Intelligent evaluation method and system for large deformation of surrounding rock of deep-buried tunnel based on digital twinning

Abstract

The application discloses a kind of based on digital twinning's deep-buried tunnel surrounding rock large deformation intelligent evaluation method and system, comprising: first, the tunnel surrounding rock digital twinning model of fusion geometry and mechanical property is established. Then section level mapping is carried out to model with construction monitoring data, and structured data is formed. Then execute risk assessment, by extracting section deformation characteristics, dividing mechanical and spatial feature set, constructing topology structure and carrying out coupling calculation, generating target features to assess risk. Subsequently, based on the result-driven digital twinning model, mechanical evolution simulation is carried out to predict the deformation trend. Finally, according to the simulation data, the risk level is determined and the warning information is generated. The application realizes the deep combination of data, mechanism and spatial relationship, improves the accuracy and timeliness of risk assessment, and provides intelligent decision support for construction safety.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method and system for intelligent assessment of large deformation of surrounding rock in deep-buried tunnels based on digital twins. Background Technology

[0002] In the construction of deep-buried tunnels, large deformation of the surrounding rock is one of the major geological hazards threatening construction safety and progress. Traditional methods for assessing surrounding rock deformation mainly rely on simple threshold comparisons of on-site monitoring data and engineers' experience in finite element simulation. These methods have significant shortcomings: on the one hand, monitoring data and geomechanical models are isolated from each other, making it difficult to reflect the complex mechanical interactions and spatiotemporal evolution of the surrounding rock system; on the other hand, risk assessments are mostly based on static or offline analysis, failing to achieve dynamic, real-time early warning of risks during construction. Digital twin technology provides a framework for integrating physical entities and virtual models, but existing applications are mostly focused on visualization and data mapping, lacking intelligent assessment models that deeply integrate real-time monitoring data, surrounding rock mechanical mechanisms, and spatial topological relationships. This results in insufficient accuracy in early identification of large deformation risks in surrounding rock and limited ability to predict development trends, making it difficult to provide timely and accurate guidance for dynamic construction decisions. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for intelligent assessment of large deformation of surrounding rock in deep-buried tunnels based on digital twins.

[0004] In a first aspect, embodiments of the present invention provide an intelligent assessment method for large deformation of surrounding rock in deeply buried tunnels based on digital twins, including: A digital twin model of the tunnel surrounding rock is established based on the tunnel's geological structure parameters and design and construction parameters. The digital twin model of the tunnel surrounding rock is used to characterize the spatial positional relationship of each monitoring section and the mechanical properties of the surrounding rock. Acquire monitoring data during the tunnel construction period, and perform cross-section-level mapping and association between the monitoring data during the tunnel construction period and the digital twin model of the tunnel surrounding rock to obtain structured monitoring data corresponding to each monitoring section; Based on the structured monitoring data, a risk assessment of large deformation of the surrounding rock is performed, and the risk assessment result of large deformation of the surrounding rock is obtained. Based on the risk assessment results of large deformation of the surrounding rock, the digital twin model of the tunnel surrounding rock is driven to perform mechanical evolution simulation of the surrounding rock to obtain data on the development trend of surrounding rock deformation. The risk level of large deformation of the surrounding rock is determined based on the data on the development trend of surrounding rock deformation, and corresponding construction early warning information is generated.

[0005] In a second aspect, embodiments of the present invention provide a server system, including a server, the server being used to execute the method described in the first aspect.

[0006] Compared to existing technologies, the beneficial effects of this invention include: The invention employs a digital twin-based intelligent assessment method and system for large deformation of surrounding rock in deeply buried tunnels. This involves establishing a digital twin model of the tunnel's surrounding rock that integrates geometric and mechanical properties. Next, construction monitoring data is mapped to the model at the cross-sectional level to form structured data. Then, a risk assessment is performed by extracting cross-sectional deformation features, dividing mechanical and spatial feature sets, constructing a topological structure, and performing coupled calculations to generate target features for risk assessment. Subsequently, the results drive the digital twin model to perform mechanical evolution simulations to predict deformation trends. Finally, the risk level is determined based on the simulation data, and early warning information is generated. This invention achieves a deep integration of data, mechanisms, and spatial relationships, improving the accuracy and timeliness of risk assessment and providing intelligent decision support for construction safety. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A flowchart illustrating the steps of the intelligent assessment method for large deformation of surrounding rock in deep-buried tunnels based on digital twins provided in this embodiment of the invention; Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0010] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the intelligent assessment method for large deformation of surrounding rock in deep-buried tunnels based on digital twins, as provided in this embodiment. The following is a detailed description of this intelligent assessment method for large deformation of surrounding rock in deep-buried tunnels based on digital twins.

[0012] Step S201: Establish a digital twin model of the tunnel surrounding rock based on the tunnel geological structure parameters and design and construction parameters. The digital twin model of the tunnel surrounding rock is used to characterize the spatial positional relationship of each monitoring section and the mechanical properties of the surrounding rock. Step S202: Obtain monitoring data during tunnel construction, and perform cross-section-level mapping and association between the monitoring data during tunnel construction and the digital twin model of the surrounding rock of the tunnel to obtain structured monitoring data corresponding to each monitoring cross-section. Step S203: Perform a risk assessment of large deformation of surrounding rock based on the structured monitoring data to obtain the risk assessment result of large deformation of surrounding rock; Step S204: Based on the risk assessment results of large deformation of the surrounding rock, drive the digital twin model of the tunnel surrounding rock to perform mechanical evolution simulation of the surrounding rock to obtain data on the development trend of surrounding rock deformation. Step S205: Determine the risk level of large deformation of the surrounding rock based on the data on the development trend of surrounding rock deformation, and generate corresponding construction early warning information.

[0013] In this embodiment of the invention, for example, firstly, the server establishes a digital twin model of the tunnel surrounding rock based on the tunnel's geological structure parameters and design and construction parameters. Specifically, the server receives and integrates data from geological survey reports, design drawings, and construction plans. This data includes, but is not limited to, rock mass type, joint and fracture development, geostress field information, tunnel axis coordinates, cross-sectional shape and size, and support parameters. Using these parameters, the server constructs a three-dimensional solid model in a 3D modeling engine that accurately reflects the actual tunnel's geometry and spatial relationship. Simultaneously, the server assigns corresponding material constitutive models (such as elastoplastic models) to different rock strata and regions in the model based on the surrounding rock's mechanical properties (such as elastic modulus, Poisson's ratio, cohesion, and internal friction angle), thus forming a digital twin model of the tunnel surrounding rock that integrates geometric and mechanical properties. The core value of this model lies in its ability to characterize the precise spatial location of each monitoring section distributed along the tunnel axis, as well as the mechanical behavior characteristics of the surrounding rock at each section, providing a high-fidelity virtual environment for subsequent simulation and analysis.

[0014] Next, the server acquires monitoring data from the tunnel construction period and maps this data to the established digital twin model of the tunnel surrounding rock at the section level to obtain structured monitoring data. During actual construction, sensors (such as multi-point displacement gauges, stress gauges, and convergence piles) deployed on various monitoring sections within the tunnel continuously generate monitoring data, including displacement, stress, and convergence values. The server receives these raw, multi-source monitoring data streams in real-time or periodically through a data interface. Then, based on the sensor installation location information, the server accurately maps each set of monitoring data to the corresponding monitoring section spatial location in the digital twin model. For example, the server binds the crown settlement data and sidewall convergence data of a monitoring section at a certain station number to the virtual representation node of that section in the digital twin model. Through this section-level mapping, the server transforms the raw, discrete monitoring data into a structured monitoring dataset that strictly corresponds to the model structure, is organized by section, and includes spatiotemporal labels, providing a standardized data foundation for subsequent analysis.

[0015] Then, the server performs a risk assessment of large deformation of the surrounding rock based on the aforementioned structured monitoring data, obtaining the risk assessment results. This is a core analysis step. The server first performs feature extraction processing on the structured monitoring data of each monitoring section. For example, the server reads the displacement observation sequence of a section at multiple time points and calculates its deformation rate, cumulative deformation, etc., as the observation features of that section. Simultaneously, based on the spatial topological relationships of each section in the digital twin model (such as the adjacent relationships along the tunnel axis and the geometric relationships between measuring points within the same section), the server generates a structured representation describing the spatial relationships between sections and between measuring points within a section, i.e., the geometric relationship features of the sections. Subsequently, the server couples the observation features with the geometric relationship features to generate section deformation features that comprehensively reflect the deformation state and spatial constraints.

[0016] To conduct a more in-depth evaluation, the server further decomposes the deformation characteristics of each cross-section into a first set of rock mass parameter features strongly correlated with mechanical response (such as features directly related to stress redistribution and plastic zone development) and a second set of rock mass parameter features with closer spatial correlation. The server calculates the mechanical response correlation degree between features in the first set of rock mass parameter features (e.g., by analyzing the covariant relationship of feature values ​​under a mechanical model) and constructs a first cross-section correlation topology reflecting the mechanical coupling relationship between them. This structure can be a weighted graph, where nodes are features and edge weights represent the correlation degree. Simultaneously, based on the spatial topological correlation of the monitored cross-section data, the server constructs a second cross-section correlation topology reflecting the spatial adjacency or influence relationships between the second set of rock mass parameter features.

[0017] Subsequently, the server performs coupling parameter calculations on both topologies. For a specific first rock mass parameter feature in the first cross-section associated topology, the server identifies other features directly associated with it in the topology (i.e., associated cross-section unit features), calculates the comprehensive response (e.g., weighted average) of these associated features, and the offset between the feature itself and this comprehensive response. Then, the server combines the feature's own value, offset, and comprehensive response value, and performs information fusion and transformation through a trained fully connected neural network (i.e., full-dimensional parameter mapping calculation) to output a first coupling parameter feature that better represents its state in the current mechanical coupling network. The server performs this operation on all first rock mass parameter features in the topology, obtaining a set of first coupling parameter features. A similar operation is performed on the second rock mass parameter features in the second cross-section associated topology, calculating a set of second coupling parameter features based on spatial association relationships.

