, reducing transmission time, and ensuring faster response speed.

[0019] (2) The shield digital twin system of the present invention can accurately reflect the status and behavior of the shield machine entity by constructing a mapping relationship between the physical world and the digital space. It uses a variety of machine learning and deep learning algorithms for data analysis and mining, and realizes functions such as formation parameter prediction and tunneling parameter optimization. It provides powerful data-driven capabilities for the simulation, prediction and optimization of complex shield machine systems, and accelerates the process of intelligent construction and construction automation.

[0020] (3) The federated learning system in the present invention provides a platform for mutual communication and learning for different shield machine manufacturers, tunnel construction parties, scientific research institutions and colleges. On the premise of ensuring the data privacy of all parties, it learns the model parameters and experience of all parties and improves the prediction and simulation capabilities of the local digital twin system.

[0021] A compliant smart contract is signed between the shield machine manufacturer, the tunnel construction party, and the scientific research institutions and colleges and universities, using encrypted engineering data from the construction party and encrypted equipment data from the shield machine manufacturer.

[0022] For shield machine manufacturers, by analyzing the equipment status of the entire construction area and incorporating important learning experiences into the subsequent product design stage, the operating status of the shield machine throughout its life cycle can be considered during the design stage.

[0023] For different construction parties, by learning from each other the model parameters of different projects and integrating richer construction experience, the ability of shield machines to cope with complex geological environments has been improved, and the intelligent development of shield construction has been promoted.

[0024] For scientific research institutions and colleges and universities, by strengthening learning and exchanges with other parties, they can propose solutions to specific problems on site, develop more intelligent and scientific prediction models, and provide more efficient decision-making guidance for on-site shield construction decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 Conceptual diagram of a federated learning system in an embodiment of the present invention.

[0027] Figure 2 A diagram showing the composition of the edge computing system in an embodiment of the present invention.

[0028] Figure 3 Flowchart of data processing and transmission of edge computing system in an embodiment of the present invention.

[0029] Figure 4 Functional framework diagram of the digital twin system module in an embodiment of the present invention.

[0030] Figure 5 An overall flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0032] The widespread application of digital twin technology, which can accurately reflect the state and behavior of shield machine entities by building a mapping relationship between the physical world and the digital space, provides a powerful data-driven capability for the simulation, prediction and optimization of complex shield machine systems, and accelerates the process of intelligent construction and construction automation. The effective implementation of digital twin technology depends on massive real-time data, which comes from sensors installed on the shield machine and the surrounding construction environment. Since shield construction adopts an all-weather construction operation mode, sensors usually collect data in seconds, and a large amount of raw data is generated during the construction process, which inevitably contains abnormal data and noise. The two processes of tunneling construction and segment assembly are carried out alternately, and the time used is close. The downtime data during the segment assembly process will also be collected by sensors, which is of little significance for model establishment. If all the data collected on site are directly transmitted to the cloud platform, the cloud computing platform will face high network latency, massive equipment access, difficulty in processing massive data, insufficient bandwidth and excessive power consumption. The existence of a large amount of downtime data, abnormal data and noise not only increases the pressure of data transmission, but also causes certain interference to subsequent analysis and modeling.

[0033] Edge computing technology has emerged to solve the problems of high latency and lack of real-time data analysis capabilities in traditional data processing methods. Edge computing technology provides edge intelligence services at the edge of the network close to the data source through a distributed open platform that integrates network, computing, storage, and application core capabilities. Edge computing processes data directly at the data source and only transmits necessary data to the data cloud platform, reducing bandwidth requirements and transmission costs, reducing transmission time, and ensuring faster response speed.

[0034] Shield construction is facing an increasingly complex geological environment. Complete historical data and valuable construction experience are extremely important for ensuring construction safety and improving construction efficiency. However, due to industry competition, privacy security, and complex administrative procedures, data integration between participants related to tunnel construction (such as shield machine design and production parties, construction parties, scientific research institutions, etc.) faces many obstacles, and the data between the parties exist in the form of data islands. Federated learning enables multiple participants to continue machine learning while protecting data privacy and meeting legal and compliance requirements, solving the problem of data islands. In the process of machine learning, each participant can use the data of other participants for joint modeling, establish a shared machine learning model, and adjust their own model parameters based on the shared model to improve the accuracy of the model.

[0035] Embodiment 1

[0036] This embodiment provides a shield tunnel risk warning method based on digital twins and federated learning;

[0037] like Figure 5 As shown in the figure, the risk warning method for shield tunneling based on digital twin and federated learning includes:

[0038] S101: The shield tunnel construction client, as an edge computing terminal, obtains geological parameters and construction parameters; the shield tunnel construction client transmits the obtained data to the shield machine production client and the third-party client after homomorphic encryption; the shield machine production client transmits the shield equipment parameters to the shield tunnel construction client and the third-party client after homomorphic encryption; the geological parameters, construction parameters and shield equipment parameters are collectively referred to as basic parameters;

[0039] S102: The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively according to the basic parameters; each shield digital twin system is equipped with a shield construction risk warning model;

[0040] S103: The shield machine manufacturer client, the shield tunnel construction client, the third-party client and the central server together constitute a federated learning system; the shield machine manufacturer client, the shield tunnel construction client and the third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server;

[0041] S104: The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant; each participant updates its own model according to the average value of the parameters, and implements shield construction risk warning based on the updated model.

[0042] Furthermore, the geological parameters refer to the geological information of the tunnel design section obtained through geological exploration and advanced geological prediction technology; the construction parameters refer to the real-time excavation parameters of the shield machine, such as the propulsion speed, propulsion force, cutter head torque, penetration rate, etc.; the shield machine equipment parameters refer to the detailed equipment parameters, such as the cutter head diameter, tool type and installation position, and the number of propulsion cylinders.

