A tunnel disaster prevention safety risk prediction system and method and a storage medium

Abstract

The application discloses a tunnel disaster prevention safety risk prediction system and method and a storage medium, and relates to the technical field of tunnel disaster prevention safety risk prediction, and specifically discloses a tunnel disaster prevention safety risk prediction system, which comprises a space-time analysis module, a time analysis module and a risk prediction module.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk prediction, in particular to a tunnel disaster prevention safety risk prediction system, method and storage medium. BACKGROUND

[0002] As an important part of transportation infrastructure, the tunnel disaster prevention safety is facing severe challenges. The tunnel environment is complex, and the risk has significant spatio-temporal coupling characteristics. The abnormal changes of structural deformation (such as cracks and settlement) and environmental parameters (temperature, humidity, and harmful gas concentration) influence each other, which may cause major accidents such as fire and collapse. However, the traditional tunnel risk analysis system has significant defects.

[0003] Currently, the tunnel risk is interwoven and influenced in the spatial and temporal dimensions. The traditional tunnel risk analysis system usually regards the tunnel structure as an independent mechanical system, focuses on the mechanical properties of the structure itself, and calculates the stress and strain distribution by dividing the grid. However, it does not consider the changes of environmental parameters (such as temperature, humidity, and harmful gas concentration) over time, which leads to the fact that in the actual environment, small deformation of the tunnel structure may change the ventilation path, cause local temperature and humidity anomalies, and long-term high humidity environment will accelerate the corrosion of structural materials and reduce the bearing capacity. This model cannot capture the secondary influence of such environmental parameter changes on the structure, and it is not comprehensive and difficult to effectively respond to complex tunnel risk situations. Therefore, a tunnel disaster prevention safety risk prediction system, method and storage medium are proposed. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application proposes the following technical scheme:

[0005] A tunnel disaster prevention safety risk prediction system, comprising:

[0006] A data acquisition module acquires comprehensive state data in the tunnel through a multi-modal lightweight perception network, and the comprehensive state data includes spatial feature data and temporal feature data.

[0007] A space-time analysis module analyzes the spatial correlation of the spatial feature data to obtain spatial state data and construct an optimized tunnel three-dimensional grid, and obtains a first risk mode from the spatial feature data by combining the optimized tunnel three-dimensional grid with a graph neural network. The optimized tunnel three-dimensional grid is realized by introducing a dynamic adaptive optimization mechanism based on the spatial state data from a basic three-dimensional grid. The dynamic adaptive optimization mechanism obtains a spatial correlation coefficient based on the spatial state data, and is realized by adjusting a preset threshold value through the spatial correlation coefficient.

[0008] The time analysis module: analyze the time characteristic data to obtain the time state data and construct an improved long short-term memory network, and obtain the second risk mode from the time characteristic data through the improved long short-term memory network; wherein the improved long short-term memory network is realized by introducing a new cycle gate into the LSTM memory unit;

[0009] The risk prediction module: based on the first risk mode and the second risk mode, the optimal risk data is obtained by weight optimization distribution through an attention mechanism, and the risk prediction result is obtained based on the optimal risk data.

[0010] The space state data acquisition process is:

[0011] The diffusion distance method of graph signal processing is used, the monitoring points at different positions in the tunnel are regarded as nodes of a graph, the connection relationship between the nodes is determined based on the spatial position relationship to construct a graph structure;

[0012] The diffusion distance between two nodes is obtained based on the Laplacian matrix of the graph structure, the diffusion distances between all nodes in the space are counted to form the space state data.

[0013] The optimization tunnel three-dimensional grid construction process is:

[0014] The internal structure of the tunnel is represented as a three-dimensional grid matrix, and the spatial correlation coefficient is obtained based on the diffusion distance;

[0015] The grid elements in the three-dimensional grid matrix are dynamically optimized based on the spatial correlation coefficient;

[0016] A preset correlation threshold is set, the size of the grid element in the region where the spatial correlation coefficient is greater than the correlation threshold is reduced, the number of longitudinally divided elements is reduced by half, the elements are redivided and the correlation feature information is updated;

[0017] For the region where the spatial correlation coefficient is less than or equal to the correlation threshold, the size of the grid element is increased, the number of longitudinally divided elements is increased by half, the elements are redivided and the correlation feature information is updated, and the optimized tunnel three-dimensional grid is obtained.

