Tunnel construction safety monitoring and early warning method and system based on multi-dimensional data fusion

Abstract

The invention provides a tunnel construction safety monitoring and early warning method and system based on multi-dimensional data fusion, and relates to the technical field of construction safety early warning, and the method comprises the steps: obtaining first information which comprises tunnel construction parameters and image information; performing space-time fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space of tunnel construction; performing cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space, and constructing a dynamic heterogeneous graph network; performing space-time diagram convolution processing on the dynamic heterogeneous graph network, and generating a composite risk manifold based on a result obtained by processing and a multi-head attention mechanism; coupling processing is carried out according to the composite risk manifold, stability analysis and critical state judgment are carried out based on the risk phase change hypersurface obtained through processing, and real-time early warning is carried out based on a dynamic early warning boundary of tunnel construction obtained through judgment. According to the invention, the risk identification accuracy and the early warning response timeliness in the tunnel construction environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of construction safety early warning technology, and in particular to a tunnel construction safety monitoring and early warning method and system based on multi-dimensional data fusion. Background Art

[0002] In current tunnel construction, due to complex geological conditions, severe construction disturbances, and volatile environmental factors, sudden collapses, gas leaks, and other safety incidents frequently occur during tunnel construction. To achieve early warning and risk control, the industry generally adopts threshold discrimination methods based on single-sensor monitoring data, or uses finite physical models to conduct deformation prediction and risk assessment. However, existing technologies generally suffer from the following problems: First, the monitoring data is single-dimensional, making it difficult to fully reflect the evolution of multi-physics fields under construction disturbances; second, fusion processing methods are limited, mostly relying on static data analysis or linear modeling, lacking the ability to deeply model nonlinear and dynamic coupling phenomena in the construction environment; third, risk discrimination relies primarily on empirical rules or shallow classification models, making it difficult to identify critical transitions and potential mutation structures, resulting in delayed warnings and insufficient accuracy. Especially under complex working conditions, existing methods lack systematic modeling of the spatial distribution and evolution trends of potential risks, limiting their application in dynamic construction scenarios.

[0003] Therefore, there is an urgent need for a tunnel construction safety monitoring and early warning method and system based on multi-dimensional data fusion to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a tunnel construction safety monitoring and early warning method and system based on multi-dimensional data fusion to improve the above-mentioned problems. To achieve the above-mentioned purpose, the technical solutions adopted by the present invention are as follows:

[0005] In a first aspect, the present application provides a tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion, comprising:

[0006] Acquiring first information, where the first information includes tunnel construction parameters and image information;

[0007] Performing spatiotemporal fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space for tunnel construction;

[0008] Performing cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space, and constructing a dynamic heterogeneous graph network;

[0009] Performing spatiotemporal graph convolution processing on the dynamic heterogeneous graph network, and generating a composite risk manifold based on the processing results and a multi-head attention mechanism;

[0010] Coupling processing is performed according to the composite risk manifold, and stability analysis and critical state discrimination are carried out based on the risk phase transition hypersurface obtained through processing. Real-time warning is given based on the dynamic warning boundary of tunnel construction obtained by discrimination.

[0011] Secondly, this application also provides a tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion, including:

[0012] an acquiring unit, configured to acquire first information, wherein the first information includes tunnel construction parameters and image information;

[0013] a processing unit, configured to perform spatiotemporal fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space for tunnel construction;

[0014] A construction unit, configured to perform cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space and construct a dynamic heterogeneous graph network;

[0015] A generation unit, configured to perform spatiotemporal graph convolution processing on the dynamic heterogeneous graph network, and generate a composite risk manifold based on the processing result and a multi-head attention mechanism;

[0016] The discrimination unit is used to perform coupling processing based on the composite risk manifold, and to perform stability analysis and critical state discrimination based on the risk phase change hypersurface obtained through the processing, and to provide real-time warning based on the dynamic warning boundary of tunnel construction obtained through the discrimination.

[0017] The beneficial effects of the present invention are:

[0018] This paper utilizes a variety of mathematical methods and intelligent algorithms to construct a high-dimensional risk modeling system for complex disturbance propagation paths. This system systematically implements the entire process from raw sensory data to risk phase transition identification and dynamic warning boundary generation. This method transcends the limitations of traditional safety monitoring, which relies on empirical rules and static analysis. It possesses multimodal, dynamic, cross-scale, and strong topological recognition capabilities, significantly improving the accuracy of risk identification and the timeliness of warning responses in tunnel construction environments.

[0019] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the embodiments of the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 Schematic diagram of the process of a tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion according to an embodiment of the present invention;

[0022] Figure 2 Schematic diagram of the structure of the tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion described in an embodiment of the present invention.

[0023] In the figure: 701, acquisition unit; 702, processing unit; 703, construction unit; 704, generation unit; 705, discrimination unit. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and should not be understood as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion.

[0028] See also Figure 1 , the figure shows that the method includes step S1, step S2, step S3, step S4 and step S5.

[0029] Step S1: Acquire first information, where the first information includes tunnel construction parameters and image information;

[0030] It can be understood that the parameters and image information of the tunnel construction in this step include the three-dimensional geological parameter matrix of tunnel construction, the time series parameter set of construction machinery, the deformation sequence of the support structure, the environmental gas concentration data and the thermodynamic image of the construction surface. Among them, the three-dimensional geological parameter matrix is ​​derived from the previous geological exploration, drilling sampling and geological radar inversion data, reflecting the spatial distribution characteristics of the stratum structure, rock property changes and joints and fissures, providing a benchmark for subsequent geological disturbance modeling; the multimodal time series parameter set of construction machinery is obtained in the form of high-frequency sampling by multiple types of sensors (such as vibration, torque, propulsion, current, voltage, etc.) on the tunnel boring machine (TBM) or drilling and blasting equipment, reflecting the continuous disturbance process of the construction behavior on the surrounding environment; the support structure deformation monitoring sequence is obtained through the deployment of lining, anchor rods or steel frames and other structures. Fiber Bragg gratings, total stations, GNSS, and other equipment are used for dynamic data collection to characterize the evolution of stability states during structural response. Ambient gas concentration gradient distribution data relies on a distributed network of point-based or mobile hazardous gas detection modules, such as CO, CH4, and H2S gas sensors, to capture the diffusion of hazardous gases caused by ventilation anomalies or sudden geological releases during construction. Construction surface thermodynamic imaging combines infrared thermal imagers with machine vision systems to identify potential risk factors such as abnormal hot spots and thermal disturbances caused by leakage in real time during high-temperature operations and closed sprayed anchor environments. By constructing a multi-source raw data pool covering space, time, modality, and physical domains, a complete and realistic input space is provided for subsequent tensor decomposition, manifold mapping, and graph modeling.