[0043] It should be understood that the construction parameters of the shield machine are collected by sensors. The geological parameters are collected and summarized through geotechnical surveys before construction and advanced geological forecasts during construction. The shield machine equipment parameters are provided by the shield machine manufacturer. The sensor and the edge computing client are connected through a communication network.

[0044] like Figure 2 As shown, the edge computing terminal pushes computing resources and services from the cloud data center to the edge of the network, that is, closer to the data source. This architecture can increase data processing speed, reduce latency, enhance privacy protection, and reduce bandwidth usage. The edge computing terminal includes: a physical layer, an information layer, and a communication layer.

[0045] The physical layer is the actual hardware equipment deployed at the edge of the network, including various sensors, edge servers, storage devices, etc. arranged in the shield machine and the surrounding geological environment. The physical layer is responsible for the collection of multi-source heterogeneous data and is also the operating platform of the information layer.

[0046] The information layer is the data processing center in the edge computing system, which includes a variety of software applications and services such as data processing algorithms. The information layer depends on the physical layer and is responsible for data processing, analysis and storage.

[0047] The communication layer is the part of the edge computing system responsible for data transmission and communication, including network protocols, interfaces, routers, switches and other network equipment. The communication layer is responsible for ensuring the data flow between the physical layer and the information layer, and between the information layer and the cloud data center, to achieve rapid data transmission and synchronization.

[0048] The information layer processes and analyzes the data collected by the physical layer, including deleting the non-excavation state data of the shield machine, correcting abnormal data, and processing noise, so as to simplify the massive amount of original data.

[0049] Furthermore, if Figure 3 As shown, the S101: further includes:

[0050] The shield tunnel construction client pre-processes the shield machine working parameters, and the pre-processing includes:

[0051] Step (11) retains the parameters of the shield machine in the excavation state, and deletes the parameters of the shield machine in the non-excavation state;

[0052] Step (12) correcting the abnormal data;

[0053] Step (13) processes the noise data.

[0054] Furthermore, the step (11) retains the parameters of the shield machine in the excavation state and deletes the parameters of the shield machine in the non-excavation state, specifically including:

[0055] The deletion of non-excavation state data is achieved through a binary discriminant function. By analyzing the on-site construction situation, when the propulsion speed, cutter head torque, and cutter head speed are 0, the shield machine is in a shutdown state. Therefore, the excavation parameters of the shutdown state are deleted to reduce the pressure of data transmission. Because the excavation parameters at this time have little effect on subsequent analysis and learning, and may even interfere with the accuracy of the model;

[0056] x=T×n×v; G(x)=g·F(x);

[0057] Among them, T is the cutter head torque, n is the cutter head speed, v is the excavation speed, x is the product of the cutter head torque, cutter head speed and excavation speed, g is the original data set, and G(x) is the data set of normal excavation.

[0058] It should be understood that the deletion of the non-excavation state data is achieved through a binary discriminant function. This is because during the shield construction process, forward excavation and segment assembly are carried out in series. When the excavation reaches a ring width, the shield stops and starts segment assembly. At this time, the sensor still works at the set frequency, and many shield machine excavation-related parameters are zero values, which are meaningless for analysis. The excavation parameters with zero values ​​can be deleted to reduce the pressure of data transmission.

[0059] Furthermore, the step (12) corrects the abnormal data, specifically by finding the abnormal value through the isolation forest algorithm, and using the average value of the adjacent data of the abnormal value as the replacement data of the abnormal value.

[0060] It should be understood that the abnormal data correction is achieved through the isolation forest algorithm, which is a tree-based machine learning algorithm that "isolates" observations by randomly selecting features and randomly selecting the cut value of the feature to achieve anomaly detection. The outliers selected by the isolation forest are statistically analyzed in combination with expert experience, and the abnormal threshold of the isolation forest algorithm is appropriately adjusted to ensure the accuracy and rationality of abnormal data processing and avoid misjudging normal data as abnormal. In order to ensure the integrity of the data stream, the abnormal data is replaced by the average value of the adjacent data.

[0061] Furthermore, the step (13) processes the noise data by means of a weighted average filtering algorithm, wherein the weighted average filtering algorithm assigns higher weights to the nearest points and lower weights to the farther points, thereby retaining as much information of the current state as possible and improving the accuracy and reliability of noise processing.

[0062] It should be understood that the noise processing is implemented by weighted moving average filtering technology, which is a signal processing technology used to smooth time series data, reduce the impact of noise, and retain the trend information of the data.

[0063] The edge computing terminal provides data support for the shield digital twin system. The digital twin system uses the transmitted data to update the model and learn the data, providing online decision-making guidance for on-site construction.

[0064] It should be understood that the first, second and third shield digital twin systems are established according to geological parameters, construction parameters and shield machine equipment parameters. The first, second and third shield digital twin systems can analyze the data from the edge computing system in real time and update the first, second and third shield digital twin systems in a timely manner. Deeply explore key parameters such as tunneling speed, shield posture, settlement deformation, etc., predict equipment status and stratum conditions through a variety of machine learning and deep learning algorithms, timely repair and replace tools and reasonably adjust construction strategies. At the same time, the first, second and third digital twin systems have parameter prediction and optimization modules, which enable the shield machine to work according to the optimized parameters through the automatic control module, providing a more scientific and efficient tunneling construction plan.

