A multi-point coordinated tunnel gas sensing alarm method and device

Abstract

The present invention provides a multi-point coordinated tunnel gas sensing alarm method and device, which relates to the field of sensing alarm technology, including: establishing distributed sensing optical fibers in the tunnel, obtaining preset monitoring accuracy, and establishing initial point distribution results; establishing point correction parameters according to traffic characteristics, construction activity characteristics, and tunnel structure characteristics, and establishing updated point distribution results; establishing a dynamic pre-capture period, and establishing parent-child nodes; establishing position concentration data, performing long and short feature extraction, and establishing gas concentration feature extraction results; establishing independent position warning results; configuring a dependent identification network, generating dependent position warning results; and performing gas sensing alarm processing. The present invention solves the technical problems that the static monitoring point distribution of the prior art is difficult to adapt to dynamic environmental changes, resulting in inaccurate monitoring data, and that a single warning analysis method cannot fully reflect the actual situation of the gas, resulting in incomplete and inaccurate warning results.

Description

Technical Field

[0001] The present invention relates to the field of sensing alarm technology, and in particular to a multi-point coordinated tunnel gas sensing alarm method and device. Background Art

[0002] Tunnels play a key role in the transportation system. Due to their closed structure and specific environmental conditions, gas accumulation in tunnels, such as carbon monoxide, carbon dioxide, nitrogen oxides, etc., poses a potential threat to driving safety and personnel health. Therefore, accurate monitoring and timely alarm are important measures to ensure tunnel safety. However, the existing technology for tunnel gas sensing alarm still has certain technical problems. First, factors such as vehicle flow and construction activities in tunnels may cause dynamic changes in the environment. The existing technology usually uses fixed monitoring points and cannot adapt to these changes, resulting in inaccurate monitoring data and affecting the effect of early warning. Second, the existing method cannot fully extract the characteristics of gas concentration changes, resulting in insufficient recognition of gas concentration changes, and cannot accurately identify subtle fluctuations in gas concentration changes, thereby affecting the sensitivity and reliability of gas early warning. Third, the existing technology usually uses a single early warning analysis method, which cannot fully reflect the actual situation of gas concentration. Relying only on a single early warning result may lead to misjudgment or omission of potential dangers, thereby affecting the effect of safety protection. Summary of the invention

[0003] The present application provides a multi-point coordinated tunnel gas sensing alarm method and device, aiming to solve the technical problems that the static monitoring point distribution in the prior art is difficult to adapt to dynamic environmental changes, resulting in inaccurate monitoring data, and a single early warning analysis method cannot fully reflect the actual situation of the gas, resulting in incomplete and inaccurate early warning results.

[0004] The first aspect disclosed in the present application provides a multi-point coordinated tunnel gas sensing alarm method, the method comprising: establishing a distributed sensing optical fiber of the tunnel, and obtaining a preset monitoring accuracy of the tunnel, performing uniform point distribution according to the distributed sensing optical fiber and the preset monitoring accuracy, and establishing an initial point distribution result; performing traffic flow monitoring and construction monitoring on the tunnel, establishing traffic flow characteristics and construction activity characteristics, obtaining structural characteristics of the tunnel, establishing point correction parameters based on the traffic flow characteristics, construction activity characteristics, and structural characteristics, performing position compensation on the initial point distribution result based on the point correction parameters, and establishing an updated point distribution result; establishing a dynamic pre-capture period, and updating the point distribution based on the dynamic pre-capture period. The parent-child nodes of the distribution results are established, and the parent-child nodes have parent-child association weights and are multimodal parent-child nodes; according to the updated point distribution results, the distributed sensing optical fiber is called to monitor the gas concentration, and the position concentration data is established. The length characteristics of the gas concentration changes in the position concentration data are extracted through a multi-scale time series to establish a gas concentration feature extraction result; based on the gas concentration feature extraction result, an independent point concentration warning analysis is performed to establish an independent position warning result; according to the parent-child node configuration dependency identification network, the position concentration data is input into the configured dependency identification network to generate a dependent position warning result; according to the independent position warning result and the dependent position warning result, the gas sensing alarm processing of the tunnel is performed.

