A tunnel intelligent supporting method and system based on a graph structure

Abstract

The application belongs to the field of tunnel engineering intelligent construction, and discloses a tunnel intelligent supporting method and system based on a graph structure, which comprises a time-varying graph structure model; in response to the occurrence of an engineering event, the time-varying graph structure model is dynamically updated to form a current subgraph matching the current engineering state; based on the subgraph sequence of the time-varying graph structure model at the current and historical time slices, a predicted value of a supporting parameter in the next stage is output; when the risk of the current supporting effect exceeds a threshold value, a risk factor chain is extracted to update the correlation relationship weight in the time-varying graph structure model; based on the updated weight, the predicted value of the supporting parameter is iteratively optimized to output a final supporting scheme, and the model is updated by using the actual engineering data after the implementation of the scheme and enters the next decision cycle. The application solves the problems of mismatching of supporting schemes, high safety risks and poor economy caused by insufficient data correlation expression, static design and lagging risk feedback in the existing tunnel supporting technology.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction technology for tunnel engineering, and in particular to an intelligent tunnel support method and system based on graph structure. Background Technology

[0002] Tunnel support is a crucial element in ensuring the safety of tunnel construction and the long-term stability of the structure. Traditional support design mainly relies on empirical analogy methods or static numerical simulation methods based on limited exploration data, which are insufficient to address the challenges of deep burial, large spans, and complex geological conditions. Existing technologies generally suffer from the following shortcomings: 1) Data silos and weak correlation: Exploration, construction, and monitoring data are stored in a scattered manner, lacking a unified model to express the topology of geological bodies, the spatial extension of structural structures, and the correlation of multi-source data; 2) Static design and delayed feedback: Support schemes are mostly determined at once, unable to perform dynamic closed-loop optimization based on geological information revealed during construction and real-time monitoring data, resulting in a significant time lag from risk identification to decision adjustment; 3) Insufficient model expressive power: Existing intelligent support research is mostly based on Euclidean space or simple data integration, failing to effectively characterize the complex non-Euclidean topological relationships and temporal evolution dependencies between "geology-support-monitoring," leading to insufficient foresight in risk warning and poor interpretability of the scheme.

[0003] For example, while existing technologies include BIM-integrated monitoring data or machine learning-based prediction of support parameters, the former focuses on visualization rather than topological relationship modeling of multi-source data, while the latter, due to its Euclidean space modeling, cannot effectively express the graph structure relationships and causal chains between engineering entities, making it difficult to achieve true adaptive iteration and quantitative risk warning.

[0004] Therefore, there is an urgent need for an intelligent support technology solution that can uniformly model multi-source heterogeneous data, dynamically characterize the evolution process of the surrounding rock-support system, and realize real-time prediction of support parameters, early warning of risks, and closed-loop optimization of solutions. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a graph-based intelligent tunnel support method and system. It constructs a tunnel graph structure model that can integrate multi-source data and adaptively evolve with the construction process, and drives an intelligent decision-making closed loop based on this model.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a tunnel intelligent support method based on graph structure, comprising the following steps: A time-varying graph structure model for tunnel multi-source data fusion is constructed. The time-varying graph structure model includes an attribute time layer, which is used to generate multiple time slices based on engineering events. Each time slice corresponds to a subgraph representing the engineering state at a specific moment. In response to engineering events, new time slices are generated, real-time engineering data is collected, and the time-varying graph structure model is dynamically updated to form a current subgraph that matches the current engineering state. Based on the time-varying graph structure model, the subgraph sequences of the current and historical time slices are used to learn and predict using a graph neural network model, and the predicted values ​​of the support parameters for the next stage are output. Based on the time-varying graph structure model and real-time monitoring data, the current support effect is evaluated and risk warning is given. When the risk exceeds the threshold, the risk factor chain that leads to the risk is extracted, and the correlation weight in the time-varying graph structure model is updated accordingly. Based on the updated association weights, the predicted values ​​of support parameters are iteratively optimized to output the final support scheme. The model is then updated using the actual engineering data after the scheme is implemented, and the next decision-making loop begins.

[0007] As an alternative implementation, the time-varying graph structure model also includes a node layer and an edge layer. The node layer includes at least spatial partition nodes for representing spatial location relationships, as well as geological nodes, support nodes, and monitoring nodes. The edge layer includes at least mapping edges for associating geological nodes, support nodes, and monitoring nodes with their corresponding spatial locations.

[0008] As an alternative implementation, the side layer also includes: Partition adjacency edges are used to express the adjacency relationship between nodes in a spatial partition; Causal association edges are used to connect pairs of nodes that have a causal relationship, and their weights are determined based on statistical causal tests and model interpretability analysis. Constraint-related edges are used to connect engineering management nodes and support nodes to express interval constraints or discrete set constraints on support parameters.