[0065] According to the geological survey results, a 3D geological model of the entire tunnel design interval construction is established in the 3D geological modeling and analysis software EVS (Earth Volumetric Studio). The visualization of the stratigraphic interface of the entire interval is realized through the analysis and interpolation of the drilling information. The model is imported into Unity, and the stratigraphic parameter information in the 3D geological model is displayed, such as the compressive strength, bearing capacity, and natural gravity of the rock. The above parameters are derived from field tests and indoor tests. According to the shield machine design drawings, a shield machine model is established in 3DMAX, and the shield machine model is imported into the Unity system where the 3D geological model is located to realize the tunneling simulation of the shield machine in the 3D geological model. Relying on the data interaction between the edge computing system and the above model, a shield digital twin system is established to realize the real-time synchronization of construction dynamics and digital twin models.

[0066] Furthermore, the shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively according to the basic parameters, specifically including:

[0067] The shield machine manufacturer client built the first shield machine digital twin system;

[0068] The client of the shield tunnel construction party built the second shield digital twin system;

[0069] The third-party client builds the third shield digital twin system.

[0070] Furthermore, if Figure 4 As shown, the first shield digital twin system includes: a first data storage module, a first data encryption module, a first formation parameter prediction module, a first surface settlement prediction module, a first tool wear prediction module, a first shield posture deviation prediction module, a first excavation parameter prediction module, a first excavation parameter optimization module, a first shield construction risk warning module, a first automatic control module, and a first visualization module.

[0071] The second shield digital twin system includes: a second data storage module, a second data encryption module, a second formation parameter prediction module, a second surface settlement prediction module, a second tool wear prediction module, a second shield posture deviation prediction module, a second excavation parameter prediction module, a second excavation parameter optimization module, a second shield construction risk warning module, a second automatic control module, and a second visualization module.

[0072] The third shield digital twin system includes: a third data storage module, a third data encryption module, a third formation parameter prediction module, a third surface settlement prediction module, a third tool wear prediction module, a third shield posture deviation prediction module, a third excavation parameter prediction module, a third excavation parameter optimization module, a third shield construction risk warning module, a third automatic control module, and a third visualization module.

[0073] Furthermore, the first, second and third data storage modules are all used to store real-time data streams from the shield tunnel construction client and update the shield digital twin model.

[0074] Furthermore, the first, second and third data encryption modules use homomorphic encryption algorithms to encrypt the data required by the federated learning contract parties. Homomorphic encryption supports computing and learning on encrypted data, while protecting the privacy of the data, ensuring the learning effect, and providing security for cross-organizational data cooperation and sharing. The shield machine manufacturer's client provides the basic parameters of the shield equipment, and the shield tunnel construction client provides geological parameters and construction parameters, which are encrypted and shared with other participants. The data encryption module has a decryption function, and each participant applies decryption technology to process the encrypted data, including basic parameters from other participants and model parameters from the central server.

[0075] Furthermore, the first, second and third formation parameter prediction modules establish a database based on the geological survey data and the historical excavation data of the shield machine. The geological survey data is used to obtain the formation category label and the surrounding rock grade label, and the excavation parameters such as the cutter head speed and penetration rate are selected for model training. The trained model is used to realize the formation category prediction and the surrounding rock grade prediction.

[0076] Furthermore, the formation category prediction is achieved based on the trained support vector machine. During the training process of the support vector machine, the input values ​​of the support vector machine are cutter head torque, cutter head speed, thrust, penetration, tunneling speed, torque cutting depth index (TPI) and field cutting depth index (FPI) values, and rock uniaxial compressive strength, and the output value of the support vector set is the formation category; the formation category specifically includes: ultra-soft formation, relatively homogeneous soft rock, fully weathered soft and hard uneven formation, strongly weathered and moderately weathered soft and hard uneven formation, medium-soft formation, soft and hard uneven soft formation, soft and hard uneven hard formation, medium-hard formation, relatively homogeneous hard rock, and ultra-hard formation; when the loss function value of the support vector machine drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained support vector machine.

[0077]

[0078] Where: T is the cutter head torque; P is the penetration rate;

[0079]

[0080] Where: F is the propulsion force.

[0081] It should be understood that since support vector machine (SVM) is a commonly used classification algorithm, it can be used to process multi-category classification problems, and the SVM algorithm is selected for stratum category prediction. According to the previous geological drilling data and the historical excavation parameters of the shield machine, a database of the stratum category prediction model is established, and the cutter head torque, cutter head speed, propulsion force, penetration, excavation speed, TPI, FPI, and rock uniaxial compressive strength are selected as the input variables of the SVM model. Ultra-soft strata, relatively homogeneous soft rock, fully weathered soft and hard uneven strata, strongly weathered medium weathered soft and hard uneven strata, medium soft strata, soft and hard uneven soft strata, soft and hard uneven hard strata, medium hard strata, relatively homogeneous hard rock, and ultra-hard strata are 10 categories as the output variables of the SVM model. The SVM model is trained using the established database to obtain a stratum category prediction model based on the SVM algorithm. In the shield digital twin system, real-time excavation parameters are input to achieve accurate prediction of 10 stratum categories.

[0082] Furthermore, the surrounding rock grade prediction is achieved based on a trained multilayer perceptron. During the training process, the input values ​​of the multilayer perceptron are disk rotation speed, penetration rate, gripper pressure, gripper pump pressure, gripper pitch angle, control pump pressure, propulsion speed and gripper roll angle; the output values ​​of the multilayer perceptron are surrounding rock grades I, II, III, IV and V; when the loss function value of the multilayer perceptron drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained multilayer perceptron.

[0083] It should be understood that the prediction of surrounding rock grade is based on a neural network, and since the multi-layer perceptron (MLP for short) is a classification algorithm that can learn complex nonlinear relationships and handle multi-category classification tasks, the MLP algorithm is selected for surrounding rock grade prediction. A database of the stratum category prediction model is established based on the previous geological survey data and the historical tunneling parameters of the shield machine, and the cutterhead speed, penetration, gripper pressure, gripper pump pressure, gripper pitch angle, control pump pressure, propulsion speed and gripper rolling angle are selected as the input variables of the MLP model, and the surrounding rock grades I, II, III, IV, and V are used as the output variables of the model. The MLP model is trained using the established database to obtain the surrounding rock grade prediction based on the MLP algorithm. In the shield digital twin system, real-time tunneling parameters are input to achieve accurate prediction of 5 surrounding rock grades.

