A deep buried tunnel early warning system and method based on wireless sensor network

Abstract

The present invention relates to the technical field of tunnel water inrush monitoring and early warning, and discloses a deep buried tunnel early warning system and method based on a wireless sensor network, including: a data acquisition module, which is used to collect unit feature data in the tunnel; a data preprocessing module, which is used to preprocess the unit feature data to obtain a second feature sequence; a unit graph structure construction module, which is used to construct a unit graph structure according to the second feature sequence; a water inrush prediction module, which is used to construct a water inrush prediction model according to the unit graph structure and predict the probability of water inrush in the tunnel; a prevention and control measures generation module, which is used to generate prevention and control measures according to the probability of water inrush. The present invention constructs a water inrush prediction model, comprehensively considers the changes in multidimensional feature data such as water pressure and water seepage of each subunit in the tunnel in time and space, accurately predicts the risk of tunnel water inrush, and generates prevention and control measures according to the probability of water inrush, thereby realizing early warning and intelligent prevention and control of tunnel water inrush phenomenon.

Description

Technical Field

[0001] The invention belongs to the technical field of tunnel water inrush monitoring and early warning, and in particular relates to a deep buried tunnel early warning system and method based on a wireless sensor network. Background Art

[0002] During the construction and operation of deep buried tunnels, they often pass through complex geological layers, such as aquifers, underground rivers, or geological bodies with high water pressure. Groundwater activity in these areas is relatively frequent, and the water pressure is high, which can easily lead to catastrophic accidents such as water and mud inrush. The occurrence of tunnel water inrush will not only cause water accumulation inside the tunnel, damage to equipment, and stagnation of the project, but may also cause secondary disasters such as failure of the support structure and tunnel collapse, seriously threatening the safety of tunnel construction and operation. Therefore, early monitoring and early warning of tunnel water inrush are particularly important.

[0003] Existing tunnel monitoring technologies mainly rely on a single type of sensor, such as water pressure sensors, water seepage sensors, and groundwater level sensors. Sensors installed inside and around the tunnel monitor water pressure, water seepage, groundwater level and other data, and combine data threshold settings to determine the risk of water inrush. However, these traditional monitoring methods mostly rely on a single characteristic parameter for early warning. For example, if judgments are made based only on changes in water pressure or water seepage, it is difficult to accurately assess the comprehensive state of the surrounding rock before water inrush occurs.

[0004] Therefore, there is an urgent need for a deep buried tunnel early warning system based on wireless sensor networks. Summary of the invention

[0005] The present invention provides a deep buried tunnel early warning system and method based on a wireless sensor network, which solves the technical problems in the related technology that there is a lack of refined division for different structures or areas of the tunnel, and it is difficult to cope with complex spatiotemporal changes and the correlation between different areas.

[0006] The present invention provides a deep buried tunnel early warning system based on a wireless sensor network, including a data acquisition module, a data preprocessing module, a unit graph structure building module, a water inrush prediction module and a prevention and control measures generation module:

[0007] A data acquisition module, which is used to collect unit characteristic data of M subunits in the tunnel at a preset time interval t within a first preset time period T, and represent the unit characteristic data at each time point by a first characteristic sequence, wherein the mth sequence unit of the first characteristic sequence at the nth time point represents the unit characteristic data of the mth subunit at the nth time point, wherein 1≤m≤M, 1≤n≤N, N=T / t; the unit characteristic data includes unit type, water pressure, stress data, water seepage, temperature and humidity;

[0008] A data preprocessing module, which is used to preprocess the first feature sequence to obtain a second feature sequence;

[0009] A unit graph structure construction module is used to construct a unit graph structure according to the second feature sequence, wherein the unit graph structure includes: nodes, initial features of nodes and edges; wherein a mapping relationship is established between subunits and nodes; unit feature data of subunits is used as initial features of nodes that establish a mapping relationship with subunits; and edges represent connection relationships between nodes;

[0010] A water inrush prediction module, which is used to construct a water inrush prediction model according to the unit graph structure and predict the probability of water inrush in the tunnel; the input of the water inrush prediction model is the unit graph structure, and the output is the probability of water inrush in M ​​subunits within the second preset time period G; the value range of the water inrush probability is 0 to 1;

[0011] The prevention and control measures generation module is used to generate prevention and control measures according to the water inrush probabilities of the M subunits within the second preset time period G.

[0012] Furthermore, T, t, M, m, N, n and G are all custom parameters, and the system dynamically adjusts the custom parameters during operation through a real-time feedback mechanism.