[0031] Step S2: performing spatiotemporal fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space for tunnel construction;

[0032] It can be understood that this step achieves cross-modal and cross-scale coupled fusion processing by constructing a unified tensor representation and high-dimensional spatial alignment strategy, achieving unified modeling and dimensionality reduction embedding of heterogeneous and multimodal data, effectively removing redundant interference information and strengthening the inherent correlation of various construction disturbances, thus laying a solid foundation for the subsequent construction of dynamic graph structures and extraction of topological information. In this step, step S2 includes steps S21, S22, S23, and S24.

[0033] Step S21: extracting principal components from the spatial dimension and the temporal dimension of the first information based on Tucker tensor decomposition, and constructing a local low-rank subspace expression of the first information to obtain an initial third-order tensor set;

[0034] It can be understood that this step is centered on multidimensional tensor modeling. For five types of data, namely geological parameters, mechanical behavior, support deformation, environmental gas, and thermal images, three-order tensor representations based on spatial distribution and temporal evolution dimensions are constructed. Among them, the three-dimensional geological parameter matrix constructs the spatial dimension through spatial gridding and hierarchical coding, the construction machinery and support structure data set the time label and channel dimension according to the monitoring node and equipment working status, and the environmental gas concentration and thermal image data are indexed uniformly through regional location and time frame sequence. Then, the Tucker tensor decomposition model is used to perform low-rank approximation on each type of tensor, extracting its core feature components in the main spatial dimension (such as stratum structure, sensor layout location) and time dimension (such as excavation stage, construction equipment operation cycle). This decomposition not only compresses data redundancy, but also retains the key disturbance evolution trend, effectively constructing a local low-rank subspace expression. The final output of the initial three-order tensor set has a unified index structure, which facilitates subsequent alignment and fusion between modalities.

[0035] Step S22: aligning the construction machinery timing parameters of the initial third-order tensor set with the support deformation dynamic trajectory in the manifold space to obtain a fourth-order fusion tensor representation;

[0036] It can be understood that this step introduces a shared embedding mapping strategy based on manifold alignment, mapping the time series parameter tensor of the construction machinery and the dynamic deformation sequence of the support structure into the same low-dimensional embedding space, using the Gaussian kernel function to retain the local structure within each mode and align the common geometric features between modes to the greatest extent. At the same time, the three-dimensional geological parameter matrix and thermodynamic image data have stable spatial structural characteristics. Therefore, during the alignment process, the geometric constraints on the scene background are achieved by maintaining the continuity of their spatial distribution to prevent structural distortion caused by embedding distortion. This step achieves global spatiotemporal alignment between modalities on the basis of maintaining the inherent local structure of various types of data, and obtains a unified fourth-order fusion tensor representation with consistent spatiotemporal indexes, transferable spatial topological relationships, and nested dynamic evolution patterns. This result lays the foundation for high-dimensional geometric consistency for subsequent multimodal tensor decomposition and dynamic graph modeling.

[0037] Step S23: extracting the common change trend of the fourth-order fusion tensor representation under a fixed mode based on the preset PARAFAC tensor decomposition method to obtain a sparse tensor structure;

[0038] It can be understood that this step uses the PARAFAC tensor decomposition method as its core tool, leveraging its property of preserving a unique decomposition structure between tensor modes. It selectively operates on some fixed modes in the geological, support, construction machinery, and environmental modes, extracting common variation trends between different modes, such as the synchronous structural deformation and gas change response caused by construction progress. On this basis, the original high-dimensional tensor is projected into a local low-rank subspace using the QR decomposition method, thereby suppressing the recurring collinear parameters in each mode and highlighting the main response characteristics. In particular, for the frequently recorded redundant operation data in the construction machinery mode (such as repeated startups and energy consumption in non-operating states) and high-frequency fluctuation interference signals in gas concentration (such as noise waves caused by sudden ventilation changes), this step can significantly improve their compression ratio and reduce the impact of interference on subsequent modeling. The resulting sparse tensor structure not only has higher information density and structural clarity, but also maintains the dynamic synergy characteristics across modalities.

[0039] Step S24: construct a node hypergraph structure based on the sparse fusion tensor structure, and use the time evolution sequence as the edge weight to map it into a four-dimensional hypergraph tensor space.

[0040] It can be understood that this step constructs a node hypergraph structure composed of the principal component distribution of the thermodynamic image and the main path of the support structure deformation, and further maps it into a four-dimensional hypergraph tensor space with nonlinear topological relationships using the time evolution sequence as the edge weight. The vertices represent entity-type multimodal observation points, and the edge weights represent the dynamic correlation strength of potential disturbances. This structure serves as the basis for the spectral wavelet and topological feature extraction in the next stage. This step first uses the principal component distribution in the thermodynamic image of the construction surface as the spatial hierarchy basis, combined with the main path extracted by the K-Means algorithm in the support structure deformation monitoring sequence, to determine a representative set of observation nodes and represent them as vertices in the hypergraph. These vertices include not only the observation points arranged in the actual physical space, but also the fusion dimensions derived from tensor decomposition and alignment in the virtual space, such as the changing level of geological stratification or the working section of construction machinery. Next, using the construction time sequence as the evolutionary sequence of edge weights, a dynamic Bayesian network was used to calculate the time-varying correlations between multimodal variables. The perturbation paths of parameters such as thermal, mechanical, gas, and shape were dynamically encoded as strength values ​​between edges, thereby constructing a heterogeneous edge set with multi-scale coupling. Ultimately, within the hypergraph framework, this node-edge relationship was mapped into a four-dimensional hypergraph tensor space with a nonlinear topological connection structure. This not only possesses high-dimensional modal fusion capabilities but also preserves the dynamic traces of construction disturbance propagation in the spatiotemporal evolution dimension.

[0041] Step S3: performing cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space, and constructing a dynamic heterogeneous graph network;

[0042] It can be understood that this step utilizes advanced algorithms for spectral and topological analysis, enabling in-depth analysis of the spatiotemporal characteristics of construction safety monitoring data at multiple scales. Through multi-resolution analysis and persistent homology topological extraction, the propagation paths of disturbances, structural changes, and high-order correlation patterns can be accurately identified within complex construction environments, ultimately generating a heterogeneous graph network with high-dimensional dynamic expression capabilities. In this step, step S3 includes steps S31, S32, S33, and S34.