[0084] Furthermore, the first, second and third surface subsidence prediction modules realize the prediction of surface subsidence through the trained CNN-BiLSTM model;

[0085] The CNN-BiLSTM model includes: a CNN network and a BiLSTM network connected in sequence;

[0086] During the training process, the input values ​​of the CNN-BiLSTM model are bearing capacity, natural gravity, tunnel burial depth, distance between the monitoring point and the cutterhead, penetration, thrust, incision pressure, working chamber pressure, cutterhead speed and cutterhead torque; the output value of the CNN-BiLSTM model is the surface settlement prediction result; when the loss function value of the CNN-BiLSTM model drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained CNN-BiLSTM model.

[0087] It should be understood that the convolutional neural network is introduced to improve the bidirectional long short-term memory neural network, so as to better capture the local characteristics of the surface settlement data and establish a surface settlement prediction model based on the CNN-BiLSTM algorithm. A database is established based on the geological survey report, the historical excavation data of the shield machine and the surface settlement monitoring value. The bearing capacity, natural gravity, tunnel burial depth, the distance between the monitoring point and the cutterhead, penetration, thrust, incision pressure, working chamber pressure, cutterhead speed, and cutterhead torque are selected as the input parameters of the surface settlement prediction model, and the surface settlement value is the output parameter of the prediction. For a specific monitoring point, the distance between the monitoring point and the cutterhead is constantly changing. Specifically, before the shield machine passes through, the distance between the monitoring point and the cutterhead is constantly decreasing, and the distance at this time is set to a positive value; when the shield machine cutterhead just reaches the section where the monitoring point is located, the distance is 0; when the shield machine cutterhead passes through the section, the distance continues to increase, but in order to facilitate the distinction between the two different states before and after the crossing, the distance is set to a negative value, and the negative sign is only to distinguish the two position states. The established database is used for model training to obtain the surface settlement value prediction based on the CNN-BiLSTM algorithm. In the shield digital twin system, the real-time excavation parameters and the bearing capacity, natural gravity, and tunnel burial depth of the stratum where the shield machine is located in the digital twin system are input to realize the surface settlement prediction of different locations before, during, and after the shield crossing.

[0088] Furthermore, the first, second and third tool wear prediction modules predict tool wear based on the trained BP neural network;

[0089] During the training process, the BP neural network relies on the historical excavation parameters of the shield, formation parameter information, and tool wear information to establish a database, and selects cutter head torque, cutter head speed, thrust, excavation speed, penetration, hob diameter, installation radius, hob spacing, rock uniaxial compressive strength, formation type and surrounding rock grade as input values; the output value of the BP neural network is the tool wear amount; when the loss function value of the BP neural network drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained BP neural network.

[0090] It should be understood that a tool wear prediction model based on the BP neural network algorithm is established, and a database is established based on the historical excavation parameters of the shield, the formation parameter information, and the tool wear information. The cutter torque, cutter speed, propulsion force, excavation speed, penetration, hob diameter, installation radius, hob spacing, rock uniaxial compressive strength, formation category, and surrounding rock grade are selected as input features, and the tool wear amount is the output variable. The tool wear prediction model is trained based on the historical database, and a tool wear prediction model based on the BP neural network is established to predict the tool wear in real time. Referring to the tool replacement standard, the remaining life of the hob is predicted according to the wear trend of a single hob, and the tool is repaired and replaced in time to ensure the construction progress.

[0091] Furthermore, the first, second and third shield attitude deviation prediction modules realize the prediction of shield attitude deviation based on the trained long short-term memory neural network model LSTM;

[0092] During the training process of the model, the input values ​​of the long short-term memory neural network model LSTM are cutter head torque, cutter head rotation speed, excavation speed, propulsion force, thrust difference between upper and lower cylinders, cohesion, internal friction angle, lateral pressure coefficient and tunnel burial depth; the output values ​​are incision horizontal deviation, incision vertical deviation, shield tail horizontal deviation and shield tail vertical deviation; when the loss function value of the model drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained long short-term memory neural network model LSTM.

[0093] It should be understood that first, a long short-term memory neural network model (LSTM for short) is established to realize the prediction of shield posture deviation. The cutter head torque, cutter head speed, excavation speed, propulsion force, upper and lower cylinder thrust difference, cohesion, internal friction angle, lateral pressure coefficient, and tunnel burial depth are selected as input variables of the posture deviation prediction model. The output variables include the horizontal deviation of the incision, the vertical deviation of the incision, the horizontal deviation of the shield tail, and the vertical deviation of the shield tail.

[0094] Furthermore, the first, second and third tunneling parameter prediction modules realize tunneling parameter prediction based on the trained XGBoost model;

[0095] During the training process, the input variables of the XGBoost model are cutterhead extrusion pressure, penetration, incision pressure, tunneling speed, grouting pressure, grouting flow rate, grouting flow rate, formation type, surrounding rock grade, torque cutting depth index (TPI) and field cutting depth index (FPI) values. The output variables of the XGBoost model are cutterhead torque, cutterhead speed and total thrust. When the loss function value of the XGBoost model drops to the minimum value or the number of iterations reaches the set number, the training is stopped to obtain the trained XGBoost model.

[0096] Among them, the formation category and surrounding rock grade come from the output values ​​of the corresponding formation parameter prediction module.