[0018] The first risk mode acquisition process is:

[0019] The optimized tunnel three-dimensional grid is converted into a new graph structure;

[0020] The feature vector in each grid element in the new graph structure is taken as the initial feature input of a multi-layer graph convolution network, and the spatial risk feature vector is obtained based on the stacking operation of the multi-layer graph convolution network;

[0021] The first risk mode is formed based on the spatial risk feature vector.

[0022] The time sequence state data acquisition process is:

[0023] The time characteristic data is converted into frequency domain by using discrete Fourier transform, and a time characteristic sequence is obtained;

[0024] The autocorrelation coefficient is obtained by measuring the correlation between different time points of the time characteristic sequence through the autocorrelation function;

[0025] The information of the periodic component in the time characteristic sequence is extracted based on the autocorrelation coefficient, and the information is integrated with the time characteristic data to form time sequence state data; wherein the information of the periodic component at least includes: amplitude, phase and period length.

[0026] The second risk mode acquisition process is:

[0027] The time sequence state data is encoded and spliced with the original sensor data as a new input vector after encoding, and a new gate is introduced;

[0028] In the process of updating the state of the traditional long short-term memory network cell, the periodic cell state and the updated weight of the traditional cell state are obtained by controlling the periodic analysis result through the periodic gate;

[0029] The optimized updated cell state is obtained based on the updated weight of the periodic cell state and the traditional cell state;

[0030] The new input vector is sequentially input into the improved long short-term memory network in time sequence, and the cell state and hidden state are sequentially updated through the forget gate, input gate and periodic gate, and the time risk feature vector at the selected time is output through the output gate;

[0031] The second risk mode is formed based on the time risk feature vectors of all time points in the selected time period.

[0032] The optimal risk data acquisition process is:

[0033] The first risk mode and the second risk mode are respectively mapped through two different fully connected layers using the attention mechanism, and the mapped first risk vector and second risk vector are obtained;

[0034] The similarity score between the first risk vector and the second risk vector is obtained through the dot product attention in the attention mechanism;

[0035] The attention weight of the first risk mode and the second risk mode is obtained based on the similarity score;

[0036] The optimal risk data is obtained by weighted fusion based on the attention weight of the first risk mode and the second risk mode.

[0037] A tunnel disaster prevention safety risk prediction method, the prediction method steps are:

[0038] S1: collecting comprehensive state data in the tunnel through a multi-modal lightweight perception network, the comprehensive state data including spatial feature data and time feature data;

[0039] S2: analyzing the spatial correlation of the spatial feature data to obtain spatial state data and constructing an optimized tunnel three-dimensional grid, and obtaining a first risk mode from the spatial feature data through the combination of the optimized tunnel three-dimensional grid and a graph neural network;

[0040] S3: analyzing the time sequence periodicity of the time feature data to obtain time sequence state data and constructing an improved long short-term memory network, and obtaining a second risk mode from the time feature data through the improved long short-term memory network;

[0041] S4: performing weight optimization distribution based on the first risk mode and the second risk mode through an attention mechanism to obtain optimal risk data, and obtaining a risk prediction result based on the optimal risk data.

[0042] A computer readable storage medium, the computer readable storage medium has a computer program stored thereon, the computer program is executed by a processor to implement any one of the tunnel disaster prevention safety risk prediction methods.

[0043] The present application has the following beneficial effects:

[0044] In the present application, first, the spatial correlation coefficient is calculated based on the diffusion distance method of graph signal processing, the tunnel three-dimensional grid is dynamically optimized, the high correlation area is encrypted with the grid to improve the monitoring accuracy, and the low correlation area is expanded with the grid to reduce the data amount, thereby reducing the computing burden by 40% while ensuring the monitoring effect, and the neighbor node features are aggregated layer by layer through the graph neural network (GCN) to capture the spatial propagation characteristics of the structural risk, the influence of the deformation of a certain area on the surrounding 50-meter range can be identified, and the positioning error is less than 5 meters;

