Information safety evaluation and construction optimization method for shield wall grinding underneath passing of existing station

Abstract

The invention relates to the technical field of tunnel engineering intelligent construction and safety monitoring, and discloses an information safety evaluation and construction optimization method for shield wall grinding underneath passing of an existing station, and the method comprises the steps: constructing a system state vector evolving along with time, and enabling the system state vector to be in a unified mathematical structure, the physical state, the control input, the unmeasurable state and the information health degree of the shield construction system are jointly described; based on an extended Kalman filtering framework, using a graph neural network as a state transfer function, and performing prior prediction on the system state vector at the current moment according to the system state vector at the previous moment and control input; and correcting the prior prediction in combination with actual measurement data at the current moment. A graph topology dynamic adaptive mechanism is designed, so that the whole prediction model has self-evolution and real-time correction capabilities, and the adaptability of the model to various uncertain risks in the construction process is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent construction and safety monitoring of tunnel engineering, and in particular to an information security assessment and construction optimization method for a shield machine grinding wall underpass through an existing station. Background Art

[0002] As the main construction technology for the development of underground space in modern cities, the shield method has been widely used in the construction of subway tunnels, cross-river tunnels, and integrated pipeline corridors. With the acceleration of urbanization, the number of scenarios in which tunnels pass under or approach existing buildings and structures is increasing. Among them, the "grinding wall underpass" working condition, in which the shield machine excavates with a very small clearance close to or even rubs against the side walls of existing station structures, constitutes one of the most challenging technical problems in the current tunnel engineering field due to its extremely high construction risks and strict deformation control requirements. Under this working condition, the stress release and redistribution caused by shield tunneling can easily lead to uneven settlement, tilting, and even cracking of existing station structures, seriously threatening the safety of operating lines and passengers.

[0003] To ensure construction safety, existing technologies usually deploy a large number of monitoring instruments to conduct real-time monitoring of key physical quantities of the ground surface, existing station structures, and the surrounding environment, and set graded warning thresholds. However, this approach, which mainly relies on observable physical quantities for risk control, has inherent limitations. Its risk judgment is often based on deformations that have already occurred, which is a hysteresis. When the monitoring data exceeds the warning threshold, it usually means that the potential risk has accumulated to a certain extent and has produced external, irreversible effects. Existing technologies lack effective means to penetrate these surface data and accurately estimate the key unpredictable states deep in the stratum or inside the structure that drive the evolution of risks in real time, making it difficult to achieve advanced warning and root cause diagnosis of risks.

[0004] At the same time, to predict the impact of construction, the engineering community often uses empirical formulas or numerical simulation techniques based on finite element and finite difference methods. While these methods can provide important references during the engineering design and early analysis stages, their limitations are also more prominent during the construction process. For one thing, these predictive models are mostly based on geological survey data acquired before construction and idealized parameter settings. Once established, their structures tend to be static. Faced with the complex, changing, and uncertain geological conditions encountered during construction, these static models struggle to dynamically adjust themselves in real time to changing physical realities, resulting in a significant decline in their predictive accuracy over time.

[0005] On the other hand, the value of the massive, multi-source, and heterogeneous monitoring data collected at construction sites has not been fully tapped. Existing technologies often analyze data from different sources in isolation or perform simple data overlays, lacking a unified theoretical framework that can deeply integrate physical mechanisms and monitoring data. More importantly, in harsh construction environments, sensors may malfunction, drift, or be subject to signal interference, resulting in inconsistent measurement data quality. Existing technologies generally lack mechanisms for real-time "health" assessments of the data stream itself, often using all data indiscriminately for analysis. This can lead to erroneous judgments due to the introduction of "dirty data," seriously impacting the reliability of decision-making.

[0006] Furthermore, current shield tunneling parameter control still relies heavily on the operator's personal experience and on-the-spot judgment. While this passive, feedback-based control model, based on "monitoring-alarm-adjustment," can correct deviations to a certain extent, it remains essentially a post-event fix. In high-risk conditions like "grinding wall underpass," the lack of a feedforward control method that can proactively optimize current construction parameters based on accurate predictions of future risks makes it difficult to systematically eliminate risks before they occur. Summary of the Invention

[0007] In response to the shortcomings of the existing technology, the present invention provides an information security assessment and construction optimization method for shield tunneling through existing stations. It solves the problems of existing shield construction safety monitoring technology, such as static models, inability to adapt to dynamic uncertainties in the construction process, and risk perception relying on lagged apparent deformation data, making it difficult to predict deep risks and actively optimize and control them.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: an information security assessment and construction optimization method for shield tunneling through an existing station, comprising the following steps:

