A method and system for identifying stratum disturbance in shield tunnel construction based on muon detection

Abstract

The application provides a muon detection-based stratum disturbance identification method and system in shield tunnel construction, and relates to the technical field of shield tunnel construction.The method comprises the following steps: obtaining a muon flux scanning image of a shield construction influence area; determining a plurality of stratum disturbance suspected areas based on gas diffusion response data and stratum deformation response data of the shield construction influence area; determining an artificial vibration loading sequence of each stratum disturbance suspected area based on the plurality of stratum disturbance suspected areas; applying vibration excitation to each stratum disturbance suspected area based on the artificial vibration loading sequence of each stratum disturbance suspected area, and obtaining vibration response data and density change data of each stratum disturbance suspected area; determining an abnormal stratum disturbance area based on the vibration response data and the density change data of each stratum disturbance suspected area, and sending stratum disturbance early warning information.The method can efficiently and accurately identify an abnormal stratum disturbance area in a shield construction influence range.

Description

Technical Field

[0001] This invention relates to the field of shield tunnel construction technology, specifically to a method and system for identifying ground disturbances during shield tunnel construction based on muon detection. Background Technology

[0002] In existing shield tunnel construction, it is typically necessary to analyze the geological conditions within the construction impact area by combining engineering design and construction data, geological survey data, and on-site monitoring results. Because the shield machine continuously alters the stress state of the surrounding strata during excavation, the response of different areas to construction disturbances varies significantly, making it difficult to quickly and accurately determine the location of potential abnormal strata within a large construction impact area. Furthermore, shield construction impact areas are usually characterized by large spatial extent, complex geological conditions, and the gradual propagation of disturbances. In some areas, subsequent excavation disturbances may lead to problems such as loosening of the strata, increased voids, and localized structural damage. Current analysis methods, when facing complex construction environments, usually require a comprehensive assessment of the geological conditions using multiple monitoring data. However, due to the large area affected by shield construction, significant monitoring resources are needed, and the differences in geological responses across different areas make it difficult to quickly and accurately determine the location of abnormal strata requiring close monitoring. This can easily lead to insufficient efficiency in identifying risk areas, affecting safety risk control during shield construction.

[0003] Therefore, how to efficiently and accurately identify abnormal geological disturbance areas within the impact range of shield tunneling construction is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem this invention addresses is how to efficiently and accurately identify abnormal geological disturbance areas within the influence range of shield tunneling construction.

[0005] According to a first aspect, the present invention provides a method for identifying ground disturbance during shield tunnel construction based on muon detection, comprising: acquiring a muon flux scanning image of the area affected by shield construction; determining multiple inert gas injection monitoring areas based on the muon flux scanning image of the area affected by shield construction; determining an inert gas injection test plan based on the muon flux scanning image of the area affected by shield construction and the multiple inert gas injection monitoring areas; executing the inert gas injection test plan and acquiring gas diffusion response data and ground deformation response data of the area affected by shield construction; determining multiple suspected ground disturbance areas based on the gas diffusion response data and ground deformation response data of the area affected by shield construction; determining an artificial vibration loading sequence for each suspected ground disturbance area based on the multiple suspected ground disturbance areas; applying vibration excitation to each suspected ground disturbance area based on the artificial vibration loading sequence for each suspected ground disturbance area and acquiring vibration response data and density change data for each suspected ground disturbance area; determining abnormal ground disturbance areas based on the vibration response data and density change data of each suspected ground disturbance area and sending ground disturbance early warning information.

[0006] In one possible implementation, determining the inert gas injection test scheme based on the muon flux scanning image of the shield tunneling area and the multiple inert gas injection monitoring areas includes: determining multiple candidate inert gas injection point information based on the muon flux scanning image of the shield tunneling area and the multiple inert gas injection monitoring areas; clustering the multiple candidate inert gas injection point information to obtain multiple candidate injection point clusters; and determining the inert gas injection test scheme based on the multiple candidate injection point clusters.

[0007] In one possible implementation, determining the artificial vibration loading sequence for each suspected formation disturbance region based on the plurality of suspected formation disturbance regions includes: determining the formation compaction, formation structural stability level, and formation vibration sensitivity of each suspected formation disturbance region based on gas diffusion response data and formation deformation response data; determining multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity of each suspected formation disturbance region based on the formation compaction, formation structural stability level, and formation vibration sensitivity of each suspected formation disturbance region; and constructing multiple vibration loading maps, each vibration loading map corresponding to a suspected formation disturbance region. Each vibration loading map, representing a suspected region, includes multiple nodes and edges. The nodes consist of multiple fixed vibration frequency nodes and one random vibration frequency node. Each fixed vibration frequency node is connected to the random vibration frequency node via an edge. The fixed vibration frequency nodes are characterized by a fixed vibration loading frequency and a maximum vibration intensity, while the random vibration frequency nodes are characterized by a random vibration loading frequency range and a maximum vibration intensity. The edges represent the stratigraphic compaction, stratigraphic structural stability level, and stratigraphic vibration sensitivity of the suspected disturbance region. Based on a vibration sequence determination model, the multiple vibration loading maps are processed to obtain a fixed vibration loading sequence and a random vibration loading sequence for each suspected disturbance region.

[0008] In one possible implementation, the artificial vibration loading sequence for each suspected formation disturbance area includes a fixed vibration loading sequence and a random vibration loading sequence for each suspected formation disturbance area.

[0009] According to a second aspect, the present invention provides a ground disturbance identification system based on muon detection during shield tunnel construction, comprising: an image acquisition module for acquiring muon flux scanning images of the area affected by shield construction; a monitoring area determination module for determining multiple inert gas injection monitoring areas based on the muon flux scanning images of the area affected by shield construction; a scheme determination module for determining an inert gas injection test scheme based on the muon flux scanning images of the area affected by shield construction and the multiple inert gas injection monitoring areas; a response data acquisition module for executing the inert gas injection test scheme and acquiring gas diffusion response data and ground deformation response data of the area affected by shield construction; and a suspected area... The system includes a determination module for identifying multiple suspected ground disturbance areas based on gas diffusion response data and ground deformation response data of the shield tunneling impact area; a sequence determination module for determining an artificial vibration loading sequence for each suspected ground disturbance area based on the multiple suspected ground disturbance areas; an excitation acquisition module for applying vibration excitation to each suspected ground disturbance area based on the artificial vibration loading sequence of each suspected ground disturbance area, and acquiring vibration response data and density change data of each suspected ground disturbance area; and an anomaly warning module for identifying abnormal ground disturbance areas based on the vibration response data and density change data of each suspected ground disturbance area, and sending ground disturbance warning information.

[0010] In one possible implementation, the scheme determination module is further configured to: determine multiple candidate inert gas injection point information based on the muon flux scanning image of the shield tunneling impact area and the multiple inert gas injection monitoring areas; cluster the multiple candidate inert gas injection point information to obtain multiple candidate injection point clusters; and determine an inert gas injection test scheme based on the multiple candidate injection point clusters.

[0011] In one possible implementation, the sequence determination module is further configured to: determine the formation compaction, formation structural stability level, and formation vibration sensitivity of each suspected formation disturbance area based on gas diffusion response data and formation deformation response data; determine multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity of each suspected formation disturbance area based on the formation compaction, formation structural stability level, and formation vibration sensitivity of each suspected formation disturbance area; and construct multiple vibration loading maps, each vibration loading map corresponding to a suspected formation disturbance area, and each vibration loading map including multiple... Each node consists of several nodes and multiple edges. The nodes include several fixed vibration frequency nodes and one random vibration frequency node. Each fixed vibration frequency node is connected to the random vibration frequency node to establish an edge. The node characteristics of the fixed vibration frequency node are fixed vibration loading frequency and maximum vibration intensity. The node characteristics of the random vibration frequency node are random vibration loading frequency range and maximum vibration intensity. The edges represent the stratigraphic compaction degree, stratigraphic structural stability level, and stratigraphic vibration sensitivity of the suspected stratigraphic disturbance area. Based on the vibration sequence determination model, the multiple vibration loading maps are processed to obtain the fixed vibration loading sequence and the random vibration loading sequence for each suspected stratigraphic disturbance area.