[0097] Furthermore, the first, second and third tunneling parameter optimization modules select cutterhead speed, cutterhead torque, total thrust and penetration as optimization variables based on multi-objective optimization theory, and use NSGA-II algorithm to achieve multi-objective optimization of maximum propulsion speed and minimum tunneling specific energy. The multi-objective optimization framework is as follows:

[0098] F = (min(f1), max(f2));

[0099] f2 = XGBoost(v(n,F,T,P));

[0100] stn min ≤n≤n max ;v min ≤v≤v max ; F min ≤F≤F max ;

[0101] T min ≤T≤T max ;P min ≤P≤P max ;

[0102] Among them, E s is the tunneling specific energy (kJ / m3), n is the cutterhead speed (r / min), T is the cutterhead torque (MN·m), F is the total thrust (unit MN), R is the cutterhead radius of the shield machine (m), v is the tunneling speed (mm / min), and P is the cutterhead penetration of the shield machine (mm / rot).

[0103] F is the objective function of the model. This model has two objectives, including minimizing the excavation specific energy f1 and maximizing the excavation speed f2. The excavation specific energy can be directly calculated through real-time parameters such as the cutterhead penetration. There is no direct calculation formula for the advancement speed. It is necessary to establish an excavation speed prediction model based on XGBoost, input four parameters: cutterhead speed, total thrust, cutterhead torque, and cutterhead penetration, and output the excavation speed. Due to the large differences in excavation parameters under different geological conditions, taking the cutterhead penetration as an example, the cutterhead penetration in hard rock formations can reach 30mm / rot, while the soft soil formation generally fluctuates around 2mm / rot. Therefore, the maximum and minimum values ​​of the parameters in the constraint conditions need to be determined according to the specific engineering characteristics.

[0104] The parameters of the NSGA-II algorithm are initially set to a population size of 250, a maximum number of iterations of 100, a crossover probability of 0.8, and a mutation probability of 0.1, which can be adjusted according to specific conditions. A multi-objective optimization model for shield tunneling parameters based on the NSGA-II algorithm is established to obtain the optimal solution. Since the values ​​of the optimization variables such as cutterhead speed, total thrust, cutterhead torque, and cutterhead penetration are one-to-one corresponding to the optimal values ​​of the targets (tunneling speed, tunneling specific energy), the corresponding cutterhead speed, total thrust, cutterhead torque, and cutterhead penetration are displayed in the digital twin system to control the construction parameters within the optimization range.

[0105] Furthermore, the first, second and third shield construction risk warning modules are used to give warning results for various construction risks during the construction process. The construction risk warning modules include: geological risk warning, equipment risk warning and environmental impact warning;

[0106] Geological risk warning: When the shield machine is physically excavating, the shield machine digital twin is synchronously carried out in the digital space. Since the shield digital twin system has imported the three-dimensional geological information of the construction area during modeling, the geological information in front of the excavation face can be intuitively displayed in the digital twin system. Based on the prediction results of the formation parameter prediction module and the visualization of the geological characteristics in front of the excavation face, early warning information can be issued before the geological conditions change. For example, when encountering weak surrounding rock and fault fracture zones, an alarm will be issued in advance to avoid construction accidents.

[0107] Equipment risk warning, based on the tool wear prediction module, monitors the tool status, predicts failures and anomalies, and takes measures in advance to avoid risks. At the same time, the shield digital twin model integrates the real-time data of the shield machine. Based on the visualization module, it can monitor the operating status of key equipment such as the main bearing and hydraulic system. When the equipment parameters exceed the rated value or do not work, the equipment risk warning module sends a warning signal and gives the abnormal location.

[0108] Environmental risk warning, including surface settlement warning and toxic gas warning. Surface settlement warning is based on the surface settlement deformation module, which can realize the settlement value prediction for a long time before, during and after the shield crossing. During construction, the change trend of the surface settlement value at that location is predicted in advance. If it exceeds the standard value, a warning signal is given to make construction adjustments (such as grouting reinforcement) to avoid excessive settlement in the later stage and reduce the impact on the surrounding environment. The harmful gas warning function relies on the gas monitoring sensor carried by the shield machine, analyzes the sensor information in real time, and pays attention to the concentration of harmful gases such as methane, carbon monoxide, and hydrogen sulfide. When the concentration changes abnormally, reasonable ventilation and exhaust measures are taken to ensure construction safety and personnel health.

[0109] Safety accident warning, the shield machine entity and supporting systems are equipped with monitoring equipment, combined with on-site monitoring video, when a safety accident occurs, the accident location is given according to the monitoring information, timely rescue is carried out, and the accident loss is reduced. A large amount of video or image data is collected through monitoring cameras, and these data are annotated, especially the injured scenes and the postures of personnel. The database is enriched through data enhancement (such as rotation, scaling, adjusting brightness, etc.) to ensure that the model has better robustness to different environments. The human posture estimation technology OpenPose is used to detect the key points of the human skeleton to determine whether there are posture abnormalities, such as falls or twists. After the model training is completed, it can be deployed in the monitoring system to analyze the monitoring screen in real time and detect whether anyone is injured. When the system detects abnormal behavior or injury, it can trigger an alarm mechanism to notify relevant personnel to take emergency measures. The alarm conditions can be set according to characteristics such as abnormal posture and abnormal behavior. At the same time, the system needs to continuously collect new data to fine-tune the model to adapt to the changing monitoring environment and improve the detection accuracy. To reduce false alarms, the different perspectives of multiple monitoring cameras can be combined to enhance the system's judgment ability.