[0009] S1. Construct a system state vector that evolves over time, wherein the system state vector describes the physical state, control input, unmeasurable state, and information health of the shield construction system within a unified mathematical structure;

[0010] S2. Based on the extended Kalman filter framework and using a graph neural network as a state transfer function, a priori prediction of the system state vector at the current moment is made based on the system state vector and control input at the previous moment; and the a priori prediction is corrected based on the actual measurement data at the current moment to obtain the posterior optimal estimate at the current moment;

[0011] S3. Monitoring the unmeasurable latent state vector in the posterior optimal estimate in real time, and automatically adjusting the graph topology structure on which the graph neural network relies when the unmeasurable latent state vector exhibits a preset abnormal pattern;

[0012] S4. Based on the posterior optimal estimate and the adjusted graph topology, a model predictive control method is used to solve an optimal control input sequence that minimizes a preset cost function in the future prediction time domain, and the current instructions in the optimal control input sequence are used to optimize the control of shield construction parameters.

[0013] Preferably, the system state vector X t The construction method is:

[0014]

[0015] in,

[0016] P t is the observable physical state vector, including the monitored physical quantities of the station and the surface;

[0017] U t is the control input vector, which contains shield construction parameters;

[0018] M t It is an unmeasurable implicit state vector, including physical quantities that cannot be directly measured, such as formation stress and structural internal force;

[0019] H t is an information health vector, which includes a quantitative evaluation value of the reliability of each data source in the observable physical state vector.

[0020] Preferably, the graph topology dynamic self-adaptation step specifically includes:

[0021] diagnosing whether the rate of change or the value of the unmeasurable latent state vector exceeds a preset expected range to identify a risk source area;

[0022] When the risk source area is identified, the current graph topology structure is determined to be invalid, and based on the location and impact range of the risk source area, the connection weights between corresponding nodes in the graph topology structure are enhanced, or new connection relationships are established between nodes to generate a new graph topology structure that can reflect the current physical impact path.

[0023] Preferably, the adjustment of the graph topology is performed by updating its weight matrix A t Implementation satisfies the following relationship:

[0024] A t =F update (A t-1 ,M t|t );

[0025] in,

[0026] A t-1 is the weight matrix of the previous moment;

[0027] M t|t is the posterior optimal estimate of the unmeasurable hidden state vector at the current moment;

[0028] F update Update functions for the preset graph structure.

[0029] Preferably, in the system state dynamic estimation step, the updating method of performing a posteriori optimal estimation on the system state vector is:

[0030] X t|t =X t|t-1 +K t (Z t -h obs (X t|t-1 ));

[0031] in,

[0032] X t|t is the posterior optimal estimate;

[0033] X t|t-1 is a priori prediction;

[0034] K t is the Kalman gain;

[0035] Z t is the actual measurement data;

[0036] h obs is the observation function;

[0037] The updating process uses the measurement residual (Z t -h obs (X t|t-1 )), synchronously correct the observable physical state vector, the unmeasurable hidden state vector and the information health vector.

[0038] Preferably, the updating of the information health vector is linked to the system state dynamic estimation step; when the measurement residual corresponding to specific measurement data continues to exceed a preset range, the reliability quantitative evaluation value corresponding to the data in the information health vector is automatically reduced.

[0039] Preferably, in the construction parameter feedforward optimization step, the preset cost function J is constructed as follows:

[0040]

[0041] in,

[0042] The cost function aims to minimize the prediction time domain N p Predicting physical state With the reference trajectory P ref The deviation of the control input U t+k|t energy consumption;

[0043] Q mpc and R mpc R mpc is the corresponding weight matrix.

[0044] Preferably, the graph neural network acts as a state transfer function and performs message passing operations on the graph topology to simulate the spatiotemporal transmission process in which the control input affects the entire system state through the formation medium.

[0045] Preferably, the method further comprises an initial graph construction step, which is performed before construction begins, and establishes an initial graph topology structure based on the three-dimensional spatial coordinates of each monitoring point and shield machine and geological survey data.

[0046] The information security assessment and construction optimization system for shield tunneling through existing stations includes:

[0047] A system state space construction unit is used to construct a system state vector that evolves over time. The system state vector describes the physical state, control input, unmeasurable state, and information health of the shield construction system within a unified mathematical structure.