[0012] In one possible implementation, the artificial vibration loading sequence for each suspected formation disturbance area includes a fixed vibration loading sequence and a random vibration loading sequence for each suspected formation disturbance area.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method including: acquiring a muon flux scan image of a shield tunneling area affected by tunneling construction; determining a plurality of inert gas injection monitoring areas based on the muon flux scan image of the shield tunneling area affected by tunneling construction; determining an inert gas injection test plan based on the muon flux scan image of the shield tunneling area affected by tunneling construction and the plurality of inert gas injection monitoring areas; executing the inert gas injection test plan and acquiring the shield tunneling impact... The system collects gas diffusion response data and ground deformation response data for the affected area. Based on these data, multiple suspected ground disturbance areas are identified. An artificial vibration loading sequence is determined for each suspected ground disturbance area. Vibration excitation is applied to each suspected ground disturbance area based on the artificial vibration loading sequence, and vibration response data and density change data are obtained for each suspected ground disturbance area. Abnormal ground disturbance areas are identified based on the vibration response data and density change data of each suspected ground disturbance area, and ground disturbance early warning information is sent.

[0014] According to the fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned method for identifying ground disturbance during shield tunnel construction based on muon detection. The method includes: acquiring a muon flux scanning image of the shield construction affected area; determining multiple inert gas injection monitoring areas based on the muon flux scanning image of the shield construction affected area; determining an inert gas injection test plan based on the muon flux scanning image of the shield construction affected area and the multiple inert gas injection monitoring areas; executing the inert gas injection test plan and acquiring the gas flux in the shield construction affected area. The system collects gas diffusion response data and stratum deformation response data. Based on these data, multiple suspected stratum disturbance areas are identified within the shield tunneling area. An artificial vibration loading sequence is determined for each of these suspected stratum disturbance areas. Vibration excitation is applied to each suspected stratum disturbance area based on the artificial vibration loading sequence, and vibration response data and density change data are acquired for each suspected stratum disturbance area. Based on the vibration response data and density change data of each suspected stratum disturbance area, abnormal stratum disturbance areas are identified, and stratum disturbance early warning information is sent.

[0015] The present invention has at least the following beneficial effects:

[0016] 1. This invention integrates information on the impact area of ​​shield tunneling construction, engineering design and construction data, and geological survey data to determine the inert gas injection monitoring area and detection location. It also combines ground response data to conduct targeted detection of the impact area of ​​shield tunneling construction, avoiding the insufficient detection range caused by relying on experience to set up detection locations in traditional methods, and improving the accuracy of ground condition acquisition.

[0017] 2. This invention utilizes gas diffusion response data and stratum deformation response data after inert gas injection to identify suspected stratum disturbance areas within the shield tunneling impact area. It also optimizes subsequent analysis processes by combining stratum state information from different areas, thereby enabling early location of potential stratum disturbance areas and improving the reliability of abnormal stratum state identification.

[0018] 3. Based on the vibration response data and density change data of suspected ground disturbance areas, this invention further analyzes the abnormal ground disturbance areas and sends ground disturbance early warning information. It realizes a continuous processing process from ground condition detection and disturbance area identification to construction risk early warning. It can detect areas that may cause ground loosening, increased voids or local structural damage during subsequent tunnel boring machine excavation in advance, thus improving the safety of tunnel boring machine construction. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a method for identifying ground disturbances during shield tunnel construction based on muon detection, provided as an embodiment of the present invention;

[0020] Figure 2 A schematic diagram illustrating the acquisition of muon flux scanning images of the shield tunneling affected area according to an embodiment of the present invention;

[0021] Figure 3 A schematic diagram illustrating the distribution of an inert gas injection monitoring area provided in an embodiment of the present invention;

[0022] Figure 4 This is a flowchart illustrating a method for determining an inert gas injection test scheme, provided by an embodiment of the present invention.

[0023] Figure 5 A schematic flowchart illustrating an artificial vibration loading sequence for determining each suspected disturbance zone in the formation, provided as an embodiment of the present invention;

[0024] Figure 6 This is a schematic diagram illustrating the changes in the density, stability level, and sensitivity to ground vibration in multiple suspected areas of ground disturbance, as provided in an embodiment of the present invention.

[0025] Figure 7 This is a schematic diagram comparing a fixed vibration loading sequence and a random vibration loading sequence provided in an embodiment of the present invention;

[0026] Figure 8 A schematic diagram illustrating the probability distribution of excavation disturbance risk in a suspected geological disturbance area, provided as an embodiment of the present invention;

[0027] Figure 9 This is a schematic diagram of a ground disturbance identification system based on muon detection during shield tunnel construction, provided as an embodiment of the present invention. Detailed Implementation

[0028] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0029] In this embodiment of the invention, the following are provided: Figure 1 The method for identifying ground disturbances during shield tunnel construction based on muon detection is shown, and the method includes steps S1 to S8:

[0030] Step S1: Obtain the muon flux scan image of the area affected by the shield tunneling construction.

[0031] The muon flux scanning image of the shield tunneling impact area is generated by continuously detecting muons passing through the shield tunneling impact area using muon detection instruments deployed around the area, and the images are produced according to the scanning positions. In some embodiments, Figure 2 This is a schematic diagram illustrating the acquisition of muon flux scanning images of the affected area during shield tunneling construction, as provided in an embodiment of the present invention. (See attached diagram.) Figure 2 , attached Figure 2 The lower part of the shield tunnel is equipped with a muon detector array, and multiple upward-extending dotted arrows indicate the monitoring field of view and path of the detector array.

[0032] Mun flux scan images of the shield tunneling area can reveal the differences in the distribution density of muon particles after penetrating the strata. Each pixel in the mun flux scan image of the shield tunneling area represents the number of muons that penetrate along the corresponding stratum penetration path.

[0033] Step S2: Based on the muon flux scan image of the shield tunneling affected area, determine multiple inert gas injection monitoring areas.

[0034] In some embodiments, a monitoring region determination model can be used to determine multiple inert gas injection monitoring regions. The monitoring region determination model is a convolutional neural network. The input to the monitoring region determination model is a muon flux scan image of the shield tunneling impact area, and the output of the monitoring region determination model is the multiple inert gas injection monitoring regions.

[0035] A convolutional neural network (CNN) is a neural network model that uses a convolutional kernel to perform calculations within a local region of an image. A CNN consists of convolutional layers, activation layers, pooling layers, and fully connected layers. Convolutional layers can extract boundaries, textures, and intensity distributions in an image; pooling layers compress the calculation results; and fully connected layers can map the extracted results to region recognition results.

[0036] Multiple inert gas injection monitoring areas are the spatial ranges in the strata surrounding the shield tunnel that require inert gas injection to monitor the strata's tightness and disturbance state, as determined by a monitoring area determination model. In some embodiments, Figure 3 This is a schematic diagram of the distribution of an inert gas injection monitoring area provided in an embodiment of the present invention.

[0037] The muon flux scan images of the shield tunneling area reveal the differences in the number and density of muon particles after penetrating the strata. Since variations in stratum density lead to changes in the number of muons penetrating along the stratum's penetration path, and these variations are directly related to stratum tightness and disturbance conditions, the muon flux scan images of the shield tunneling area can clearly and quantitatively reflect spatial regions in the strata surrounding the tunnel that exhibit density anomalies or signs of loose structure. This allows the model to pinpoint the spatial area requiring inert gas injection.