[0045] Secondly, the improved LSTM fuses the periodic cell state through the period gate, the identification accuracy of the periodic risk such as peak vibration reaches 92%, which is improved by 28% compared with the traditional LSTM, the spatio-temporal risk is deeply described, and the periodic fluctuation trend of the environmental parameters is accurately captured;

[0046] Finally, the weight of the spatio-temporal risk mode is dynamically calculated through the dot product attention mechanism, when the spatial deformation and the time sequence anomaly appear at the same time, the weight of the key risk mode is automatically improved, the spatio-temporal periodic integrated analysis of the tunnel risk is realized, and the prediction accuracy of the tunnel disaster prevention safety risk is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1A system block diagram of a tunnel disaster prevention safety risk prediction system, method and storage medium is provided.

[0048] Figure 2 A method step diagram of a tunnel disaster prevention safety risk prediction system, method and storage medium is provided. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0050] Embodiment one:

[0051] As shown in Figure 1 , a tunnel disaster prevention safety risk prediction system provided by the present application comprises:

[0052] The data acquisition module acquires comprehensive state data in the tunnel through a multi-modal lightweight perception network, and the comprehensive state data includes spatial feature data and temporal feature data.

[0053] The multi-modal lightweight perception network comprises a distributed optical fiber sensor and a multi-modal environmental sensor.

[0054] The distributed optical fiber sensor adopts optical time domain reflection and phase-sensitive optical time domain reflection technology, and obtains the loss distribution and physical change of the optical fiber along the line by detecting the intensity of backscattered light. The expression of backscattered light power is:

[0055] ;

[0056] Wherein, is the incident light power, is the optical fiber attenuation coefficient, is the optical fiber length, is the Rayleigh scattering coefficient, is the effective scattering cross-sectional area of the optical fiber.

[0057] Specifically, the optical fiber attenuation coefficient is usually set to ≤ 3.5dB / km. In the local area network scene where the transmission distance requirement is not particularly long, the attenuation coefficient of the common multimode optical fiber is controlled to ≤ 3dB / km at this wavelength.

[0058] By monitoring and analyzing the backscattered light power at different positions, the detection of tunnel structure deformation (such as cracks, settlement, etc.) can be realized.

[0059] A multi-modal environment sensor is deployed to monitor the environmental parameters in the tunnel, which includes a temperature and humidity sensor, a harmful gas sensor (such as an MQ-7 sensor for detecting CO), and a smoke sensor.

[0060] The temperature and humidity sensor is connected to the edge computing node through an I2C communication interface, with a humidity measurement accuracy of ±2% RH. The harmful gas sensor MQ-7 detects CO concentration using electrochemical principles, with a detection range of 0-2000 ppm. The detected analog signal is converted to a digital signal by an ADC sampling circuit and transmitted to the edge computing node. The smoke sensor uses an infrared light-emitting diode and a phototransistor to detect smoke particles, with a sensitivity of 0.5 V / 100 μg / m 3 , and a DC bias circuit is used to eliminate interference factors such as dust.

[0061] These environmental sensors are evenly distributed in the tunnel every 50 meters to comprehensively collect environmental parameter information in the tunnel.

[0062] Based on the collection of distributed optical fiber sensors and multi-modal environment sensors, the collected data is divided into spatial feature data and temporal feature data .

[0063] Spatial feature data Through optical time domain reflection and phase-sensitive optical time domain reflection technology, the state of each position in the tunnel is recorded with spatial position as the index. For the monitoring position of the distributed optical fiber sensor, the backscattering light power is recorded. For the position of the environmental sensor, the corresponding temperature and humidity, CO concentration, and smoke value are recorded, representing the state of different spatial positions in the tunnel, and constituting the spatial feature data .

[0064] Temporal feature data For each specific spatial position, the data changes are recorded in chronological order. For the backscattering light power sequence of the distributed optical fiber sensor at position t , and the environmental sensor data at position t, the normalized data is obtained, representing the state evolution of different times in the tunnel, and constituting the temporal feature data .

[0065] In the data collection process, the collected spatial feature data and temporal feature data are time-synchronized, and then the comprehensive state data is preliminarily fused through a fusion algorithm.