[0009] As an alternative implementation method, engineering events include at least one of the following: face advancement events, geological condition change events, monitoring data mutation events, construction method change events, and system risk warning triggering events.

[0010] As an alternative implementation, the graph neural network model adopts a hybrid architecture combining graph convolutional networks and graph attention networks to extract the topological features of subgraphs, and further uses temporal convolutional networks or gated recurrent units to model the temporal evolution features of the subgraph sequence. The graph neural network model adopts a multi-task learning framework to output the joint prediction parameters of anchor bolts, steel arch frames and shotcrete in parallel.

[0011] As an alternative implementation, iterative optimization includes a feasible region projection operation, wherein: For support parameters with continuous value ranges, perform interval projection to constrain their values ​​within the upper and lower limits allowed by the specifications; For support parameters with a set of discrete candidate values, perform set projection to map their values ​​to the candidate values ​​in the set that are closest to the predicted values ​​and meet the engineering requirements.

[0012] Secondly, the present invention provides a graph-based intelligent tunnel support system, comprising: The graph structure model building module is configured to: build a time-varying graph structure model for multi-source data fusion of tunnels. The time-varying graph structure model includes an attribute time layer, which is used to generate multiple time slices based on engineering events. Each time slice corresponds to a subgraph representing the engineering state at a specific moment. The graph structure model update module is configured to: generate new time slices in response to the occurrence of engineering events, collect real-time engineering data, and dynamically update the time-varying graph structure model to form a current subgraph that matches the current engineering state; The support parameter prediction module is configured to: learn and predict the support parameters for the next stage based on the subgraph sequence of the time-varying graph structure model in the current and historical time slices, and output the predicted values ​​of the support parameters for the next stage. The correlation weight update module is configured to: evaluate the current support effect and provide risk warning based on the time-varying graph structure model and real-time monitoring data; and when the risk exceeds the threshold, extract the risk factor chain that leads to the risk and update the correlation weight in the time-varying graph structure model accordingly. The support scheme output module is configured to: iteratively optimize the predicted values ​​of support parameters based on the updated correlation weights, output the final support scheme, update the model using the actual engineering data after the scheme is implemented, and enter the next decision loop.

[0013] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0014] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0015] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The proposed intelligent tunnel support method based on graph structure achieves unified association and dynamic representation of multi-source heterogeneous data. By constructing a tunnel graph structure model (TGM) containing multi-level nodes and multi-element edges, it for the first time unifies discrete geological information, support components, monitoring data, construction events, and management constraints within a non-Euclidean topological framework. In particular, the introduction of spatial organization nodes based on "mileage segment-circular partition" explicitly represents the spatial heterogeneity of tunnel surrounding rock and support response, solving the problems of weak data association and information silos in traditional methods.

[0017] An event-driven dynamic evolution mechanism was established to improve the timeliness of state perception: abandoning the fixed-interval update method, time slices are generated based on key events such as construction progress, geological changes, and monitoring anomalies, so that the TGM update is synchronized with the evolution of engineering risks. This mechanism can capture the turning points of the surrounding rock state in a timely manner, avoid information loss due to update lag, and provide a precise "digital twin" foundation for real-time decision-making.

[0018] The proposed intelligent tunnel support method based on graph structures achieves intelligent, collaborative, and interpretable prediction of support parameters. By learning the TGM using a graph neural network (GNN), the model naturally captures the complex topological relationships and spatial dependencies between geology, support, and monitoring. The multi-task learning architecture ensures the synergy and overall optimality of the predictions for parameters such as anchor bolts, steel frames, and shotcrete. Simultaneously, the model possesses interpretability, providing a basis for the prediction results.

[0019] The intelligent tunnel support method based on graph structure proposed in this invention constructs a proactive risk management closed loop of "assessment-early warning-calibration": through a four-dimensional assessment system and risk quantification model, it achieves quantitative evaluation of support effectiveness and forward-looking graded early warning of risks. It innovatively proposes a model self-calibration mechanism based on risk factor chains. When high risk is identified, the system can automatically trace the root causes of the risk (key geological conditions, weak support zones, etc.) and reverse-calibrate the causal weights and constraints in the TGM, making subsequent prediction and optimization more focused on risk weaknesses, thus realizing a shift from passive response to proactive intervention.