[0038] Convolutional neural networks (CNNs) can continuously slide convolutional kernels of different sizes across muon flux scanning images of the shield tunneling area. Shallow convolutional operations of CNNs can extract differences in muon penetration quantity, numerical change boundaries, and local distribution patterns between adjacent pixels. Deeper convolutional layers can expand the analysis range, determine whether local anomalies persist in adjacent scanning positions, and identify interconnected anomaly distribution ranges and the degree of continuity within these ranges. Based on the learned correspondence between muon flux distribution patterns and the formation's tightness and disturbance state, CNNs can determine various spatial locations within the shield tunneling area. When a muon flux distribution significantly differs from the surrounding formations within a continuous range, and the anomaly boundary remains continuous across multiple adjacent locations, the CNN classifies the corresponding range as an area requiring inert gas injection testing. When scattered pixel changes cannot form stable boundaries, the CNN reduces the impact of these changes on area identification. The model can merge interconnected identification results into independent spatial ranges and determine multiple inert gas injection monitoring areas based on boundary positions in the image.

[0039] Step S3: Based on the muon flux scanning image of the shield tunneling affected area and the multiple inert gas injection monitoring areas, determine the inert gas injection test plan.

[0040] In some embodiments, Figure 4 This is a flowchart illustrating the process of determining an inert gas injection test scheme according to an embodiment of the present invention. The process of determining the inert gas injection test scheme includes steps S31 to S33:

[0041] Step S31: Based on the muon flux scan image of the shield tunneling affected area and the multiple inert gas injection monitoring areas, determine information on multiple candidate inert gas injection points.

[0042] In some embodiments, an injection point determination model can be used to determine information on multiple candidate inert gas injection points. The injection point determination model is a Transformer model. The inputs to the injection point determination model are the muon flux scan image of the shield tunneling impact area and the multiple inert gas injection monitoring areas; the output of the injection point determination model is information on multiple candidate inert gas injection points.

[0043] The Transformer model is a deep network architecture that abandons traditional recursive computation and relies entirely on a self-attention mechanism. The multi-head attention mechanism in the Transformer model can compute the association weights between any two elements within the input sequence in parallel, thereby accurately capturing long-distance contextual dependencies. Combined with feedforward computation units and residual connection design, the Transformer model can significantly improve the accuracy of global modeling of high-dimensional multimodal data.

[0044] The information on multiple candidate inert gas injection points is determined by the injection point determination model. It includes the attribute information of candidate points selected for inert gas injection tests within each inert gas injection monitoring area. The information on each candidate inert gas injection point may include the spatial location, estimated formation density, formation porosity, injection pressure requirement, and expected injection diffusion radius.

[0045] Muon flux scanning images of the shield tunneling impact area can reflect the differences in muon particle number distribution density at the underlying physical field level and the amount of formation penetration. The inert gas injection monitoring area delineates the three-dimensional spatial boundary range where key monitoring and injection operations need to be implemented. The muon flux scanning images of the shield tunneling impact area, together with multiple inert gas injection monitoring areas, define the spatial constraints and micro-density conditions for the test, enabling the model to calculate suitable candidate point information within the defined spatial boundaries based on the actual density and penetration characteristics of the formation medium.

[0046] The Transformer model divides the muon flux scan image of the shield tunneling impact area into continuous image regions according to the scan location, and incorporates the spatial boundaries corresponding to multiple inert gas injection monitoring areas into the attention calculation. The Transformer model uses a multi-head attention mechanism to compare the muon flux differences between any image region and its surrounding image regions, while also paying attention to the distribution relationships between distant locations within the same inert gas injection monitoring area. This allows local muon flux variations, regional uniformity, and regional edge variations to participate in the selection of candidate points. The model's feedforward calculation layer, based on the correspondence between muon flux distribution and estimated formation density and porosity, can infer selectable locations within the inert gas injection monitoring area on a location-by-location basis. Locations with lower estimated formation density and significant flux variations correspond to lower injection pressure requirements and can form a larger diffusion range; the Transformer model calculates the injection pressure requirement and expected injection diffusion radius accordingly. Conversely, locations with higher estimated formation density or significant regional edge variations need to consider the limitation of inert gas diffusion, and the Transformer model adjusts the injection pressure requirement and expected injection diffusion radius for these locations. Attention calculation also suppresses duplicate locations that are too close in space and have similar predicted conditions, and retains locations that can cover different formation conditions within each inert gas injection monitoring area, thereby identifying multiple alternative inert gas injection point information.

[0047] Step S32: Cluster the multiple candidate inert gas injection points to obtain multiple candidate injection point clusters.

[0048] In some embodiments, the K-means clustering algorithm can be used to cluster the multiple candidate inert gas injection points to obtain multiple candidate injection point clusters. The K value in the K-means clustering algorithm can be preset manually.

[0049] K-means clustering is a clustering algorithm that groups samples based on their numerical distance. It divides samples into a predetermined number of clusters, repeatedly calculates the centroid of each cluster and adjusts sample assignments until the distance between samples within a cluster is small and the distance between samples in different clusters is large. K-means clustering maintains a high degree of internal similarity among data samples grouped within the same cluster.

[0050] Multiple candidate injection point clusters are groups of point information formed by clustering multiple candidate inert gas injection point information according to the similarity between estimated formation density, formation porosity, expected injection diffusion radius, and injection pressure requirement.

[0051] As an example, specifically: The K-means clustering algorithm first sets a predetermined number of K initial cluster centers, each corresponding to a set of estimated formation density, formation porosity, expected injection diffusion radius, and injection pressure requirements. The algorithm calculates the numerical difference between each candidate inert gas injection point and each initial cluster center, and assigns each candidate inert gas injection point to the group corresponding to the initial cluster center with the smallest difference. After one assignment, the K-means clustering algorithm recalculates the cluster centers based on all candidate inert gas injection point information within each group, making the updated cluster centers close to the common level of estimated formation density, formation porosity, expected injection diffusion radius, and injection pressure requirements within the group. The candidate inert gas injection point information is then regrouped according to the updated cluster centers. The cluster center calculation and point assignment are repeated until the grouping results of the candidate inert gas injection point information remain stable. Candidate inert gas injection points with similar estimated formation density and porosity, and relatively small differences in expected injection diffusion radius and injection pressure requirements, will stably be grouped into the same candidate injection point cluster. When any indicator shows a significant difference and the overall distance is closer to the remaining cluster center, the corresponding candidate inert gas injection point information will be transferred to different groups, thus obtaining multiple candidate injection point clusters with similar intra-cluster states and different inter-cluster states.

[0052] Step S33: Determine the inert gas injection test scheme based on the multiple candidate injection point clusters.

[0053] In some embodiments, a test scheme determination model can be used to determine the inert gas injection test scheme. The test scheme determination model is a deep neural network. The input to the test scheme determination model is the plurality of candidate injection point clusters, and the output of the test scheme determination model is the inert gas injection test scheme.

[0054] Deep neural networks are biomimetic multilayer perceptron networks containing numerous hidden layers. They rely on stacked neurons and nonlinear activation functions to perform in-depth transformations and mappings of complex input signals. Deep neural networks optimize their internal connection weights through backpropagation training on massive amounts of data, demonstrating extremely high fitting accuracy in pattern discrimination tasks of complex nonlinear systems.

[0055] An inert gas injection test plan is an operational procedure for conducting inert gas injection tests within the area affected by tunnel boring machine (TBM) construction. The inert gas injection test plan may include the location of the inert gas injection point for each test, the inert gas injection volume, the inert gas injection pressure, the inert gas injection duration, the execution sequence of each test, the interval between adjacent tests, and the duration of gas diffusion monitoring for each test.

[0056] Multiple candidate injection point clusters reflect the distribution of candidate sites with similar geological properties and injection pressure requirements. Meanwhile, the differences between different candidate injection point clusters demonstrate the heterogeneity and spatial differentiation of geological conditions within the area affected by tunnel boring machine (TBM) construction. Data within the candidate injection point clusters enable the model to calculate specific test parameters and execution strategies based on the specific characteristics of different groupings to ensure the coverage of the test plan.