[0066] Spatiotemporal analysis module: Analyzes the spatial correlation of spatial feature data to obtain spatial state data and constructs an optimized tunnel 3D mesh. By combining the optimized tunnel 3D mesh with a graph neural network, the first risk pattern is obtained from the spatial feature data.

[0067] The process of acquiring spatial state data is as follows:

[0068] Using the diffusion distance method in graph signal processing, monitoring points at different locations in the tunnel are considered as nodes in a graph. A graph structure is constructed by determining the connections between nodes based on their spatial positions. Let the graph structure... ,in This is a set of nodes (corresponding to tunnel monitoring points). Define the Laplacian matrix for a graph structure, given the set of edges (representing connections between nodes). ,in It is a degree matrix. It is a weight matrix;

[0069] Based on the graph's Laplacian matrix, a heat kernel function is introduced to obtain the relationship between two nodes (from node). Diffusion to nodes diffusion distance The formula is expressed as:

[0070] ;

[0071] in, Laplace matrix eigenvalues, Laplace matrix Feature vector Indicates the Laplace matrix at the nodes The value, Indicates the Laplace matrix at the nodes The value of , where n is the number of nodes;

[0072] Diffusion distance The value reflects the node and nodes The degree of similarity in spatial association, when When smaller, it represents a node. and nodes They exhibit high similarity during diffusion, meaning they have strong spatial correlation; conversely, when When the distance is large, the spatial correlation is weak, and the diffusion distance between nodes in all spaces is statistically analyzed. Composition of spatial state data ;

[0073] Specifically, by calculating the diffusion distance between different monitoring points in the tunnel and analyzing the spatial correlation between monitoring data at different locations in the tunnel, it is possible to determine which locations in the space may have a state change that affects the surrounding locations, as well as the degree and direction of the influence. By starting from the topology of the graph and the signal diffusion perspective, compared with the traditional spatial autocorrelation method, it can capture the correlation characteristics of monitoring points under the complex spatial structure of the tunnel more flexibly.

[0074] The process of introducing a dynamic adaptive optimization mechanism is as follows:

[0075] The internal structure of the tunnel is represented by a three-dimensional grid matrix. The tunnel is divided into units along its longitudinal direction (mileage direction) at fixed intervals of 10m. Let the number of units in the longitudinal division of the tunnel be... And divided into Each region, constructing a The three-dimensional mesh matrix of the structure;

[0076] Based on spatial state data diffusion distance Further calculate the spatial correlation coefficient, assuming known nodes and diffusion distance and the average diffusion distance between all nodes. The spatial correlation coefficient is expressed as:

[0077] ;

[0078] in, It is an exponential function. Spatial correlation coefficient;

[0079] The grid cells are dynamically optimized based on the strength of spatial correlation, and the spatial correlation coefficient is... Greater than ( If a region is defined as having a set correlation threshold, indicating a strong spatial correlation between monitoring points within that region, then the size of the grid cells should be reduced, and the number of vertically divided cells should be decreased. Reduce by half, re-divide into units, and update the association feature information for spatial association coefficients. Less than or equal to In areas with weak spatial correlation, increasing the grid cell size and the number of vertically divided cells... By increasing the number of units by half and re-dividing them, and updating the associated feature information, the monitoring effect is guaranteed while reducing the amount of data and the computational burden, resulting in an optimized three-dimensional mesh for the tunnel.

[0080] For example:

[0081] After re-dividing the unit, for example, the average depth of burial, average surrounding rock grade information in the original 100-meter unit is recorded, and after dividing into 50-meter units, the depth of burial and surrounding rock grade of each 50-meter unit need to be re-measured or calculated;

[0082] Convert the optimized tunnel three-dimensional grid into a new graph structure wherein, is a node set, and the number of nodes corresponding to each grid unit is , is an edge set, and if two grid units are directly adjacent in the longitudinal or transverse direction, then there is an edge between the corresponding two nodes in the graph;

[0083] For example, for the longitudinally adjacent grid units and , and the transversely adjacent grid units and , the nodes and , and the nodes and are connected by edges, respectively, wherein, , ;