[0048] a system state dynamic estimation unit, connected to the system state space construction unit, for performing a priori prediction of the system state vector at the current moment based on the system state vector and control input at the previous moment, based on an extended Kalman filter framework and using a graph neural network as a state transfer function; and correcting the a priori prediction based on actual measurement data at the current moment to obtain a posterior optimal estimate at the current moment;

[0049] A graph topology dynamic adaptive unit, connected to the system state dynamic estimation unit, is used to monitor the unmeasurable latent state vector in the posterior optimal estimate in real time, and automatically adjust the graph topology structure on which the graph neural network relies when the unmeasurable latent state vector shows a preset abnormal pattern;

[0050] The construction parameter feedforward optimization unit is connected to the system state dynamic estimation unit and the graph topology dynamic adaptive unit, and is used to solve an optimal control input sequence that minimizes a preset cost function in the future prediction time domain based on the posterior optimal estimate and the adjusted graph topology structure, using the model predictive control method, and use the current instructions in the optimal control input sequence to optimize the control of shield construction parameters.

[0051] The present invention provides an information security assessment and construction optimization method for a shield tunneling machine grinding a wall under an existing station.

[0052] It has the following beneficial effects:

[0053] 1. This invention employs a dynamic graph topology adaptive mechanism to enable the entire prediction model to self-evolve and adapt in real time, significantly enhancing the model's adaptability to various uncertain risks during the construction process. This mechanism can identify risk sources caused by sudden geological changes, unknown obstacles, and other unexpected factors online based on real-time estimates of unpredictable latent states. It also automatically adjusts the graph topology upon which the graph neural network relies, ensuring that the model can promptly and accurately reflect the system's latest physical impact paths, avoiding prediction failures caused by model rigidity.

[0054] 2. The present invention introduces a feedforward optimization strategy based on model predictive control. This strategy uses an established and dynamically adaptive system model to perform risk deduction over a longer time domain in the future and solve an optimal control parameter sequence that can balance safety and efficiency. It can proactively and proactively adjust current construction parameters to avoid potential risks that will occur in the future, thereby improving the active safety of the construction process.

[0055] 3. This invention innovatively introduces an information health vector into the system state vector and establishes a linked update mechanism for it and the measurement residuals in the extended Kalman filter. This effectively suppresses the negative impact of poor-quality data, such as sensor failure, noise, or signal interference, on overall system decision-making, ensuring the robustness and reliability of the entire assessment and optimization system. This mechanism automatically identifies and quantitatively assesses the real-time reliability of field sensor data. When data anomalies are detected, it dynamically and intelligently reduces their weight in the state estimation process, enabling online assessment and management of information security. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flow chart of the method of the present invention;

[0057] Figure 2 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] Example:

[0060] Please see the attached Figure 1 The embodiment of the present invention provides an information security assessment and construction optimization method for a shield tunneling machine passing through an existing station, including the following steps:

[0061] S1. Construct a system state vector that evolves over time. The system state vector describes the physical state, control input, unmeasurable state, and information health of the shield construction system within a unified mathematical structure.

[0062] The core purpose of this step is to transform the complex, multi-physics coupled construction site and its inherent mechanical connections into a structured data model—the initial spatial coupling graph—that can be processed and analyzed by computers before actual shield tunneling begins. This abandons the traditional approach of treating each monitoring data source as an isolated information source. By establishing a network of connections between entities, the present invention forms the basis for the subsequent state evolution prediction and dynamic graph topology adaptation based on graph neural networks.

[0063] Specifically, this step first requires comprehensive identification and node definition of the key physical entities in the construction scene. The selection of nodes is directly related to the level of detail of the physical world that can be described by the subsequent model. In a preferred embodiment, the node set V includes at least the following types:

[0064] Shield machine node: The shield machine is regarded as a single, centralized disturbance source node. This node is unique in the graph, and its attributes include the real-time position and posture of the shield machine in three-dimensional space, and serve as the subsequent control input vector U t The object of action.

[0065] Existing station monitoring nodes: These nodes are the key monitoring targets of this invention and correspond to all automated monitoring sensors deployed in key locations of existing station structures. For example, displacement gauges, strain gauges, inclinometers, etc. deployed on the station floor, side walls, roof, and internal columns are defined as independent nodes. The properties of these nodes will be directly related to the subsequent observable physical state vector P. t The corresponding components in .

[0066] Surface and environmental monitoring nodes: To comprehensively assess construction impacts, monitoring sensors deployed within the surface subsidence monitoring network, adjacent important buildings, and underground pipelines must be defined as nodes. This ensures the model's assessment scope covers all objects requiring protection.

[0067] After completing the node definition of all key entities, it is necessary to construct the initial edge set E0 and the initial weight matrix A0, which together constitute the initial spatial coupling graph G0 = (V, E0, A0).