[0020] The intelligent tunnel support method based on graph structure proposed in this invention forms a constraint optimization and sustainable iteration capability that integrates engineering knowledge: engineering knowledge such as design specifications, costs, and schedules are explicitly represented as constraint-related edges in the graph structure, and interval projection and discrete set projection are strictly executed during the optimization process to ensure that the generated solution not only has excellent performance but also conforms to specifications and is constructible. By writing back the actual data of each iteration to the TGM and triggering incremental model learning, the system can accumulate engineering experience, continuously evolve, and adapt to different projects and geological conditions, achieving a continuous improvement in its intelligence level.

[0021] In summary, this invention forms a complete intelligent tunnel support decision-making system through unified graph structure modeling, event-driven dynamic updates, GNN intelligent prediction, risk source tracing calibration, and constraint closed-loop optimization, which significantly improves the scientific nature, safety, economy, and adaptability of support design.

[0022] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart of a graph-based intelligent tunnel support method according to the present invention; Figure 2 This is an architectural diagram of the time-varying graph structure model of the present invention; Figure 3 This is an architecture diagram of the multi-task graph neural network model of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed description is exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0029] Example 1 like Figure 1As shown, this embodiment provides a graph-based intelligent tunnel support method, including the following steps: A time-varying graph structure model for tunnel multi-source data fusion is constructed. The time-varying graph structure model includes an attribute time layer, which is used to generate multiple time slices based on engineering events. Each time slice corresponds to a subgraph representing the engineering state at a specific moment. In response to engineering events, new time slices are generated, real-time engineering data is collected, and the time-varying graph structure model is dynamically updated to form a current subgraph that matches the current engineering state. Based on the time-varying graph structure model, the subgraph sequences of the current and historical time slices are used to learn and predict using a graph neural network model, and the predicted values ​​of the support parameters for the next stage are output. Based on the time-varying graph structure model and real-time monitoring data, the current support effect is evaluated and risk warning is given. When the risk exceeds the threshold, the risk factor chain that leads to the risk is extracted, and the correlation weight in the time-varying graph structure model is updated accordingly. Based on the updated association weights, the predicted values ​​of support parameters are iteratively optimized to output the final support scheme. The model is then updated using the actual engineering data after the scheme is implemented, and the next decision-making loop begins.

[0030] The specific solution of the present invention is as follows: Step S1: Construct a time-varying graph structure model (TGM) for multi-source tunnel data fusion.

[0031] The Time-Varying Graph Structure Model (TGM) is the core data carrier of this method. Through a three-layer architecture of "node layer - edge layer - attribute time layer," it achieves a structured representation of multi-source data in tunnel engineering. Its construction process includes four parts: data acquisition, node definition, edge association, and attribute assignment, as detailed below: S1.1 Multi-source data acquisition and preprocessing.

[0032] Collect multi-source data throughout the tunnel's entire lifecycle, including but not limited to: Geological exploration data: borehole columnar section, geostress test, geological profile; Working face image data: RGB images of the working face and, if necessary, multispectral images were acquired using a high-definition camera; Surrounding rock monitoring data: crown settlement, perimeter convergence, anchor bolt axial force, support internal force, etc.; Support construction data: excavation advance, excavation method, support type and parameters, construction time; Project management data: project schedule, cost constraints, and safety level requirements.

[0033] Data preprocessing includes: Data cleaning and missing value imputation: Time series monitoring data If missing intervals exist, time-series interpolation combined with Kalman filtering is used for correction and interpolation. (Time interval not specified) The original observations are The state variable is Then we have a linear state-space model: , in , The noise is Gaussian, and the optimal estimate is obtained through Kalman filtering. Used to fill in missing data This indicates a node. State estimate Composition of state estimation time series In this article This represents the raw monitoring time series output by the sensor (which may contain missing data), while the state-space model contains... It represents the state variable of the monitored quantity (or its implicit true state).

[0034] Data standardization: Standardize mechanical parameters (such as elastic modulus E, cohesion c, internal friction angle φ) and geometric parameters.

[0035] Feature extraction from the face of the palm: Crack identification and crack density extraction were performed using improved YOLOv8+ edge detection. Average fracture length , fracture orientation and distribution, etc.

[0036] BIM Geometric Feature Extraction: Extracting the Center Coordinates of the Steel Arch from the BIM / IFC Model Shotcrete coverage area wait.

[0037] S1.2: Node layer definition of the graph structure model (TGM).

[0038] The node layer is used to characterize entity objects in tunnel engineering. Based on data type and engineering logic, nodes are classified into at least the following categories: (1) Geological nodes: including lithological nodes and structural nodes; used to express lithological parameters, structural types, fracture characteristics and advanced prediction results, etc.

[0039] (2) Support nodes: including anchor nodes, steel arch nodes, and shotcrete nodes; used to express the type, specifications, parameters and construction status of support components.