[0018] Finally, the server combines the first coupling parameter feature and the second coupling parameter feature obtained from the decomposition of the deformation feature of the same cross-section and the above-mentioned topological coupling calculation (e.g., vector concatenation) to form the target cross-section deformation feature. This feature integrates information from both mechanical response and spatial correlation after in-depth processing. The server inputs this target cross-section deformation feature into a pre-trained risk assessment classification model (such as a gradient boosting decision tree or a deep learning classifier). The model outputs a probability or level assessment of the current large deformation risk of the cross-section, thus obtaining the surrounding rock large deformation risk assessment result. This result identifies which cross-sections and to what extent large deformation risk exists.

[0019] Furthermore, based on the obtained risk assessment results of large deformation of the surrounding rock, the server drives the digital twin model of the tunnel surrounding rock to perform a simulation of the mechanical evolution of the surrounding rock, obtaining data on the development trend of surrounding rock deformation. The server uses the high-risk section information identified in the risk assessment results, along with the corresponding deformation characteristics and load conditions, as boundary conditions or disturbance factors, and loads them onto the corresponding parts of the digital twin model. Then, the server calls the integrated finite element analysis or discrete element analysis solver to perform a time-step mechanical simulation calculation on the digital twin model. The simulation process simulates the evolution of the stress field and displacement field over time under the current construction progress and surrounding rock conditions. By running the simulation, the server predicts the possible development path of surrounding rock deformation, the extent of plastic zone expansion, and potential instability modes in the future (such as the next construction cycle or a few hours later), thereby outputting a series of data on the deformation development trend, such as displacement-time prediction curves for key points and cloud maps of stress concentration zones.

[0020] Finally, the server determines the risk level of large deformation of the surrounding rock based on the simulation data of the surrounding rock deformation development trend and generates corresponding construction early warning information. The server analyzes the simulation output data and extracts key indicators, such as the maximum predicted displacement value, deformation acceleration trend, and plastic zone penetration status. The server compares these indicators with preset risk level thresholds (e.g., deformation control standards corresponding to minor, moderate, severe, and extremely severe levels set according to specifications and experience) to determine the final risk level of each high-risk section. Based on the determined risk level, the server automatically generates structured construction early warning information. This information may include the warning section station number, risk level, main risk characteristics (e.g., "expected crown settlement exceeds limit"), and suggested treatment measures (e.g., "strengthen support" and "temporarily suspend tunneling"), and is promptly sent to on-site construction management personnel through the monitoring center's visual interface, SMS, or application push, providing direct support for dynamic and precise construction decision-making and risk management.

[0021] In this embodiment of the invention, the step of performing a risk assessment of large deformation of surrounding rock based on the structured monitoring data to obtain the risk assessment result of large deformation of surrounding rock can be implemented through the following example.

[0022] In this embodiment of the invention, for example, the server acquires structured monitoring data that has been mapped and associated with a digital twin model. This data is organized by monitoring cross-section, for example, including time-series observations from multiple cross-sections such as chainage K10+100 and K10+120. The server first performs a segmentation process, classifying each data record into different monitoring cross-section data sets based on the cross-section identifier, thus obtaining multiple independent monitoring cross-section data objects. Next, the server performs feature extraction processing on each monitoring cross-section data. Taking the chainage K10+100 cross-section as an example, the server reads the observation sequences of multiple sensors for crown settlement, sidewall convergence, etc., over the past 24 hours for this cross-section, and calculates its mean, variance, recent rate of change, and other statistical quantities as observation features. Simultaneously, the server obtains from the digital twin model the spatial distance and orientation relationship between this cross-section and adjacent cross-sections (such as K10+080 and K10+120), as well as the geometric layout between various measuring points within the cross-section, encoding these spatial topological relationships into structured feature vectors, i.e., cross-section geometric relationship features. Finally, the server uses a feature fusion module to couple observed features with geometrically correlated features, generating a comprehensive vector that fully characterizes the current deformation state and spatial context information of the cross-section—the cross-section deformation feature. Subsequently, the server classifies each cross-section deformation feature separately. Based on predefined feature classification rules, the server divides the various dimensions (i.e., sub-features) constituting the cross-section deformation feature into two categories. For example, sub-features directly related to stress, strain, and deformation rate, such as "maximum principal stress change rate" and "estimated plastic strain energy density," are classified to form the first rock mass parameter feature set associated with the cross-section deformation feature. Simultaneously, sub-features related to spatial location, relative distance, and topological connectivity strength, such as "distance from upstream fault zone" and "correlation coefficient with adjacent cross-section deformation," are classified to form the second rock mass parameter feature set associated with the cross-section deformation feature. Next, the server constructs two associated topologies based on mechanical principles and data analysis. For the first rock mass parameter feature set, the server calculates the mechanical response correlation degree between any two first rock mass parameter features. For example, the server calculates the covariance and physical correlation strength between the two features "internal displacement of surrounding rock" and "stress of support structure" by analyzing historical data or based on elastoplastic mechanics formulas, obtaining a quantified correlation value. The server iterates through all feature pairs, calculates all correlation degrees, and constructs a weighted undirected graph, namely the first cross-section correlation topology, based on this. In this graph, nodes represent various first rock mass parameter features, and the weights of the connecting edges are the calculated mechanical response correlation degrees, which characterize the tightness of coupling between features in terms of mechanical mechanisms. Simultaneously, for the second set of rock mass parameter features, the server directly constructs the structure based on the inherent spatial topological correlation relationships of each monitoring cross-section data.For example, based on the spatial adjacency relationships of each section in the digital twin model, the server determines whether there are spatial influence links between second rock mass parameter features (such as the "axis offset" feature of different sections), and constructs a second section-related topology structure accordingly. This structure focuses more on reflecting the spatial correlation of features. Then, the server performs coupling parameter calculations on both topologies to deepen feature representation. In the first section-related topology structure, the server selects one of the first rock mass parameter features, such as "displacement rate of measuring point A". The server finds all other feature nodes directly connected to this feature node in the topology graph; these are its associated section unit features. The server calculates the weighted sum of these associated feature values ​​(weighted by the edge weights) to obtain the first comprehensive response feature. Simultaneously, the server calculates the difference between the "displacement rate of measuring point A" feature value and this first comprehensive response feature value to obtain the first response offset feature. Next, the server concatenates the original "displacement rate of measuring point A" feature value, the calculated first response offset feature value, and the first comprehensive response feature value into a new vector, which is then input into a pre-trained fully connected neural network (i.e., a full-dimensional parameter mapping calculation module). This network learns to fuse these three pieces of information and map them to a new feature vector that better reflects the dynamic state of the feature in the local mechanical coupling network, namely the first coupling parameter feature. The server performs the above operation on each first rock mass parameter feature in the topology, thereby generating the corresponding first coupling parameter feature for each feature in the first rock mass parameter feature set. Similarly, on the second cross-section associated topology, the server performs a similar operation on each second rock mass parameter feature: finding its associated cross-section unit features based on spatial association, calculating the second comprehensive response feature and the second response offset feature, and obtaining the second coupling parameter feature corresponding to each second rock mass parameter feature through the same full-dimensional parameter mapping calculation. Finally, the server performs joint construction and final risk assessment. For the same monitoring section, the server sequentially concatenates and combines all the first and second coupling parameter features corresponding to the section deformation characteristics, obtained after the aforementioned complex coupling calculations, to form a new, highly integrated feature vector—the target section deformation feature. This target feature deeply integrates high-level abstract information from both mechanical response mechanisms and spatial correlation constraints. The server inputs this target section deformation feature into a pre-trained classification model for large deformation risk assessment of surrounding rock. This model (e.g., a deep neural network classifier) ​​analyzes the feature vector and outputs a risk assessment result, such as "high risk," or a risk probability value like 0.92. The server sequentially executes this process for all monitoring sections, ultimately summarizing the risk assessment results for large deformation of surrounding rock for each section corresponding to the entire tunnel monitoring interval, thus completing the intelligent analysis process from raw structured monitoring data to refined risk assessment.

[0023] In this embodiment of the invention, the feature extraction process to obtain the deformation features of each cross section can be implemented through the following example.

[0024] In this embodiment of the invention, exemplarily, the server performs feature extraction processing to obtain the deformation features of each cross-section. The server first acquires the observation features of the data for each monitoring cross-section. Taking the K10+100 cross-section as an example, the server reads the raw data sequence of multiple sensors deployed at this cross-section within the latest time window, including readings from the crown settlement gauge and the sidewall convergence survey line. The server processes this time-series data, calculating its statistical characteristics, such as the average deformation rate, cumulative deformation, and data fluctuation variance over the past 6 hours. These calculated values ​​are arranged in a predetermined order to form a numerical vector, thus obtaining the cross-sectional observation features of the cross-section. Next, the server acquires the spatial topological relationships of the monitoring cross-section data and represents them in a structured manner. The server queries the digital twin model of the tunnel surrounding rock to extract the spatial attributes of the K10+100 cross-section. This includes the cross-section's coordinates in three-dimensional space, its precise distance and azimuth along the tunnel axis to the upstream K10+080 and downstream K10+120 cross-sections, and the relative geometric positions of monitoring points within the cross-section (such as the arch apex and left arch waist). The server encodes these complex spatial relationships into a structured, machine-readable feature vector, for example, using a set of distance values, angle cosine values, and connection identifiers, to obtain the cross-section's geometric correlation features. Finally, the server couples the obtained cross-section observation features and the corresponding cross-section geometric correlation features to generate a representation. The server takes the two feature vectors as input and feeds them into a feature fusion module. This module fuses and compresses the information from the two vectors through a pre-defined transformation layer (such as a fully connected neural network layer), outputting a new, fixed-dimensional feature vector. This new vector simultaneously contains the real-time deformation state information of the cross-section and its unique spatial context information, thus generating the final cross-section deformation features for subsequent in-depth analysis.