[0057] The deep neural network analyzes the concentration and variation range of candidate inert gas injection point information within each candidate injection point cluster through hidden layers, and determines which candidate inert gas injection points within the same cluster can represent the formation state of the group. Multiple candidate inert gas injection points with similar estimated formation density, formation porosity, expected injection diffusion radius, and injection pressure requirements do not need to undergo repeated testing. The deep neural network retains the locations of inert gas injection points that cover the spatial range of the group and are close to the group's common level. The hidden layers can convert the injection pressure requirements and expected injection diffusion radius corresponding to the retained locations into inert gas injection volume, inert gas injection pressure, inert gas injection duration, and gas diffusion monitoring duration. When the expected injection diffusion radius within a candidate injection point cluster is generally small and the injection pressure requirements are generally high, a longer diffusion observation process is required. Consequently, the model will correspondingly extend the gas diffusion monitoring duration and avoid significant overlap with adjacent tests in terms of diffusion range. When the expected injection diffusion radius within a candidate injection point cluster is generally large, sufficient intervals need to be maintained after the initial test. The deep neural network arranges the interval time between adjacent tests accordingly. The output layer determines the execution order of each test according to the spatial relationship between the locations of each inert gas injection point and the influence of gas diffusion, and finally obtains the inert gas injection test plan.

[0058] Step S4: Execute the inert gas injection test plan and obtain gas diffusion response data and ground deformation response data in the shield tunneling affected area.

[0059] The gas diffusion response data in the shield tunneling impact area is obtained by collecting gas diffusion measurement data through gas concentration sensors deployed within the shield tunneling impact area during the execution of the inert gas injection test plan. The gas diffusion response data in the shield tunneling impact area may include the location of each gas concentration sensor, acquisition time, inert gas concentration sequence, first arrival time of inert gas, peak inert gas concentration, and inert gas concentration decay rate.

[0060] Ground deformation response data are measurements of ground deformation collected during the inert gas injection test program using displacement sensors, settlement sensors, and strain sensors deployed within the shield tunneling area. The data may include the location coordinates of each sensor, acquisition time, ground displacement, surface settlement, ground strain, and ground deformation rate.

[0061] Step S5: Based on the gas diffusion response data and stratum deformation response data of the shield tunneling affected area, identify multiple suspected stratum disturbance areas.

[0062] In some embodiments, a disturbance-prone area determination model can be used to determine multiple suspected ground disturbance areas. The disturbance-prone area determination model is a deep neural network. The inputs to the disturbance-prone area determination model are gas diffusion response data and ground deformation response data of the shield tunneling impact area, and the output of the disturbance-prone area determination model is multiple suspected ground disturbance areas.

[0063] Multiple suspected areas of ground disturbance are several spatial areas within the shield tunneling area where the soil structure may change due to the disturbance caused by the shield tunneling construction.

[0064] Gas diffusion response data in the shield tunneling impact area reflects the arrival, accumulation, and attenuation of inert gases at different strata locations, indicating whether there are any anomalies in the gas permeability and sealing conditions within the strata. Strata deformation response data can represent the changes in strata displacement, surface subsidence, strata strain, and strata deformation rate during inert gas injection, quantifying whether the strata structure is prone to morphological changes under external forces.

[0065] The first hidden layer of the deep neural network can extract features from data segments where the inert gas arrives too early or the peak concentration of inert gas is abnormally high, assessing the spatial distribution of hidden channels with abnormal internal connectivity within the strata. Another parallel hidden layer specifically performs nonlinear activation on data regions with surges in stratum displacement and accelerated stratum deformation rates to pinpoint the coordinate ranges of intense stratum stress release. Subsequently, the deep network structure of the model uses a feature fusion layer to calculate the gradient characteristics of the product of gas concentration decay rate and surface subsidence in the same coordinate system. The deep neural network can then identify dangerous nodes where gas diffusion rate and stratum deformation are strongly coupled in the high-dimensional latent space. The network's internal connection weights, after forward propagation, map the strong coupling features of gas diffusion and stratum deformation into disturbance risk probability scores. Finally, the output layer of the deep neural network applies a set threshold function to extract spatial grid cells with scores exceeding the safety limit, accurately defining multiple stratum spatial regions where the stratum soil structure may change due to shield tunneling disturbances, thus outputting multiple suspected stratum disturbance areas.

[0066] Step S6: Based on the multiple suspected formation disturbance areas, determine the artificial vibration loading sequence for each suspected formation disturbance area.

[0067] In some embodiments, Figure 5 This is a schematic flowchart illustrating an artificial vibration loading sequence for determining each suspected formation disturbance area according to an embodiment of the present invention. The artificial vibration loading sequence for determining each suspected formation disturbance area includes steps S61 to S64:

[0068] Step S61: Based on the gas diffusion response data and formation deformation response data of each suspected formation disturbance area, determine the formation compaction degree, formation structure stability level, and formation vibration sensitivity of each suspected formation disturbance area.

[0069] The gas diffusion response data for each suspected ground disturbance area is the gas diffusion response data corresponding to the spatial range of each suspected ground disturbance area in the gas diffusion response data of the shield tunneling affected area.

[0070] The ground deformation response data for each suspected ground disturbance area is the ground deformation response data corresponding to the spatial range of each suspected ground disturbance area in the ground deformation response data of the shield tunneling affected area.

[0071] In some embodiments, a formation state determination model can be used to determine the formation compaction, formation stability level, and formation vibration sensitivity of each suspected formation disturbance area. The formation state determination model is a deep neural network. The inputs to the formation state determination model are gas diffusion response data and formation deformation response data for each suspected formation disturbance area, and the outputs are the formation compaction, formation stability level, and formation vibration sensitivity of each suspected formation disturbance area.

[0072] The stratigraphic compaction degree of a suspected area of ​​stratigraphic disturbance is a numerical indicator representing the degree of density of stratigraphic particles within each suspected area of ​​stratigraphic disturbance. The higher the stratigraphic compaction degree value, the smaller the gaps between stratigraphic particles within the corresponding suspected area of ​​stratigraphic disturbance.

[0073] The stratigraphic stability grade of a suspected area of ​​stratigraphic disturbance indicates the level of its ability to maintain its original spatial morphology when subjected to external disturbances. A higher stratigraphic stability grade indicates a lower probability of displacement and subsidence in the corresponding suspected area.

[0074] The ground vibration sensitivity of suspected ground disturbance areas is a numerical value representing the ease with which each suspected ground disturbance area will produce a vibration response and density change after being subjected to vibration excitation. The higher the ground vibration sensitivity value, the easier it is for the same vibration excitation to produce a larger vibration response and density change in the corresponding suspected ground disturbance area.

[0075] In some embodiments, Figure 6 This is a schematic diagram illustrating the changes in the density, structural stability, and vibration sensitivity of multiple suspected geological disturbance areas, as provided in an embodiment of the present invention.

[0076] Gas diffusion response data from suspected disturbance areas reveal the variation patterns of inert gases as they propagate within the suspected disturbance area, influenced by the formation's spatial structure. This data can provide a reference for analyzing the porosity and particle arrangement within the formation. Formation deformation response data from suspected disturbance areas demonstrate the overall response characteristics of the formation after inert gas injection, thus reflecting the formation's stability against external forces and its sensitivity to morphological changes.