[0084] The feature vector in each grid unit in the new graph structure is taken as an initial feature input of a multi-layer graph convolution network, and a stacking operation is performed based on the multi-layer graph convolution network. In the first layer, the feature representation of node is , and the update formula is as follows:

[0085] ;

[0086] wherein, is an activation function, and represents the updated neighborhood aggregation;

[0087] Specifically, the neighborhood aggregation calculation formula is: wherein, denotes the neighbor node set of node , and are the degrees (i.e., the number of edges connected to the node) of node and node , respectively, is the feature representation of node in the first layer, is the weight matrix of the first layer;

[0088] After Layer graph convolution network stacking operation, each node can capture the information of the neighbor nodes, and finally obtain the final feature vector of the node in each grid cell The final feature vector of the node in each grid cell is input into the full connection layer for dimension reduction processing, and the spatial risk feature vector of each grid cell is output, wherein the weight matrix of the full connection layer is , the bias vector is , and the spatial risk feature vector output by the full connection layer is: , wherein is an activation function, , and the spatial risk feature vector of the i-th node is denoted as

[0089] Specifically, the spatial risk feature vectors comprehensively reflect the relative risk state of each grid cell in the entire tunnel space structure and the correlation degree with the surrounding area, and the spatial risk feature vector composed of all the nodes in the grid cell determines the optimal risk distribution of the tunnel structure in space, forming a first risk mode

[0090] Time analysis module: analyze the time sequence periodicity of the time characteristic data to obtain time sequence state data and construct an improved long short-term memory network, and obtain a second risk mode from the time characteristic data through the improved long short-term memory network;

[0091] For the feature data in the time characteristic data , discrete Fourier transform (DFT) is used to convert the time domain signal to the frequency domain to obtain a time characteristic sequence , and a autocorrelation function is used to measure the correlation between the time characteristic sequence at different times, and the autocorrelation coefficient of the converted time characteristic sequence (time characteristic data discrete Fourier transform feature) is expressed as:

[0092]

[0093] , wherein is a time delay, and when takes different values, the autocorrelation coefficient is calculated, and if there is a significant peak at a certain , it indicates that the signal has strong correlation in the time interval

[0094] Specifically, the significant peak refers to that when the autocorrelation coefficient is calculated, with the time delay ​​​Taking different values ​​will yield a series of results for the autocorrelation coefficient. When there is a value among these results that makes the corresponding autocorrelation coefficient significantly higher than the value near it, this higher value is called a significant peak.

[0095] For example:

[0096] Suppose there is a time feature series Calculate its autocorrelation coefficient ,when hour, The value is 0.2, when hour, The value is 0.1, when hour, The value suddenly rose to 0.8, when hour, The value dropped to 0.3 again, here Time corresponding This value is significantly higher than the values ​​in its vicinity, forming a clear peak.

[0097] Based on the results of autocorrelation function analysis, key information such as the amplitude, phase, and period length of the periodic components of the time feature series is extracted. This information is then integrated with the corresponding time feature data to form time series state data. ;

[0098] Traditional Long Short-Term Memory (LSTM) networks only accept sensor monitoring data as input. Improved LSTM networks, however, incorporate temporal state data obtained through temporal periodicity analysis. After encoding, the data is concatenated with the original sensor data (initial data collected by distributed fiber optic sensors and multimodal environmental sensors) to form a new input vector. This allows the model to obtain the temporal periodicity information of the data in the initial stage;

[0099] Then, a new gating mechanism is introduced into the LSTM memory unit (traditional long short-term memory networks consist of three gates: input gate, forget gate, and output gate), adding a new period gate. Periodic gate The formula for obtaining it is expressed as:

[0100] ;

[0101] in, For activation function, For the new periodic gate The weight matrix, Indicates the hidden state in the previous moment. and the new input vector at the current time, For bias terms;

[0102] During cell state renewal, according to the cycle gate The output optimizes and improves the cell update state;

[0103] Through the periodic gate By controlling the periodic cell states obtained from the periodicity analysis results and the weights of the traditional cell states after updating, the model can more effectively capture and remember periodic changes in time-series data. The improved formula is expressed as:

[0104] ;