[0040] (3) Monitoring nodes: including displacement monitoring nodes, stress / internal force monitoring nodes, and environmental monitoring nodes; in addition to the monitoring values, the monitoring node attributes also include spatial location marking information, the spatial location markings include at least mileage location and circumferential partition markings; the circumferential partition markings are one or more of the following: arch crown, arch shoulder, side wall, and invert arch.

[0041] (4) Spatial partition nodes: including mileage segment nodes and circumferential partition nodes; mileage segment nodes are used to express longitudinal segments (start and end mileage, segment number), and circumferential partition nodes are used to express circumferential positions (arch crown / arch shoulder / sidewall / inverted arch) within the same mileage segment, and are used to carry the summary of geological features, support status and monitoring response within the partition.

[0042] (5) Construction event node: used to express construction / risk events that trigger map updates. The event node attributes include at least event type, occurrence time, corresponding mileage segment, and event intensity level. The event type includes at least one or more of the following: construction progress event, geological revelation change event, and monitoring response mutation event.

[0043] (6) Project management nodes: including schedule nodes, cost nodes and safety level nodes, used to express plans and constraints.

[0044] S1.3: Edge layer definition for the graph structure model (TGM).

[0045] Edge layers are used to characterize various relationships between nodes, mainly including topological association edges, partition mapping edges, partition adjacency edges, causal association edges, and constraint association edges.

[0046] To unify the notation representation, the time-varying graph structure model is defined in time slices. subgraph The edge layer consists of the union of multiple edge sets:

[0047] in For the set of topologically related edges, For the set of partitioned mapping edges, For the set of adjacent edges of a partition, For a set of causal edges, For the set of constrained associated edges; any edge e∈ Both can include attribute vectors It is used to express information such as distance, weight, confidence, and significance.

[0048] (1) Topologically related edges : Spatial topological edge: If the distance between two nodes in three-dimensional space is... Less than the threshold Then establish spatial topological edges The edge attributes are: in The overlapping area, This represents the azimuth difference.

[0049] Temporal topology edge: If two construction cycles are adjacent or have a clear sequential relationship, establish a temporal edge. : in For time intervals, This refers to the pass rate of process connection.

[0050] (2) Partition mapping edge : Partition mapping edges are used to establish a mapping relationship between "entity nodes (monitoring / support / geology, etc.) → spatial partition nodes". If entity nodes... If a point is located in a certain mileage segment and its circumferential marker belongs to a preset zone (such as crown / shoulder / sidewall / invert), then a mapping edge is established. ,in This represents a spatial partition node. The mapped edge attribute vector is defined as follows: ; in This represents the mapping distance from the entity node to the location represented by the partition. The mapping confidence level is (0~1).

[0051] (3) Adjacent edges of partitions : Partition adjacency edges are used to express the adjacency relationships between nodes in a spatial partition, including circumferential and vertical adjacency: ; in This is the set of circumferential adjacent edges of different circumferential partitions within the same mileage segment. It is the set of vertically adjacent edges of the same-named circumferential partitions of adjacent mileage segments.

[0052] For two partition nodes , Establish adjacent edges Its attribute vector is defined as: ; in This is used to identify the adjacency type (circular / vertical). This is the adjacency weight.

[0053] (4) Causal Relationship Edge : This is used to characterize the causal relationships between geology, support, monitoring, and risk. For any pair of nodes (i,j), if the correlation is significant and passes the Granger causality test, then a causal edge is established. And it is believed that there is a direction. Causal effects: First, calculate the correlation coefficient: ,in They are nodes With nodes The corresponding state estimation time series, To find the standard deviation, perform a Granger causality test: if the result is at a significance level... If the following is passed, it is considered that There is a causal relationship.

[0054] Edge attributes include: in The SHAP weights are derived from the model interpretation (see S4 below for details). This is the p-value for the Granger test. For direction encoding (if) but If j→i, then If only one-way causal edges are retained, then take... ).

[0055] To support online calibration updates driven by the risk factor chain (see step S4.3), the causal edge influence weights are defined as updatable dynamic quantities:

[0056] Where τ is the time slice index. (x) is a normalized function that maps "explanatory contribution - causal significance - correlation" to edge weights; when a risk warning is triggered, the risk factor chain is adjusted accordingly. Perform calibration This enables adaptive updating of causal edge weights.

[0057] (5) Constraining associated edges : Used to express the constraints of design specifications, schedule, cost, etc., on support parameters: Constraint types include upper bound constraints, lower bound constraints, set constraints, and interval constraints; : A set of discrete candidate values ​​used to store set constraints (e.g., collection of steel arch frame models, collection of concrete strength grades, etc.) This indicates the degree to which the current support parameters satisfy the constraints (0–1).