[0077] A deep neural network converts gas diffusion response data and formation deformation response data for each suspected disturbance region into corresponding numerical vectors. Then, it uses deep hidden units to perform in-depth cross-validation calculations on the variance of gas concentration sequence fluctuations and formation displacement. The model calculates the average pore size and tortuosity parameters of soil pore channels by analyzing the nonlinear combination relationship between the first arrival time of inert gas and the inert gas concentration decay rate, thereby outputting the formation compaction degree, representing the density of formation particles within each suspected disturbance region. The deep neural network continues to perform time-dimensional partial derivative calculations on surface subsidence, formation strain, and formation deformation rate in the hidden layers to assess the rheological characteristics of local formations under sustained stress. The model relies on deformation evolution pattern knowledge accumulated through backpropagation to map the calculated rheological characteristics into classification labels, thus outputting the formation structural stability level, representing the ability of each suspected disturbance region to maintain its original spatial morphology under external disturbance. Deep neural networks can further integrate the computational features of hidden layers to compare the synchronous rates of gas diffusion and formation deformation, and deduce the threshold marginal energy parameter for critical instability of the soil structure. After linear layer normalization of the model output layer, the threshold marginal energy parameter is finally transformed into the formation vibration sensitivity, representing the ease with which each suspected disturbance area of ​​the formation will generate a vibration response and density change after being subjected to vibration excitation.

[0078] Step S62: Based on the density of the strata, the stability level of the strata structure, and the sensitivity of the strata to vibration of each suspected strata disturbance area, determine multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity of each suspected strata disturbance area.

[0079] In some embodiments, a vibration parameter determination model can be used to determine multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity for each suspected formation disturbance area. The vibration parameter determination model is a deep neural network. The inputs to the vibration parameter determination model are the formation compaction degree, formation structural stability level, and formation vibration sensitivity of each suspected formation disturbance area. The outputs of the vibration parameter determination model are multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity for each suspected formation disturbance area.

[0080] The multiple fixed vibration loading frequencies for suspected formation disturbance areas are several constant vibration frequency values ​​set for each suspected formation disturbance area. Each fixed vibration loading frequency is used to continuously apply vibration excitation to each suspected formation disturbance area at the corresponding frequency for a period of time.

[0081] The random vibration loading frequency range for suspected formation disturbance areas is a range of random vibration frequency values ​​set for each suspected formation disturbance area. The random vibration loading frequency range for each suspected formation disturbance area includes a lower frequency limit and a higher frequency limit.

[0082] Each vibration frequency used during random vibration loading is located between the lower and upper frequency limits.

[0083] The maximum vibration intensity in a suspected area of ​​formation disturbance is an upper limit for vibration intensity defined for each suspected area of ​​formation disturbance.

[0084] The vibration intensity during both the fixed and random vibration loading processes in each suspected area of ​​formation disturbance does not exceed the corresponding maximum vibration intensity.

[0085] The density of the stratigraphic unit in a suspected disturbance area indicates the compactness of the stratigraphic particles, which limits the range of vibration energy transmission within the stratigraphy. The stratigraphic structural stability level reflects the ability of the stratigraphy to maintain its original spatial morphology and determines the safety margin required for vibration loading. The stratigraphic vibration sensitivity indicates the ease with which the stratigraphy responds and undergoes density changes under vibration excitation. The density, structural stability level, and vibration sensitivity collectively characterize the vibration bearing capacity and resonance-induced tendency of each suspected disturbance area, enabling the model to specifically deduce safe vibration parameter boundaries that can generate effective test signals without causing substantial damage.

[0086] Deep neural networks can establish parameter correlations based on the soil compaction, soil structural stability level, and soil vibration sensitivity of each suspected disturbance area. The hidden layers of the deep neural network establish a nonlinear mapping relationship between soil compaction and wave propagation velocity, identifying the distribution spectrum of the soil's potential natural frequencies through nonlinear neuron nodes. Based on this, the deep neural network calculates multiple constant frequencies that can excite an effective response within the soil without causing structural damage, thus outputting multiple fixed vibration loading frequencies. The model further reads the soil vibration sensitivity parameters and uses a set penalty function to determine a safe frequency sweep interval around the potential natural frequency spectrum to avoid extremely sensitive areas entering a runaway resonance state. The frequency sweep interval, defined by the output layer's boundary settings, forms the random vibration loading frequency range, along with its lower and upper frequency limits, for each suspected disturbance area. Finally, the deep neural network converts the soil structural stability level into an energy tolerance coefficient and performs nonlinear calculations based on the soil vibration sensitivity, thereby deriving the peak excitation force limit corresponding to the soil before significant structural changes occur. The model transformed the peak excitation force limit and finally confirmed it as the upper limit value of vibration intensity for each suspected area of ​​formation disturbance, i.e., the maximum vibration intensity.

[0087] Step S63: Construct multiple vibration loading maps, each corresponding to a suspected area of ​​formation disturbance. Each vibration loading map includes multiple nodes and multiple edges. The multiple nodes include multiple fixed vibration frequency nodes and one random vibration frequency node. Each fixed vibration frequency node is connected to the random vibration frequency node to establish an edge. The node characteristics of the fixed vibration frequency node are fixed vibration loading frequency and maximum vibration intensity. The node characteristics of the random vibration frequency node are random vibration loading frequency range and maximum vibration intensity. The edges represent the formation compaction degree, formation structure stability level, and formation vibration sensitivity of the corresponding suspected area of ​​formation disturbance.

[0088] Multiple vibration loading maps are map data constructed for multiple suspected areas of strata disturbance. Each vibration loading map is used to represent the connection relationship between the fixed vibration loading frequency, random vibration loading frequency range, maximum vibration intensity of the corresponding suspected area of ​​strata disturbance and the strata compaction degree, strata structure stability level, and strata vibration sensitivity of the corresponding suspected area of ​​strata disturbance.

[0089] Multiple fixed vibration frequency nodes are nodes in the vibration loading map that correspond to multiple fixed vibration loading frequencies. Each fixed vibration frequency node records a fixed vibration loading frequency and the maximum vibration intensity of the corresponding suspected ground disturbance area.

[0090] A random vibration frequency node is a node in the vibration loading map corresponding to a random vibration loading process. The random vibration frequency node records the random vibration loading frequency range and maximum vibration intensity of the suspected area of ​​formation disturbance.

[0091] Each fixed vibration frequency node is connected to a random vibration frequency node to establish an edge. The edge records the stratigraphic compaction, stratigraphic structural stability level, and stratigraphic vibration sensitivity of the suspected area of ​​stratigraphic disturbance.

[0092] Step S64: Based on the vibration sequence determination model, process the multiple vibration loading maps to obtain the fixed vibration loading sequence and random vibration loading sequence for each suspected area of ​​formation disturbance.

[0093] In some embodiments, the vibration sequence determination model is a graph neural network. The input to the vibration sequence determination model is the plurality of vibration loading maps, and the output of the vibration sequence determination model is a fixed vibration loading sequence and a random vibration loading sequence for each suspected area of ​​formation disturbance.

[0094] Graph Neural Networks (GNNs) are learning models capable of processing non-Euclidean topological graph data. GNNs update the hidden state of each node by performing information aggregation and message passing mechanisms along the edges connecting nodes. GNNs can deeply fuse the global topological shape of the graph with local node features, thereby accurately inferring the inherent association attributes and optimal node order within the graph structure.

[0095] The fixed vibration loading sequence for each suspected area of ​​formation disturbance consists of multiple fixed vibration loading segments arranged in chronological order. Each fixed vibration loading segment includes the fixed vibration loading frequency, vibration intensity, vibration duration, start time, and end time.

[0096] The random vibration loading sequence for each suspected area of ​​formation disturbance consists of multiple random vibration loading segments arranged in the order of execution. Each random vibration loading segment includes a random vibration frequency change sequence, vibration intensity, vibration duration, start time, and end time. The frequency values ​​in the random vibration frequency change sequence are within the corresponding random vibration loading frequency range.

[0097] In some embodiments, Figure 7 This is a schematic diagram comparing a fixed vibration loading sequence and a random vibration loading sequence, provided as an embodiment of the present invention.