[0068] The establishment of edges is not arbitrary but based on the principles of physical proximity and structural relevance. A preferred implementation is to set an influence radius threshold. For any two nodes whose spatial distance is less than this threshold, an initial edge is established, indicating the potential for mutual influence between them. For directly connected structural components, such as different measurement points on the same column, edges are established regardless of distance.

[0069] The most critical step in constructing the initial spatial coupling graph is to quantize the initial weight matrix A0. Each element A0[i,j] in the matrix is ​​used to represent the node v i With v j The intensity of the initial physical impact between the two should be set scientifically and reasonably by comprehensively considering a variety of prior information to reflect the initial physical characteristics of the system before the disturbance occurs as accurately as possible.

[0070] In this embodiment, the initial weight value is set by comprehensively considering the following two core factors:

[0071] Spatial distance decay effect: According to Saint-Venant's principle, the influence of local effects decays with increasing distance. Therefore, part of the weight is modeled as a decreasing function of the spatial distance between nodes. Preferably, a Gaussian kernel function or an inverse distance weighting function can be used to quantify this decay relationship. For example, its spatial influence component can be expressed as:

[0072]

[0073] in,

[0074] d(i,j) is the node v i and v j The Euclidean distance between

[0075] σ is a distance scaling factor that can be preset based on engineering experience or site size to control the speed of impact decay.

[0076] Differences in impact transmission media: The efficiency of impact transmission between two nodes depends on the properties of the medium between them. For example, the efficiency of impact transmission through hard rock or reinforced concrete structures is much higher than that through soft silty soil layers. To this end, the present invention introduces a geological or structural transmission coefficient γ ij The value of this coefficient is mainly based on the geological survey report and structural design drawings. For example, if two nodes are located in the same continuous, high elastic modulus rock layer, then the transfer coefficient γ ij Set to a higher value; if two nodes are separated by a weak interlayer, fault or structural settlement joint, the transfer coefficient γ ij Should be set to a lower value.

[0077] Finally, the element A0[i,j] in the initial weight matrix can be obtained by combining the above spatial influence components with the medium transfer coefficient. A preferred combination is to multiply the two:

[0078]

[0079] The weights set in this way not only reflect the general rule that the closer the distance, the greater the impact, but also take into account the differentiated effects of different geological and structural conditions on the transmission of impact.

[0080] Through the above steps, a weighted graph G0 = (V, E0, A0) is generated that comprehensively and meticulously reflects the initial physical characteristics of the construction system. It should be understood that this initial graph is not static; rather, it serves as the starting point for the graph neural network operations in subsequent steps, particularly the dynamic estimation of the system state step (S3) and the basis for adjustments in the dynamic adaptation of the graph topology step (S4). The rationality and sophistication of its construction have a significant impact on the ultimate performance of the entire method.

[0081] S2. Based on the extended Kalman filter framework, using a graph neural network as the state transfer function, the system state vector at the current moment is predicted a priori based on the system state vector and control input at the previous moment. The a priori prediction is then corrected based on the actual measurement data at the current moment to obtain the posterior optimal estimate at the current moment.

[0082] This step is the logical starting point and core foundation for implementing the cyber-physical fusion method of this invention. Its purpose is to systematically integrate traditionally independent, heterogeneous, and multi-source construction information, including physical world responses, human control instructions, and the quality of the information itself, to construct a comprehensive state space that can be fully described and precisely evolved within a unified mathematical framework.

[0083] Specifically, the core of this step is to construct a system state vector X that evolves over time. tThe design of this vector fully considers the complexity and uncertainty of the shield construction system. It describes the physical state, control input, unmeasurable state and information health of the shield construction system in a unified mathematical structure. In a preferred embodiment, the system state vector X t How to build:

[0084]

[0085] Among them, each component of the vector has a clear physical meaning and technical connotation, which are explained one by one below:

[0086] First, the observable physical state vector It is a direct and explicit representation of the system's physical response. This vector aggregates all physical quantity readings that can be directly measured and acquired in real time by sensors deployed on-site. Preferably, this includes, but is not limited to, 3D displacement readings of all monitoring points on the existing station structure, concrete strain values ​​at key structural locations, and overall or local inclination of the structure.

[0087] Secondly, the control input vector It is the external cause and direct source of changes in the drive system's state. This vector comprehensively encompasses the key construction parameters currently applied to the shield machine that can be controlled manually or by the system. Examples include the shield machine's total thrust, cutterhead drive torque, cutterhead speed, actual tunneling speed, soil bin pressure balance, and the grouting pressure and single-ring grouting volume of the synchronous grouting system.