[0013] Furthermore, the unit types include tunnel top, tunnel sidewall and tunnel bottom, and the unit types are represented by real number coding.

[0014] Furthermore, for the division of subunits, a method combining longitudinal division and cross-sectional division is adopted.

[0015] Furthermore, the pre-processing step includes:

[0016] Step S201, construct the seepage pressure feature according to the water pressure and the seepage amount, and after splicing it with the unit feature data, the calculation formula of the seepage pressure is: , where PPC represents the osmotic pressure, WP represents the water pressure, and IR represents the water seepage;

[0017] Step S202: construct stress seepage characteristics according to stress data and seepage amount, and concatenate them with unit characteristic data. The calculation formula of stress seepage is: , where SSC represents stress seepage and S represents stress;

[0018] Step S203, using a mean filtering algorithm to perform noise reduction processing on the first feature sequence;

[0019] Step S204: normalize the first feature sequence using a maximum-minimum normalization method to obtain a second feature sequence.

[0020] Furthermore, the edge of the unit graph structure is constructed in the following manner:

[0021] When two subunits are directly adjacent in space, an edge is constructed between the two subunits and a first weight is set;

[0022] Calculate the cosine similarity of water pressure and water seepage between two adjacent sub-units. When it is greater than a first threshold, build an edge between the two sub-units and set a second weight. The value range of the first threshold is 0 to 1.

[0023] When two sub-units have the same tasks in the tunnel, an edge is constructed between the two sub-units and a third weight is set.

[0024] Further, the water inrush prediction model includes a first hidden layer, a combination layer, a second hidden layer, a reduction layer and a classifier;

[0025] The first hidden layer includes N first units, the nth first unit inputs the unit graph structure at the nth time point, and outputs the first updated features of M nodes;

[0026] The combination layer includes N combination units, and the nth combination unit is used to combine the first updated features output by the N first units into a combination vector;

[0027] The second hidden layer includes N second units, the nth second unit inputs the combination vector output by the nth combination unit, and outputs a second updated feature;

[0028] The reduction layer is used to split the second update feature output by the Nth second unit into M third update features, where each third update feature corresponds to a node update feature of a node;

[0029] The node update feature of each node is input into the classifier, and the classification space of the classifier represents the water inrush probability of each subunit.

[0030] Furthermore, the calculation formula of the first hidden layer includes:

[0031] ,

[0032] in represents the first weight parameter, , and represents the initial features of node v, node u and node k, represents the second weight parameter, represents the concatenation operation, exp represents the exponential function, express Activation function, T represents the transposition operation, Represents the set of all neighbor nodes of node v;

[0033] The feature update formula of node v is:

[0034] ,

[0035] in represents the initial features of node v at the t+1th time point, represents the third weight parameter, represents the initial features of node v’s neighbor node u at the tth time point, express Activation function.

[0036] Furthermore, the second unit of the second hidden layer is constructed based on a long short-term memory network unit.

[0037] Furthermore, a digital twin model of the tunnel is established through the SEEP / W, SIGMA / W and SLOPE / W modules in the GeoStudio simulation software to simulate the interaction between the seepage field and the stress field inside the tunnel, and the water inrush risk of the tunnel is evaluated. These models are used as sample labels for the training samples of the water inrush prediction model, and the simulation data generated by the digital twin model of the tunnel under different environments are used as sample data for the training samples of the water inrush prediction model.

[0038] The present invention provides a deep buried tunnel early warning method based on a wireless sensor network, comprising the following steps:

[0039] Step S501, collecting unit characteristic data of M sub-units in the tunnel;

[0040] Step S502, preprocessing the unit feature data to obtain a second feature sequence;

[0041] Step S503, constructing a unit graph structure according to the second feature sequence;

[0042] Step S504, constructing a water inrush prediction model according to the unit graph structure and predicting the probability of water inrush in the tunnel;

[0043] Step S505, generating prevention and control measures according to the probability of water inrush in the tunnel.