[0098] The artificial vibration loading sequence for each suspected formation disturbance area consists of a fixed vibration loading sequence and a random vibration loading sequence for each suspected formation disturbance area, arranged according to execution time. It is used to control the vibration loading device to apply vibration excitation to the corresponding suspected formation disturbance area.

[0099] The vibration loading map represents multiple fixed vibration frequency nodes and one random vibration frequency node in graphical form, along with the connecting edges between the nodes. This allows for the organization of the spatial correlation between various vibration loading parameters and the formation's physical state. The fixed vibration frequency nodes are characterized by a fixed vibration loading frequency and a maximum vibration intensity, reflecting the numerical value and upper limit of the applied vibration excitation at the corresponding constant frequency. The random vibration frequency nodes are characterized by a random vibration loading frequency range and a maximum vibration intensity, reflecting the value range and upper limit of the intensity during frequency sweep vibration. The edges between the nodes represent the formation compaction, formation structural stability level, and formation vibration sensitivity of the suspected disturbance area, reflecting the microscopic compactness, macroscopic stability, and ease of response to vibration under excitation. By using fixed and random vibration parameters as node features and formation state indicators as edges, the vibration loading map transforms the originally dispersed excitation parameters and formation physical properties into structured graphical data that can be processed by graph neural networks. Graph neural networks can thus analyze the response transmission patterns of different vibration frequencies under specific formation density and sensitivity conditions along the edges between nodes, avoiding the isolated evaluation of the excitation effect of a single frequency. Vibration loading maps allow for the effective organization of excitation parameters and formation conditions according to topological relationships, facilitating the graph neural network's capture of energy accumulation and formation adaptation characteristics during frequency switching. By comprehensively analyzing node characteristics and the edges between nodes, the graph neural network can identify multiple fixed and random vibration loading segments that are most effective, suitable for the formation condition, and meet strength constraints, thereby generating fixed and random vibration loading sequences for each suspected formation disturbance area.

[0100] The graph neural network (Graph Neural Network) executes information aggregation and message passing mechanisms along the connections between nodes in multiple vibration loading maps. In each message passing iteration, the Graph Neural Network reads the ground compaction, ground structural stability level, and ground vibration sensitivity from the edges, and calculates the association weights between nodes based on the feature differences between adjacent nodes. The Graph Neural Network updates the hidden layer feature states of each fixed vibration frequency node and random vibration frequency node according to the association weights, enabling each vibration frequency node to integrate the structural state information of the corresponding suspected ground disturbance area. The Graph Neural Network further integrates the overall topological relationship of the vibration loading maps with the local features of the nodes, analyzes the ability of different vibration frequencies and intensities to generate effective detection responses under different ground conditions, and evaluates the impact of different vibration loading combinations on the stability of the ground structure. The attention ranking layer inside the Graph Neural Network determines the combination of vibration frequencies that can obtain significant response changes without exceeding the ground's bearing capacity based on the updated hidden layer features, and generates an adapted execution order for multiple fixed vibration loading segments. The graph neural network further combines the maximum vibration intensity corresponding to each fixed vibration frequency node to constrain and adjust the vibration intensity of each fixed vibration loading segment, and determines the corresponding vibration duration, start time, and end time, thereby generating a fixed vibration loading sequence for each suspected area of ​​formation disturbance. The graph neural network then searches for dynamic frequency change processes within the random vibration loading frequency range that can supplement the fixed frequency detection effect, based on the hidden layer characteristics of the random vibration frequency nodes. It adjusts the range and duration of vibration intensity changes during the random vibration process according to the formation vibration sensitivity, ultimately generating multiple random vibration loading segments arranged in the order of execution, and outputting a random vibration loading sequence for each suspected area of ​​formation disturbance.

[0101] Step S7: Apply vibration excitation to each suspected formation disturbance area based on the artificial vibration loading sequence of each suspected formation disturbance area, and obtain vibration response data and density change data of each suspected formation disturbance area.

[0102] Vibration response data for each suspected formation disturbance area are obtained by vibration sensors deployed within each suspected disturbance area during the period when a hydraulic vibration excitation device, controlled according to the artificial vibration loading sequence of each suspected formation disturbance area, applies vibration excitation to that area. The vibration response data for each suspected formation disturbance area may include the vibration sensor's location coordinates, acquisition time, vibration acceleration, vibration velocity, vibration displacement amplitude, principal vibration frequency, and vibration response duration.

[0103] The density change data for each suspected area of ​​formation disturbance is obtained by scanning and converting the formation density data of each suspected area of ​​formation disturbance using a muon detector before and after vibration excitation. The density change data for each suspected area of ​​formation disturbance may include the coordinates of the scanning location, the formation density value before vibration excitation, the formation density value after vibration excitation, the amount of formation density change, and the rate of formation density change.

[0104] Step S8: Based on the vibration response data and density change data of each suspected formation disturbance area, determine the abnormal formation disturbance area and send formation disturbance early warning information.

[0105] In some embodiments, an abnormal disturbance region determination model can be used to determine abnormal formation disturbance regions. The abnormal disturbance region determination model is a deep neural network. The input to the abnormal disturbance region determination model is the vibration response data and density change data of each suspected formation disturbance region, and the output of the abnormal disturbance region determination model is the abnormal formation disturbance region.

[0106] An abnormal geological disturbance area is a geological space within the shield tunneling construction area that has a high risk of loosening of the strata, increased voids, or local structural damage during the subsequent excavation process by the shield machine.

[0107] Vibration response data in suspected ground disturbance areas reflect the vibration propagation and energy absorption characteristics of soil structure under excited state and the degree of ground response to external vibration. Density change data in suspected ground disturbance areas reflect the changing trend of ground compaction after vibration. Based on this, the model can deduce the structural deterioration tendency and instability evolution law of ground soil under subsequent shield excavation disturbance, thus providing quantitative data basis for early identification of spatial areas that may be at risk of ground loosening, increased voids or local structural damage during subsequent shield excavation.

[0108] A deep neural network jointly analyzes the vibration response data and density change data of each suspected disturbance area. Through multiple layers of neurons and nonlinear activation functions, the deep neural network can perform dimensionality reduction and nonlinear mapping on the vibration propagation process, response decay process, and response change characteristics at different loading stages in the vibration response data. It also establishes a correspondence between vibration response changes and formation structure changes, thereby calculating the stability of the formation response after external vibration. The deep neural network can combine density change data and use network weights to perform weighted analysis of density change characteristics. It can also extract the amplitude of changes in formation compaction before and after vibration, thus assessing the disturbance impact of vibration loading on the density of formation particles. Subsequently, the deep neural network uses feature fusion to jointly calculate the response stability and density change characteristics, analyzing the structural change trends of different suspected disturbance areas under artificial vibration. Based on the correlation between continuous changes in vibration response, abnormal response decay, and density reduction trends, it quantifies the stability state change indicators of the formation after external disturbance. Based on the stability state change indicators of each region and combined with the resulting risk extrapolation capabilities, the model can estimate the probability of ground loosening, increased voids, or local structural damage in different regions under the disturbance of subsequent tunnel boring machine excavation. The model then maps these probability values ​​to a three-dimensional spatial grid using corresponding spatial coordinates. Finally, the output layer determines the risk area range based on the distribution of risk probability values, calibrates the corresponding spatial region, and outputs it as an area of ​​abnormal ground disturbance.

[0109] In some embodiments, determining the abnormal formation disturbance region based on the vibration response data and density change data of each suspected formation disturbance region includes steps S81 to S83:

[0110] Step S81: Based on the vibration response data and density change data of each suspected formation disturbance area, determine the vibration duration range, vibration response attenuation coefficient, and density disturbance change amplitude of each suspected formation disturbance area.

[0111] In some embodiments, deep neural networks can be used to determine the duration of vibration impact, vibration response attenuation coefficient, and density disturbance change amplitude of each suspected formation disturbance area.