[0058] S1.4: Attribute Time Layer.

[0059] In addition to node and edge attributes, the attribute time layer of this invention introduces the concept of time slice τ, which discretizes the tunnel construction process into multiple time slices generated by event triggering, denoted as τ=1,2,…,T.

[0060] Each time slice corresponds to a subgraph TGM is a dynamic graph sequence composed of all subgraphs concatenated in time. .

[0061] The time slice boundary is determined by the triggering event criterion, which includes at least one of the following: a) Construction progress event: The advance distance of the working face reaches the preset threshold ΔL0 or one excavation-initial support cycle is completed; b) Geologically revealed changes: Changes in the grade of the surrounding rock, the degree of development of structural planes, or the water-bearing state revealed at the working face; c) Monitoring response to sudden events: The increment, rate of change, or acceleration of any monitored quantity exceeds the corresponding preset threshold; d) Construction method / support change event: Excavation method or support type / parameters change; e) Risk warning trigger event: The risk level reaches the preset threshold.

[0062] When a triggering event occurs, a new time slice is generated and a new construction event node is added to the graph. At the same time, event impact edges are established to connect the affected spatial partition nodes, support nodes, and monitoring nodes. Event evolution edges are established between construction event nodes of adjacent time slices to express the event sequence and evolution relationship.

[0063] Step S2: Dynamically update the graphical structure model based on the construction progress.

[0064] S2.1: Event criterion calculation and time slice generation.

[0065] Event criteria are calculated for construction progress data, face reveal data and monitoring time series; when any criterion meets the threshold condition, a new time slice τ+1 is generated, construction event nodes are created and the event type and intensity level are recorded, and the newly added event nodes are associated with the corresponding mileage segment nodes and circumferential partition nodes to provide an index for subsequent node / edge updates.

[0066] When the event trigger criteria are met, a TGM update is performed. The event trigger criteria include one or more of the following: construction progress, geological changes, sudden changes in monitoring response, changes in construction methods / support, or risk warning triggers.

[0067] S2.2: Real-time data acquisition.

[0068] Working face image acquisition and crack identification: Cracks are detected based on YOLOv8, followed by skeletonization and length statistics, and then updated. , wait; Ground-penetrating radar data: towards the front of the working face Identify potential weak or aquifer zones within the area and add new construction nodes; Monitoring data: Displacement, stress, pore pressure, etc. are updated in real time; Construction data: Record the excavation method, advance, support type, and materials delivered for the current cycle.

[0069] S2.3: Adding new nodes and updating their status.

[0070] New geological nodes ; Add support nodes (anchor bolts, steel arch frames, shotcrete) and monitoring nodes; The status of existing nodes changes from "in progress" to "completed" or "pending construction" to "in progress". The status can be encoded as a one-hot feature and added to the node attribute vector.

[0071] S2.4: Update while updating.

[0072] Add new topological edges: Updated based on spatial distance and time relationship. .

[0073] Update causal edges: Calculate the latest monitoring quantity Y(t) and the geological and support quantities X(t). The correlation and causality of k) are evaluated, and if the significance changes, the corresponding values ​​are updated. , .

[0074] Update constraint edges: Update constraint satisfaction based on the latest schedule deviation ΔT and cost deviation ΔC. .

[0075] S2.5: Attribute Update.

[0076] Update the attribute vectors for all affected nodes and edges to form a subgraph at time slice τ+1. .

[0077] Step S3: Support parameter prediction model based on multi-task graph neural network.

[0078] This invention employs a hybrid architecture GNN consisting of GCN, GAT, temporal convolution, and multi-task decoupling output to achieve the mapping from TGM to support parameters.

[0079] S3.1: Dataset Construction and Preprocessing.

[0080] The event-triggered time slice τ is used as the unit, and each time slice corresponds to a subgraph. and its corresponding actual support parameter vector: Data augmentation was performed using methods such as node / edge perturbation, subgraph pruning, and random edge dropping. The dataset was divided into a training set: validation set: test set ratio of 7:2:1.

[0081] S3.2: Overall Model Architecture.

[0082] (1) Graph feature extraction layer (spatial): For each time slice graph GCN and GAT are used to extract local and global topological features.

[0083] GCN convolution: Let The initial node feature matrix, Then the L1 layer GCN is updated as follows: Where σ is the activation function, This is a learnable weight matrix.

[0084] GAT attention mechanism: focusing on the features of node i Features of neighbor j First, calculate the attention coefficient: Then normalize: The final update node indicates: .

[0085] (2) Temporal feature extraction layer (time).