[0005] The second aspect disclosed in the present application provides a multi-point coordinated tunnel gas sensing alarm device, which is used for the above-mentioned multi-point coordinated tunnel gas sensing alarm method, and the device includes: a uniform point distribution module, which is used to establish a distributed sensing optical fiber of the tunnel and obtain a preset monitoring accuracy of the tunnel, and perform uniform point distribution according to the distributed sensing optical fiber and the preset monitoring accuracy to establish an initial point distribution result; a position compensation module, which is used to perform traffic flow monitoring and construction monitoring on the tunnel, establish traffic flow characteristics and construction activity characteristics, obtain structural characteristics of the tunnel, establish point correction parameters based on the traffic flow characteristics, construction activity characteristics, and structural characteristics, perform position compensation on the initial point distribution result based on the point correction parameters, and establish an updated point distribution result; a parent-child node establishment module, which is used to establish a dynamic front capture period, and update the dynamic front capture period based on the dynamic front capture period. The parent-child nodes of the point distribution results are established, and the parent-child nodes have parent-child association weights and are multimodal parent-child nodes; a long-short feature extraction module, which is used to call the distributed sensing optical fiber to monitor the gas concentration according to the updated point distribution results, establish position concentration data, extract the long and short features of the gas concentration changes in the position concentration data through multi-scale time series, and establish the gas concentration feature extraction result; an independent position warning module, which is used to perform independent point concentration warning analysis based on the gas concentration feature extraction result and establish an independent position warning result; a dependent position warning module, which is used to configure a dependent identification network according to the parent-child nodes, input the position concentration data into the configured dependent identification network, and generate a dependent position warning result; an alarm processing module, which is used to perform gas sensing alarm processing in the tunnel according to the independent position warning result and the dependent position warning result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] By laying distributed sensing optical fibers in the tunnel and distributing them evenly according to the preset monitoring accuracy, the coverage and accuracy of gas monitoring are ensured, and the comprehensiveness of the data is improved. By integrating traffic flow monitoring, construction activity monitoring and structural characteristics, the initial point distribution is dynamically adjusted so that the monitoring points can adapt to changes in actual conditions, thereby providing more reliable monitoring data. A dynamic pre-capture period is established to respond to real-time environmental changes, so that the system can be dynamically adjusted, enhancing the real-time and adaptability of the monitoring system. A multi-modal parent-child node is established based on the dynamic pre-capture period to update the point distribution results, which can process different types of data and improve the analysis ability and accuracy of complex environments. The long-term and short-term characteristics of gas concentration are extracted through multi-scale time series analysis, which can comprehensively capture the different time scales of gas concentration changes. The system can obtain detailed gas change trend data based on the degree of information, realize more accurate identification of gas concentration change patterns, and improve the accuracy and response speed of the early warning system; based on the gas concentration feature extraction results, the concentration warning analysis of independent points can be carried out, which can accurately warn at specific points, detect potential gas leaks or abnormal conditions early, and provide independent warning capabilities for each monitoring point; through the dependency identification network, the position concentration data is comprehensively analyzed, the dependency relationship between different points can be identified, and a comprehensive warning result can be generated, which effectively integrates the data of each point and enhances the accuracy of the overall warning; the gas sensing alarm processing is carried out based on the independent warning and dependent warning results, which provides a comprehensive alarm mechanism, enhances the response capability of the system, can effectively protect the safety in the tunnel, and reduce the possibility of accidents.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A schematic flow chart of a multi-point coordinated tunnel gas sensing alarm method provided in an embodiment of the present application;

[0010] Figure 2 A schematic diagram of the structure of a multi-point coordinated tunnel gas sensing alarm device provided in an embodiment of the present application.