[0112] The duration of vibration impact in each suspected area of ​​formation disturbance represents the spatial range within which vibration loading can continuously exert an effect in each suspected area of ​​formation disturbance.

[0113] The vibration response attenuation coefficient for each suspected area of ​​formation disturbance is a quantitative indicator representing the degree to which the vibration response of the corresponding suspected area of ​​formation disturbance weakens with increasing propagation distance under artificial vibration. The larger the vibration response attenuation coefficient, the faster the vibration response weakens within the suspected area of ​​formation disturbance, and the weaker the impact of external vibration on areas far from the excitation location.

[0114] The magnitude of density disturbance change in each suspected area of ​​formation disturbance is a quantitative indicator representing the extent of change in formation density before and after vibration in the corresponding suspected area of ​​formation disturbance. The larger the magnitude of the density disturbance change, the more significant the change in formation compaction after vibration in the corresponding suspected area of ​​formation disturbance.

[0115] A deep neural network extracts the vibration propagation velocity and acceleration attenuation rate from the vibration response data of each suspected ground disturbance area through multiple layers of neurons, calculating the dissipation gradient of vibration energy in the soil medium. The deep neural network uses a nonlinear activation function to spatially map the dissipation gradient, deducing the boundary distance at which the excitation force can maintain an effective response within the area, and identifying the attenuation boundary of effective energy transfer within this boundary distance. This accurately calibrates and outputs the sustained influence range of vibration in each suspected ground disturbance area. The deep network structure of the model further analyzes the vibration amplitude variation process, vibration response duration, and response attenuation at different stages in the vibration response data of each suspected ground disturbance area. A cross-layer attention mechanism is used to fuse and extrapolate the vibration response changes at different stages, establishing a nonlinear mapping relationship between vibration response intensity and propagation time. Based on the nonlinear mapping results, the model calculates the degree of vibration response attenuation with propagation distance and action time, maps the calculation results to a continuous numerical range, and outputs the vibration response attenuation coefficient for each suspected ground disturbance area. The deep neural network further utilizes a dynamic weight matrix to analyze density change data in each suspected area of ​​formation disturbance. It jointly calculates the amplitude, rate, and continuous change process of formation density before and after vibration excitation, thereby establishing a nonlinear correlation between vibration intensity and changes in formation compaction. The model can perform feature fusion on the density change process, quantify the magnitude of formation density change caused by vibration, and continuously output the density disturbance amplitude of each suspected area of ​​formation disturbance through a regression prediction layer.

[0116] Step S82: Based on the vibration response data, density change data, vibration duration range, vibration response attenuation coefficient, and density disturbance change amplitude of each suspected stratum disturbance area, determine the excavation disturbance risk probability distribution map and the expected stratum instability impact depth of each suspected stratum disturbance area.

[0117] In some embodiments, deep neural networks can be used to determine the probability distribution map of excavation disturbance risk in each suspected area of ​​ground disturbance and the expected depth of ground instability impact from excavation in each suspected area of ​​ground disturbance.

[0118] The excavation disturbance risk probability distribution map for each suspected ground disturbance area is an image showing the level of disturbance risk at various locations within that area during tunnel boring machine (TBM) excavation. The excavation disturbance risk probability distribution map can be used to represent the probability of ground loosening, increased voids, and localized structural damage occurring at different spatial locations within each suspected ground disturbance area during subsequent TBM excavation. In some embodiments, Figure 8 This is a schematic diagram illustrating the probability distribution of excavation disturbance risk in a suspected geological disturbance area, provided as an embodiment of the present invention.

[0119] The expected depth of ground instability impact during excavation of each suspected ground disturbance area is the vertical spatial extension distance that indicates structural damage to the ground in that area under the disturbance of subsequent tunnel boring machine construction.

[0120] The hidden layers of the deep neural network project high-dimensional spatial features onto the vibration duration of each suspected disturbance area, the vibration response attenuation coefficient of each suspected disturbance area, and the density perturbation variation amplitude of each suspected disturbance area. The model converts the vibration response attenuation coefficient of each suspected disturbance area into a nonlinear attenuation weight characterizing the vibration transmission capacity, and converts the density perturbation variation amplitude of each suspected disturbance area into a dynamic influence factor characterizing the degree of change in the soil compaction state. The deep neural network uses a cross-layer multiplier to perform element-wise matrix multiplication of the attenuation weights and influence factors to calculate the probability of local soil stress imbalance in that area under the cutting thrust of a tunnel boring machine. Using this probability as the core benchmark, the model performs high-density tensor interpolation calculations within the three-dimensional spatial coordinate topology defined by the vibration duration of each suspected disturbance area. The model assigns a corresponding predicted disturbance risk value to each 3D grid node within the defined area based on the interpolation results. Then, a dimensionality reduction projection mechanism is applied to map the assigned 3D grid nodes into a 2D slice image structure. Pixel chromaticity differences are then used to calibrate the risk level, resulting in an output disturbance risk probability distribution map for each suspected ground disturbance area. An additional parallel, fully connected deep neural network branch is specifically designed to analyze the variation of local stress imbalance probability at different underground depths. The model can combine the depth-direction response changes in vibration response data and the depth-direction compaction changes in density change data for each suspected ground disturbance area to determine the influence boundary location of the local stress imbalance probability as a function of underground depth. Finally, the model can calculate the vertical depth distance between this influence boundary location and the planned excavation top surface of the tunnel boring machine, and extrapolate the expected ground instability influence depth for each suspected ground disturbance area.

[0121] Step S83: Based on the probability distribution map of excavation disturbance risk of each suspected stratum disturbance area and the expected depth of stratum instability impact of excavation in each suspected stratum disturbance area, determine the abnormal stratum disturbance area.

[0122] In some embodiments, deep neural networks can be used to determine regions of anomalous formation disturbance.

[0123] The deep neural network analyzes the excavation disturbance risk probability distribution map of each suspected geological disturbance area according to its spatial location, and maps the expected geological instability impact depth of each suspected geological disturbance area to the same spatial location. The model can analyze whether the probability of higher excavation disturbance risk remains continuous between adjacent spatial locations, and determine whether the expected geological instability impact depth at the corresponding location indicates that the instability effect has been able to extend to deeper strata. When a few isolated locations show a high probability of excavation disturbance risk, but the probability of excavation disturbance risk in the surrounding spatial locations is significantly lower, and the corresponding expected geological instability impact depth is small, the deep neural network will reduce the possibility of these isolated locations being classified as abnormal geological disturbance areas. When a high probability of excavation disturbance risk appears continuously in adjacent spatial locations, and the expected geological instability impact depths corresponding to these spatial locations all reach large values, it indicates that subsequent tunnel boring machine excavation disturbances are not only likely to cause geological changes in local locations, but may also have a lasting impact on surrounding locations and deeper strata. Deep neural networks determine the spatial locations to be included based on the continuous spatial distribution of the excavation disturbance risk probability and the expected depth of the impact of strata instability during excavation. They then connect spatially continuous strata locations with high risks of loose strata, increased voids, and local structural damage to form a complete region, thereby identifying the abnormal strata disturbance area.

[0124] Based on the same inventive concept Figure 9 This is a schematic diagram of a ground disturbance identification system based on muon detection during shield tunnel construction, provided by an embodiment of the present invention. The ground disturbance identification system based on muon detection during shield tunnel construction includes:

[0125] Image acquisition module 91 is used to acquire muon flux scan images of the area affected by shield tunneling construction;

[0126] The monitoring area determination module 92 is used to determine multiple inert gas injection monitoring areas based on the muon flux scanning image of the shield tunneling impact area;

[0127] The scheme determination module 93 is used to determine the inert gas injection test scheme based on the muon flux scanning image of the shield tunneling impact area and the multiple inert gas injection monitoring areas.