[0044] The beneficial effects of the present invention are as follows: the present invention constructs a unit graph structure, based on the spatial, task and other associations between each sub-unit, and uses a graph neural network model combined with a long short-term memory network unit to predict the risk of water inrush. It can calculate the probability of water inrush in a tunnel based on the multidimensional characteristics of the sub-units and their associations, thereby achieving a more accurate spatial and temporal combined water inrush prediction, especially dynamically adjusting the weights between nodes through an attention mechanism to achieve flexible feature updates; the present invention also supports dynamic adjustment of custom monitoring parameters through a real-time feedback mechanism, which enables the system to adaptively adjust according to the actual conditions of different tunnels, ensuring the flexibility and accuracy of monitoring and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a module schematic diagram of a deep buried tunnel early warning system based on a wireless sensor network of the present invention;

[0046] Figure 2 is a flow chart of preprocessing unit characteristic data of the present invention;

[0047] Figure 3 is a flow chart of processing a unit graph structure using a random walk sampling method of the present invention;

[0048] Figure 4 is a schematic diagram of a water inrush prediction model of the present invention;

[0049] Figure 5 The present invention is a flowchart of a deep buried tunnel early warning method based on a wireless sensor network.

[0050] In the figure: data acquisition module 101, data preprocessing module 102, unit graph structure construction module 103, water inrush prediction module 104, prevention and control measures generation module 105, first hidden layer 401, combination layer 402, second hidden layer 403, reduction layer 404 and classifier 405. DETAILED DESCRIPTION

[0051] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0052] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0053] like Figure 1-Figure 5 As shown, a deep buried tunnel early warning system and method based on a wireless sensor network includes a data acquisition module, a data preprocessing module, a unit graph structure building module, a water inrush prediction module and a prevention and control measures generation module:

[0054] The data acquisition module 101 is used to collect unit characteristic data of M subunits in the tunnel at a preset time interval t within a first preset time period T, and represent the unit characteristic data at each time point by a first characteristic sequence, wherein the mth sequence unit of the first characteristic sequence at the nth time point represents the unit characteristic data of the mth subunit at the nth time point, wherein 1≤m≤M, 1≤n≤N, N=T / t; the unit characteristic data includes unit type, water pressure, stress data, water seepage, temperature and humidity;

[0055] A data preprocessing module 102, which is used to preprocess the first feature sequence to obtain a second feature sequence;

[0056] The unit graph structure construction module 103 is used to construct a unit graph structure according to the second feature sequence, wherein the unit graph structure includes: nodes, initial features of nodes and edges; wherein a mapping relationship is established between subunits and nodes; unit feature data of subunits is used as initial features of nodes that establish a mapping relationship with subunits; and edges represent connection relationships between nodes;

[0057] A water inrush prediction module 104 is used to construct a water inrush prediction model according to the unit graph structure and predict the probability of water inrush in the tunnel; the input of the water inrush prediction model is the unit graph structure, and the output is the probability of water inrush in M ​​subunits within the second preset time period G; the value range of the water inrush probability is 0 to 1;

[0058] The control measures generating module 105 is used to generate control measures according to the water inrush probabilities of the M subunits within the second preset time period G.

[0059] In one embodiment of the present invention, T, t, M, m, N, n and G are custom parameters. The system dynamically adjusts the custom parameters during operation through a real-time feedback mechanism. Preferably, T is set to 2 hours and t is set to 5 minutes.

[0060] In one embodiment of the present invention, the unit types include tunnel top, tunnel sidewall and tunnel bottom, and the unit types are represented by real number codes 1 to 3. For example, when the unit type is tunnel sidewall, it is represented by 2.

[0061] In one embodiment of the present invention, a method combining longitudinal division and cross-sectional division is adopted for the division of sub-units; first, longitudinal division is performed along the length direction of the tunnel, and then cross-sectional division is performed within each longitudinal unit according to the structure of the tunnel (such as the top, left and right side walls and bottom), thereby finely dividing the monitoring area of ​​the tunnel.

[0062] In one embodiment of the present invention, a water pressure sensor is arranged in the cavity behind the lining inside the tunnel to collect water pressure data; a stress sensor is arranged inside the tunnel to collect stress data; a water seepage sensor is installed in the cavity behind the tunnel support to monitor the water flow and water flow rate that penetrates along the rock mass into the back of the support structure and collect water seepage data; temperature and humidity sensors are arranged in each subunit of the tunnel to collect temperature and humidity.

[0063] In one embodiment of the present invention, the pre-processing step comprises:

[0064] Step S201, construct the seepage pressure feature according to the water pressure and the seepage amount, and after splicing it with the unit feature data, the calculation formula of the seepage pressure is: , where PPC represents the osmotic pressure, WP represents the water pressure, and IR represents the water seepage;

[0065] Step S202: construct stress water seepage characteristics according to stress data and water seepage amount, and concatenate them with unit characteristic data. The calculation formula of stress water seepage is: , where SSC represents stress seepage and S represents stress;

[0066] Step S203, using a mean filtering algorithm to perform noise reduction processing on the first feature sequence;

[0067] Step S204: normalize the first feature sequence using a maximum-minimum normalization method to obtain a second feature sequence.