[0011] Explanation of the accompanying drawings: uniform point distribution module 10, position compensation module 20, parent-child node establishment module 30, long and short feature extraction module 40, independent position warning module 50, dependent position warning module 60, alarm processing module 70. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a multi-point coordinated tunnel gas sensing alarm method and device, which solves the technical problems that the static monitoring point distribution in the prior art is difficult to adapt to dynamic environmental changes, resulting in inaccurate monitoring data, and a single early warning analysis method cannot fully reflect the actual situation of the gas, resulting in incomplete and inaccurate early warning results.

[0013] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0014] Embodiment 1, as Figure 1 As shown, the embodiment of the present application provides a multi-point coordinated tunnel gas sensing alarm method, the method comprising:

[0015] A distributed sensing optical fiber of the tunnel is established, and a preset monitoring accuracy of the tunnel is obtained. Points are evenly distributed according to the distributed sensing optical fiber and the preset monitoring accuracy, and an initial point distribution result is established.

[0016] The sensing optical fiber is laid inside the tunnel. The sensing optical fiber can be laid along the tunnel wall, tunnel roof, or other key locations to ensure that the entire monitoring area can be covered. The distributed sensing optical fiber laid in the tunnel can monitor the changes in gas concentration at different locations in the tunnel in real time through the optical fiber scattering signal. According to the use scenario of the tunnel, gas type, tunnel structure and other factors, the preset monitoring accuracy is determined, including the detection accuracy of gas concentration, the distance between sensor points, data collection frequency, etc.

[0017] Uniform point distribution is carried out according to the distributed sensing optical fiber and the preset monitoring accuracy. Uniform point distribution refers to the use of a uniform distribution method, that is, the sensing points are evenly distributed in the longitudinal and transverse directions of the tunnel to ensure full coverage of the monitoring area and meet the preset monitoring accuracy. The initial point distribution results are determined after analysis, including the location of each sensing point and the corresponding monitoring accuracy.

[0018] Carry out traffic flow monitoring and construction monitoring on the tunnel, establish traffic flow characteristics and construction activity characteristics, obtain structural characteristics of the tunnel, establish point correction parameters based on the traffic flow characteristics, construction activity characteristics and structural characteristics, perform position compensation on the initial point distribution result based on the point correction parameters, and establish an updated point distribution result.

[0019] Traffic flow monitoring equipment, such as traffic flow sensors, speed monitors and vehicle identification systems, are deployed in the tunnel. These devices can be installed at multiple key locations at the tunnel entrance, exit and inside to comprehensively capture the traffic flow information and vehicle type information in the tunnel. Traffic flow monitoring equipment can be used to collect traffic flow data in real time, including vehicle type, number of vehicles, speed, direction and other information. Based on the collected data, traffic flow characteristics are analyzed and established. These characteristics include traffic flow distribution, frequency of heavy-loaded vehicles, average speed, etc. These characteristics have a direct impact on the distribution of gas in the tunnel. For example, areas where heavy-loaded vehicles frequently pass through may have a greater impact on air flow.

[0020] When there is construction activity in the tunnel, construction monitoring equipment is deployed, such as vibration sensors, noise sensors and construction equipment operation status monitoring devices. These devices can monitor the impact of construction on the tunnel environment, such as the release of dust and gas and changes in air flow, and monitor data such as the intensity, duration, and construction location of construction activities in real time. This information can help identify the areas where construction affects the air quality inside the tunnel. Based on the collected construction data, the characteristics of construction activities are analyzed and established. The characteristics of construction activities include construction frequency, types of construction equipment, and the types and concentrations of gas pollutants they produce.

[0021] Obtain the structural design drawings and related data of the tunnel, determine the geometry of the tunnel, the layout of the ventilation system, the tunnel slope and other key structures in the tunnel, such as air vents, inspection wells, etc., and establish a structural characteristic model of the tunnel based on the structural design. This model can simulate the behavior of air flow, gas diffusion and accumulation in the tunnel.