[0105] in, To pass through the periodic gate Optimize the updated cell state. Let t represent the cell state at time t. for Cellular state at any given moment This refers to the element-wise multiplication operation of matrices or vectors in Long Short-Term Memory (LSTM) networks. For the Gate of Oblivion For input gate, Candidate cell state, To determine the periodic cell states obtained from periodicity analysis, which reflect the periodic characteristics in time series data, a periodicity gate is used. control In addition to the weights of the traditional cell state update component, the model is able to more effectively capture and remember periodic changes in time-series data;

[0106] Specifically, periodic gates The output is between 0 and 1, when When it approaches 1, at this time In this context, the effect of the forgetting gate on historical cell states is regulated by the periodic gate, and simultaneously... It will cause the cyclical cell state The contribution to updating the current cell state increases, and when because As the input value approaches 0, the traditional input gate's effect on updating the candidate cell state weakens; conversely, when... When the value approaches 0, the traditional cell state update part (the forget gate updates the historical state and the input gate updates the candidate state) plays a dominant role, and the influence of periodic cell states will become smaller. In this way, when updating cell states, the long short-term memory network can selectively incorporate these periodic information according to the control of the periodic gate, so as to better capture and remember the periodic change patterns in the data.

[0107] Through the aforementioned cell renewal mechanism, the preprocessed (time-series state data) new input vector (spliced with the original sensor data after encoding) is sequentially input into the improved long short-term memory network in chronological order, at each time , the network updates the cell state and the hidden state according to the current input, the previous time hidden state and cell state through a gating mechanism (in the order of forget gate, input gate, and cycle gate) , as the data is continuously input, the improved long short-term memory network gradually learns the periodic variation law of the time feature data over time, and finally outputs the time risk feature vector at the time through the output gate , which contains the risk trend of the tunnel state over time, as well as the risk change trend caused by the fluctuation of environmental parameters over time (fire, structural damage, etc.), and the second risk pattern is formed based on the time risk feature vectors of all time points within a specific time period

[0108] Risk prediction module: based on the first risk pattern and the second risk pattern, the optimal risk data is obtained through a weight optimization distribution of an attention mechanism, and the risk prediction result is obtained based on the optimal risk data;

[0109] After obtaining the first risk pattern (based on spatio-temporal feature data) and the second risk pattern (based on time feature data), the attention mechanism is used to map and through two different fully connected layers to obtain the mapped first risk vector and the second risk vector , and then the similarity score and of the mapped vectors is obtained through dot product attention in the attention mechanism ;

[0110] After obtaining the similarity score e, the attention weight of the first risk pattern is calculated based on the similarity score , which is expressed by the formula:

[0111] ;

[0112] wherein, is the score of the first risk pattern itself, is the score of the second risk pattern itself, is an exponential function;

[0113] Specifically, the exponential function The similarity scores are nonlinearly transformed so that the differences between the scores are amplified, thereby more obviously distinguishing the importance of different risk patterns;

[0114] According to the calculated first risk pattern attention weight The two risk patterns are weighted and fused to obtain optimal risk data The calculation formula is:

[0115] ;

[0116] Wherein, is the attention weight of the second risk pattern;

[0117] Specifically, if is close to 1, it means that the first risk pattern (based on spatiotemporal feature data) is more critical in the current risk assessment, and the optimal risk data after fusion will more reflect the characteristics of the first risk pattern, if is close to 1, the influence of the second risk pattern (based on time feature data) is greater, through the attention mechanism to fully utilize the advantages of the two risk patterns, comprehensive consideration of the risk information of the tunnel in the space and time dimension, get more comprehensive optimal risk data;

[0118] Based on the optimal risk data, set the warning mechanism, determine the threshold of different risk levels through a large number of historical data and actual cases, divide the risk level into low, medium and high three levels, and get the corresponding threshold through clustering analysis , , ;

[0119] Use the optimal risk data to perform linear transformation to obtain a risk assessment value, and perform risk prediction according to the assessment value, for example:

[0120] If the risk assessment value exceeds but is lower than , it is determined as low risk, no warning is triggered, only data recording and continuous monitoring are performed, and the data acquisition frequency of the sensor is appropriately reduced to save energy and resources;