[0068] In one embodiment of the present invention, the edge of the unit graph structure is constructed in the following manner:

[0069] When two sub-units are directly adjacent in space, such as upper and lower, left and right units inside a tunnel, an edge is constructed between the two sub-units, and a first weight is set. Preferably, the first weight is set to 1;

[0070] Calculate the cosine similarity of water pressure and water seepage between two adjacent subunits. When the cosine similarity is greater than a first threshold, construct an edge between the two subunits and set a second weight. The value range of the first threshold is 0 to 1. Preferably, the first threshold is set to 0.7 and the second weight is set to 5.

[0071] The formula for cosine similarity is: , a and b represent the combined vector formed by the water pressure and seepage of the two sub-units, respectively.

[0072] When two subunits have the same tasks in the tunnel, such as drainage, support, stress dispersion, etc., an edge is constructed between the two subunits, and a third weight is set. Preferably, the third weight is set to 5;

[0073] There can only be one edge between two sub-units, and the comprehensive weight of the edge is obtained by summing the weights of the above three types of edges.

[0074] In one embodiment of the present invention, a random walk sampling method may be used to process the unit graph structure at the N time points, and the specific steps include:

[0075] Step S301, randomly select a node as the starting node of the walk;

[0076] Step S302, setting the step length and transition probability. The step length indicates the maximum number of nodes passed during the walk. Preferably, the step length is set to 3. The transition probability indicates the probability of transferring from the current node to the neighboring node. Weighted sampling can be performed according to the weight of the edge. The calculation formula of the transition probability based on the weight of the edge is:

[0077] ,in represents the weight of the edge between node v and its neighbor node u;

[0078] Step S303, starting from the starting node, walking in the unit graph structure according to the set step size and transition probability;

[0079] Step S304: Generate a sampling subgraph based on the nodes and edges sampled during the walk, for subsequent feature aggregation and model training.

[0080] In one embodiment of the present invention, through the random walk sampling method, a representative subgraph can be generated from a large-scale tunnel unit graph structure for efficient graph neural network model training. This method can reduce the computational complexity while retaining the graph structure features, so that the tunnel water inrush prediction model can efficiently and accurately assess the water inrush risk inside the tunnel in practical applications and take corresponding prevention and control measures.

[0081] In one embodiment of the present invention, step S303 specifically includes:

[0082] Step S3031, initialize the walking path and add the starting node to the path;

[0083] Step S3032, at each step of walking, transfer from the current node to a neighboring node according to the transfer probability;

[0084] Step S3033, adding the transferred node to the walking path until the set step length is reached.

[0085] In one embodiment of the present invention, the water inrush prediction model includes a first hidden layer 401, a combination layer 402, a second hidden layer 403, a reduction layer 404 and a classifier 405;

[0086] The first hidden layer includes N first units, the nth first unit inputs the unit graph structure at the nth time point, and outputs the first updated features of M nodes;

[0087] The combination layer includes N combination units, and the nth combination unit is used to combine the first updated features output by the N first units into a combination vector;

[0088] The second hidden layer includes N second units, the nth second unit inputs the combination vector output by the nth combination unit, and outputs a second updated feature;

[0089] The reduction layer is used to split the second update feature output by the Nth second unit into M third update features, where each third update feature corresponds to a node update feature of a node;

[0090] The node update feature of each node is input into the classifier, and the classification space of the classifier represents the water inrush probability of each subunit.

[0091] In one embodiment of the present invention, the water inrush prediction model first updates the node features of the unit graph structure at each time point, and then makes predictions based on the feature changes at different time points, effectively combining the spatial structure and the trend of time changes, thereby improving the prediction accuracy and reliability.

[0092] In one embodiment of the present invention, the calculation formula of the first hidden layer includes:

[0093] ,

[0094] in represents the first weight parameter, , and represents the initial features of node v, node u and node k, represents the second weight parameter, represents the concatenation operation, exp represents the exponential function, express Activation function, T represents the transposition operation, Represents the set of all neighbor nodes of node v;

[0095] The feature update formula of node v is:

[0096] ,

[0097] in represents the initial features of node v at the t+1th time point, represents the third weight parameter, represents the initial features of node v’s neighbor node u at the tth time point, express Activation function.