[0110] Furthermore, the first, second and third automatic control modules rely on the Internet of Things to realize the control of the shield machine entity by the shield digital twin system. In order to make full use of the function of the excavation parameter optimization module, the automatic control module converts the optimization strategy into control instructions. The control instructions are in a form that the shield machine can recognize, such as PLC control and PID control. They are sent to the controller of the shield machine entity through the Internet of Things, thereby realizing precise control and optimized operation of the shield machine entity, adjusting the construction parameters in real time, and improving construction efficiency.

[0111] Furthermore, the first, second and third visualization modules are used to display the established shield digital twin model, which is tightly integrated with all the above-mentioned functional modules, and display the real-time collected construction parameters and the shield digital twin model parameters constructed in the early stage in the form of charts, and visualize the predicted optimization results of each functional module.

[0112] In order to make full use of the construction experience of different units and improve data diversity, the concept of federated learning is introduced to learn the model parameters and experience of all parties under the premise of protecting the data privacy of all parties. The federated learning system mainly includes multiple shield digital twin models established locally by the participating parties and the corresponding edge control system, and a central server.

[0113] The participants establish their own shield digital twin model and system based on their respective construction characteristics, geological conditions, shield parameters, etc. The shield digital twin system collects data from the edge control system and has the above-mentioned data storage, tunneling parameter prediction, tunneling parameter optimization and other module functions. The participants transmit their own model updates and gradients to the central server after homomorphic encryption, and integrate the latest model data and gradients issued from the central server into the local shield digital twin system.

[0114] The federated learning method is a continuous and dynamic learning and updating process between each participant and the central server. First, the participant transmits the updated parameters and gradients of the digital twin model to the central processor through homomorphic encryption. Next, the central processor processes the encrypted data and updates the global model through decryption and aggregation, and sends the new parameters and gradient information to the participants through homomorphic encryption. Finally, the participants adjust the local digital twin system according to the latest information, and compare and analyze it with the measured values. The above process is repeated. Through the joint participation of multiple parties in federated learning, the construction of the shield digital twin model is collaboratively optimized, the prediction and simulation capabilities of the digital twin systems of all parties are continuously enhanced, and the tunneling efficiency of the shield machine is continuously improved.

[0115] like Figure 5 As shown, it should be understood that S103: the shield machine manufacturer's client, the shield tunnel construction party's client, the third-party client and the central server together constitute a federated learning system; establish a federated learning system, set up a central server, determine the participants of the federated learning, and build a connection between the participants and the central server.

[0116] Participants who do not have direct construction data, such as scientific research institutions and shield machine manufacturers, can sign legal and compliant smart contracts with construction units, determine data transmission protocols, and encrypt and transmit the required data through the data encryption module in the shield machine digital twin system.

[0117] The participating parties establish a unique shield digital twin model and system based on their respective construction characteristics, geological conditions, shield parameters, etc. The shield digital twin system collects data from the client of the shield tunnel construction party and has the above-mentioned data storage, excavation parameter prediction, excavation parameter optimization and other module functions.

[0118] Furthermore, the shield machine manufacturer client, shield tunnel construction client and third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server, including: the participants transmit their own model updates and gradients to the central server after homomorphic encryption.

[0119] Furthermore, if Figure 1As shown, S104: the central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant, including:

[0120] The central server is responsible for receiving and decrypting the model update parameters and gradients from each participant, verifying the integrity and correctness of the data, and preventing data from being damaged or maliciously tampered with during transmission. The central server uses a weighted average aggregation method to assign higher weights to model parameters with larger data volumes and higher prediction accuracy, build a new global model, and distribute the new parameters and gradient information to each participant through homomorphic encryption.

[0121] Furthermore, each participant updates its own model according to the average value of the parameters, and implements shield construction risk warning based on the updated model, including:

[0122] The participants receive and decrypt the latest model data and gradient information sent from the central server, and integrate the information into the local shield construction risk warning model. After multiple debugging and tests, the accuracy of the local shield construction risk warning model is verified.

[0123] S103-S104 is a continuous cycle process. Through the joint participation of multiple parties in federated learning, the construction of the shield digital twin model is optimized collaboratively, the prediction and simulation capabilities of the digital twin systems of all parties are continuously enhanced, and the tunneling efficiency of the shield machine is continuously improved. When the accuracy of the local digital twin system of a participant reaches the expected value, it can exit automatically, and other participants can continue the learning task.

[0124] The edge computing system provides data support for the shield digital twin system. The digital twin system uses the transmitted data to update the model and learn the data, providing online decision-making guidance for on-site construction.

[0125] The shield digital twin system can analyze data from the edge computing system in real time, deeply explore key parameters such as tunneling speed, shield posture, settlement deformation, etc., predict equipment status and stratum conditions through a variety of machine learning and deep learning algorithms, timely repair and replace tools, and reasonably adjust construction strategies. At the same time, the digital twin system has a parameter prediction and optimization module, which enables the shield machine to work according to the optimized parameters through the automatic control module, providing a more scientific and efficient tunneling construction plan.

[0126] The federated learning system can break the limitations of data silos, strengthen the links between shield machine manufacturers, tunnel construction parties, scientific research institutions and colleges, make full use of the parameter information of each digital twin model, improve the prediction accuracy and model generalization ability of the local digital twin system of each participant, and promote the scientific, efficient and intelligent development of shield construction.

[0127] Embodiment 2

[0128] This embodiment provides a shield tunnel risk early warning system based on digital twins and federated learning;

[0129] The shield tunnel risk early warning system based on digital twin and federated learning includes: shield tunnel construction party client, shield tunnel construction party client, third-party client and central server;

[0130] The shield tunnel construction client, as an edge computing terminal, obtains geological parameters and construction parameters; the shield tunnel construction client transmits the obtained data to the shield machine production client and third-party clients after homomorphic encryption; the shield machine production client transmits the shield equipment parameters to the shield tunnel construction client and third-party clients after homomorphic encryption; geological parameters, construction parameters and shield equipment parameters are collectively referred to as basic parameters;

[0131] The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively based on basic parameters; each shield digital twin system is equipped with a shield construction risk warning model;

[0132] The shield machine manufacturer client, shield tunnel construction client, third-party client and central server together constitute a federated learning system; the shield machine manufacturer client, shield tunnel construction client and third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server;

[0133] The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant; each participant updates its own model according to the average value of the parameters and implements shield construction risk warning based on the updated model.