[0025] In this embodiment of the invention, the construction of the first cross-section association topology structure corresponding to the first rock mass parameter feature set based on the mechanical response correlation degree between each first rock mass parameter feature in the first rock mass parameter feature set can be implemented through the following example.

[0026] In an embodiment of the invention, for example, the server first loads a first set of rock mass parameter features for a specific monitoring section (e.g., station K10+100). This set includes specific features such as "displacement rate inside the surrounding rock," "maximum principal stress change," and "estimated value of plastic strain energy density." The server calculates the correlation degree between the characteristic mechanical responses of these features pairwise. Specifically, the server calls the built-in correlation analysis module to calculate the Pearson correlation coefficient between each pair of feature values ​​based on the monitoring data sequence of the section over a historical period. This coefficient is then weighted and corrected using pre-set mechanical knowledge rules (e.g., according to elastoplastic theory, displacement rate and stress change should be strongly correlated in the yielding stage), ultimately assigning a quantified correlation degree value to each pair of features.

[0027] Based on the calculated correlation of characteristic mechanical responses, the server determines the mechanical coupling relationship between each first rock mass parameter feature. The server sets a correlation threshold, for example, 0.7. When the calculated correlation between two features is greater than this threshold, the server determines that there is a significant mechanical coupling relationship between the two features; if it is lower than the threshold, the relationship is considered weak or negligible. For example, if the calculated correlation between "displacement rate inside the surrounding rock" and "stress of the support structure" is 0.85, the server determines that there is a strong mechanical coupling relationship between the two.

[0028] Subsequently, the server treats each first rock mass parameter feature as an independent cross-sectional feature unit (i.e., a node in the topological network). Based on the mechanical coupling relationships determined in the previous step, the server establishes a bidirectional connection mapping between strongly correlated feature units. The server formally constructs all feature units and their connections into a weighted undirected graph data structure. In this graph, nodes represent the various feature units, connecting edges represent the mechanical coupling relationships between them, and the weight of the edge is the previously calculated feature mechanical response correlation degree value. This complete weighted undirected graph is the final first cross-sectional correlation topology structure, quantitatively characterizing the network relationships of mutual influence and mutual constraint among the key mechanical parameters within the cross-section.

[0029] In this embodiment of the invention, the construction of the second cross-section association topology structure corresponding to the second rock mass parameter feature set based on the spatial topological association relationship of each monitoring cross-section data can be implemented through the following example.

[0030] In this embodiment of the invention, for example, the server first determines the cross-sectional spatial representation of the second rock mass parameter features associated with each cross-section deformation feature based on the inherent spatial topological relationships of each monitoring cross-section data in the digital twin model. For example, for the K10+100 cross-section, its second rock mass parameter feature set includes features such as "distance from the upstream fault zone" and "correlation coefficient with the deformation of the K10+080 cross-section". The server queries the model to obtain the specific spatial attribute values ​​corresponding to these features, such as a distance of 15 meters and a correlation coefficient of 0.6. These values ​​constitute the cross-sectional spatial representation of the features. Based on these cross-sectional spatial representations, the server determines the spatial relationships between the features in the second rock mass parameter feature set. The server calculates the spatial distribution relationship of different features. For example, by analyzing the gradient of the value of the "distance from the upstream fault zone" feature of multiple cross-sections as a function of the tunnel axis position, it determines whether there is a continuous spatial influence relationship between these features and assigns a spatial correlation strength value to the feature pair accordingly. Subsequently, the server treats each second rock mass parameter feature as a cross-sectional feature unit. Based on the spatial relationships determined in the previous step, the server establishes connection mappings between feature units with significant spatial relationships. Finally, the server constructs all feature units and their connection relationships into a graph structure, which is the second cross-sectional relational topology structure, depicting the networked connections between various spatially related features.

[0031] In this embodiment of the invention, the coupling parameter calculation based on the first rock mass parameter features and the associated cross-section unit features corresponding to the first rock mass parameter features in the first cross-section associated topology to obtain the first coupling parameter features associated with each first rock mass parameter feature in the first rock mass parameter feature set can be implemented through the following example.

[0032] In this embodiment of the invention, for example, the server loads a first cross-section associated topology structure already constructed for a certain monitoring cross-section (e.g., station K10+100). The server selects a first rock mass parameter feature in this topology structure as the current processing object, such as "displacement rate of measuring point A". In this topology graph, the server finds all other feature nodes that are directly connected to the feature node through edges. The features represented by these nodes (e.g., "stress of the support structure" and "displacement inside the surrounding rock") are the associated cross-section unit features corresponding to the current feature. Then, the server calculates the comprehensive response feature of these associated cross-section unit features to obtain the first comprehensive response feature. Specifically, the server reads the latest value of each associated feature and uses the weight of the edge connecting it to the current feature node (i.e., the mechanical response correlation degree) as a weighting coefficient to calculate the weighted sum of all associated feature values. For example, if the correlation degree of "stress of the support structure" is 0.85 and its value is 2.1 MPa, and the correlation degree of "displacement inside the surrounding rock" is 0.78 and its value is 5.3 mm, then the calculated first comprehensive response feature is a weighted scalar value that integrates these two associated features. Simultaneously, the server calculates the first response offset feature. The server takes the currently processed "displacement rate of measurement point A" feature value (e.g., 0.25 mm / h), subtracts the previously calculated first comprehensive response feature value, and the difference is the first response offset feature, which quantifies the degree of deviation of the current feature from the overall response of its local mechanical coupling network. Then, the server concatenates the original "displacement rate of measurement point A" feature value, the calculated first response offset feature value, and the first comprehensive response feature value in sequence to form a three-dimensional vector, i.e., the first joint constructed feature. The server inputs this joint constructed feature into a pre-trained fully connected neural network layer, defined as a full-dimensional parameter mapping calculation module. This module performs nonlinear transformation and information fusion on the input three-dimensional vector, outputting a new feature vector with designed dimensions (e.g., a five-dimensional vector), which is the first coupling parameter feature corresponding to the "displacement rate of measurement point A" feature. Finally, following the above process, the server sequentially traverses and processes each first rock mass parameter feature node in the associated topology of the first cross-section, such as "maximum principal stress change" and "estimated plastic strain energy density," generating a corresponding first coupling parameter feature for each feature. Ultimately, the server obtains the first coupling parameter features associated with each feature in the first rock mass parameter feature set of this cross-section, completing this round of coupling calculation.

[0033] In this embodiment of the invention, the coupling parameter calculation based on the second rock mass parameter features and the associated cross-sectional unit features corresponding to the second rock mass parameter features in the second cross-sectional associated topology, to obtain the second coupling parameter features associated with each second rock mass parameter feature in the second rock mass parameter feature set, can be implemented through the following example.

[0034] In this embodiment of the invention, for example, the server loads a second cross-section associated topology structure already constructed for a certain monitoring cross-section (e.g., station K10+100). The server selects a second rock mass parameter feature in this topology structure as the current processing object, such as "distance from the upstream fault zone". In this topology graph, the server finds all other feature nodes that are directly connected to the feature node through edges. The features represented by these nodes (e.g., the "axis offset" of the adjacent K10+080 cross-section, and the "deformation correlation coefficient with the upstream cross-section" of this cross-section) are the associated cross-section unit features corresponding to the current feature. Next, the server calculates the comprehensive response features of these associated cross-section unit features to obtain the second comprehensive response features. Specifically, the server reads the latest value of each associated feature and uses the weight of the edge connecting it to the current feature node (i.e., spatial correlation strength) as a weighting coefficient to calculate the weighted sum of all associated feature values. For example, if the correlation strength of "K10+080 section axis offset" is 0.6 and its value is 12 mm, and the correlation strength of "correlation coefficient with upstream section deformation" is 0.9 and its value is 0.75, then the calculated second comprehensive response feature is a weighted scalar value that integrates these two correlation features. Simultaneously, the server calculates the second response offset feature. The server takes the currently processed "distance from upstream fault zone" feature value (e.g., 15 meters), subtracts the just-calculated second comprehensive response feature value, and the difference is the second response offset feature, which quantifies the degree of deviation of the current feature relative to the overall state of its local spatial correlation network. Then, the server concatenates the original "distance from upstream fault zone" feature value, the calculated second response offset feature value, and the second comprehensive response feature value in sequence to form a three-dimensional vector, i.e., the second jointly constructed feature. The server inputs this jointly constructed feature into another fully connected neural network layer (i.e., the full-dimensional parameter mapping calculation module) with the same structure but independent parameters as when processing the first type of feature. This module performs nonlinear transformation and information fusion on the input 3D vector, outputting a new feature vector (e.g., a 4D vector), which is the second coupling parameter feature corresponding to the "distance from upstream fault zone" feature. Finally, following the above process, the server sequentially traverses and processes each second rock mass parameter feature node in the associated topology of the second fault section, generating a corresponding second coupling parameter feature for each feature. Ultimately, the server obtains the second coupling parameter features associated with each feature in the second rock mass parameter feature set of this fault section, completing this round of spatial topology-based coupling calculation.

[0035] In this embodiment of the invention, the first coupling parameter feature and the second coupling parameter feature associated with each cross-sectional deformation feature are jointly constructed to obtain the target cross-sectional deformation feature associated with each cross-sectional deformation feature. Based on the target cross-sectional deformation feature associated with each cross-sectional deformation feature, a risk assessment of large deformation of surrounding rock is performed to obtain the risk assessment result of large deformation of surrounding rock corresponding to the structured monitoring data. This can be implemented through the following example.