[0043] Step S31, performing multi-resolution analysis on the spatiotemporal perturbation path of the four-dimensional hypergraph tensor space based on the spectral wavelet transform method to obtain a cross-scale frequency domain response map;

[0044] It can be understood that after constructing the four-dimensional hypergraph tensor space, this step uses the spectral wavelet transform method to perform spectral domain feature expansion processing based on the cross-scale propagation characteristics of the construction disturbance signal. Among them, based on the spatiotemporal correlation network in the hypergraph tensor space, its nodes include geological body units (generated by discretization of the three-dimensional geological parameter matrix), mechanical state clusters (derived from the cluster analysis of the multimodal time series parameter set of construction machinery) and structural deformation domains (constructed based on the support structure deformation monitoring sequence). The hypergraph topology is mapped to the spectral domain space through the graph Laplacian operator, and a Morlet wavelet kernel function group with frequency band adaptation is constructed. Among them, the low-frequency kernel function captures the global redistribution trend of the geological stress field (such as the low-frequency energy accumulation caused by rock creep), and the high-frequency kernel function analyzes the transient shock wave of mechanical vibration in the support structure (such as the high-frequency vibration mode generated by the impact of the tunnel boring machine cutter head). During this processing step, a spectral wavelet optimization algorithm with non-uniform sampling points is introduced for the spatiotemporal correlation matrix constructed from the thermodynamic image data of the construction surface. By adjusting the dynamic matching mechanism of scale parameters and translation parameters, the mechanical vibration-geological response coupling propagation path (such as the resonant frequency band of the cutterhead rotation speed and the expansion of surrounding rock cracks) is separated in the frequency domain.

[0045] Step S32: performing persistent coherence analysis based on the cross-scale frequency domain response graph to identify high-order topological connection patterns in the disturbance diffusion and obtain a topological coding set;

[0046] It can be understood that after obtaining the cross-scale frequency domain response diagram, this step uses persistent homology analysis to extract high-order topological structures based on the topological evolution characteristics of the construction disturbance diffusion. This step uses the energy distribution in the 0.1-100Hz frequency band in the frequency domain response diagram as input data. The low-frequency component (0.1-1Hz) comes from the stress redistribution characteristics of the three-dimensional geological parameter matrix (such as the energy gradient change caused by rock creep), and the high-frequency component (10-100Hz) corresponds to the modal decomposition result of the cutterhead vibration acceleration spectrum of the construction machinery multimodal time series parameter set. The specific processing process is to first construct a weighted simplicial complex constrained by temperature gradient based on the thermodynamic image of the construction surface: each frequency domain response unit is regarded as a multidimensional simplex vertex, and the connection weights of the simplicial complex edges are defined by jointly calculating the thermodynamic gradient and the frequency domain energy decay rate (such as the product of the temperature change rate and the energy decay coefficient), forming a topological skeleton with physical field coupling characteristics. Subsequently, the disturbance intensity scale is used as a filter function parameter, and the alpha-shape algorithm is used to dynamically track ring structures (such as the H1 homology ring formed by the closure of the rock fracture network) and chain structures (such as the H0 homology chain caused by the deformation of the support anchor). The filter function is used to generate a bar chart on the disturbance intensity scale and map it into a topological feature vector, obtaining a topological coding set with spatial nesting hierarchy and structural continuity characteristics.

[0047] Step S33: performing graph space fusion processing according to the topological coding set to obtain a graph node feature set with time series labels and modal labels;

[0048] It can be understood that this step constructs a multi-resolution matching framework based on the Weisfeiler-Lehman kernel method. At the low-frequency scale (0.1-1 Hz), a diffusion kernel is used to capture the slow evolution path of the geological stress field (such as the cascading fracture mode of the H0 chain caused by rock creep). At the high-frequency scale (10-100 Hz), an impact response kernel is designed to analyze transient topological mutations caused by mechanical vibration (such as the rapid generation and annihilation of the H1 ring under the impact load of the cutterhead). The weight of each kernel is dynamically adjusted by the frequency band energy proportion of the construction machinery hydraulic pressure time series spectrum (for example, the weight of the high-frequency kernel corresponding to the main frequency of the propulsion cylinder pressure pulsation of 35 Hz is increased to 0.7).

[0049] Then, a graph embedding space mapping is performed by constructing a time evolution graph based on the convergence rate sequence monitored by the laser profiler and a spatial correlation graph of the temperature rise area >5°C captured by the infrared thermal imager. Through the modal attention gating mechanism in the hypergraph neural network, the coupling coefficient of the displacement gradient and temperature gradient of the support node is calculated to generate a joint embedding vector with physical constraints.

[0050] Finally, the timestamp mark of the construction machinery operation cycle (such as the 2.5s rotation period of the tunnel boring machine cutterhead) is introduced into the embedding space. The deformation phase characteristics of the support structure (such as the phase difference between the anchor stress peak and the cutterhead torque curve) are extracted through a temporal convolutional network. At the same time, the spatial grid encoding of the thermodynamic image (such as 2cm×2cm resolution) is used to annotate the spatial mode type of the abnormal area (such as annular hot spots are marked as Type-A, and strip heat fluxes are marked as Type-B). This results in a graph node feature set with time series marks and modal labels.

[0051] The calculation formula for the coupling coefficient between the displacement gradient and temperature gradient of the support node is as follows:

[0052]

[0053] Among them, α is the coupling coefficient between the displacement gradient and the temperature gradient, T is the temperature of the node, u is the displacement of the support node, C is the thermal conductivity coefficient, which represents the thermal conductivity of the material, E is the elastic modulus of the support material, which characterizes the rigidity of the material, β is the thermal expansion coefficient of the support material, which represents the volume or linear expansion of the material per unit temperature change, and k is the thermal conductivity, which represents the ability of the material to conduct heat.

[0054] Step S34: performing dynamic graph generation processing based on the node feature set, wherein a dynamic heterogeneous graph network structure is obtained by constructing a dynamic graph network and encoding the time dimension as the graph evolution path.