[0134] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A risk warning method for shield tunneling based on digital twins and federated learning, characterized by: include: The shield tunnel construction client, as an edge computing terminal, obtains geological parameters and construction parameters; The shield tunnel construction client transmits the acquired data to the shield machine production client and the third-party client after homomorphic encryption; the shield machine production client transmits the shield equipment parameters to the shield tunnel construction client and the third-party client after homomorphic encryption; geological parameters, construction parameters and shield equipment parameters are collectively referred to as basic parameters; The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively based on basic parameters; each shield digital twin system is equipped with a shield construction risk warning model; The shield machine manufacturer client, shield tunnel construction client, third-party client and central server together constitute a federated learning system; the shield machine manufacturer client, shield tunnel construction client and third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server; The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant; each participant updates its own model according to the average value of the parameters and implements shield construction risk warning based on the updated model.

2. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 1 is characterized in that: The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively according to the basic parameters, specifically including: the shield machine manufacturer client constructs the first shield digital twin system; the shield tunnel construction client constructs the second shield digital twin system; the third-party client constructs the third shield digital twin system; The first shield digital twin system includes: a first data storage module, a first data encryption module, a first formation parameter prediction module, a first surface settlement prediction module, a first tool wear prediction module, a first shield posture deviation prediction module, a first tunneling parameter prediction module, a first tunneling parameter optimization module, a first shield construction risk warning module, a first automatic control module, and a first visualization module; The second shield digital twin system includes: a second data storage module, a second data encryption module, a second formation parameter prediction module, a second surface settlement prediction module, a second tool wear prediction module, a second shield posture deviation prediction module, a second tunneling parameter prediction module, a second tunneling parameter optimization module, a second shield construction risk warning module, a second automatic control module, and a second visualization module; The third shield digital twin system includes: a third data storage module, a third data encryption module, a third formation parameter prediction module, a third surface settlement prediction module, a third tool wear prediction module, a third shield posture deviation prediction module, a third excavation parameter prediction module, a third excavation parameter optimization module, a third shield construction risk warning module, a third automatic control module, and a third visualization module.

3. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 2 is characterized in that: The first, second and third data encryption modules use a homomorphic encryption algorithm to encrypt the data required by the federated learning contract parties. Homomorphic encryption supports calculation and learning on encrypted data. The data encryption module is also used to decrypt the encrypted data.

4. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 3 is characterized in that: The first, second and third formation parameter prediction modules establish a database based on geological survey data and historical tunneling data of the shield machine. The geological survey data is used to obtain the formation category label and surrounding rock grade label. The cutter head speed and penetration tunneling parameters are selected for model training. The trained model is used to realize the formation category prediction and surrounding rock grade prediction. or, The formation category prediction is realized based on the trained support vector machine. During the training process of the support vector machine, the input values ​​of the support vector machine are the cutter head torque, cutter head rotation speed, propulsion force, penetration, tunneling speed, torque cutting depth index and field cutting depth index value, and rock uniaxial compressive strength, and the output value of the support vector set is the formation category; the formation category specifically includes: ultra-soft formation, relatively homogeneous soft rock, fully weathered soft and hard uneven formation, strongly weathered and moderately weathered soft and hard uneven formation, medium-soft formation, soft and hard uneven soft formation, soft and hard uneven hard formation, medium-hard formation, relatively homogeneous hard rock, and ultra-hard formation; when the loss function value of the support vector machine drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained support vector machine; or, The surrounding rock grade prediction is achieved based on a trained multilayer perceptron. During the training process, the input values ​​of the multilayer perceptron are disk rotation speed, penetration, gripper pressure, gripper pump pressure, gripper pitch angle, control pump pressure, propulsion speed and gripper rolling angle; the output values ​​of the multilayer perceptron are surrounding rock grades I, II, III, IV and V; when the loss function value of the multilayer perceptron drops to a minimum value, or the number of iterations reaches a set number, the training is stopped to obtain the trained multilayer perceptron.

5. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 2 is characterized in that: The first, second and third surface subsidence prediction modules realize the prediction of surface subsidence through the trained CNN-BiLSTM model; The CNN-BiLSTM model includes: a CNN network and a BiLSTM network connected in sequence; during the training process, the input values ​​of the CNN-BiLSTM model are bearing capacity, natural gravity, tunnel burial depth, distance between the monitoring point and the cutterhead, penetration, thrust, incision pressure, working chamber pressure, cutterhead speed and cutterhead torque; the output value of the CNN-BiLSTM model is the surface settlement prediction result; when the loss function value of the CNN-BiLSTM model drops to a minimum value, or the number of iterations reaches a set number, the training is stopped to obtain the trained CNN-BiLSTM model; or, The first, second and third tool wear prediction modules predict tool wear based on the trained BP neural network; during the training process, the BP neural network relies on the historical excavation parameters of the shield machine, the formation parameter information, and the tool wear information to establish a database, and selects the cutter head torque, cutter head speed, thrust, excavation speed, penetration, hob diameter, installation radius, hob spacing, rock uniaxial compressive strength, formation type and surrounding rock grade as input values; the output value of the BP neural network is the tool wear amount; when the loss function value of the BP neural network drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained BP neural network.

6. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 2 is characterized in that: The first, second and third shield attitude deviation prediction modules are based on the trained long short-term memory neural network model LSTM to predict the shield attitude deviation; during the model training process, the input values ​​of the long short-term memory neural network model LSTM are cutter head torque, cutter head speed, excavation speed, propulsion force, upper and lower cylinder thrust difference, cohesion, internal friction angle, lateral pressure coefficient and tunnel burial depth; The output values ​​are the horizontal deviation of the incision, the vertical deviation of the incision, the horizontal deviation of the shield tail, and the vertical deviation of the shield tail. When the loss function value of the model drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained long short-term memory neural network model LSTM. or, The first, second and third tunneling parameter prediction modules realize tunneling parameter prediction based on the trained XGBoost model; During the training process, the input variables of the XGBoost model are cutterhead extrusion pressure, penetration, incision pressure, excavation speed, grouting pressure, grouting flow rate, grouting flow rate, formation type, surrounding rock grade, torque cutting depth index and field cutting depth index value; the output variables of the XGBoost model are cutterhead torque, cutterhead speed and total thrust; when the loss function value of the XGBoost model drops to the minimum value, or the number of iterations reaches the set number, the training is stopped to obtain the trained XGBoost model.

7. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 2 is characterized in that: The first, second and third tunneling parameter optimization modules are based on multi-objective optimization theory, and select cutterhead speed, cutterhead torque, total thrust and penetration as optimization variables. The NSGA-II algorithm is used to achieve multi-objective optimization of maximum propulsion speed and minimum tunneling specific energy. The multi-objective optimization framework is as follows: F = (min(f1), max(f2)); s.t.n min ≤n≤n max ;v min ≤v≤v max ;F min ≤F≤F max ; T min ≤T≤T max ;P min ≤P≤P max ; Among them, E s is the tunneling specific energy, n is the cutterhead speed, T is the cutterhead torque, F is the total thrust, R is the cutterhead radius of the shield machine, v is the tunneling speed, and P is the penetration of the shield cutterhead; or, The first, second and third shield construction risk warning modules are used to give warning results for various construction risks during the construction process. The construction risk warning modules include: geological risk warning, equipment risk warning, and environmental impact warning; geological risk warning: when the shield machine is physically excavating, the digital twin of the shield machine is synchronously carried out in the digital space. Since the shield digital twin system has imported the three-dimensional geological information of the construction area during modeling, the geological information in front of the excavation face can be intuitively displayed in the digital twin system. Based on the prediction results of the formation parameter prediction module and the visual display of the geological characteristics in front of the excavation face, warning information can be issued in advance before the geological conditions change; Equipment risk warning, based on the tool wear prediction module, monitors the tool status, predicts failures and abnormalities, and takes measures in advance to avoid risks; at the same time, the shield digital twin system integrates the real-time data of the shield machine, and monitors the operating status of the main bearing and hydraulic system based on the visualization module. When the equipment parameters exceed the rated value or do not work, the equipment risk warning module issues a warning signal and gives the abnormal location; environmental risk warning, including surface subsidence warning and toxic gas warning; safety accident warning, the shield machine entity and supporting systems are equipped with monitoring equipment, combined with on-site monitoring video, when a safety accident occurs, the accident location is given according to the monitoring information.

8. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 2 is characterized in that: The first, second and third automatic control modules: The automatic control module relies on the Internet of Things to realize the control of the shield machine entity by the shield digital twin system. In order to make full use of the function of the tunneling parameter optimization module, the automatic control module converts the optimization strategy into control instructions. The control instructions are sent to the controller of the shield machine entity through the Internet of Things to achieve precise control and optimized operation of the shield machine entity and adjust the construction parameters in real time; or, The first, second and third visualization modules are used to display the established shield digital twin model. They are closely integrated with all the above-mentioned functional modules, display the real-time collected construction parameters and the shield digital twin model parameters constructed in the early stage in the form of charts, and visualize the predicted optimization results of each functional module.

9. The shield tunnel risk early warning method based on digital twin and federated learning as claimed in claim 1 is characterized in that: The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant, including: the central server is responsible for receiving and decrypting the model update parameters and gradients from each participant, verifying the integrity and correctness of the data, and preventing the data from being damaged or maliciously tampered with during transmission; or, Each participant updates its own model according to the average value of the parameters, and implements shield construction risk warning based on the updated model, including: the participant receives and decrypts the latest model data and gradient information sent from the central server, and integrates the information into the local shield construction risk warning model. After multiple debugging and tests, the accuracy of the local shield construction risk warning model is verified.

10. The shield tunnel risk early warning system based on digital twin and federated learning is characterized by: include: Shield tunnel construction client, shield machine manufacturer client, third-party client and central server; The shield tunnel construction client, as an edge computing terminal, obtains geological parameters and construction parameters; the shield tunnel construction client transmits the obtained data to the shield machine production client and third-party clients after homomorphic encryption; the shield machine production client transmits the shield equipment parameters to the shield tunnel construction client and third-party clients after homomorphic encryption; geological parameters, construction parameters and shield equipment parameters are collectively referred to as basic parameters; The shield machine manufacturer client, shield tunnel construction client and third-party client all construct the first, second and third shield digital twin systems respectively based on basic parameters; each shield digital twin system is equipped with a shield construction risk warning model; The shield machine manufacturer client, shield tunnel construction client, third-party client and central server together constitute a federated learning system; the shield machine manufacturer client, shield tunnel construction client and third-party client are all participants, and each participant sends the model parameters of its own shield construction risk warning model to the central server; The central server takes the average value of the parameters of the three models and sends the average value of the parameters to each participant; each participant updates its own model according to the average value of the parameters and implements shield construction risk warning based on the updated model.

Images