[0022] Based on the characteristics of traffic flow, construction activities and tunnel structure, the impact of each feature on the gas concentration distribution in the tunnel is analyzed. For example, areas with heavy traffic require more dense monitoring points, while areas with frequent construction activities require enhanced monitoring during specific periods. Based on the above analysis results, point correction parameters are formulated, including density adjustment of monitoring points, adjustment of monitoring weights in specific areas, etc.

[0023] The correction parameters are applied to the initial point distribution results to adjust the point positions, including adding or reducing monitoring points, adjusting the positions of monitoring points, etc. After applying the correction parameters, an updated point distribution result is generated. The updated result shows the position and density distribution of the monitoring points after compensation adjustment, so that the monitoring system can more accurately and effectively capture the changes in gas concentration in the tunnel, thereby improving the response accuracy of the tunnel gas sensing alarm system.

[0024] A dynamic front capture period is established, and a parent-child node for updating the point distribution result is established based on the dynamic front capture period. The parent-child node has a parent-child association weight and is a multimodal parent-child node.

[0025] The dynamic pre-capture period is set up to perceive possible gas concentration anomalies in advance. The change rate characteristics, prediction error characteristics and abnormal characteristics are extracted from the historical monitoring data. For example, certain time periods in the historical data show sharp changes in gas concentration. These time periods need to be monitored in the dynamic pre-capture period. The length and start time of the pre-capture period are dynamically adjusted according to the matching of real-time data and historical characteristics. For example, when an increase in traffic volume or abnormal construction activities is detected, the capture process of certain monitoring points can be automatically started in advance.

[0026] Based on the updated point distribution results, a parent-child node network is constructed. In this network, each monitoring point is regarded as a node. According to the spatial and data correlation between it and other points, the association between the parent node and the child node is established. Since different types of data, including gas concentration, temperature, humidity, etc., have different influences in the prediction of gas concentration changes, these data categories can be divided into different modes. Specifically, the gas mode is used to process the concentration data of different types of gases and analyze the interaction of different gases and their impact on the tunnel air quality; the temperature mode is used to process temperature data and analyze the impact of temperature on gas diffusion and concentration; the humidity mode is used to process humidity data. The impact of humidity changes on the concentration of some gases may be greater, such as the impact of water vapor on the response sensitivity of gas sensors.

[0027] Based on multimodal fusion technology, the information of parent and child nodes under different modes is processed comprehensively, and different weights can be assigned to different modes. The size of the weight depends on the contribution of each mode to the prediction result. The fusion process can be realized through the multimodal network architecture in deep learning, such as inputting data of different modes at each node at the same time, and adaptively adjusting the impact of these inputs through network weights. Through multimodal fusion technology, multiple factors such as temperature, humidity, and gas type can be integrated to make gas concentration prediction and alarm more accurate and reliable, and enhance the expression ability and adaptability of the model.

[0028] The distributed sensing optical fiber is called to monitor the gas concentration according to the updated point distribution result, and the position concentration data is established. The length feature extraction of the gas concentration change in the position concentration data is performed through a multi-scale time series to establish the gas concentration feature extraction result.

[0029] According to the updated point distribution results, the distributed sensing optical fiber collects gas concentration data at different monitoring points. Each monitoring point corresponds to a specific geographical location. The collected gas concentration data is recorded with the position as the index to form position concentration data. The position concentration data is the gas concentration value at different monitoring points in the tunnel, reflecting the distribution of gas concentration in the tunnel space.