[0036] In this embodiment of the invention, for example, the server first loads a preset first response modulation parameter from the storage module. This parameter is a weight matrix trained using historical data. Based on this parameter, the server performs response modulation processing on the first coupling parameter features associated with each feature in the first rock mass parameter feature set. Specifically, the server performs matrix multiplication on each first coupling parameter feature vector and the first response modulation parameter matrix, and obtains a new feature vector through an activation function. For example, for the five-dimensional first coupling parameter feature corresponding to "displacement rate of measuring point A", after this transformation, a three-dimensional vector is output, which is the first response modulation feature associated with that feature. The server performs this operation sequentially on all features in the first rock mass parameter feature set of the cross section. Next, the server uses the same first response modulation parameter to perform response modulation processing on the second coupling parameter features associated with each feature in the second rock mass parameter feature set. The server also multiplies each second coupling parameter feature vector with the modulation parameter matrix and activates it. For example, for the four-dimensional second coupling parameter feature corresponding to "distance from upstream fault zone", after transformation, a three-dimensional vector is output, which is the second response modulation feature associated with that feature. The server performs this operation sequentially on all features in the second rock mass parameter feature set of the cross section. Then, the server performs joint construction. For the same monitoring section (e.g., K10+100), the server concatenates all its first-response modulation features and all its second-response modulation features according to the original classification order of the features into a longer, unified feature vector. This new vector is the final, deeply modulated deformation feature of the section. Finally, the server performs a large deformation risk assessment of the surrounding rock based on this modulated deformation feature. The server inputs this feature vector into a pre-trained large deformation risk classification model for the surrounding rock (e.g., a support vector machine or neural network classifier). The model analyzes the feature vector and outputs a clear risk level label (e.g., "high risk") or a risk probability value. The server repeats the above complete process from modulation to assessment for each monitoring section data included in the structured monitoring data, and finally summarizes the risk assessment results corresponding to each monitoring section, thereby completing the intelligent assessment of the large deformation risk of the surrounding rock in the entire tunnel monitoring section and generating the target large deformation risk assessment result.

[0037] In this embodiment of the invention, the first coupling parameter feature associated with each first rock mass parameter feature in the first rock mass parameter feature set is subjected to response modulation processing based on the first response modulation parameter to obtain the first response modulation feature associated with each first rock mass parameter feature in the first rock mass parameter feature set. This can be implemented through the following example.

[0038] In this embodiment of the invention, for example, the server loads a preset first response modulation parameter, which comprises two parts: a matrix (W_map) for state mapping and a matrix (W_weight) for weight modulation. First, the server performs state mapping reconstruction processing. The server multiplies the first coupling parameter feature (a five-dimensional vector (V_cpl)) corresponding to a feature in the first rock mass parameter feature set (e.g., the displacement rate of measuring point A) with the matrix (W_map) to obtain a new feature vector, namely the first state mapping feature (V_map). Subsequently, the server calculates the Euclidean distance between this first state mapping feature (V_map) and a predefined reference feature vector (V_ref) representing a "normal stable state." The calculated distance value is the first response bias assessment (D_bias) corresponding to this feature, which quantifies the degree to which the current feature state deviates from the benchmark. Next, the server performs weight modulation processing. The server multiplies the same first coupling parameter feature (V_cpl) with another matrix (W_weight) and passes it through a Sigmoid activation function to obtain a new feature vector, namely the first weight modulation feature (V_weight). This operation aims to recalibrate the feature values ​​based on their global importance. Finally, the server performs coupling operations. The server calculates the first weighted modulation feature (V_weight) and the first response bias estimate (D_bias). Specifically, the server performs element-wise multiplication (i.e., broadcasting) of the scalar (D_bias) and the vector (V_weight) to obtain a completely new feature vector. This resulting vector is the first response modulation feature (V_mod) ultimately corresponding to the "displacement rate of measuring point A" feature. The server processes each of the other features in the first rock mass parameter feature set in the same manner, such as "maximum principal stress change" and "estimated plastic strain energy density," generating corresponding first response modulation features for each of them, thus completing the entire modulation process.

[0039] In this embodiment of the invention, the second coupling parameter feature associated with each second rock mass parameter feature in the second rock mass parameter feature set is subjected to response modulation processing based on the first response modulation parameter to obtain the second response modulation feature associated with each second rock mass parameter feature in the second rock mass parameter feature set. This can be implemented through the following example.

[0040] In this embodiment of the invention, for example, the server loads and processes the same preset first response modulation parameters as when processing the first type of features. These parameters include a matrix (W_map) for state mapping and a matrix (W_weight) for weight modulation. First, the server performs state mapping reconstruction processing. The server multiplies the second coupling parameter feature (a four-dimensional vector (V_cpl2)) corresponding to a certain feature (e.g., "distance from the upstream fault zone") in the second rock mass parameter feature set with the matrix (W_map) to obtain a new feature vector, namely the second state mapping feature (V_map2). Subsequently, the server calculates the Euclidean distance between the second state mapping feature (V_map2) and a predefined reference feature vector (V_ref_spatial) representing the "typical spatial distribution state". The calculated distance value is the second response bias evaluation quantity (D_bias2) corresponding to the feature, which quantifies the degree to which the current spatial feature state deviates from the typical spatial distribution pattern. Next, the server performs weight modulation processing. The server multiplies the same second coupling parameter feature (V_cpl2) with the matrix (W_weight) and applies a sigmoid activation function to obtain a new feature vector, namely the second weight modulation feature (V_weight2). This operation aims to recalibrate the weight contributions of these spatially correlated features based on the global model. Finally, the server performs coupling operations. The server operates on the obtained second weight modulation feature (V_weight2) with the second response bias evaluation (D_bias2). Specifically, the server multiplies the scalar (D_bias2) element-wise with the vector (V_weight2) to obtain a new feature vector. This resulting vector is the final second response modulation feature (V_mod2) corresponding to the "distance from upstream fault zone" feature. The server processes each of the other features in the second rock mass parameter feature set in the same manner, generating corresponding second response modulation features for each of them, thus completing the modulation processing of the entire second type of features.

[0041] In this embodiment of the invention, the risk assessment of large deformation of surrounding rock based on the modulated cross-sectional deformation features associated with each cross-sectional deformation feature, and the resulting target surrounding rock large deformation risk assessment result corresponding to the structured monitoring data, can be implemented through the following example.

[0042] The modulation cross-sectional deformation features associated with each cross-sectional deformation feature are divided to obtain a first modulation rock mass parameter feature set, a second modulation rock mass parameter feature set, and a third modulation rock mass parameter feature set. The sum of the number of parameter dimensions of the second modulation rock mass parameter features in the second modulation rock mass parameter feature set and the number of parameter dimensions of the third modulation rock mass parameter features in the third modulation rock mass parameter feature set is the same as the number of parameter dimensions of the second rock mass parameter features. Based on the mechanical response correlation degree between each first modulated rock mass parameter feature in the first modulated rock mass parameter feature set, a first constraint-enhanced correlation topology structure corresponding to the first modulated rock mass parameter feature set is constructed, and based on the spatial topology correlation relationship of each monitoring section data, a second constraint-enhanced correlation topology structure corresponding to the second modulated rock mass parameter feature set is constructed. Based on the spatial topological association of each monitoring section data, determine the directly associated modulated rock mass parameter features of each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set. Based on the mechanical response association degree between the directly associated modulated rock mass parameter features of each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set, construct the third constraint-strengthened association topology structure corresponding to the third modulated rock mass parameter feature set. Based on the first modulated rock mass parameter features and the associated cross section unit features corresponding to the first modulated rock mass parameter features in the first constraint-enhanced associated topology, coupling parameters are calculated to obtain the first coupled modulation features associated with each first modulated rock mass parameter feature in the first modulated rock mass parameter feature set. Based on the associated cross section unit features associated with the second modulated rock mass parameter features and the second modulated rock mass parameter features in the second constraint-enhanced associated topology, coupling parameters are calculated to obtain the second coupled modulation features associated with each second modulated rock mass parameter feature in the second modulated rock mass parameter feature set. Based on the third constraint-enhanced correlation topology, the coupling parameters are calculated for the third modulated rock mass parameter features and the associated cross section unit features associated with each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set, to obtain the third coupled modulation features associated with each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set. Based on the first coupled modulation feature, the second coupled modulation feature and the third coupled modulation feature associated with the same cross-sectional deformation feature, the target modulation cross-sectional deformation feature associated with each cross-sectional deformation feature is jointly constructed to obtain the target modulation cross-sectional deformation feature associated with each cross-sectional deformation feature. Based on the target modulation section deformation features associated with each section deformation feature, a risk assessment of large deformation of the surrounding rock is performed, and the constrained correction risk assessment result of large deformation of the surrounding rock corresponding to the structured monitoring data is obtained.