[0121] If the risk assessment value exceeds but is lower than , it is determined as medium risk, triggering a primary warning, starting risk mitigation measures, for example:

[0122] Strengthen the operation of the ventilation system in the tunnel to reduce the concentration of harmful gases and increase the frequency of inspection of the tunnel structure and equipment;

[0123] If the risk assessment value exceeds If yes, it is determined as high risk, a high-level warning is triggered, and an emergency contingency plan is immediately started, for example:

[0124] People and vehicles in the tunnel are quickly evacuated, the tunnel exits are closed partially or entirely, professional emergency rescue teams and equipment such as fire engines, ambulances, engineering rescue vehicles, etc. are mobilized, and the disaster in the tunnel is disposed of. Through various channels such as broadcasting and electronic display screens, the tunnel danger information is released to the public, and the surrounding traffic is guided to detour, so as to avoid causing greater influence.

[0125] In the application, several formulas involved are calculated by taking their numerical values after being dimensionless, and the establishment of the formula is obtained by collecting a large amount of data to simulate a formula of the nearest real situation, and part of the coefficients or weights in the formula are set by the person skilled in the art according to the actual situation, so here is not described.

[0126] Embodiment two:

[0127] As shown in Figure 2 The tunnel disaster prevention safety risk prediction method provided by the application comprises:

[0128] S1: acquiring comprehensive state data in the tunnel through a multi-modal lightweight perception network, wherein the comprehensive state data comprises spatial feature data and time feature data;

[0129] S2: analyzing the spatial correlation of the spatial feature data to obtain spatial state data and constructing an optimized tunnel three-dimensional grid, and acquiring a first risk mode from the spatial feature data by combining the optimized tunnel three-dimensional grid with a graph neural network;

[0130] S3: analyzing the time sequence periodicity of the time feature data to obtain time sequence state data and constructing an improved long short-term memory network, and acquiring a second risk mode from the time feature data by the improved long short-term memory network;

[0131] S4: performing weight optimization distribution based on the first risk mode and the second risk mode through an attention mechanism to obtain optimal risk data, and acquiring a risk prediction result based on the optimal risk data.

[0132] Embodiment three:

[0133] In this embodiment, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium. It should be noted that the functions or steps that can be achieved by the computer readable storage medium or the computer device are described in the foregoing method embodiments, and the related descriptions of the server side and the client side are described. To avoid repetition, they will not be described one by one here.

[0134] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art can be aware that units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on the specific application and design constraints of the technical solutions.

[0135] Although embodiments of the present application have been shown and described, it would be appreciated by those skilled in the art that changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A tunnel disaster prevention and safety risk prediction system, characterized in that, include: Data acquisition module: Collects comprehensive state data in the tunnel through a multimodal lightweight sensing network. The comprehensive state data includes spatial feature data and temporal feature data. The spatiotemporal analysis module analyzes the spatial correlation of spatial feature data to obtain spatial state data and constructs an optimized tunnel 3D mesh. By combining the optimized tunnel 3D mesh with a graph neural network, the first risk pattern is obtained from the spatial feature data. The optimized tunnel 3D mesh is implemented by introducing a dynamic adaptive optimization mechanism from the basic 3D mesh based on spatial state data. The dynamic adaptive optimization mechanism obtains spatial correlation coefficients based on spatial state data and adjusts them by setting a threshold. The process of obtaining the first risk pattern is as follows: The optimized 3D tunnel mesh is transformed into a new graph structure; The feature vector in each grid cell of the new graph structure is used as the initial feature input of a multi-layer graph convolutional network, and the spatial risk feature vector is obtained by stacking operations based on the multi-layer graph convolutional network. The first risk pattern is formed based on spatial risk feature vectors; The time analysis module analyzes the temporal periodicity of time feature data to obtain time-series state data and constructs an improved long short-term memory network. The improved long short-term memory network is used to obtain a second risk pattern from the time feature data. The improved long short-term memory network is implemented by introducing new periodic gates into the LSTM memory cells. The process of obtaining the second risk model is as follows: The temporal state data is encoded using a traditional long short-term memory network and then concatenated with the original sensor data to form a new input vector, and a new gating mechanism is introduced. In the process of cell state update in traditional long short-term memory networks, the periodic cell state and the weights after the traditional cell state update are obtained by controlling the periodicity analysis results through periodic gates. The optimized updated cell state is obtained based on the weights of the periodic cell state and the traditional cell state. The new input vector is fed into the improved long short-term memory network in chronological order. The cell state and hidden state are updated in the order of forget gate, input gate and periodic gate. The time risk feature vector at the selected time is output through the output gate. A second risk pattern is formed based on the time risk feature vectors of all moments within a selected time period; Risk prediction module: Based on the first risk model and the second risk model, the module uses an attention mechanism to optimize the weight allocation to obtain the optimal risk data, and then obtains the risk prediction result based on the optimal risk data.