[0128] The response data acquisition module 94 is used to execute the inert gas injection test plan and acquire gas diffusion response data and ground deformation response data in the shield tunneling construction affected area.

[0129] The suspected area determination module 95 is used to determine multiple suspected areas of ground disturbance based on the gas diffusion response data and ground deformation response data of the shield tunneling construction affected area.

[0130] The sequence determination module 96 is used to determine the artificial vibration loading sequence for each suspected formation disturbance region based on the multiple suspected formation disturbance regions;

[0131] The excitation acquisition module 97 is used to apply vibration excitation to each suspected formation disturbance area based on the artificial vibration loading sequence of each suspected formation disturbance area, and to acquire vibration response data and density change data of each suspected formation disturbance area.

[0132] The anomaly warning module 98 is used to determine the abnormal stratum disturbance area based on the vibration response data and density change data of each suspected stratum disturbance area, and to send stratum disturbance warning information.

[0133] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0134] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and are considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for identifying ground disturbance during shield tunnel construction based on muon detection, characterized in that, include: Obtain muon flux scan images of the area affected by shield tunneling construction; Multiple inert gas injection monitoring areas were identified based on the muon flux scanning images of the shield tunneling impact area. Based on the muon flux scanning image of the shield tunneling area and the multiple inert gas injection monitoring areas, an inert gas injection test plan is determined. The inert gas injection test plan was executed, and gas diffusion response data and ground deformation response data of the shield tunneling affected area were obtained. Based on the gas diffusion response data and stratum deformation response data of the shield tunneling affected area, several suspected stratum disturbance areas were identified. Based on the multiple suspected formation disturbance areas, an artificial vibration loading sequence is determined for each suspected formation disturbance area; Based on the artificial vibration loading sequence of each suspected formation disturbance area, vibration excitation is applied to each suspected formation disturbance area, and vibration response data and density change data of each suspected formation disturbance area are obtained; Based on the vibration response data and density change data of each suspected formation disturbance area, abnormal formation disturbance areas are identified, and formation disturbance early warning information is sent.

2. The method for identifying ground disturbance during shield tunnel construction based on muon detection as described in claim 1, characterized in that, The determination of the inert gas injection test scheme based on the muon flux scanning image of the shield tunneling impact area and the multiple inert gas injection monitoring areas includes: Based on the muon flux scanning image of the shield tunneling impact area and the multiple inert gas injection monitoring areas, information on multiple candidate inert gas injection points is determined. Multiple candidate injection point clusters are obtained by clustering based on the information of the multiple candidate inert gas injection points; An inert gas injection test scheme is determined based on the multiple candidate injection point clusters.

3. The method for identifying ground disturbance during shield tunnel construction based on muon detection as described in claim 1, characterized in that, The artificial vibration loading sequence for determining each suspected formation disturbance region based on the multiple suspected formation disturbance regions includes: Based on the gas diffusion response data and formation deformation response data of each suspected formation disturbance area, the formation compaction degree, formation structure stability level, and formation vibration sensitivity of each suspected formation disturbance area are determined. Based on the stratigraphic compaction, stratigraphic structural stability level, and stratigraphic vibration sensitivity of each suspected stratigraphic disturbance area, multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity are determined for each suspected stratigraphic disturbance area. Multiple vibration loading maps are constructed, each corresponding to a suspected area of ​​formation disturbance. Each vibration loading map includes multiple nodes and multiple edges. The multiple nodes include multiple fixed vibration frequency nodes and one random vibration frequency node. Each fixed vibration frequency node is connected to the random vibration frequency node to establish an edge. The node characteristics of the fixed vibration frequency node are the fixed vibration loading frequency and the maximum vibration intensity. The node characteristics of the random vibration frequency node are the random vibration loading frequency range and the maximum vibration intensity. The edges represent the formation compaction degree, formation structure stability level, and formation vibration sensitivity of the corresponding suspected area of ​​formation disturbance. The vibration sequence determination model is used to process the multiple vibration loading maps to obtain the fixed vibration loading sequence and random vibration loading sequence for each suspected area of ​​formation disturbance.

4. The method for identifying ground disturbance during shield tunnel construction based on muon detection as described in claim 1, characterized in that, The artificial vibration loading sequence for each suspected formation disturbance area includes a fixed vibration loading sequence and a random vibration loading sequence for each suspected formation disturbance area.

5. A ground disturbance identification system based on muon detection during shield tunnel construction, characterized in that, include: The image acquisition module is used to acquire muon flux scan images of the area affected by shield tunneling construction; The monitoring area determination module is used to determine multiple inert gas injection monitoring areas based on the muon flux scanning image of the shield tunneling construction impact area; The scheme determination module is used to determine the inert gas injection test scheme based on the muon flux scanning image of the shield tunneling impact area and the multiple inert gas injection monitoring areas. The response data acquisition module is used to execute the inert gas injection test plan and acquire gas diffusion response data and ground deformation response data in the shield tunneling construction affected area. The suspected area determination module is used to determine multiple suspected areas of ground disturbance based on the gas diffusion response data and ground deformation response data of the shield tunneling affected area. The sequence determination module is used to determine the artificial vibration loading sequence for each suspected geological disturbance area based on the multiple suspected geological disturbance areas; The excitation acquisition module is used to apply vibration excitation to each suspected formation disturbance area based on the artificial vibration loading sequence of each suspected formation disturbance area, and to acquire vibration response data and density change data of each suspected formation disturbance area. The anomaly warning module is used to determine the abnormal geological disturbance area based on the vibration response data and density change data of each suspected geological disturbance area, and to send geological disturbance warning information.

6. The ground disturbance identification system based on muon detection during shield tunnel construction as described in claim 5, characterized in that, The scheme determination module is also used for: Based on the muon flux scanning image of the shield tunneling impact area and the multiple inert gas injection monitoring areas, information on multiple candidate inert gas injection points is determined. Multiple candidate injection point clusters are obtained by clustering based on the information of the multiple candidate inert gas injection points; An inert gas injection test scheme is determined based on the multiple candidate injection point clusters.

7. The ground disturbance identification system based on muon detection during shield tunnel construction as described in claim 5, characterized in that, The sequence determination module is also used for: Based on the gas diffusion response data and formation deformation response data of each suspected formation disturbance area, the formation compaction degree, formation structure stability level, and formation vibration sensitivity of each suspected formation disturbance area are determined. Based on the stratigraphic compaction, stratigraphic structural stability level, and stratigraphic vibration sensitivity of each suspected stratigraphic disturbance area, multiple fixed vibration loading frequencies, random vibration loading frequency ranges, and maximum vibration intensity are determined for each suspected stratigraphic disturbance area. Multiple vibration loading maps are constructed, each corresponding to a suspected area of ​​formation disturbance. Each vibration loading map includes multiple nodes and multiple edges. The multiple nodes include multiple fixed vibration frequency nodes and one random vibration frequency node. Each fixed vibration frequency node is connected to the random vibration frequency node to establish an edge. The node characteristics of the fixed vibration frequency node are the fixed vibration loading frequency and the maximum vibration intensity. The node characteristics of the random vibration frequency node are the random vibration loading frequency range and the maximum vibration intensity. The edges represent the formation compaction degree, formation structure stability level, and formation vibration sensitivity of the corresponding suspected area of ​​formation disturbance. The vibration sequence determination model is used to process the multiple vibration loading maps to obtain the fixed vibration loading sequence and random vibration loading sequence for each suspected area of ​​formation disturbance.

8. The ground disturbance identification system based on muon detection during shield tunnel construction as described in claim 5, characterized in that, The artificial vibration loading sequence for each suspected formation disturbance area includes a fixed vibration loading sequence and a random vibration loading sequence for each suspected formation disturbance area.

9. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method for identifying ground disturbances in shield tunnel construction based on muon detection as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the method for identifying ground disturbances during shield tunnel construction based on muon detection as described in any one of claims 1 to 4.

Images