[0043] In this embodiment of the invention, for example, the server first obtains the generated modulation section deformation features for a certain monitoring section (e.g., station K10+100). This feature is a fused vector containing all first and second response modulation features. The server then performs a partitioning process on this vector. Specifically, according to predefined rules, the server extracts the portion of the vector that originally corresponds to the first rock mass parameter features (i.e., all first response modulation features) to form a first modulation rock mass parameter feature set. Next, the server performs a secondary partitioning on the portion of the vector that originally corresponds to the second rock mass parameter features (i.e., all second response modulation features). For example, assuming the original second rock mass parameter features have a total dimension of 8, the server extracts the features of its first 5 dimensions to form a second modulation rock mass parameter feature set, and extracts the features of the remaining 3 dimensions to form a third modulation rock mass parameter feature set, thereby ensuring that the sum of the dimensions of the second and third sets is equal to the number of dimensions of the original second feature. Then, the server constructs three constraint-strengthening association topologies respectively. For the first set of modulated rock mass parameter features, the server calculates the mechanical response correlation degree between each feature (using a method similar to that used when constructing the first cross-section correlation topology), and constructs a first constraint-reinforced correlation topology based on this, which is a new weighted graph. For the second set of modulated rock mass parameter features, the server directly constructs a second constraint-reinforced correlation topology based on the spatial topological correlation relationships of the monitoring cross-section data (such as cross-section spatial adjacency information). For the third set of modulated rock mass parameter features, the server first determines, based on the spatial topological correlation relationships, other modulated rock mass parameter features that are directly correlated spatially with each third modulated rock mass parameter feature (these correlated features may come from the first or second set). Then, the server calculates the mechanical response correlation degree between these correlated modulated rock mass parameter features, and based on this, constructs a sub-topology network for each third feature that reflects the mechanical coupling state of its associated environment. All sub-networks of the third features together constitute the third constraint-reinforced correlation topology. Subsequently, the server performs coupling parameter calculations on these three new topologies respectively. The calculation process is consistent with the aforementioned process, but the input features and topological relationships have been updated. In the first constrained and reinforced topology, the server calculates the comprehensive response and offset of the associated features for each first modulated rock mass parameter feature, and obtains the first coupled modulation feature through full-dimensional parameter mapping. In the second constrained and reinforced topology, a similar operation is performed on each second modulated rock mass parameter feature to obtain the second coupled modulation feature. In the third constrained and reinforced topology, for each third modulated rock mass parameter feature, the server calculates the comprehensive response and offset of its directly associated other modulation features (as its associated cross-sectional unit features), and obtains the third coupled modulation feature through full-dimensional parameter mapping. Then, the server performs joint construction.For section K10+100, the server sequentially concatenates all its first, second, and third coupled modulation features into a new, highly integrated long feature vector, which is the final target modulation section deformation feature for that section. Finally, the server performs a large deformation risk assessment of the surrounding rock based on this target feature. The server inputs this feature vector into the final risk assessment classification model. This model outputs a corrected, more accurate risk level or probability. The server repeats the entire process for each section in the structured monitoring data, ultimately summarizing the risk assessment results for all sections after constraint correction, i.e., the constrained corrected large deformation risk assessment results for the surrounding rock, which serves as the final output of the entire intelligent assessment process.

[0044] In this embodiment of the invention, the construction of the third constraint-strengthened association topology corresponding to the third modulated rock mass parameter feature set based on the mechanical response correlation degree between each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set and the directly associated modulated rock mass parameter features can be implemented through the following example.

[0045] Select a reference section feature and a reference section feature from each of the third modulated rock mass parameter features; Based on the spatial topological correlation of each monitoring section data, each benchmark topological correlation feature corresponding to the benchmark section feature is determined from each third modulated rock mass parameter feature, and each benchmark topological correlation feature is subjected to topological coupling calculation to obtain the benchmark topological coupling feature. Based on the spatial topological correlation of each monitoring section data, each reference topological correlation feature corresponding to the reference section feature is determined from each third modulated rock mass parameter feature, and each reference topological correlation feature is subjected to topological coupling calculation to obtain the reference topological coupling feature. Calculate the mechanical response correlation degree between the reference topological coupling feature and the reference topological coupling feature to obtain the mechanical response correlation degree between the reference cross-sectional feature and the reference cross-sectional feature; Each third modulated rock mass parameter feature is processed sequentially to obtain the mechanical response correlation degree between each third modulated rock mass parameter feature and its directly associated modulated rock mass parameter features. The mechanical response correlation degree between each third modulated rock mass parameter feature and its directly associated modulated rock mass parameter features is taken as the target mechanical response correlation degree between each third modulated rock mass parameter feature. Based on the target mechanical response correlation degree, the target mechanical coupling correlation mapping relationship between each third modulated rock mass parameter feature is determined, and each third modulated rock mass parameter feature is used as a cross-sectional feature unit. Based on the target mechanical coupling correlation mapping relationship, a correlation mapping is established for each third modulated rock mass parameter feature to obtain the third constraint strengthening correlation topology.

[0046] In an embodiment of the invention, the exemplary process of the server constructing the third constraint-enhanced associated topology is as follows: The server loads a third modulated rock mass parameter feature set divided for a certain monitoring section (e.g., station K10+100). This set includes several features such as "correlation coefficient with upstream section deformation" and "distance from downstream fracture zone". The server first selects two features from this feature set as processing objects, for example, designating "correlation coefficient with upstream section deformation" as the baseline section feature and "distance from downstream fracture zone" as the reference section feature. Next, based on the spatial topological association relationship of the monitoring section data in the digital twin model, the server determines each baseline topological association feature corresponding to the baseline section feature. The server queries the model to find other modulated rock mass parameter features that are directly associated with the feature "correlation coefficient with upstream section deformation" in spatial topology. These associated features may come from the first or second modulated rock mass parameter feature set, such as "support stress modulation feature of upstream section K10+080" and "axis offset modulation feature of this section K10+100". The server calculates a weighted sum of the latest values ​​of the found baseline topological correlation features, weighted by their spatial correlation strength, to obtain a fused value, namely the baseline topological coupling feature. Similarly, the server determines each reference topological correlation feature corresponding to the reference section feature. For the feature "distance from downstream fracture zone", the server finds other modulation features that are directly correlated in spatial topology, such as "convergence rate modulation feature of downstream section K10+120". The server performs the same weighted summation calculation on these reference topological correlation features to obtain the reference topological coupling feature. Then, the server calculates the mechanical response correlation degree between the baseline topological coupling feature and the reference topological coupling feature. The server obtains the numerical sequences of these two coupling features over a period of time, calculates their Pearson correlation coefficient, and corrects them in conjunction with the mechanical model to finally obtain a quantified correlation degree value. This value is determined as the mechanical response correlation degree between the two third modulated rock mass parameter features: "correlation coefficient with upstream section deformation" (baseline feature) and "distance from downstream fracture zone" (reference feature). The server processes each pair of feature combinations in the third modulated rock mass parameter feature set in sequence according to the above process. For example, the correlation between the "correlation coefficient with upstream cross-section deformation" and other features within the set, as well as the correlation between "distance from downstream fracture zone" and other features, are calculated until all feature pairs are traversed. The server uses the correlation between all calculated pairs of features as their respective target mechanical response correlation. Based on the calculated target mechanical response correlation, the server determines the target mechanical coupling correlation mapping relationship between each third modulated rock mass parameter feature. The server sets a threshold, and features with correlation values ​​higher than this value are judged to have a significant target mechanical coupling correlation.Finally, the server treats each third-modulated rock mass parameter feature as a cross-sectional feature unit (i.e., a node). Based on the target mechanical coupling correlation mapping relationship determined in the previous step, the server establishes connection edges between feature units with significant correlations, and the weight of the edge is the corresponding target mechanical response correlation degree. The server constructs all third-modulated rock mass parameter feature units and their connection relationships into a complete weighted graph data structure, which is the final required third-constraint strengthened correlation topology. This structure characterizes the complex network relationship formed between these features after modulation and spatial filtering, based on the mechanical coupling state of their correlation environment.

[0047] In this embodiment of the invention, the joint construction of the first coupled modulation feature, the second coupled modulation feature, and the third coupled modulation feature associated with the same cross-sectional deformation feature to obtain the target modulation cross-sectional deformation feature associated with each cross-sectional deformation feature can be implemented through the following example.

[0048] Obtain the second response modulation parameter, and based on the second response modulation parameter, perform response modulation processing on the first coupled modulation feature associated with each first modulation rock mass parameter feature in the first modulation rock mass parameter feature set to obtain the first modulation response feature associated with each first modulation rock mass parameter feature in the first modulation rock mass parameter feature set. Based on the second response modulation parameter, the second coupled modulation feature associated with each second modulation rock mass parameter feature in the second modulation rock mass parameter feature set is subjected to response modulation processing to obtain the second modulation response feature associated with each second modulation rock mass parameter feature in the second modulation rock mass parameter feature set. Based on the second response modulation parameter, the third coupled modulation feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set is subjected to response modulation processing to obtain the third modulation response feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set. The first modulation response feature, the second modulation response feature, and the third modulation response feature associated with the same cross-sectional deformation feature are jointly constructed to obtain the target modulation cross-sectional deformation feature associated with each cross-sectional deformation feature.

[0049] In this embodiment of the invention, exemplarily, the server first loads a preset second response modulation parameter from the storage module. This parameter is a new set of parameters specifically used for feature modulation in the final stage, including a matrix (W_map2) for state mapping and a matrix (W_weight2) for weight modulation. Based on this second response modulation parameter, the server performs response modulation processing on the first coupled modulation feature associated with each feature in the first modulated rock mass parameter feature set. Taking the K10+100 section as an example, the server selects a feature from its first modulated rock mass parameter feature set, such as the first coupled modulation feature (a multi-dimensional vector) corresponding to the "displacement rate modulation feature of measuring point A". The server multiplies this vector with the matrix (W_map2) to obtain a new state mapping vector and calculates the deviation evaluation quantity between this vector and a stable state reference vector in the final stage. Simultaneously, the server multiplies the first coupled modulation feature vector with the matrix (W_weight2) and obtains the weight modulation vector through a Sigmoid activation function. Finally, the server multiplies the calculated deviation evaluation quantity element-wise with the weight modulation vector to obtain the final first modulation response feature of the feature. The server repeats this operation for each feature in the first modulated rock mass parameter feature set, obtaining a set of first modulated response features. Next, using the same second response modulation parameters, the server performs the exact same response modulation processing flow on the second coupled modulation features associated with each feature in the second modulated rock mass parameter feature set. The server sequentially operates on each second coupled modulation feature vector with matrices (W_map2) and (W_weight2), combined with bias evaluation calculations, ultimately generating a corresponding second modulation response feature for each feature in the second modulated rock mass parameter feature set. Then, the server again uses the same second response modulation parameters to perform the same response modulation processing on the third coupled modulation features associated with each feature in the third modulated rock mass parameter feature set. The server inputs each third coupled modulation feature vector into the same processing pipeline, undergoing state mapping reconstruction, bias evaluation calculations, and weighted modulation coupling, generating a corresponding third modulation response feature for each feature in the third modulated rock mass parameter feature set. Finally, the server performs the final joint construction. For the monitoring section K10+100, the server concatenates all generated first, second, and third modulation response features into a complete, unified long vector according to the classification order of the original feature set. This final synthesized feature vector integrates all high-level abstract information after two-level topological coupling calculation and two-level response modulation processing, representing the most comprehensive and in-depth digital representation of the surrounding rock state of this section. This is the target modulation section deformation feature ultimately associated with the deformation features of this section.The server performs this process for each monitoring section of data, generating unique target modulation section deformation characteristics for each section, which are then used by the final risk assessment model.