[0055] It can be understood that this step achieves network structure stability control based on topological invariant constraints. It adopts a dual mechanism of dynamic monitoring of Betti numbers and thermodynamic correction of Euler characteristics (for example, when the cutterhead vibration triggers a surge in ring structure, the abnormal edge weight is attenuated by the Lyapunov exponent). It combines three heterogeneous edge generation strategies: mutual information entropy time-lag correlation analysis (geology-equipment edges are established when the mutual information between rock stress and equipment torque is greater than 0.7), thermodynamically constrained random walk (geology-structure edges are generated along the permeability field in areas with a temperature greater than 3°C), and graph attention mechanism (equipment-structure edges are generated when the attention score of hydraulic pressure gradient and support displacement rate is greater than 0.6). This forms a multidimensional coupled network with 12 node types and 9 edge types. The timestamps encoded in the construction cycle are used to drive graph snapshot updates (preferably at 0.5s intervals) and memory-enhanced graph neural networks (retaining the topological patterns of the previous 10 cycles). Ultimately, a heterogeneous graph model that can accurately characterize the dynamic interaction between "geology, equipment, and structure" is constructed.

[0056] Step S4: performing spatiotemporal graph convolution processing on the dynamic heterogeneous graph network, and generating a composite risk manifold based on the processing results and the multi-head attention mechanism;

[0057] It can be understood that this step, through the combination of graph convolution and the attention mechanism, can effectively improve the spatial resolution and temporal response sensitivity of risk state expression, thereby accurately modeling the deep dependencies between complex disturbance factors during tunnel construction, providing highly reliable data support and a foundation for risk description for subsequent critical state identification and warning boundary construction. In this step, step S4 includes steps S41, S42, S43, and S44.

[0058] Step S41: performing spatiotemporal structure modeling based on the dynamic heterogeneous graph network to obtain a spatiotemporal fused graph node state sequence;

[0059] It can be understood that in this step, the multimodal graph constructed in the previous stage is first functionally divided, and the graph nodes are split into two channels: geological disturbance subgraph and structural response subgraph according to physical properties and observation indicators. The nodes represent key perception points in the construction area, such as crack-induced ground stress mutation points, support structure displacement monitoring points, and construction machinery earthquake source locations, etc. The edge weights reflect the physical coupling degree (such as material continuity) or dynamic correlation strength (such as vibration correlation) between nodes. In terms of spatial modeling, the spatiotemporal graph convolution module based on graph neural networks realizes spatial aggregation of node features according to the adjacency matrix through graph convolution operations, accurately identifying the propagation path and diffusion boundary of the disturbance; in terms of temporal modeling, the non-stationary evolution trend of the node state over time is captured through the gated recursive unit, effectively modeling the persistence and hysteresis of the disturbance event. In particular, the spatiotemporal graph convolution module based on graph neural networks preserves the heterogeneity of edge types and subgraph structures throughout the graph convolution process. For example, when processing the geological disturbance subgraph, it can prioritize amplifying local response changes such as those caused by sudden high-frequency stress waves, while in the structural response subgraph, it focuses more on the slow accumulation trend of continuous deformation, ensuring compatibility of information transmission paths under multiple scales and physical mechanisms within the same graph structure. Through this dual modeling mechanism, the resulting graph node state sequence not only clearly expresses the multi-level impact of the disturbance source in time and space, but also accurately depicts the response characteristics and sensitivity evolution of the structural system to external disturbances at different stages, providing highly dynamic and consistent feature input for the subsequent construction of a composite risk manifold based on the attention mechanism.

[0060] Step S42: Perform dynamic graph embedding processing according to the graph node state sequence to obtain a structural time sequence embedding graph;

[0061] It is understandable that in tunnel construction, construction machinery vibration points, surrounding rock crack nodes and support structure deformation points often constitute a typical disturbance-propagation-response triple structure, and traditional edge-to-edge modeling has difficulty identifying its evolutionary consistency. This step introduces the temporal graph regularization term and the triple consistency loss function to perform constrained optimization on the embedding vector of each node, ensuring that when the physical role of the node (such as "disturbance source", "force-bearing body" and "conduction medium") changes dynamically, its temporal embedding can still maintain semantic continuity. At the same time, this method strengthens the dynamic representation ability of edge relationships. By analyzing the changes in edge weights of the same triple in different time periods, it explores potential risk flow chain paths such as "increased mechanical vibration-aggravated structural deformation", thereby constructing a low-dimensional embedding graph with significant spatiotemporal topological characteristics. Especially between construction equipment and structural nodes, due to the existence of a large number of non-explicit physical connections, such as indirect influence paths caused by vibration, heat conduction or stress waves, traditional graph embedding is difficult to model these weak connection phenomena. This step adopts a time-aware dynamic graph embedding method, which effectively completes such implicit relationships through the triple closure mechanism, enhancing the structural integrity and dynamic consistency of the embedded representation.

[0062] Step S43: construct a multi-head heterogeneous graph attention mechanism, and perform attention weight optimization processing on the structural temporal embedding graph to perturb the propagation subgraph;

[0063] It can be understood that in this step, the multi-head heterogeneous graph attention mechanism first establishes independent attention heads for each node type (such as equipment nodes, geological nodes, and sensor nodes) in the heterogeneous graph, and calculates the intensity of its attention to neighboring nodes in the local subgraph. This avoids the problem of traditional attention mechanisms "averaging" in heterogeneous structures, which leads to the dilution of important structural information. Secondly, in the time segment dimension, the multi-head heterogeneous graph attention mechanism introduces a cross-time period attention window mechanism. By learning the semantic dependencies between node embeddings at different time points, it captures the response fluctuations of construction disturbance signals at different stages.

[0064] The calculation formula for the attention intensity of neighbor nodes is as follows:

[0065]

[0066] Among them, h i and h j represents the feature vector of node i and its neighbor node j, A is the linear transformation matrix of node features, w is the learned attention weight vector, w T is the transpose of the learned attention weight vector, || is the vector concatenation operation, Leaky ReLU is the ReLU activation function, and γ i,j is the attention weight of node i to neighbor node j, h j' represents the feature representation vector of neighbor node j.

[0067] In practical applications, such as during tunnel face advancement, the high-frequency operation of construction machinery can cause periodic disturbances in the surrounding rock ahead. These disturbances often manifest as subtle deformation changes in the support structure after 1-2 time steps. The multi-head heterogeneous graph attention mechanism demonstrates significant advantages in modeling this type of delayed response. Its temporal attention mechanism can identify these asynchronous signal propagation paths and assign higher weight coefficients, making the disturbance propagation paths more prominent in the graph. Furthermore, for structural instability caused by support transitions, the model dynamically adjusts node type attention to enhance sensitivity to coordinated strain changes between structural nodes, forming a disturbance propagation subgraph centered around "structural risk focal points." This subgraph exhibits a highly concentrated star or chain structure topologically. Its edge weights are dynamically assigned by the multi-head heterogeneous graph attention mechanism, reflecting the strength of the fused response to the disturbance signal across the spatial, temporal, and physical dimensions.