[0030] Multiscale time series analysis is a method for analyzing the changing characteristics of data at different time scales. By analyzing the changes in gas concentration at different time scales, long-term trends and short-term fluctuation characteristics can be extracted. Among them, long-term feature extraction is used to analyze the changing trend of gas concentration at a longer time scale, for example, to identify the seasonal change pattern of gas concentration or the long-term trend of increasing concentration; short-term feature extraction is used to analyze the rapid change characteristics of gas concentration at a short time scale, for example, to identify the sharp fluctuations in concentration caused by emergencies such as construction activities or accidental leaks. Aggregate the features at different time scales to form a comprehensive gas concentration feature extraction result, including but not limited to the average value, fluctuation amplitude, and change trend of gas concentration, so as to more comprehensively show the dynamic changes of gas concentration in the tunnel.

[0031] Based on the gas concentration feature extraction results, independent point concentration warning analysis is performed to establish independent position warning results.

[0032] Set warning thresholds. For example, when the concentration of a certain gas reaches a certain fixed value, a warning is triggered. Different thresholds can be set according to different gas types and tunnel environments. For each point, the current gas concentration characteristics are judged according to the warning threshold to determine whether there is an abnormal situation. When the gas concentration at a certain point is detected to exceed the normal range or show an abnormal change trend, an independent location warning result is generated, including warning level, warning time and location, warning type, etc.

[0033] A dependency identification network is configured according to the parent-child nodes, the position concentration data is input into the configured dependency identification network, and a dependency position warning result is generated.

[0034] Based on the parent-child node configuration dependency identification network, this network model aims to identify possible abnormal situations by analyzing the dependency relationship between the parent node and the child node, and input the acquired gas concentration data of each monitoring point into the configured dependency identification network. In the network, the state of each point is represented by a state vector, which contains a variety of features related to the gas concentration. For different time steps, potential abnormal situations are captured by the change of the state vector.

[0035] The dependency recognition network analyzes dependencies by calculating the conditional probability of each child node under a given parent node state. This process can identify whether the abnormality of a child node is affected by the abnormality of the parent node. For example, if the gas concentration of a parent node increases abnormally, the network will analyze whether this increase will cause abnormal conditions in the child node. Based on the conditional probability between parent and child nodes, the dependency weights are dynamically adjusted. In this way, the network can adaptively strengthen or weaken the dependency between specific parent and child nodes, thereby improving the accuracy of the early warning results.

[0036] After combining the relationship between parent and child nodes, the dependency identification network conducts a joint analysis of the input gas concentration data to generate a dependent position warning result. This result not only takes into account the abnormal situation of a single point, but also considers the correlation between multiple points. It can identify potential risks that cannot be detected by independent point analysis, and provide a more comprehensive risk assessment.

[0037] Gas sensing alarm processing of the tunnel is performed according to the independent position warning result and the dependent position warning result.

[0038] The independent position warning results are integrated with the dependent position warning results, for example, by using weighted average, confidence score and other methods to determine the final warning level and response plan, and to process the gas sensing alarm in the tunnel. Specifically, according to the severity of the warning results, the scope of influence and other factors, the corresponding alarm level is set, and for different warning levels, the corresponding response strategy is selected. For example, a low-level warning only needs to remind the staff to pay attention, while a high-level warning requires the immediate initiation of emergency measures, such as evacuating vehicles and personnel in the tunnel, closing the tunnel entrance, starting the ventilation system, etc., to improve the overall safety level of the tunnel.

[0039] Furthermore, the extracting of the length characteristics of the gas concentration change in the position concentration data through the multi-scale time series and establishing the gas concentration characteristic extraction result also includes:

[0040] A scale perception model is established, wherein the scale perception model includes an equilibrium point extraction layer and a fluctuation perception layer; equilibrium value extraction of position concentration data is performed through the equilibrium point extraction layer, an equilibrium value extraction result is established, and the equilibrium value extraction result is synchronized to the fluctuation perception layer; a perception zero point of the fluctuation perception layer is established based on the equilibrium value extraction result, fluctuation perception of position concentration data is performed based on the perception zero point, and a time series scale set is output based on the fluctuation perception result; a multi-scale time series is established with the time series scale set to extract the long and short features of gas concentration changes in the position concentration data.