[0050] In this embodiment of the invention, the third coupled modulation feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set is subjected to response modulation processing based on the second response modulation parameter to obtain the third modulation response feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set. This can be implemented through the following example.

[0051] Based on the second response modulation parameters, the state mapping reconstruction process is performed on the third coupled modulation features associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set to obtain the third state mapping features associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set, and the response deviation evaluation quantity corresponding to the third state mapping feature is calculated to obtain the third response deviation evaluation quantity. The third coupled modulation feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set is subjected to weighted modulation processing to obtain the third weighted modulation feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set. The coupling operation result of the third weighted modulation feature and the third response deviation evaluation quantity is calculated to obtain the third modulation response feature associated with each third modulation rock mass parameter feature in the third modulation rock mass parameter feature set.

[0052] In this embodiment of the invention, for example, the server first loads a preset second response modulation parameter, which includes a matrix (W_map2) for state mapping and a matrix (W_weight2) for weight modulation. The server selects a feature from the third modulation rock mass parameter feature set as the current processing object, for example, the third coupling modulation feature (a multi-dimensional vector (V_cpl3)) corresponding to the "correlation coefficient with upstream section deformation" feature of section K10+100. The server performs state mapping reconstruction processing, multiplying the vector (V_cpl3) by the matrix (W_map2) and passing it through an activation function to obtain a new feature vector, namely the third state mapping feature (V_state3) associated with this feature. Next, the server calculates the response deviation evaluation value corresponding to the third state mapping feature. The server obtains a predefined reference vector (V_ref_final) that characterizes the numerical distribution pattern that the feature should have under ideal stable conditions. The server calculates the Euclidean distance between the third state mapping feature (V_state3) and this reference vector (V_ref_final). The calculated scalar distance value is the third response bias assessment (D_bias3) corresponding to this feature. This assessment quantifies the final deviation between the current feature's high-level abstract state and the ideal benchmark. Subsequently, the server performs weight modulation processing. The server multiplies the same third-coupling modulation feature vector (V_cpl3) with the matrix (W_weight2) and passes it through a sigmoid activation function to obtain a new feature vector, which is the associated third-weight modulation feature (V_weight3). The purpose of this operation is to perform a final calibration and scaling of the contribution of each dimension of the feature according to the global requirements of the final risk assessment model. Finally, the server performs coupling operations. The server performs operations on the obtained third-weight modulation feature (V_weight3) with the third response bias assessment (D_bias3). Specifically, the server multiplies the scalar (D_bias3) element-wise with the vector (V_weight3) to generate a completely new feature vector. This resulting vector is the final third-modulated response feature (V_resp3) corresponding to the feature "correlation coefficient with upstream cross-sectional deformation". Following the exact same process, the server sequentially processes each of the other features in the third modulated rock mass parameter feature set, such as the "distance from downstream fracture zone" feature, generating corresponding third modulated response features for each. At this point, the server has completed the final response modulation processing for all three types of modulation features (first, second, and third), preparing for the next step of joint construction.

[0053] In this embodiment of the invention, the following implementation methods are also provided.

[0054] The steps of taking the target modulation section deformation feature as the modulation section deformation feature and returning to divide the modulation section deformation features associated with each section deformation feature to obtain the first modulation rock mass parameter feature set, the second modulation rock mass parameter feature set, and the third modulation rock mass parameter feature set are executed. The number of parameter dimensions of the second modulation rock mass parameter feature in the second modulation rock mass parameter feature set is adjusted by expansion mapping according to the benchmark parameter quota. The number of parameter dimensions of the third modulation rock mass parameter feature in the third modulation rock mass parameter feature set is adjusted by contraction mapping according to the benchmark parameter quota. Under the condition of meeting the preset cycle adjustment requirements, the final state cross-sectional deformation characteristics associated with each cross-sectional deformation characteristic are obtained, and the risk assessment of large deformation of surrounding rock is carried out based on the final state cross-sectional deformation characteristics associated with each cross-sectional deformation characteristic, so as to obtain the risk assessment result of large deformation of surrounding rock corresponding to the structured monitoring data.

[0055] In this embodiment of the invention, for example, the server acquires the new modulation cross-section deformation feature and performs another partitioning process. However, in this and subsequent possible loops, the server dynamically adjusts the partitioning rules. The server performs an expansion mapping adjustment on the number of parameter dimensions of the second modulation rock mass parameter feature set according to a predefined baseline parameter quota. For example, if the baseline quota indicates that the dimension of the second set should increase in this loop, the server correspondingly allocates more dimension features from the feature vectors to that set. Simultaneously, the server performs a contraction mapping adjustment on the number of parameter dimensions of the third modulation rock mass parameter feature set, that is, correspondingly reduces the feature dimensions allocated to that set to ensure that the total dimension of the two sets remains unchanged. The server repeatedly executes the complete process from partitioning, topology construction, coupling calculation to joint construction, generating a new round of target modulation cross-section deformation features. The server has a built-in loop control module that monitors the degree of difference between the new features generated after each loop and the features of the previous round, and records the number of loops. When the server determines that the preset loop adjustment requirements are met, for example, if the change in Euclidean distance between the feature vectors generated in two consecutive loops is less than a certain threshold, or if the preset maximum number of loops (e.g., 3 times) has been reached, the server terminates the loop. At this point, the server identifies the deformation features of the target modulation section generated in the last round of iteration as the final-state deformation features of that section. Finally, based on the final-state deformation features corresponding to each monitoring section, the server performs a final risk assessment of large deformation of the surrounding rock. The server inputs these final-state feature vectors into the risk assessment classification model, and the model outputs the final risk level determination. The server summarizes the assessment results of all sections to obtain the final-state risk assessment results of large deformation of the surrounding rock corresponding to the entire structured monitoring data, after multiple rounds of constraint reinforcement and iterative adjustment optimization.

[0056] In this embodiment of the invention, the step of conducting a risk assessment of large deformation of surrounding rock based on the target cross-sectional deformation features associated with each cross-sectional deformation feature, and obtaining the risk assessment result of large deformation of surrounding rock corresponding to the structured monitoring data, can be implemented through the following example.

[0057] The target cross-sectional deformation features associated with each of the cross-sectional deformation features are divided to obtain a first target rock mass parameter feature set, a second target rock mass parameter feature set, and a third target rock mass parameter feature set. The sum of the number of parameter dimensions of the second target rock mass parameter features in the second target rock mass parameter feature set and the number of parameter dimensions of the third target rock mass parameter features in the third target rock mass parameter feature set is the same as the number of parameter dimensions of the second rock mass parameter features. Based on the mechanical response correlation degree between each first target rock mass parameter feature in the first target rock mass parameter feature set, a first reference section feature correlation topology structure corresponding to the first target rock mass parameter feature set is constructed, and based on the spatial topology correlation relationship of each monitoring section data, a second reference section feature correlation topology structure corresponding to the second target rock mass parameter feature set is constructed. Based on the spatial topological association of each monitoring section data, the associated target rock mass parameter features of each third target rock mass parameter feature in the third target rock mass parameter feature set are determined. Based on the mechanical response association degree between the associated target rock mass parameter features of each third target rock mass parameter feature in the third target rock mass parameter feature set, the feature association topology of the third reference section corresponding to the third target rock mass parameter feature set is constructed. Based on the first target rock mass parameter features and the associated section unit features corresponding to the first target rock mass parameter features in the first reference section feature association topology, coupling parameters are calculated to obtain the first coupled reference section features associated with each first target rock mass parameter feature in the first target rock mass parameter feature set. Based on the associated cross-section unit features associated with the second target rock mass parameter features and the second target rock mass parameter features in the associated topology of the second reference cross-section features, coupling parameters are calculated to obtain the second coupled reference cross-section features associated with each second target rock mass parameter feature in the set of second target rock mass parameter features. Based on the third target rock mass parameter features and the associated cross section unit features associated with each third target rock mass parameter feature in the third reference cross section feature association topology, coupling parameters are calculated to obtain the third coupled reference cross section features associated with each third target rock mass parameter feature in the third target rock mass parameter feature set. The current cross-sectional deformation feature associated with each cross-sectional deformation feature is obtained by jointly constructing the first coupled reference cross-sectional feature, the second coupled reference cross-sectional feature, and the third coupled reference cross-sectional feature associated with the same cross-sectional deformation feature. Based on the current cross-sectional deformation characteristics associated with each cross-sectional deformation characteristic, a risk assessment of large deformation of the surrounding rock is performed to obtain the current risk assessment result of large deformation of the surrounding rock corresponding to the structured monitoring data.