[0068] Step S44: construct a set of disturbance trajectories of construction disturbances propagating among multiple types of nodes by using the disturbance propagation subgraph and random walk diffusion mechanism, and embed the set into a high-dimensional disturbance state space to obtain a composite risk manifold.

[0069] It can be understood that in constructing a set of disturbance trajectories that propagate construction disturbances across multiple types of nodes, the disturbance source node (such as a high-frequency vibration shield cutterhead node or a sudden temperature increase in the surrounding rock region) is first set as the initial wandering starting point. At each hop, a biased decision is made based on the edge type, timestamp, and historical propagation probability, prioritizing diffusion to node groups with high structural sensitivity or significant historical disturbance responses. Secondly, the wandering path records not only the node sequence but also changes in disturbance intensity, directional flow trends, and local aggregation (calculated through node cohesion and path overlap), thereby forming a disturbance trajectory map for spatiotemporal evolution. This trajectory set effectively represents the nonlinear transmission chain of disturbances from "source-intermediary-response," such as the cross-modal dynamic path by which mechanical disturbances propagate through weak strata and induce deformation anomalies at the support end.

[0070] Then, multiple disturbance trajectories are projected into a unified high-dimensional disturbance state space. A graph trajectory embedding approach is used to jointly encode the structural features, node semantics, and propagation dynamics of the disturbance paths into a risk vector representation. This results in a composite risk manifold that integrates the propagation characteristics of mechanical and geological bidirectional disturbances. This manifold not only includes the spatial propagation range and time delay characteristics of the disturbance, but also reflects the combined impact of different disturbance source types on system stability.

[0071] Step S5: performing coupling processing according to the composite risk manifold, and performing stability analysis and critical state discrimination based on the risk phase change hypersurface obtained by the processing, and performing real-time warning based on the dynamic warning boundary of the tunnel construction obtained by the discrimination.

[0072] It can be understood that this step, through the risk phase transition hypersurface mapping mechanism, not only dynamically determines whether the system is approaching a critical state of instability, but also accurately delineates the dynamic warning boundary under multi-source coupling risks, thereby providing a quantitative basis and early intervention window for construction scheduling optimization, equipment operating condition adjustment, and intelligent support strategies. In this step, step S5 includes steps S51, S52, S53, and S54.

[0073] Step S51: performing structural domain mapping processing on the composite risk manifold, and using local cross-section construction to map the local response transformation of the construction disturbance under different dimensional variables to obtain the fiberization disturbance manifold;

[0074] It is understandable that the Northern Road Standard first uses a local section construction method to select specific disturbance sources (such as construction equipment operation, surrounding rock stress fluctuations, and construction process changes). These disturbance sources are used as sections in the bottom space to form response mapping relationships under local disturbance conditions. Each local section reflects the structural response changes in the construction environment under specific disturbance source conditions. Then, by performing geometric projection transformations on these local sections, disturbance response maps are generated on the structural fiber. These maps can effectively capture the different impacts of disturbances on tunnel construction at different construction stages. For example, the impact of equipment operating status on geological condition fluctuations and stress redistribution caused by process changes are expanded as the local geometric structure of the fiber on the high-dimensional disturbance manifold. Finally, by constructing the disturbance propagation path through the fiber bundle geometry, an engineering semantic layering based on variables such as construction process, equipment, and geological conditions can be implemented on this map. In other words, each different disturbance factor is assigned a clear engineering meaning on the high-dimensional disturbance manifold, thereby more clearly identifying the impact of each disturbance source on construction safety and risk assessment.

[0075] Step S52: constructing a continuous state transformation model of disturbance between different working conditions based on the fiberized disturbance manifold, and using the exponential mapping of Lie algebra to characterize the differential evolution law of the disturbance factor to obtain the dynamic driving manifold field;

[0076] It can be understood that this step first defines the disturbance states under different construction stages or working conditions by using construction state variables (such as the operating status of construction equipment, process change parameters, geological conditions, etc.) as elements of Lie group orbits. Lie group orbits represent the continuous state transition of disturbance factors during changes in working conditions. This transition process maintains the symmetry and continuity between the disturbance source and the construction structure. In this way, the changes in the disturbance factors in the system in the structural field can be regarded as the flow and transformation between Lie group orbits.

[0077] Then, the differential evolution law of the disturbance factor in the structural dynamic field is characterized by the exponential mapping of Lie algebra. The exponential mapping of Lie algebra maps the elements of Lie algebra (describing small changes in the disturbance factor) into the Lie group, thus providing a continuous dynamic evolution path. This mapping can not only describe the immediate response of the disturbance factor during the construction process, but also reveal the cumulative effect and long-term evolution process of the disturbance factor under different construction conditions. Specifically, the disturbance factor may cause a series of complex dynamic responses in the structure through nonlinear coupling effects, and the exponential mapping of Lie algebra can effectively capture the dynamic patterns of these changes and provide a continuous and precise mathematical description for risk assessment.

[0078] Ultimately, the resulting dynamic driving manifold field, constructed through this model, possesses a continuous and symmetrical structure, capable of reflecting the driving effects of various disturbance factors (such as equipment operation, process changes, and geological disturbances) on the dynamic system of tunnel construction during construction. This structured model not only describes how disturbance factors interact under different construction conditions but also provides a multi-level, dynamically evolving modeling foundation for construction safety prediction.

[0079] Step S53: extracting features based on the curvature response of the dynamic driving manifold field, and performing direction-sensitive modeling based on the covariant derivative operator to obtain a geometric feature field that describes the degree of distortion of the disturbance response;

[0080] It can be understood that this step introduces two important tools in geometry: the Riemann curvature tensor and the covariant derivative operator, to analyze and model high-dimensional disturbance responses. Specifically, when disturbance factors (such as mechanical vibrations, geological stresses, etc.) propagate in the structure, their paths may bend due to changes in the spatial structure. Through the Riemann curvature tensor, we can quantify this bending trend, that is, identify the degree to which the disturbance factor deviates from the space-time manifold. This process reveals whether disturbances in different regions will produce nonlinear responses or local stress concentrations in the structure.