[0086] The temporal feature extraction layer (time) captures the evolutionary relationship of "mileage segment-circular partition" between different time slices. This invention employs a temporal modeling layer to model the node embedding sequence of spatial partition nodes (which can be implemented as 1DTCN or GRU). Specifically, for each time slice subgraph... After completing GCN+GAT message passing, read the node representation of the spatial partition node set Q. (q∈Q represents the partition nodes such as the crown / shoulder / sidewall / inverted arch).

[0087] Representation sequence of the same partition node q at different time slices Perform timing modeling to obtain a partition-level timing representation:

[0088] Then, the temporal representations of all partition nodes are read out / aggregated to obtain the temporal feature vectors used for subsequent multi-task prediction:

[0089] (3) Multi-task feature fusion layer.

[0090] The time-series feature vector obtained after time-series modeling of partition nodes The fusion vector is obtained by concatenating the features (schedule, cost, safety level) of the project management nodes: .

[0091] (4) Multi-task parameter prediction layer.

[0092] Support parameter prediction was broken down into three related tasks: anchor bolts, steel arches, and shotcrete. The model adopted a shared-dedicated structure. ; in It consists of a multi-layered, fully connected network.

[0093] S3.3: Loss Function and Constraint Optimization.

[0094] This invention employs a comprehensive loss function consisting of multi-task MSE, constraint penalties, and regularization terms: .

[0095] (1) Multi-task MSE loss: ,in This represents the task weight.

[0096] (2) Constraint loss For each prediction parameter Define engineering constraints The constraint loss is defined as: .

[0097] (3) Regularization term: L2 regularization is used to prevent overfitting.

[0098] S3.4: Model Training and Evaluation.

[0099] The optimizer uses Adam with an initial learning rate of 0.001 and employs a cosine annealing strategy. Set epoch=500, batchsize=32, and adopt an early stopping strategy; Use MAE and RMSE as evaluation indicators; Compared with traditional random forests, XGBoost, and single-task GNNs, the superiority of the multi-task GNN model of this invention under complex geological conditions is verified.

[0100] Step S4: Support effect assessment and risk warning module.

[0101] S4.1: Support effect evaluation index system.

[0102] Construct a three-dimensional evaluation system that includes "stability indicators", "economic indicators", and "schedule adaptability indicators", among which: (1) Stability index Based on the comparison between monitored deformation and numerical simulation, the following definition is made: ,in To monitor the maximum displacement, To allow displacement, For the actual safety factor, The target safety factor.

[0103] (2) Economic indicators ,in, Given the target cost / budget cost for project management data, This represents the actual cost or estimated cost of the candidate support scheme.

[0104] (3) Schedule adaptability index ,in, The planned duration given for project management data, This represents the estimated construction period for the actual project or the candidate project.

[0105] (4) Risk redundancy index: .

[0106] The overall score is defined as follows: , in .

[0107] Rating scale: S≥0.9: Excellent; 0.75≤S<0.9: Good; S<0.75: Poor.

[0108] S4.2: Risk Probability and Loss Quantification.

[0109] (1) Risk probability Based on the output features of the GNN and the monitoring time series, a small classification sub-network is used to estimate the probability of instability events occurring within the next Δt. ,in, This represents the Sigmoid activation function; This represents a classification subnetwork used for risk probability estimation.

[0110] (2) Risk loss Based on the instability scenarios (partial collapse, support failure, water inrush, etc.), the estimated repair costs and project time losses are calculated. An empirical model is used to model the expected losses. , Indicates the type of instability scenario (such as partial collapse, support failure, water inrush, etc.). Let represent the probability or conditional probability of scene s occurring in time slice τ. This indicates the cost of loss in the corresponding scenario (repair costs, project time loss, etc.).

[0111] (3) The risk level is determined by the “risk probability-risk loss” matrix, which divides the risk level into three levels: “low risk (blue), medium risk (yellow) and high risk (red)”.

[0112] S4.3: Risk factor chain extraction and graph structure / constraint self-calibration.

[0113] When the risk level reaches the preset threshold, the current time slice image is... Perform interpretability contribution calculation to obtain the set of nodes and edges that contribute the most to the risk probability output; based on the set of nodes and edges, extract the risk factor chain (Top-K path) consisting of "geological nodes / spatial partition nodes - support nodes - monitoring nodes".