[0058] In this embodiment of the invention, the server performs an in-depth risk assessment process based on the deformation features of the target cross-section as follows: First, the server acquires the generated deformation features of the target cross-section for a specific monitoring cross-section (e.g., station K10+100). This feature is a vector that integrates first and second coupled parameter information. The server then partitions this feature vector. Specifically, the server partitions according to the aforementioned predefined rules, but may use different partition ratios, extracting the features corresponding to the first type of information from the vector to form a first target rock mass parameter feature set. Next, the server performs a secondary partitioning of the feature portion corresponding to the second type of information in the vector. For example, assuming the original second rock mass parameter features have a total dimension of 8, the server extracts the features of the first 4 dimensions to form a second target rock mass parameter feature set, and extracts the features of the remaining 4 dimensions to form a third target rock mass parameter feature set, thereby ensuring that the sum of the dimensions of the second and third sets equals the number of dimensions of the original second feature. Then, the server constructs three reference cross-section feature association topologies. For the first target rock mass parameter feature set, the server calculates the mechanical response correlation degree between each feature (using the same method as when constructing the topology), and constructs a first reference section feature correlation topology based on this, which is a new weighted graph. For the second target rock mass parameter feature set, the server directly constructs a second reference section feature correlation topology based on the spatial topological correlation relationship of the monitoring section data (such as section spatial adjacency information). For the third target rock mass parameter feature set, the server first determines other target rock mass parameter features directly associated with each third target rock mass parameter feature based on the spatial topological correlation relationship (these associated features may come from the first or second set). Then, the server calculates the mechanical response correlation degree between these associated target rock mass parameter features, and based on this, constructs a sub-topology network for each third feature that reflects the mechanical coupling state of its associated environment. All sub-networks of the third features together constitute the third reference section feature correlation topology. Subsequently, the server performs coupling parameter calculations on these three new topologies respectively. The calculation process is consistent with the aforementioned scheme. On the first reference topology, the server calculates the comprehensive response and offset of the associated features for each first target rock mass parameter feature, and obtains the first coupled reference section feature through full-dimensional parameter mapping. On the second reference topology, a similar operation is performed on each second target rock mass parameter feature to obtain the second coupled reference section feature. On the third reference topology, for each third target rock mass parameter feature, the server calculates the comprehensive response and offset of its directly associated other target features (as its associated section unit features), and obtains the third coupled reference section feature through full-dimensional parameter mapping. Then, the server performs joint construction.For section K10+100, the server sequentially concatenates all its first, second, and third coupled reference section features into a new long feature vector, which represents the current deformation feature associated with that section. Finally, the server performs a large deformation risk assessment of the surrounding rock based on this current deformation feature. The server inputs this feature vector into a risk assessment classification model. This model outputs a risk level or probability. The server repeats the entire process for each section in the structured monitoring data, ultimately summarizing the risk assessment results for all sections based on the current round of analysis, i.e., the current large deformation risk assessment result of the surrounding rock.

[0059] In this embodiment of the invention, the following implementation methods are also provided.

[0060] The structured monitoring data is input into the surrounding rock large deformation risk assessment model. The structured monitoring data is divided and processed by the surrounding rock large deformation risk assessment model to obtain multiple monitoring section data. Feature extraction processing is then performed to obtain the deformation characteristics of each section. The deformation features of each cross section are divided and processed by the surrounding rock large deformation risk assessment model to obtain the first rock mass parameter features associated with each cross section deformation feature, forming a first rock mass parameter feature set; and the second rock mass parameter features associated with each cross section deformation feature are obtained to form a second rock mass parameter feature set. The surrounding rock large deformation risk assessment model uses the mechanical response correlation degree between each first rock mass parameter feature in the first rock mass parameter feature set to construct the first cross-section correlation topology structure corresponding to the first rock mass parameter feature set, and constructs the second cross-section correlation topology structure corresponding to the second rock mass parameter feature set based on the spatial topology correlation relationship of each monitoring cross-section data. The surrounding rock large deformation risk assessment model uses the first rock mass parameter features and the associated cross section unit features corresponding to the first rock mass parameter features in the first cross section associated topology to perform coupling parameter calculation, thereby obtaining the first coupling parameter features associated with each first rock mass parameter feature in the first rock mass parameter feature set. Based on the second rock mass parameter features and the associated cross section unit features corresponding to the second rock mass parameter features in the second cross section associated topology, coupling parameter calculation is performed to obtain the second coupling parameter features associated with each second rock mass parameter feature in the second rock mass parameter feature set. The surrounding rock large deformation risk assessment model is used to jointly construct the first coupling parameter feature and the second coupling parameter feature associated with the deformation feature of the same cross section, so as to obtain the target cross section deformation feature associated with each cross section deformation feature. Based on the target cross section deformation feature associated with each cross section deformation feature, the surrounding rock large deformation risk assessment is performed, and the surrounding rock large deformation risk assessment result corresponding to the structured monitoring data is generated.

[0061] In this embodiment of the invention, the risk assessment model, by way of example, first divides the input structured monitoring data. The model parses the monitoring section identifier (e.g., station number information) carried by each data record in the data, and classifies all data into different data subsets based on this identifier. For example, all sensor data marked with station number K10+100 are grouped together to form K10+100 monitoring section data; all data marked with station number K10+120 are grouped into another group to form K10+120 monitoring section data, and so on, thus obtaining multiple independent monitoring section data objects. Next, the model performs feature extraction processing on each monitoring section data. Taking K10+100 section data as an example, the model reads the observation value sequence of various sensors (such as displacement gauges and stress gauges) at that section within a specific time window. The model calculates the statistical characteristics of these sequences, such as the average deformation rate, cumulative deformation, and data fluctuation variance over the past 12 hours, forming the section observation characteristics of that section. Simultaneously, the model obtains the spatial distance, azimuth, and geometric relationships between the cross-section and adjacent cross-sections, as well as the measurement points within the cross-section, from the integrated digital twin model interface. These spatial topological relationships are encoded into structured feature vectors, i.e., cross-section geometric correlation features. Finally, the model couples and transforms the cross-section observation features and cross-section geometric correlation features through a built-in feature fusion layer, outputting a comprehensive feature vector, i.e., the cross-section deformation features. Subsequently, the model classifies each cross-section deformation feature separately. Based on pre-loaded feature classification rules, the model divides the various dimensions (sub-features) constituting the cross-section deformation features into two categories. For example, sub-features directly related to mechanical response, such as "displacement rate within the surrounding rock" and "maximum principal stress change," are classified to form the first rock mass parameter feature set corresponding to the cross-section. Simultaneously, sub-features closely related to spatial attributes, such as "distance from a specific geological structure" and "spatial correlation coefficient with the deformation of adjacent cross-sections," are classified to form the second rock mass parameter feature set corresponding to the cross-section. Then, the model begins to construct the correlation topology. For the first set of rock mass parameter features, the model calculates the mechanical response correlation between any two features. For example, by analyzing the covariance of the feature value sequence and combining the correlation weights derived from the physical equations, the model calculates the correlation between the features "support structure stress" and "surrounding rock displacement". Based on the correlation between all feature pairs, the model constructs a weighted undirected graph, i.e., the first cross-section correlation topology, where nodes are features and edge weights are correlation degrees. For the second set of rock mass parameter features, the model directly constructs the second cross-section correlation topology based on the inherent spatial topological correlation of the monitoring cross-section data (such as the adjacency order of the cross-section on the tunnel axis). Then, the model performs coupling parameter calculations on the constructed topology.In the first cross-section associated topology, for a certain first rock mass parameter feature (such as "displacement rate at measuring point A"), the model finds other feature nodes directly connected to it in the graph (i.e., associated cross-section unit features). The model calculates the weighted sum of these associated feature values ​​(weights being edge weights) as the first comprehensive response feature, and calculates the difference between the feature's own value and this weighted sum as the first response offset feature. The model concatenates the original feature value, the first response offset feature value, and the first comprehensive response feature value, and inputs them into a fully connected neural network layer (i.e., the full-dimensional parameter mapping calculation module) to obtain the first coupling parameter feature corresponding to that feature. The model performs this operation on all features in the first feature set. Similarly, in the second cross-section associated topology, the model performs the same calculation process for each second rock mass parameter feature, generating the corresponding second coupling parameter feature based on its spatial associated unit features. Finally, the model performs joint construction and final evaluation. For the same monitoring cross-section, the model sequentially concatenates all its first coupling parameter features and second coupling parameter features to form a completely new, deeply information-fused target cross-section deformation feature. The model inputs the deformation features of this target cross-section into its end-effector, a risk assessment classifier (such as a trained deep learning classification network). This classifier analyzes the features and outputs a probability or level assessment of the risk of large deformation of the surrounding rock at that cross-section, such as "high risk" or a probability value of 0.88. The model sequentially completes the above-described process for all input monitoring cross-section data, and finally, the server aggregates and outputs the risk assessment results for large deformation of the surrounding rock for all cross-sections.

[0062] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned intelligent assessment method for large deformation of surrounding rock in deeply buried tunnels based on digital twins. Figure 2 As shown, Figure 2 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0063] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A smart assessment method for large deformation of surrounding rock in deeply buried tunnels based on digital twins, characterized in that, include: A digital twin model of the tunnel surrounding rock is established based on the tunnel's geological structure parameters and design and construction parameters. The digital twin model of the tunnel surrounding rock is used to characterize the spatial positional relationship of each monitoring section and the mechanical properties of the surrounding rock. Acquire monitoring data during the tunnel construction period, and perform cross-section-level mapping and association between the monitoring data during the tunnel construction period and the digital twin model of the tunnel surrounding rock to obtain structured monitoring data corresponding to each monitoring section; Based on the structured monitoring data, a risk assessment of large deformation of the surrounding rock is performed, and the risk assessment result of large deformation of the surrounding rock is obtained. Based on the risk assessment results of large deformation of the surrounding rock, the digital twin model of the tunnel surrounding rock is driven to perform mechanical evolution simulation of the surrounding rock to obtain data on the development trend of surrounding rock deformation. The risk level of large deformation of the surrounding rock is determined based on the data on the development trend of surrounding rock deformation, and corresponding construction early warning information is generated. The process of performing a large deformation risk assessment of the surrounding rock based on the structured monitoring data yields the following results: Structured monitoring data is acquired, the structured monitoring data is divided into multiple monitoring section data, and feature extraction processing is performed to obtain the deformation features of each section. Each cross-sectional deformation feature is divided into segments to obtain the first rock mass parameter features associated with each cross-sectional deformation feature, forming a first rock mass parameter feature set; and the second rock mass parameter features associated with each cross-sectional deformation feature are obtained to form a second rock mass parameter feature set. Based on the mechanical response correlation degree between each first rock mass parameter feature in the first rock mass parameter feature set, a first cross-section correlation topology structure corresponding to the first rock mass parameter feature set is constructed, and based on the spatial topology correlation relationship of each monitoring cross-section data, a second cross-section correlation topology structure corresponding to the second rock mass parameter feature set is constructed. Based on the first rock mass parameter features and the associated cross section unit features corresponding to the first rock mass parameter features in the first cross section associated topology, coupling parameters are calculated to obtain the first coupling parameter features associated with each first rock mass parameter feature in the first rock mass parameter feature set. Based on the second rock mass parameter features and the associated cross section unit features corresponding to the second rock mass parameter features in the second cross section associated topology, coupling parameters are calculated to obtain the second coupling parameter features associated with each second rock mass parameter feature in the second rock mass parameter feature set. Based on the first coupling parameter feature and the second coupling parameter feature associated with the same cross-sectional deformation feature, a joint construction is performed to obtain the target cross-sectional deformation feature associated with each cross-sectional deformation feature. Based on the target cross-sectional deformation feature associated with each cross-sectional deformation feature, a risk assessment of large deformation of surrounding rock is performed to obtain the risk assessment result of large deformation of surrounding rock corresponding to the structured monitoring data.