[0081] Next, by establishing tensor path variability, we further analyze the spatial-temporal evolution of the disturbance response. Tensor path variability describes the differences in the propagation paths of disturbance factors between different structural nodes. This variability helps reveal whether the propagation of disturbance factors is affected by structural morphological changes at different construction stages or locations. This method accurately captures the dynamic changes at each node in the structure during disturbance transmission, helping to identify potential risk concentrations.

[0082] Furthermore, the covariant derivative operator is a tool for describing the rate of change of quantities within a manifold, particularly useful in spaces with curvature. In this step, it is used to model the directionality of the risk gradient of the disturbance. In other words, the covariant derivative accurately quantifies the rate of change of the disturbance factor in different directions within the manifold and, in conjunction with the manifold's geometric properties, assesses its impact on the risk gradient. This process reveals the directionality of disturbance propagation and its changing trends during different construction phases, providing more refined directional information for risk analysis.

[0083] Ultimately, by combining these geometric tools, a geometric characteristic field is generated that describes the degree of distortion in response to a disturbance factor. This characteristic field not only reveals the impact of local disturbances on the structure but also identifies the distortion of the disturbance factor at different spatial and temporal scales, further supporting dynamic risk prediction and early warning during construction.

[0084] Step S54: constructing a phase change critical surface based on the geometric characteristic field to obtain a risk phase change hypersurface that describes the drastic change boundary in the construction disturbance evolution stage.

[0085] As you can understand, this step first uses the curvature tensor to calculate the local curvature trend of the disturbance on the spacetime manifold, thereby identifying areas where the curvature changes suddenly, which often indicate a sharp change in the structural or geological response. Covariant derivative discontinuities reveal discontinuous changes in the propagation of the disturbance factor, which often occur at critical moments when the structural state changes. By identifying these areas, potential risk points can be marked.

[0086] Next, a dual-scalar function construction method was used to extract the risk of sudden changes. This method defines two scalar functions: one describes the evolution of the perturbation factor, and the other describes the changes in the structural response. Through the interaction of these two scalar functions, bifurcation structures can be accurately extracted in the state space—the transition points between different states of the perturbation factor and the structural response. These bifurcation points correspond to moments in the construction process where sudden changes may occur, revealing the nonlinear evolution of the perturbation factor and the structural response under different conditions.

[0087] The evolution formula of the perturbation factor is as follows:

[0088] φ1(t,x)=ρ E ·e -B·x sin(ωt)

[0089] Among them, φ1(t,x) represents the evolution intensity of the perturbation factor at time t and position x, ρ E It represents the disturbance energy per unit volume, e represents the strain variable during the disturbance or structural response process, B represents the attenuation coefficient of the disturbance in space, x represents the spatial position variable, t represents the time variable, and ω represents the frequency variable.

[0090] The formula describing the change in structural response is as follows:

[0091]

[0092] Among them, φ2(t,x) represents the comprehensive change of the structural response at time t and position x, α1 represents the weight coefficient, which is used to adjust the influence of displacement change on the structural response, u(t,x) represents the displacement function of the node or structural unit in time and space, α2 represents the weight coefficient, which is used to adjust the influence of temperature gradient on the structural response, and T(t,x) represents the temperature field distribution of the structure at a certain position and time.

[0093] To extend these bifurcation points to a continuous risk boundary, high-dimensional surface fitting technology is further introduced. This technology constructs a smooth hypersurface in high-dimensional space by performing surface fitting on bifurcation points. This hypersurface can reflect the critical stages of construction disturbance evolution and the risk fluctuation trends. High-dimensional surface fitting technology utilizes the spatial relationships of multidimensional data to accurately describe the propagation path of disturbance factors and the dynamic changes in structural response during construction.

[0094] It can be understood that in this step, step S5 also includes step S55, step S56, step S57 and step S58.

[0095] Step S55: performing topological stability structure extraction processing based on the risk phase transition hypersurface to obtain a basic topological indicator set representing risk morphological stability;

[0096] It can be understood that this step first quantifies the internal topological structure of the hypersurface by introducing homological invariants from algebraic topology, such as Betti numbers and Euler characteristics. These homological invariants can effectively capture important features of the hypersurface, such as connectivity, hole structure, and ring-shaped perturbation channels.

[0097] Specifically, Betti numbers are used to measure topological features in different dimensions. For example, the Betti number measures the number of connected components in a hypersurface, reflecting the connectivity of the perturbation path in space and time. The Betti number represents the number of ring structures in the hypersurface, revealing potential ring-shaped perturbation channels that may be closely related to structural instability or risk expansion. By calculating these Betti numbers, we can further understand the topological changes within the hypersurface and identify regions that exhibit high topological complexity during the perturbation evolution process as potential critical perturbation clusters.

[0098] Furthermore, the Euler characteristic, a key topological metric that combines structural information from different dimensions within a hypersurface, further verifies and supports the conclusions of the topological stability analysis by calculating the combined values ​​of connectivity, number of holes, and other morphological features. The positive and negative variations in the Euler characteristic also provide an effective basis for assessing the morphological stability of a hypersurface, helping to identify potential mutation points or high-risk areas within the risk morphology.

[0099] Step S56: performing stability differential modeling processing on the basic topological indicator set based on a preset stochastic differential equation to obtain a stability spectrum function that characterizes the risk evolution trend;

[0100] It's understandable that this step introduces stochastic differential equations to model the randomness and uncertainty of disturbances in the system, especially when considering the effects of environmental noise and external disturbances. Stochastic differential equations provide a powerful mathematical framework for describing dynamic systems with stochastic processes, accurately capturing the evolution of disturbances, particularly their behavioral characteristics in the face of random fluctuations or external uncertainty.

[0101] The stability spectrum function is as follows:

[0102]

[0103] in, represents the stability spectrum function, ω is the frequency variable, η is the disturbance intensity, D represents the duration of the disturbance, u(x, t) represents the displacement field of the structure, e(x, t) represents the strain field of the structure, e represents the strain variable during the disturbance or structural response process, x represents the spatial position variable, t represents the time variable, and u is the displacement of the support node.