[0088] In particular, the present invention introduces the unmeasurable hidden state vector M t ∈R dm . The purpose of establishing this vector is to deeply realize that observable physical deformation is only a manifestation of system risk, and its internal driving force is the evolution of internal mechanical states that cannot be directly and comprehensively monitored. Therefore, this vector is specifically used to describe hidden physical quantities that have a decisive impact on system safety but cannot be directly measured by conventional sensors. Preferably, its content may include: additional stress distribution in the stratum below the existing station floor, internal forces in key sections of the station structure, and pore water pressure distribution deep in the stratum. Accurately estimating the real-time value of the hidden state vector is the key to the ability of the present invention to achieve early risk warning and deep cause diagnosis. It builds a bridge from observable phenomena to the essence of risk.

[0089] In addition, in order to solve the problem of uneven quality, noise and even gross errors of measurement data in actual engineering and to evaluate the security of information itself, the present invention innovatively defines the information health vector H in the system state vector. t ∈R dhThis vector is related to the observable physical state vector P t Each data source in the data set corresponds one-to-one, and each component is a normalized value in the interval [0, 1], which is used to dynamically and quantitatively assess the reliability or trustworthiness of the corresponding sensor data stream at the current moment. For example, a component value of 1 indicates that the system considers the corresponding data source to be completely reliable; if a component value approaches 0, it indicates that the system believes that the corresponding data source may be drifting, faulty, or severely interfered with by unmodeled factors, and its information value is low. The introduction of this vector enables the method of the present invention to transcend the scope of traditional data preprocessing and filtering, providing key support for subsequent developments.

[0090] The unified state vector X constructed by the above steps t , organically integrating the originally scattered and heterogeneous physical measurements, control parameters, internal stresses, information quality and other multi-dimensional information into a unified, high-dimensional mathematical framework, thereby transforming a complex, nonlinear geotechnical engineering problem into a state-space problem that can be systematically solved using modern control theory and data science methods, laying a solid data and theoretical foundation for subsequent dynamic estimation, adaptive adjustment and feedforward optimization.

[0091] S3: Real-time monitoring of the unmeasurable latent state vector in the posterior optimal estimate. When the unmeasurable latent state vector shows a preset abnormal pattern, the graph topology structure on which the graph neural network relies is automatically adjusted.

[0092] This step constitutes the central computing core and information fusion hub of the method of the present invention. Its fundamental purpose is to transform the high-dimensional, unified system state vector constructed in step S2 from a static mathematical definition to a dynamic model that can dynamically evolve and be calibrated by real-time data. This step uses a recursive, closed-loop computational process to integrate the model's prior predictions with the actual measurements of the sensors, thereby obtaining a complete system state vector X at each time step. t ——In particular, the unmeasurable hidden state vector M that cannot be directly measured t and information health vector H t ——Give the most likely and accurate posterior optimal estimate.

[0093] To achieve the above goals, this step preferably uses the Extended Kalman Filter as its core algorithm framework. The EKF is a powerful tool for handling nonlinear system state estimation problems. Its implementation in the present invention is specifically completed by alternating between the "a priori prediction" and "posteriori correction" stages.

[0094] During the a priori prediction phase, the system's task is to infer the system's state at the current moment (t) based on the optimal state estimate and control instructions at the previous moment (t-1). The core of this phase is to establish a state transition function that accurately describes the physical evolution of the system. Given the highly nonlinear, time-varying, and spatially coupled nature of influence transfer within the "shield-stratum-station" system, this paper innovatively employs a graph neural network to serve as this state transition function, f(·).

[0095] The graph neural network operates on the graph topology constructed in step S1 and dynamically adjusted in step S4. Specifically, it simulates the message transmission and aggregation process between graph nodes to replicate how the control input of the shield machine propagates in spatial and temporal dimensions through the complex strata and structural media represented by the graph structure, ultimately causing the states of all nodes in the entire system to change. This process profoundly simulates the spatiotemporal transmission process of physical influences. This prior prediction process can be mathematically described by the following state transition equation:

[0096] X t|t-1 =f(X t-1|t-1 ,U t-1 ,G t-1 )+w t-1 ;

[0097] in,

[0098] X t|t-1 That is, the prior prediction value of the system state at the current moment;

[0099] X t-1|t-1 is the posterior optimal estimate obtained at the previous moment;

[0100] U t-1 is the control input vector at the previous moment;

[0101] G t-1 Represents the graph topology at the previous moment;

[0102] And w t-1 is the process noise, which is used to describe the prediction uncertainty inherent in the nonlinear model and cannot be completely avoided. It is usually assumed to obey a Gaussian distribution with zero mean.