2. The method according to claim 1, characterized in that, The feature extraction process is performed to obtain the deformation features of each cross-section, including: The observation features of each monitoring section data are obtained to obtain the observation features of each section. The spatial topological association relationship of each monitoring section data is obtained, and the spatial topological association relationship of each monitoring section data is represented in a structured manner to obtain the geometric association features of each section; The deformation features of each cross-section are obtained by coupling and characterizing the observed features of each cross-section with the corresponding geometric correlation features.

3. The method according to claim 1, characterized in that, The construction of the first cross-section association topology structure corresponding to the first rock mass parameter feature set based on the mechanical response correlation degree between each first rock mass parameter feature in the first rock mass parameter feature set includes: Calculate the mechanical response correlation degree between each of the first rock mass parameter features, and determine the mechanical coupling correlation relationship between each of the first rock mass parameter features based on the mechanical response correlation degree; Each of the first rock mass parameter features is used as a cross-section feature unit, and an association mapping is established for each of the first rock mass parameter features based on the mechanical coupling relationship to obtain the first cross-section association topology.

4. The method according to claim 1, characterized in that, The construction of the second cross-section association topology structure corresponding to the second rock mass parameter feature set based on the spatial topological association relationship of each monitoring cross-section data includes: Based on the spatial topological correlation of each monitoring section data, the spatial representation of the second rock mass parameter features associated with each section deformation feature is determined, and the spatial correlation between each second rock mass parameter feature in the second rock mass parameter feature set is determined based on the spatial representation of the section. Each of the second rock mass parameter features is used as a cross-section feature unit, and an association mapping is established for each of the second rock mass parameter features based on the spatial association relationship to obtain the association topology of the second cross section.

5. The method according to claim 1, characterized in that, The coupling parameter calculation is performed based on the first rock mass parameter features and the associated cross-sectional unit features corresponding to the first rock mass parameter features in the first cross-sectional associated topology, to obtain the first coupling parameter features associated with each first rock mass parameter feature in the first rock mass parameter feature set, including: Calculate the comprehensive response features of the associated cross-sectional unit features corresponding to the first rock mass parameter features to obtain the first comprehensive response features, and calculate the response offset features between the first rock mass parameter features and the associated cross-sectional unit features corresponding to the first rock mass parameter features to obtain the first response offset features. The first rock mass parameter feature, the first response offset feature, and the first comprehensive response feature are jointly constructed to obtain the first joint construction feature. Based on the first joint construction feature, a full-dimensional parameter mapping calculation is performed to obtain the first coupled parameter feature corresponding to the first rock mass parameter feature. Each first rock mass parameter feature in the first cross-section associated topology is processed sequentially to obtain the first coupling parameter feature associated with each first rock mass parameter feature in the first rock mass parameter feature set.

6. The method according to claim 1, characterized in that, The coupling parameter calculation is performed based on the second rock mass parameter features and the associated cross-sectional unit features corresponding to the second rock mass parameter features in the second cross-sectional associated topology, to obtain the second coupling parameter features associated with each second rock mass parameter feature in the second rock mass parameter feature set, including: Calculate the comprehensive response characteristics of the associated cross-sectional unit characteristics corresponding to the second rock mass parameter characteristics to obtain the second comprehensive response characteristics, and calculate the response offset characteristics between the second rock mass parameter characteristics and the associated cross-sectional unit characteristics corresponding to the second rock mass parameter characteristics to obtain the second response offset characteristics; The second rock mass parameter features, the second response offset features, and the second comprehensive response features are jointly constructed to obtain the second joint construction features. Based on the second joint construction features, full-dimensional parameter mapping calculation is performed to obtain the second coupled parameter features corresponding to the second rock mass parameter features. Each second rock mass parameter feature in the second cross-section associated topology is processed sequentially to obtain the second coupling parameter feature associated with each second rock mass parameter feature in the second rock mass parameter feature set.

7. The method according to claim 1, characterized in that, The first and second coupling parameter features associated with the deformation features of the same cross-section are jointly constructed to obtain the target cross-section deformation features associated with each cross-section deformation feature. Based on the target cross-section deformation features associated with each cross-section deformation feature, a large deformation risk assessment of the surrounding rock is performed to obtain the large deformation risk assessment result of the surrounding rock corresponding to the structured monitoring data, including: Obtain the first response modulation parameter, and based on the first response modulation parameter, perform state mapping reconstruction processing on the first coupling parameter feature associated with each first rock mass parameter feature in the first rock mass parameter feature set to obtain the first state mapping feature associated with each first rock mass parameter feature in the first rock mass parameter feature set, and calculate the response deviation evaluation value corresponding to the first state mapping feature to obtain the first response deviation evaluation value. The first coupling parameter feature associated with each first rock mass parameter feature in the first rock mass parameter feature set is subjected to weight modulation processing to obtain the first weight modulation feature associated with each first rock mass parameter feature in the first rock mass parameter feature set. Calculate the coupling operation result of the first weighted modulation feature and the first response deviation evaluation quantity to obtain the first response modulation feature associated with each first rock mass parameter feature in the first rock mass parameter feature set. Based on the first response modulation parameters, the state mapping reconstruction process is performed on the second coupling parameter features associated with each second rock mass parameter feature in the second rock mass parameter feature set to obtain the second state mapping features associated with each second rock mass parameter feature in the second rock mass parameter feature set, and the response deviation evaluation quantity corresponding to the second state mapping feature is calculated to obtain the second response deviation evaluation quantity. The second coupling parameter feature associated with each second rock mass parameter feature in the second rock mass parameter feature set is subjected to weight modulation processing to obtain the second weight modulation feature associated with each second rock mass parameter feature in the second rock mass parameter feature set. The coupling operation result of the second weighted modulation feature and the second response deviation evaluation quantity is calculated to obtain the second response modulation feature associated with each second rock mass parameter feature in the second rock mass parameter feature set. The first and second response modulation features associated with the same cross-sectional deformation feature are jointly constructed to obtain the modulation cross-sectional deformation feature associated with each cross-sectional deformation feature. Based on the modulation cross-sectional deformation feature associated with each cross-sectional deformation feature, the risk assessment of large deformation of the surrounding rock is carried out to obtain the target surrounding rock large deformation risk assessment result corresponding to the structured monitoring data.

8. The method according to claim 7, characterized in that, The method of assessing the risk of large deformation of surrounding rock based on the modulated cross-sectional deformation features associated with each cross-sectional deformation feature, and obtaining the target surrounding rock large deformation risk assessment result corresponding to the structured monitoring data, includes: The modulation cross-sectional deformation features associated with each cross-sectional deformation feature are divided to obtain a first modulation rock mass parameter feature set, a second modulation rock mass parameter feature set, and a third modulation rock mass parameter feature set. The sum of the number of parameter dimensions of the second modulation rock mass parameter feature in the second modulation rock mass parameter feature set and the number of parameter dimensions of the third modulation rock mass parameter feature in the third modulation rock mass parameter feature set is the same as the number of parameter dimensions of the second rock mass parameter feature. Based on the mechanical response correlation degree between each first modulated rock mass parameter feature in the first modulated rock mass parameter feature set, a first constraint-enhanced correlation topology structure corresponding to the first modulated rock mass parameter feature set is constructed, and based on the spatial topology correlation relationship of each monitoring section data, a second constraint-enhanced correlation topology structure corresponding to the second modulated rock mass parameter feature set is constructed. Based on the spatial topological association of each monitoring section data, determine the directly associated modulated rock mass parameter features of each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set. Based on the mechanical response association degree between the directly associated modulated rock mass parameter features of each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set, construct the third constraint-strengthened association topology structure corresponding to the third modulated rock mass parameter feature set. Based on the first modulated rock mass parameter features and the associated cross section unit features corresponding to the first modulated rock mass parameter features in the first constraint-enhanced associated topology, coupling parameters are calculated to obtain the first coupled modulation features associated with each first modulated rock mass parameter feature in the first modulated rock mass parameter feature set. Based on the associated cross section unit features associated with the second modulated rock mass parameter features and the second modulated rock mass parameter features in the second constraint-enhanced associated topology, coupling parameters are calculated to obtain the second coupled modulation features associated with each second modulated rock mass parameter feature in the second modulated rock mass parameter feature set. Based on the third constraint-enhanced correlation topology, the coupling parameters are calculated for the third modulated rock mass parameter features and the associated cross section unit features associated with each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set, to obtain the third coupled modulation features associated with each third modulated rock mass parameter feature in the third modulated rock mass parameter feature set. Based on the first coupled modulation feature, the second coupled modulation feature and the third coupled modulation feature associated with the same cross-sectional deformation feature, the target modulation cross-sectional deformation feature associated with each cross-sectional deformation feature is jointly constructed to obtain the target modulation cross-sectional deformation feature associated with each cross-sectional deformation feature. Based on the target modulation section deformation features associated with each section deformation feature, a risk assessment of large deformation of the surrounding rock is performed, and the constrained correction risk assessment result of large deformation of the surrounding rock corresponding to the structured monitoring data is obtained.

9. A server system, characterized in that, Includes a server, the server being used to perform the method according to any one of claims 1-8.

Images