[0104] First, the evolution trajectory of the disturbance factor is modeled as the system dynamics subject to random influences. The basic model form is as follows:

[0105] dX(t)=μ(X(t),t)dt+σ(X(t),t)dW(t)

[0106] Where X(t) represents the state of the disturbance factor at time t, μ(X(t), t) is the drift term of the system, which describes the deterministic change of the system state over time, σ(X(t), t) is the diffusion term, which describes the random disturbance intensity of the system state, and dW(t) is the standard Brownian motion.

[0107] By incorporating Ito's lemma, this method can handle the stochastic differential evolution of nonlinear systems and further derive formulas for system stability analysis. Ito's lemma allows for effective differentiation of random processes, thereby deriving the evolution of the system state under given perturbations and helping to infer the system's long-term behavioral trends. By solving the basic model form, the probabilistic response of the system under different perturbation intensities can be derived, that is, the probability of the system stabilizing or diverging. This provides an important basis for assessing the long-term stability and security of the system.

[0108] Among them, the stability analysis is carried out using the Lyapunov index. If the Lyapunov index is less than 0, it means the system is stable. If it is greater than 0, it means the system is unstable. If it is equal to 0, it means critical stability.

[0109] Step S57: performing critical state structure identification processing on the stability spectrum function based on the Morse critical point theory, and constructing a critical point manifold structure map based on the identification result to obtain a critical state set for distinguishing risk state transitions;

[0110] It can be understood that this step first identifies critical points by performing gradient analysis on the stability spectrum function. These critical points correspond to turning points or points of structural instability in the system under specific perturbation conditions. In Morse theory, a critical point is a point where the gradient of the function is zero. When the stability spectrum function reaches a local extreme value or a saddle point, these points are considered critical points. By classifying and analyzing these critical points, it is possible to reveal key turning points in the system's evolution under perturbation.

[0111] Next, based on the identified critical points, a critical point manifold structure map is constructed. A critical point manifold is a high-dimensional structure consisting of a series of critical points and the trajectories or paths connecting these points. It illustrates the dynamic process of a system transitioning from stability to instability, and from one stable region to another. In this process, the construction of the critical point manifold utilizes the local structure and global topological relationships of the critical points, capable of displaying the transition regions of the system under different states. This structural map provides a clear visualization tool for understanding the potential instability or risk transition of the system, which is particularly important for identifying potential safety hazards in complex projects such as tunnel construction.

[0112] Step S58: quantitatively evaluate the critical state set based on the fused Lyapunov exponent spectrum, and construct a warning boundary in the parameter space based on the evaluation result.

[0113] It can be understood that in this step, the evolution trajectory of the system under different disturbance intensities is calculated by integrating the Lyapunov exponent spectrum, and a set of numerical values ​​reflecting the system's response to small disturbances is obtained. The larger these values ​​are, the more sensitive the system is to disturbances at that point, and the worse the system's stability is. For geological disturbances and structural responses in tunnel construction, the Lyapunov exponent spectrum can help assess stability risks at different construction stages. For example, equipment vibrations and changes in surrounding rock stress during construction may cause turbulence in the local system. The Lyapunov exponent spectrum can provide quantitative assessments in these key areas and identify high-risk areas that may lead to instability.

[0114] Once the Lyapunov exponent spectrum has been used to assess each critical state, it is possible to construct a warning boundary in the parameter space based on these assessment results. In this process, the parameter space is a multidimensional space of system states, with each point in this space corresponding to a construction state. The warning boundary is constructed by linking the system's critical states with the unstable regions in the Lyapunov exponent spectrum, thereby forming one or more "boundary lines" that define the critical values ​​at which the system may become unstable or experience a risk transition. For example, when the system state is in a critical state and exceeds the warning boundary, it indicates that the construction state corresponding to that point has a high risk of instability, and immediate emergency measures may be required.

[0115] Example 2:

[0116] like Figure 2 As shown, this embodiment provides a tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion, see Figure 2 The system includes an acquisition unit 701 , a processing unit 702 , a construction unit 703 , a generation unit 704 and a determination unit 705 .

[0117] An acquiring unit 701 is configured to acquire first information, where the first information includes tunnel construction parameters and image information;

[0118] A processing unit 702 is configured to perform spatiotemporal fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space for tunnel construction;

[0119] A construction unit 703 is used to perform cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space and construct a dynamic heterogeneous graph network;

[0120] A generating unit 704 is configured to perform spatiotemporal graph convolution processing on the dynamic heterogeneous graph network, and generate a composite risk manifold based on the processing result and a multi-head attention mechanism;

[0121] The discrimination unit 705 is used to perform coupling processing according to the composite risk manifold, and perform stability analysis and critical state discrimination based on the risk phase change hypersurface obtained by the processing, and provide real-time warning based on the dynamic warning boundary of the tunnel construction obtained by discrimination.

[0122] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0123] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

[0124] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion, characterized in that: include: Acquiring first information, where the first information includes tunnel construction parameters and image information; Performing spatiotemporal fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space for tunnel construction; Performing cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space, and constructing a dynamic heterogeneous graph network; Performing spatiotemporal graph convolution processing on the dynamic heterogeneous graph network, and generating a composite risk manifold based on the processing results and a multi-head attention mechanism; Coupling processing is performed according to the composite risk manifold, and stability analysis and critical state discrimination are carried out based on the risk phase transition hypersurface obtained through processing. Real-time warning is given based on the dynamic warning boundary of tunnel construction obtained by discrimination.

2. The tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion according to claim 1 is characterized in that , performing spatiotemporal fusion processing and redundancy reduction processing on the first information, including: Based on Tucker tensor decomposition, the principal components of the spatial and temporal dimensions of the first information are extracted, and a local low-rank subspace expression of the first information is constructed to obtain an initial third-order tensor set; The initial third-order tensor set of construction machinery timing parameters and support deformation dynamic trajectory are aligned in the manifold space to obtain a fourth-order fusion tensor representation; Based on the preset PARAFAC tensor decomposition method, the common change trend of the fourth-order fusion tensor representation is extracted under a fixed mode to obtain a sparse tensor structure; A node hypergraph structure is constructed based on the sparse fusion tensor structure, and the time evolution sequence is used as the edge weight to map it into a four-dimensional hypergraph tensor space.

3. The tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion according to claim 1 is characterized in that , performing cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space, including: Based on the spectral wavelet transform method, the spatiotemporal perturbation path of the four-dimensional hypergraph tensor space is analyzed at multiple resolutions to obtain the cross-scale frequency domain response map. Performing persistent homology analysis based on the cross-scale frequency domain response map to identify high-order topological connection patterns in the disturbance diffusion and obtain a topological coding set; Performing graph space fusion processing according to the topological coding set to obtain a graph node feature set with time series labels and modal labels; A dynamic graph generation process is performed based on the node feature set, wherein a dynamic heterogeneous graph network structure is obtained by constructing a dynamic graph network and encoding the time dimension as an evolution path of the graph.