[0114] Based on the risk factor chain, the weights of causal relationships in the graph structure are calibrated and updated, and the constraint weights of constrained relationships are dynamically adjusted. Let the causal edges... The dynamic edge weight in time slice τ is (from S1.3) (Calculated), let the overall contribution of the risk factor chain to this side be . (Obtained by normalizing the contribution of edges within the Top-K paths), the calibration update of causal edge weights can then be expressed as: ,

[0115] in, Calculated based on the contribution of the risk factor chain. To update step size >0 represents the calibration strength coefficient; when a certain side is not... Take within time If the edge weight remains unchanged or gradually decreases, then the constraint weight (or penalty weight) of the constrained edges is dynamically adjusted. Let the penalty weight of the k-th constraint term be... The contribution of its corresponding constraint in the risk factor chain is ,but:

[0116] in A value greater than 0 represents the constraint tightening strength coefficient. For associations with high contribution and amplifying risk, the penalty weight of the corresponding constraint is increased. Alternatively, tighten the constraint threshold; for associations with low contribution and no significant impact on risk, reduce their penalty weights or leave them unchanged. (Calibrated edge weights) With constraint weights This serves as input for the next iteration of optimization and the next time-slice modeling.

[0117] Step S5: Output and iterative optimization of support scheme.

[0118] S5.1 Support scheme generation and visualization.

[0119] according to Generate a support plan, including: (1) Parameter list: List the parameters of anchor bolts, steel arches, and shotcrete for each cycle; (2) Layout drawing: The coordinates of anchor bolt holes, arch frame positions, spraying range, and monitoring point layout are automatically marked in conjunction with the BIM model; Construction instructions: Provide construction methods, key points for quality control, and safety precautions according to the work process.

[0120] (3) Output formats include PDF+IFC model files, which can be viewed on both web and mobile devices.

[0121] S5.2: Iterative optimization algorithm for support schemes.

[0122] Based on the causal edge weights and constraint weights calibrated by S4.3, feasible region projection and iterative optimization of support parameters are performed.

[0123] If the evaluation result is "poor" or the risk level is "red", then the iterative optimization process is triggered: (1) Perform gradient-guided search for support parameters within the feasible region: The initial point is the current scheme. ; The goal is to maximize the overall score S(y) while satisfying the following constraints: .

[0124] Ascent using projective gradient: ,in This represents the parametric feasible region formed by both interval constraints and set constraints. (·) denotes the operator that projects the updated parameter vector back into the feasible region (interval projection for continuous parameters, set projection for discrete parameters). η represents the optimization iteration step size.

[0125] The feasible region projection includes continuous interval projection and discrete set projection: A) Continuous interval projection: For parameters that satisfy the interval constraints (such as anchor length, spacing, spray layer thickness, etc.), the updated parameters are clipped / mapped to the corresponding upper and lower limit intervals; B) Discrete set projection: For parameters that satisfy set constraints (such as steel arch frame type, shotcrete strength grade, etc.), the updated parameters are mapped to values ​​in the discrete candidate value set that are allowed by the specifications and available on site; when there are multiple candidate values, the candidate value that maximizes the comprehensive score S(y) and has the lowest risk level is selected.

[0126] (2) After each update, call the support effectiveness assessment module and the risk assessment module. If the following conditions are met: If the solution is correct, accept it; otherwise, continue iterating, with a maximum number of iterations. It can be set to 10-20.

[0127] S5.3: Data closed loop and incremental learning.

[0128] After each support scheme is implemented, the following data should be written back to TGM: actual construction parameters and deviations; monitoring response and extreme values; scheme evaluation results and risk management records. When the cumulative number of new construction cycles reaches... This triggers an incremental training iteration: extracting new subgraphs from the graph database. Use a smaller learning rate for local retraining or use regularization to avoid "catastrophic forgetting"; update model parameters and evaluation metrics.

[0129] Example 2 This embodiment provides a graph-based intelligent tunnel support system, including: The graph structure model building module is configured to: build a time-varying graph structure model for multi-source data fusion of tunnels. The time-varying graph structure model includes an attribute time layer, which is used to generate multiple time slices based on engineering events. Each time slice corresponds to a subgraph representing the engineering state at a specific moment. The graph structure model update module is configured to: generate new time slices in response to the occurrence of engineering events, collect real-time engineering data, and dynamically update the time-varying graph structure model to form a current subgraph that matches the current engineering state; The support parameter prediction module is configured to: learn and predict the support parameters for the next stage based on the subgraph sequence of the time-varying graph structure model in the current and historical time slices, and output the predicted values ​​of the support parameters for the next stage. The correlation weight update module is configured to: evaluate the current support effect and provide risk warning based on the time-varying graph structure model and real-time monitoring data; and when the risk exceeds the threshold, extract the risk factor chain that leads to the risk and update the correlation weight in the time-varying graph structure model accordingly. The support scheme output module is configured to: iteratively optimize the predicted values ​​of support parameters based on the updated correlation weights, output the final support scheme, update the model using the actual engineering data after the scheme is implemented, and enter the next decision loop.

[0130] It should be noted that the above modules correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above modules and their corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules can be executed in a computer system as part of the system.