[0103] In the posterior correction stage, the system obtains the actual measurement data Z at the current moment t Afterwards, its core task is to use this "ground truth" information to correct the prior predictions that are full of uncertainty.

[0104] The core of this correction process is the use of measurement residuals. Measurement residuals are defined as the difference between the actual measurement value and the predicted measurement value calculated based on the prior predicted state. The mathematical expression of this correction update process is:

[0105] X t|t =X t|t-1 +K t (Z t -h obs (X t|t-1 ));

[0106] In this equation, X t|t This is the final output of this step, the posterior optimal estimate at the current moment. obs (·) is the observation function, which is used to extract the actual measurement value Z from the complete state vector t The corresponding component. K t The Kalman gain is a weight matrix dynamically calculated based on the model prediction uncertainty and measurement noise uncertainty. Its core function is to make intelligent trade-offs: when the model prediction is more reliable, the predicted value should be trusted more; when the actual measurement data quality is high, the measurement value should be relied upon more.

[0107] Crucially, the present invention utilizes a directly observable measurement residual (Z t -h obs (X t|t-1 )), achieving synchronous correction of the entire high-dimensional state vector. This means that not only the observable physical state P t is updated, and more importantly, the unmeasurable state vectors M that cannot be directly measured t (such as deep formation stress) and information health vector H t Optimal estimation and calibration are also performed simultaneously.

[0108] What is particularly important is that the present invention has a significant impact on the information health vector H t The update mechanism is innovatively designed, and this mechanism is deeply linked to the dynamic estimation step of this state. Specifically, the system continuously monitors the measurement residual corresponding to each sensor data. When the absolute value or statistical characteristics of the measurement residual of a certain sensor continuously and significantly exceed the normal fluctuation range determined by historical data within several consecutive time steps, the system has reason to determine that the data source may no longer be reliable. At this time, the system will automatically trigger an attenuation mechanism to gradually and smoothly reduce the information health vector H t The reliability quantification evaluation value corresponding to the sensor in . This reduced evaluation value will be corrected in the posterior correction phase of the next calculation cycle by increasing the measurement noise covariance matrix R tThe value of the corresponding item in is used to automatically reduce the weight of the untrusted data in the state update process.

[0109] Through the continuous recursive cycle of prediction and correction mentioned above, this step can ultimately provide an optimal estimate of the current state of the system that is comprehensive, dynamic, and has information security assessment capabilities.

[0110] S4. Based on the posterior optimal estimate and the adjusted graph topology, the model predictive control method is used to solve an optimal control input sequence that minimizes the preset cost function in the future prediction time domain, and the current instructions in the optimal control input sequence are used to optimize the control of the shield construction parameters.

[0111] This step represents a key technological breakthrough compared to traditional static or pre-set model approaches. Its core purpose is to empower the entire cyber-physical fusion model with the ability to self-evolve and modify in real time. In the highly uncertain and dynamic process of shield construction, sudden changes in geological conditions and unforeseen structural responses are common and significant sources of risk. This step is designed to ensure that the technical solution of the present invention can perceive and adapt to these sudden changes online and in real time, rather than being constrained by the initial model established before construction, thereby ensuring the continued effectiveness of state estimation and feedforward optimization.

[0112] Specifically, the execution of this step is closely connected with step S3. After the system obtains the current moment's posterior optimal estimate X by extended Kalman filtering, t|t After that, this step will be initiated immediately. It does not continuously and indiscriminately modify the graph structure, but is implemented through a strict logical chain of "diagnosis-trigger-update".

[0113] First, the real-time diagnosis of the risk source area is carried out. The core of this process is to estimate the unmeasurable hidden state vector M in the posterior optimal estimate outputted in step S3. t|t Conduct continuous, in-depth monitoring. As claimed in claim 3, the system includes a built-in diagnostic function Ψ(·), which is specifically used to analyze whether the rate of change of the unmeasurable latent state vector or its value itself exceeds a dynamically expected range preset based on historical data, physical laws, and expert experience.

[0114] For example, in a preferred embodiment, the diagnostic function calculates in real time the spatial gradient and time rate of change of the additional stress estimate in the key area of ​​the formation. When a nonlinear, far-exceeding-expected surge in stress values ​​in a certain local area is detected, or a rapid, abnormal accumulation of pore water pressure is detected, the diagnostic function will output a high-risk signal. This signal indicates that a "risk source area" has been successfully identified. The essence of this identification process is that the system determines that the current physical phenomenon can no longer be described using the existing graph topology structure G. t-1The physical impact path implied by the graph is reasonably explained, that is, the current graph topology is judged to be ineffective in describing the emerging risk patterns.