[0041] The core purpose of the scale-aware model is to conduct in-depth analysis of gas concentration data through different analysis levels, identify the change pattern of gas concentration, and extract features at multiple time and space scales. The scale-aware model includes an equilibrium point extraction layer and a fluctuation perception layer. These two layers work together to perform multi-level and multi-dimensional feature extraction of gas concentration data.

[0042] Among them, the equilibrium point extraction layer is used to extract the equilibrium value from the position concentration data, that is, the average or equilibrium state of the gas concentration in a certain period of time. By extracting the equilibrium point, the stable state information of the concentration data can be obtained, and the basic change trend of the gas concentration can be determined. This layer can use statistical analysis methods, such as moving average, to calculate the equilibrium value of the gas concentration at each point. The equilibrium point extraction process uses different time windows to capture short-term and long-term concentration changes; the fluctuation perception layer further analyzes the fluctuation characteristics of the gas concentration based on the equilibrium point extraction. It can identify the changes in gas concentration that deviate from the equilibrium value, thereby detecting anomalies or emergencies. The fluctuation perception layer can use signal processing technology or time series analysis methods to identify the fluctuation patterns of the concentration data. These fluctuation patterns can reflect the characteristics of sudden changes and periodic fluctuations in gas concentration.

[0043] In the equilibrium point extraction layer, equilibrium values ​​are extracted from the gas concentration data at each point. The equilibrium value extraction includes time averaging the concentration data to obtain the average gas concentration level of each point in different time periods, and establish the equilibrium value extraction results. These equilibrium values ​​can reflect the overall trend and changes of the gas concentration in the tunnel.

[0044] The equilibrium value extraction results are synchronized to the fluctuation perception layer, so that the fluctuation perception layer can perform fluctuation analysis based on the latest equilibrium value data. The synchronized equilibrium value acts as a baseline in the fluctuation perception layer. The fluctuation perception layer identifies abnormal fluctuations in gas concentration based on this baseline, which can effectively improve the accuracy of fluctuation detection and reduce the false alarm rate.

[0045] The perception zero point is the reference point used in the fluctuation perception layer to evaluate concentration fluctuations. This reference point is determined based on the equilibrium value extraction result. The fluctuation perception layer uses the perception zero point as a reference to identify the fluctuation characteristics of gas concentration by comparing the deviation between the actual concentration data and the perception zero point, so as to accurately capture sudden changes or periodic fluctuations in gas concentration. For example, when the gas concentration data at a certain moment is significantly higher than the perception zero point, it is identified as an abnormal concentration fluctuation, and the corresponding fluctuation perception result is output.

[0046] Based on the fluctuation perception results, a time series scale set is output. The time series scale set contains the gas concentration fluctuation information at different time scales. Each time series scale corresponds to a fluctuation pattern or a specific concentration change period. Multi-scale analysis methods such as wavelet transform and Fourier transform are used to decompose the gas concentration data into multiple time series scale sets. Each scale represents the fluctuation characteristics under different time windows or frequencies.

[0047] Multiscale time series describes the complex pattern of gas concentration changes over time by combining multiple time series scale sets. It can capture both short-term fluctuations and long-term trends and provide comprehensive concentration change information. Specifically, based on the time series scale set, the concentration fluctuation information at different time scales is integrated into a multiscale time series. This time series can be constructed by weighted averaging, filtering or other time series processing methods to reflect the overall change of gas concentration.

[0048] Using time series analysis techniques, such as trend analysis, cycle analysis or burst detection, we can deeply explore the multi-scale time series and identify the long and short features in the changes of gas concentration, namely long-term trends and short-term fluctuations. These features are used to identify potential gas leakage risks or other abnormal situations in the tunnel.