4. The tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion according to claim 1 is characterized in that ,The dynamic heterogeneous graph network is subjected to spatiotemporal graph convolution processing, and a composite risk manifold is generated based on the processing results and the multi-head attention mechanism, including: Based on the dynamic heterogeneous graph network, spatiotemporal structure modeling is performed to obtain a spatiotemporal fusion graph node state sequence; Perform dynamic graph embedding processing based on the graph node state sequence to obtain a structural time-series embedding graph; Construct a multi-head heterogeneous graph attention mechanism, optimize the attention weights of the structured temporal embedding graph, and perturb the propagation subgraph; The disturbance propagation subgraph and random walk diffusion mechanism are used to construct a set of disturbance trajectories of construction disturbances propagating among multiple types of nodes, and the set is embedded in a high-dimensional disturbance state space to obtain a composite risk manifold.

5. The tunnel construction safety monitoring and early warning method based on multi-dimensional data fusion according to claim 1 is characterized in that ,Coupling processing is performed according to the composite risk manifold, including: The composite risk manifold is processed by structural domain mapping, and the local response transformation of the construction disturbance under different dimensional variables is mapped using local cross-section construction to obtain the fiberized disturbance manifold. Based on the fiberized disturbance manifold, a continuous state transformation model of the disturbance between different working conditions is constructed, and the exponential mapping of Lie algebra is used to characterize the differential evolution law of the disturbance factor, thereby obtaining the dynamic driving manifold field. The curvature response is extracted based on the dynamic driving manifold field, and the direction-sensitive modeling is performed based on the covariant derivative operator to obtain the geometric characteristic field that describes the degree of distortion of the disturbance response. The phase change critical surface is constructed based on the geometric characteristic field to obtain the risk phase change hypersurface that describes the drastic change boundary in the construction disturbance evolution stage.

6. A tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion, characterized in that: include: an acquiring unit, configured to acquire first information, wherein the first information includes tunnel construction parameters and image information; a processing unit, configured to perform spatiotemporal fusion processing and redundancy reduction processing on the first information to generate a four-dimensional hypergraph tensor space for tunnel construction; A construction unit, configured to perform cross-scale analysis and topology extraction on the four-dimensional hypergraph tensor space and construct a dynamic heterogeneous graph network; A generation unit, configured to perform spatiotemporal graph convolution processing on the dynamic heterogeneous graph network, and generate a composite risk manifold based on the processing result and a multi-head attention mechanism; The discrimination unit is used to perform coupling processing based on the composite risk manifold, and to perform stability analysis and critical state discrimination based on the risk phase change hypersurface obtained through the processing, and to provide real-time warning based on the dynamic warning boundary of tunnel construction obtained through the discrimination.

7. The tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion according to claim 6 is characterized in that: The processing unit includes: A first processing subunit is configured to extract principal components of the spatial dimension and the temporal dimension of the first information based on Tucker tensor decomposition, and construct a local low-rank subspace expression of the first information to obtain an initial third-order tensor set; The second processing subunit is used to align the construction machinery timing parameters of the initial third-order tensor set with the support deformation dynamic trajectory in the manifold space to obtain a fourth-order fusion tensor representation; The third processing subunit is used to extract the common change trend of the fourth-order fusion tensor representation under a fixed mode based on a preset PARAFAC tensor decomposition method to obtain a sparse tensor structure; The fourth processing subunit is used to construct a node hypergraph structure based on the sparse fusion tensor structure, and map it into a four-dimensional hypergraph tensor space using the time evolution sequence as the edge weight.

8. The tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion according to claim 6 is characterized in that: The building block comprises: The first construction subunit is used to perform multi-resolution analysis on the spatiotemporal perturbation path of the four-dimensional hypergraph tensor space based on the spectral wavelet transform method to obtain a cross-scale frequency domain response map; The second construction subunit is used to perform persistent homology analysis based on the cross-scale frequency domain response map, identify high-order topological connection patterns existing in the disturbance diffusion, and obtain a topological coding set; The third construction subunit is used to perform graph space fusion processing according to the topological coding set to obtain a graph node feature set with time series labels and modal labels; The fourth construction subunit is used to perform dynamic graph generation processing based on the node feature set, wherein a dynamic heterogeneous graph network structure is obtained by constructing a dynamic graph network and encoding the time dimension as the evolution path of the graph.

9. The tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion according to claim 6 is characterized in that: The generating unit comprises: The first generation subunit is used to perform spatiotemporal structure modeling based on the dynamic heterogeneous graph network to obtain a spatiotemporal fused graph node state sequence; The second generating subunit is used to perform dynamic graph embedding processing according to the graph node state sequence to obtain a structural time sequence embedding graph; The third generation sub-unit is used to construct a multi-head heterogeneous graph attention mechanism, optimize the attention weight of the structural temporal embedding graph, and perturb the propagation sub-graph; The fourth generation subunit is used to construct a disturbance trajectory set of construction disturbance propagation among multiple types of nodes by using the disturbance propagation subgraph and random walk diffusion mechanism, and embed it into a high-dimensional disturbance state space to obtain a composite risk manifold.

10. The tunnel construction safety monitoring and early warning system based on multi-dimensional data fusion according to claim 6, characterized in that: The discrimination unit includes: The first discriminant subunit is used to perform structural domain mapping on the composite risk manifold and use the local cross-section construction to map the local response transformation of the construction disturbance under different dimensional variables to obtain the fiberization disturbance manifold; The second discriminant subunit is used to construct a continuous state transformation model of disturbance between different working conditions based on the fiberized disturbance manifold, and use the exponential mapping of Lie algebra to characterize the differential evolution law of the disturbance factor to obtain the dynamic driving manifold field; The third discriminant subunit is used to extract features based on the curvature response of the dynamic driving manifold field and perform direction-sensitive modeling based on the covariant derivative operator to obtain a geometric feature field that describes the degree of distortion of the disturbance response; The fourth discriminant subunit is used to construct a phase change critical surface based on the geometric characteristic field to obtain a risk phase change hypersurface that describes the drastic change boundary in the construction disturbance evolution stage.

Images