2. The tunnel disaster prevention and safety risk prediction system according to claim 1, characterized in that, The process for acquiring the spatial state data is as follows: By using the diffusion distance method of graph signal processing, monitoring points at different locations in the tunnel are regarded as nodes of a graph, and a graph structure is constructed based on the spatial positional relationship to determine the connection relationship between the nodes. Based on the Laplacian matrix of the graph structure, the diffusion distance between two nodes is obtained, and the diffusion distance between nodes in all spaces is counted to form spatial state data.

3. The tunnel disaster prevention and safety risk prediction system according to claim 2, characterized in that, The optimized tunnel 3D mesh construction process is as follows: The internal structure of the tunnel is represented as a three-dimensional mesh matrix, and the spatial correlation coefficient is obtained based on the diffusion distance; Dynamic optimization of grid cells in a 3D grid matrix based on spatial correlation coefficients; A preset correlation threshold is set. For regions with spatial correlation coefficients greater than the correlation threshold, the size of the grid cells is reduced, the number of vertically divided cells is halved, and the cells are re-divided and the correlation feature information is updated. For regions where the spatial correlation coefficient is less than or equal to the correlation threshold, the grid cell size is increased, the number of vertically divided cells is increased by half, the cells are re-divided, and the correlation feature information is updated to obtain the optimized tunnel 3D mesh.

4. The tunnel disaster prevention and safety risk prediction system according to claim 1, characterized in that, The process for acquiring the time-series state data is as follows: The time-domain signal is converted to the frequency domain by using the discrete Fourier transform on the time-specific data to obtain the time-specific sequence. The autocorrelation coefficient is obtained by measuring the correlation between time feature sequences at different times using the autocorrelation function. Information on periodic components in time feature sequences is extracted based on autocorrelation coefficients, and this information is integrated with time feature data to form time series state data. The information on periodic components includes at least amplitude, phase, and period length.

5. The tunnel disaster prevention and safety risk prediction system according to claim 1, characterized in that, The process for obtaining the optimal risk data is as follows: The attention mechanism is used to map the first risk pattern and the second risk pattern through two different fully connected layers to obtain the mapped first risk vector and second risk vector. The similarity score between the first risk vector and the second risk vector is obtained by using dot product attention in the attention mechanism. Attention weights for the first and second risk patterns are obtained based on similarity scores; The optimal risk data is obtained by weighting and fusing the attention weights of the first and second risk models.

6. A method for predicting tunnel disaster prevention and safety risks, using the system described in any one of claims 1-5, characterized in that, The prediction method steps are as follows: S1: Collect comprehensive state data in the tunnel through a multimodal lightweight sensing network. The comprehensive state data includes spatial feature data and temporal feature data. S2: Analyze the spatial correlation of spatial feature data to obtain spatial state data and construct an optimized tunnel 3D mesh. By combining the optimized tunnel 3D mesh with a graph neural network, the first risk pattern is obtained from the spatial feature data. S3: Analyze the temporal periodicity of time feature data to obtain time-series state data and construct an improved long short-term memory network. Use the improved long short-term memory network to obtain the second risk pattern from the time feature data. S4: Based on the first and second risk models, the optimal risk data is obtained by weight optimization through an attention mechanism, and the risk prediction result is obtained based on the optimal risk data.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the tunnel disaster prevention and safety risk prediction system according to any one of claims 1-5.

Images