[0115] Secondly, regarding the triggering and execution of graph structure updates, once the diagnostic module identifies a risk source and triggers an update signal, the system automatically calls a graph structure update function to make precise, targeted adjustments to the existing graph topology. This adjustment is not a global reconstruction, but rather focuses on the risk source and its potential impact.

[0116] The adjustment may specifically include one or more methods: one is to enhance the connection weights between corresponding nodes in the graph topology. For example, if an abnormal stratum stress is identified in a certain area directly in front of the shield machine, the system will significantly enhance the connection weights between the nodes within the area, as well as between all nodes in the area and the shield machine nodes. The physical meaning of this in the model is that it recognizes and strengthens the consistency of the physical state within the area, as well as the intensity of the direct impact of the shield machine activities on the area. Another way is to establish new connection relationships between nodes when necessary. For example, if it is found that two areas that were originally far apart and judged to have no direct connection have highly correlated abnormal deformation responses, the system can establish new edges between representative nodes of the two areas to reflect the newly emerged remote coupling effects that were not considered by the initial model.

[0117] Finally, the entire adaptive adjustment process is encapsulated in a formal mathematical operation, that is, by updating the weight matrix A of the graph t As defined in claim 4, the update method satisfies the following relationship:

[0118] A t =F update (A t-1 ,M t|t );

[0119] In this relation, each symbol has a clear technical connotation: A t-1 is the weight matrix of the previous moment, which forms the basis of this update; M t|t is the posterior optimal estimate of the unmeasurable hidden state vector at the current moment provided by step S3. It is the core information source driving this update and provides key intelligence about "where" and "what happened". update The default graph structure update function can be understood as a set of specific rules or algorithms that transform hidden state anomalies into graph structure adjustments. In a preferred embodiment, this function can be a series of heuristic rules based on geotechnical mechanics or a lightweight neural network trained offline.

[0120] Crucially, the updated weight matrix A output by this step is t , will be immediately used in the calculation loop of the next time step (t+1). It will serve as a new substrate for the state transition prediction of the graph neural network in step S3, and as the latest model based on the feedforward optimization of the model predictive control in step S5. In this way, this step tightly bundles the system's state estimation, model structure, and optimization decision-making together to form a closed-loop, self-evolving intelligent system that can adapt to environmental changes, ensuring the robustness and reliability of the entire method in the face of the complexity and uncertainty of the construction site.

[0121] Please see the attached Figure 2 The information security assessment and construction optimization system for shield tunneling through existing stations includes:

[0122] The system state space construction unit is used to construct a system state vector that evolves over time. The system state vector describes the physical state, control input, unmeasurable state and information health of the shield construction system within a unified mathematical structure.

[0123] The system state dynamic estimation unit is connected to the system state space construction unit. It is used to make a priori prediction of the system state vector at the current moment based on the system state vector and control input at the previous moment based on the extended Kalman filter framework and using the graph neural network as the state transfer function. It also corrects the prior prediction based on the actual measurement data at the current moment to obtain the posterior optimal estimate at the current moment.

[0124] The graph topology dynamic adaptation unit is connected to the system state dynamic estimation unit and is used to monitor the unmeasurable hidden state vector in the posterior optimal estimate in real time. When the unmeasurable hidden state vector shows a preset abnormal pattern, it automatically adjusts the graph topology structure that the graph neural network relies on;

[0125] The construction parameter feedforward optimization unit is connected to the system state dynamic estimation unit and the graph topology dynamic adaptation unit. It is used to solve an optimal control input sequence that minimizes the preset cost function in the future prediction time domain based on the posterior optimal estimation and the adjusted graph topology structure, using the model predictive control method, and use the current instructions in the optimal control input sequence to optimize the control of the shield construction parameters.

[0126] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0127] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. Information security assessment and construction optimization method for shield tunneling through existing stations, characterized by: The following steps are involved: S1. Construct a system state vector that evolves over time, wherein the system state vector describes the physical state, control input, unmeasurable state, and information health of the shield construction system within a unified mathematical structure; S2. Based on the extended Kalman filter framework, using a graph neural network as a state transfer function, a priori prediction of the system state vector at the current moment is performed based on the system state vector and control input at the previous moment; The a priori prediction is corrected in combination with the actual measurement data at the current moment to obtain the posterior optimal estimate at the current moment; S3. Monitoring the unmeasurable latent state vector in the posterior optimal estimate in real time, and automatically adjusting the graph topology structure on which the graph neural network relies when the unmeasurable latent state vector exhibits a preset abnormal pattern; S4. Based on the posterior optimal estimate and the adjusted graph topology, a model predictive control method is used to solve an optimal control input sequence that minimizes a preset cost function in the future prediction time domain, and the current instructions in the optimal control input sequence are used to optimize the control of shield construction parameters.

2. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: The system state vector X t The construction method is: in, P t is the observable physical state vector, including the monitored physical quantities of the station and the surface; U t is the control input vector, which contains shield construction parameters; M t It is an unmeasurable implicit state vector, including physical quantities that cannot be directly measured, such as formation stress and structural internal force; H t is an information health vector, which includes a quantitative evaluation value of the reliability of each data source in the observable physical state vector.

3. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: The graph topology dynamic adaptation step specifically includes: diagnosing whether the rate of change or the value of the unmeasurable latent state vector exceeds a preset expected range to identify a risk source area; When the risk source area is identified, the current graph topology structure is determined to be invalid, and based on the location and impact range of the risk source area, the connection weights between corresponding nodes in the graph topology structure are enhanced, or new connection relationships are established between nodes to generate a new graph topology structure that can reflect the current physical impact path.

4. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 3 is characterized in that: The graph topology is adjusted by updating its weight matrix A t Implementation satisfies the following relationship: A t =F update (A t-1 ,M t|t ); in, A t-1 is the weight matrix of the previous moment; M t|t is the posterior optimal estimate of the unmeasurable hidden state vector at the current moment; F update Update functions for the preset graph structure.

5. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: In the system state dynamic estimation step, the updating method of performing the a posteriori optimal estimation on the system state vector is: X t|t =X t|t-1 +K t (Z t -h obs (X t|t-1 )); in, X t|t is the posterior optimal estimate; X t|t-1 is a priori prediction; K t is the Kalman gain; Z t is the actual measurement data; h obs is the observation function; The updating process uses the measurement residual (Z t -h obs (X t|t-1 )), synchronously correct the observable physical state vector, the unmeasurable hidden state vector and the information health vector.

6. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: The updating of the information health vector is linked to the system state dynamic estimation step; when the measurement residual corresponding to specific measurement data continues to exceed a preset range, the reliability quantitative evaluation value corresponding to the data in the information health vector is automatically reduced.

7. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: In the construction parameter feedforward optimization step, the preset cost function J is constructed as follows: in, The cost function aims to minimize the prediction time domain N p Predicting physical state With the reference trajectory P ref The deviation of the control input U t+k|t energy consumption; Q mpc and R mpc R mpc is the corresponding weight matrix.

8. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: The graph neural network serves as a state transfer function and performs message passing operations on the graph topology to simulate the spatiotemporal transmission process in which the control input affects the state of the entire system through the formation medium.

9. The information security assessment and construction optimization method for shield milling wall underpass of an existing station according to claim 1 is characterized in that: The method further comprises an initial graph construction step, which is performed before construction begins, and establishes an initial graph topology structure based on the three-dimensional spatial coordinates of each monitoring point and the shield machine and geological survey data.

10. An information security assessment and construction optimization system for a shield tunneling system with a grinding wall passing through an existing station, according to the information security assessment and construction optimization method for a shield tunneling system with a grinding wall passing through an existing station according to any one of claims 1 to 9, characterized in that: include: A system state space construction unit is used to construct a system state vector that evolves over time. The system state vector describes the physical state, control input, unmeasurable state, and information health of the shield construction system within a unified mathematical structure. a system state dynamic estimation unit, connected to the system state space construction unit, for performing a priori prediction of the system state vector at the current moment based on the system state vector and control input at the previous moment, based on an extended Kalman filter framework and using a graph neural network as a state transfer function; The a priori prediction is corrected in combination with the actual measurement data at the current moment to obtain the posterior optimal estimate at the current moment; A graph topology dynamic adaptive unit, connected to the system state dynamic estimation unit, is used to monitor the unmeasurable latent state vector in the posterior optimal estimate in real time, and automatically adjust the graph topology structure on which the graph neural network relies when the unmeasurable latent state vector shows a preset abnormal pattern; The construction parameter feedforward optimization unit is connected to the system state dynamic estimation unit and the graph topology dynamic adaptive unit, and is used to solve an optimal control input sequence that minimizes a preset cost function in the future prediction time domain based on the posterior optimal estimate and the adjusted graph topology structure, using the model predictive control method, and use the current instructions in the optimal control input sequence to optimize the control of shield construction parameters.

Images