[0098] In one embodiment of the present invention, the second unit of the second hidden layer is constructed based on a long short-term memory network unit.

[0099] In one embodiment of the present invention, a cross entropy loss function is used as the loss function of the water inrush prediction model.

[0100] In one embodiment of the present invention, a digital twin model of a tunnel is established through the SEEP / W, SIGMA / W and SLOPE / W modules in the GeoStudio simulation software, the interaction between the seepage field and the stress field inside the tunnel is simulated, and the water inrush risk of the tunnel is evaluated, which are used as sample labels for training samples of the water inrush prediction model, and the simulation data generated by the digital twin model of the tunnel under different environments are used as sample data for training samples of the water inrush prediction model.

[0101] In one embodiment of the present invention, the probability of water inrush is divided into four different risk levels. When the probability of water inrush is between 0 and 0.2, the risk level is a low risk level, and no measures are taken at the low risk level; when the probability of water inrush is between 0.2 and 0.5, the risk level is a medium-low risk, and preventive measures are taken at the medium-low risk level to strengthen inspection and monitoring. Instructions can be sent to the inspection robot to focus on monitoring the subunits with high probability of water inrush; when the probability of water inrush is between 0.5 and 0.8, the risk level is a medium-high risk, and emergency preventive measures are taken at the medium-high risk level to reduce the risk of water inrush. The local drainage system should be started to drain the area with large water seepage, reduce the local water pressure, and monitor the drainage volume. changes, and evacuate all on-site personnel from high-risk areas, shut down the power systems and other equipment in related areas to prevent harm to personnel when sudden water inrush accidents occur; when the probability of water inrush is between 0.8 and 1, the risk level is high, and emergency protection measures are taken at high risk levels. The comprehensive drainage system should be started immediately, and all water seepage points in the tunnel should be drained to reduce the pressure of water infiltration into the tunnel. Emergency instructions should be sent to all inspection robots, drones and emergency rescue equipment, and a large number of emergency equipment (such as forced drainage pumps and grouting devices) should be carried to the water inrush area to perform operations such as sealing water inrush points, grouting of cracks, and rock reinforcement, control water flow, and organize on-site personnel to evacuate to a safe area.

[0102] In one embodiment of the present invention, a deep buried tunnel early warning method based on a wireless sensor network is provided, comprising the following steps:

[0103] Step S501, collecting unit characteristic data of M sub-units in the tunnel;

[0104] Step S502, preprocessing the unit feature data to obtain a second feature sequence;

[0105] Step S503, constructing a unit graph structure according to the second feature sequence;

[0106] Step S504, constructing a water inrush prediction model according to the unit graph structure and predicting the probability of water inrush in the tunnel;

[0107] Step S505, generating prevention and control measures according to the probability of water inrush in the tunnel.

[0108] Finally, it should be noted that the parts of the present invention that are not described in detail are all prior art. Those of ordinary skill in the art can understand that the above description is only a preferred embodiment of the invention and is not intended to limit the invention. Although the invention is described in detail with reference to the aforementioned examples, those of ordinary skill in the art can still modify the technical solutions recorded in the aforementioned examples, or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, etc. made within the spirit and principles of the invention should be included in the scope of protection of the invention.

Claims

1. A deep buried tunnel early warning system based on wireless sensor network, characterized in that: It includes data acquisition module, data preprocessing module, unit graph structure building module, water inrush prediction module and prevention and control measures generation module: A data acquisition module, which is used to collect unit characteristic data of M subunits in the tunnel at a preset time interval t within a first preset time period T, and represent the unit characteristic data at each time point by a first characteristic sequence, wherein the mth sequence unit of the first characteristic sequence at the nth time point represents the unit characteristic data of the mth subunit at the nth time point, wherein 1≤m≤M, 1≤n≤N, N=T / t; the unit characteristic data includes unit type, water pressure, stress data, water seepage, temperature and humidity; A data preprocessing module, which is used to preprocess the first feature sequence to obtain a second feature sequence; A unit graph structure construction module is used to construct a unit graph structure according to the second feature sequence, wherein the unit graph structure includes: nodes, initial features of nodes and edges; wherein a mapping relationship is established between subunits and nodes; unit feature data of subunits is used as initial features of nodes that establish a mapping relationship with subunits; and edges represent connection relationships between nodes; A water inrush prediction module, which is used to construct a water inrush prediction model according to the unit graph structure and predict the probability of water inrush in the tunnel; the input of the water inrush prediction model is the unit graph structure, and the output is the probability of water inrush in M ​​subunits within the second preset time period G; the value range of the water inrush probability is 0 to 1; The water inrush prediction model includes a first hidden layer, a combination layer, a second hidden layer, a reduction layer and a classifier; The first hidden layer includes N first units, the nth first unit inputs the unit graph structure at the nth time point, and outputs the first updated features of M nodes; The combination layer includes N combination units, and the nth combination unit is used to combine the first updated features output by the N first units into a combination vector; The second hidden layer includes N second units, the nth second unit inputs the combination vector output by the nth combination unit, and outputs a second updated feature; The reduction layer is used to split the second update feature output by the Nth second unit into M third update features, where each third update feature corresponds to a node update feature of a node; The node update feature of each node is input into the classifier, and the classification space of the classifier represents the water inrush probability of each subunit; The second unit of the second hidden layer is constructed based on the long short-term memory network unit; The prevention and control measures generation module is used to generate prevention and control measures according to the water inrush probabilities of the M subunits within the second preset time period G.