[0131] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0132] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0133] A computer-readable storage medium for storing computer instructions that, when executed by a processor, perform the method of Embodiment 1.

[0134] The method in Example 1 can be directly executed by a hardware processor, or it can be executed by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0135] A computer program product includes a computer program that, when executed by a processor, implements the method in Embodiment 1.

[0136] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0137] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0138] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0139] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0140] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for intelligent tunnel support based on graph structure, characterized in that, Includes the following steps: A time-varying graph structure model for tunnel multi-source data fusion is constructed. The time-varying graph structure model includes an attribute time layer, which is used to generate multiple time slices based on engineering event triggers. Each time slice corresponds to a subgraph representing the engineering state at a specific moment. In response to engineering events, new time slices are generated, real-time engineering data is collected, and the time-varying graph structure model is dynamically updated to form a current subgraph that matches the current engineering state. Based on the subgraph sequence of the current and historical time slices using the time-varying graph structure model, a graph neural network model is used for learning and prediction to output the predicted values ​​of the support parameters for the next stage. Based on the time-varying graph structure model and real-time monitoring data, the current support effect is evaluated and risk warning is given. When the risk exceeds the threshold, the risk factor chain that leads to the risk is extracted, and the correlation weight in the time-varying graph structure model is updated accordingly. Based on the updated association weights, the predicted values ​​of support parameters are iteratively optimized to output the final support scheme. The model is then updated using actual engineering data after the scheme is implemented, and the process enters the next decision loop.

2. The intelligent tunnel support method based on graph structure as described in claim 1, characterized in that, The time-varying graph structure model further includes a node layer and an edge layer. The node layer includes at least spatial partition nodes for representing spatial location relationships, as well as geological nodes, support nodes, and monitoring nodes. The edge layer includes at least mapping edges for associating geological nodes, support nodes, and monitoring nodes with their corresponding spatial locations.

3. The intelligent tunnel support method based on graph structure as described in claim 2, characterized in that, The edge layer also includes: Partition adjacency edges are used to express the adjacency relationship between nodes in a spatial partition; Causal association edges are used to connect pairs of nodes that have a causal relationship, and their weights are determined based on statistical causal tests and model interpretability analysis. Constraint-related edges are used to connect engineering management nodes and support nodes to express interval constraints or discrete set constraints on support parameters.

4. The intelligent tunnel support method based on graph structure as described in claim 1, characterized in that, The engineering events include at least one of the following: face advancement events, geological condition change events, monitoring data mutation events, construction method change events, and system risk warning triggering events.

5. The intelligent tunnel support method based on graph structure as described in claim 1, characterized in that, The graph neural network model adopts a hybrid architecture combining graph convolutional networks and graph attention networks to extract the topological features of subgraphs. It further uses temporal convolutional networks or gated recurrent units to model the temporal evolution features of the subgraph sequence. The graph neural network model adopts a multi-task learning framework to output the joint prediction parameters of anchor bolts, steel arch frames and shotcrete in parallel.

6. The intelligent tunnel support method based on graph structure as described in claim 1, characterized in that, The iterative optimization includes a feasible region projection operation, wherein: For support parameters with continuous value ranges, perform interval projection to constrain their values ​​within the upper and lower limits allowed by the specifications; For support parameters with a set of discrete candidate values, perform set projection to map their values ​​to the candidate values ​​in the set that are closest to the predicted values ​​and meet the engineering requirements.

7. A graph-based intelligent tunnel support system, characterized in that, include: The graph structure model construction module is configured to: construct a time-varying graph structure model for tunnel multi-source data fusion, wherein the time-varying graph structure model includes an attribute time layer, which is used to generate multiple time slices based on engineering event triggers, and each time slice corresponds to a subgraph representing the engineering state at a specific moment; The graph structure model update module is configured to: generate new time slices in response to the occurrence of engineering events, collect real-time engineering data, and dynamically update the time-varying graph structure model to form a current subgraph that matches the current engineering state; The support parameter prediction module is configured to: learn and predict the support parameters for the next stage based on the subgraph sequence of the time-varying graph structure model in the current and historical time slices, and output the predicted values ​​of the support parameters for the next stage. The correlation weight update module is configured to: evaluate the current support effect and provide risk warning based on the time-varying graph structure model and real-time monitoring data; and when the risk exceeds the threshold, extract the risk factor chain that leads to the risk and update the correlation weight in the time-varying graph structure model accordingly. The support scheme output module is configured to: iteratively optimize the predicted values ​​of support parameters based on the updated correlation weights, output the final support scheme, update the model using the actual engineering data after the scheme is implemented, and enter the next decision loop.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images