2. The deep buried tunnel early warning system based on wireless sensor network according to claim 1 is characterized in that: T, t, M, m, N, n and G are all custom parameters. The system dynamically adjusts the custom parameters during operation through a real-time feedback mechanism.

3. The deep buried tunnel early warning system based on wireless sensor network according to claim 1 is characterized in that: For the division of subunits, a method combining longitudinal division and cross-sectional division is adopted.

4. The deep buried tunnel early warning system based on wireless sensor network according to claim 1 is characterized in that: The pre-processing steps include: Step S201, construct the seepage pressure feature according to the water pressure and the seepage amount, and after splicing it with the unit feature data, the calculation formula of the seepage pressure is: , where PPC represents the osmotic pressure, WP represents the water pressure, and IR represents the water seepage; Step S202: construct stress seepage characteristics according to stress data and seepage amount, and concatenate them with unit characteristic data. The calculation formula of stress seepage is: , where SSC represents stress seepage and S represents stress; Step S203, using a mean filtering algorithm to perform noise reduction processing on the first feature sequence; Step S204: normalize the first feature sequence using a maximum-minimum normalization method to obtain a second feature sequence.

5. The deep buried tunnel early warning system based on wireless sensor network according to claim 1 is characterized in that: The edge construction method of the unit graph structure includes: When two subunits are directly adjacent in space, an edge is constructed between the two subunits and a first weight is set; Calculate the cosine similarity of water pressure and water seepage between two adjacent sub-units. When it is greater than a first threshold, build an edge between the two sub-units and set a second weight. The value range of the first threshold is 0 to 1. When two sub-units have the same tasks in the tunnel, an edge is constructed between the two sub-units and a third weight is set.

6. The deep buried tunnel early warning system based on wireless sensor network according to claim 1 is characterized in that: The calculation formula for the first hidden layer includes: The calculation formula of the attention weight between node v and its neighbor node u is: ， in represents the first weight parameter, , and represents the initial features of node v, node u and node k, represents the second weight parameter, represents the concatenation operation, exp represents the exponential function, express Activation function, T represents the transposition operation, Represents the set of all neighbor nodes of node v; The feature update formula of node v is: ， in represents the initial features of node v at the t+1th time point, represents the third weight parameter, represents the initial features of node v’s neighbor node u at the tth time point, express Activation function.

7. The deep buried tunnel early warning system based on wireless sensor network according to claim 1 is characterized in that: A digital twin model of the tunnel is established through the SEEP / W, SIGMA / W and SLOPE / W modules in the GeoStudio simulation software to simulate the interaction between the seepage field and the stress field inside the tunnel, and to evaluate the water inrush risk of the tunnel. These models are used as sample labels for the training samples of the water inrush prediction model, and the simulation data generated by the digital twin model of the tunnel under different environments are used as sample data for the training samples of the water inrush prediction model.

8. A deep buried tunnel early warning method based on wireless sensor network, characterized in that: A deep buried tunnel early warning system based on a wireless sensor network applied to any one of claims 1 to 7, comprising the following steps: Step S501, collecting unit characteristic data of M sub-units in the tunnel; Step S502, preprocessing the unit feature data to obtain a second feature sequence; Step S503, constructing a unit graph structure according to the second feature sequence; Step S504, constructing a water inrush prediction model according to the unit graph structure and predicting the probability of water inrush in the tunnel; Step S505, generating prevention and control measures according to the probability of water inrush in the tunnel.

Images