[0049] Specifically, the multi-scale time series is as follows:

[0050] ;

[0051] in, Characterize the original time series signal, Represents a point in time, is the wavelet function, is the time scale, is the displacement parameter, which is used to control the movement of the wavelet function on the time axis. is the wavelet coefficient, Characterization function With time series signal The convolution result is used to measure the time series scale Upward displacement The characteristic strength of is a function that characterizes scale and displacement.

[0052] Specifically, the multi-scale time series is as follows:

[0053] ;

[0054] is a wavelet function used to decompose time series. Specifically, the wavelet function It is the wavelet basis function at scale a and displacement b. It is the core of multi-scale analysis and is used to capture signal characteristics at different scales and positions. a is the time series scale, which determines the degree of expansion of the wavelet function. A larger a represents a longer time scale and can capture slower changes. b is the displacement parameter, which is used to control the movement of the wavelet function on the time axis and determine the position of the wavelet function application.

[0055] Represents the wavelet coefficient, reflecting the wavelet function at a specific scale a and displacement b. With time series signal The convolution result quantifies the characteristic strength of the wavelet function on the time series signal. It is the result of wavelet transform and is used to measure the characteristics of the signal at a specific scale and position.

[0056] In this way, local and global features of time series can be extracted at different time scales, thereby identifying trends, periodic changes, and abnormal patterns of signals.

[0057] Furthermore, the establishment of the parent-child nodes for updating the point distribution result based on the dynamic front capture period also includes:

[0058] Obtain the historical data change rate characteristics, the prediction error characteristics and the abnormal characteristics of the time series; determine the pre-capture period according to the historical data change rate characteristics, the prediction error characteristics and the abnormal characteristics; configure the period data with the pre-capture period, and calculate the mutual information based on the period data; establish the parent-child nodes according to the mutual information calculation result and the weight update formula, as follows:

[0059] ;

[0060] Representing the next moment Next, from the node To Node The updated weights of Representing the current moment Next, from the node To Node The weight of Represents the learning rate, Characterizing the loss function About weight The gradient of Representation loss function, used to calculate the actual observation value and the model prediction value of error.

[0061] The historical data change rate feature describes the change in gas concentration over a period of time in the past, and is used to identify the changing trend or rate of gas concentration. Specifically, the historical gas concentration data is analyzed to calculate its change rate in different time periods. For example, the difference calculation method is used to find the concentration difference between each time point and the previous time point, and the change rate features are extracted, such as the average change rate, maximum change rate, minimum change rate, etc. These features are used to determine the fluctuation of gas concentration.

[0062] The prediction error characteristics of time series are used to evaluate the accuracy of the prediction and compare the differences between the predicted values ​​and the actual observed values. Specifically, a prediction model is established using historical data, and future gas concentration values ​​are predicted based on the long short-term memory network. The predicted values ​​are compared with the actual observed values, and the prediction error is calculated. The error metrics include mean square error, mean absolute error, etc. The prediction error characteristics are extracted, such as the mean and variance of the error, to determine the reliability and stability of the model prediction.

[0063] Anomaly features are used to detect anomalies in gas concentration data. Specifically, anomaly detection algorithms, such as statistical-based methods and machine learning methods, are applied to identify outliers in gas concentration data and extract features of abnormal data, such as the frequency, amplitude, and duration of the outliers, to identify abnormal patterns and potential risks.

[0064] Comprehensive analysis is performed using the historical data change rate characteristics, prediction error characteristics, and anomaly characteristics to identify key time windows, which are representative and can reflect the trend and abnormal pattern of gas concentration changes. Specifically, based on the change rate characteristics, time periods with significant change rates are selected to capture the law of gas concentration fluctuations; based on the prediction error characteristics, time periods with large prediction model errors are selected to ensure that the impact of inaccurate predictions is captured; based on the anomaly characteristics, time periods with high anomaly occurrence frequencies are selected to focus on the early characteristics of abnormal situations. Combined with the above characteristics, a suitable pre-capture period is determined to more accurately identify and respond to potential gas concentration anomalies.

[0065] During the pre-capture period, data related to gas concentration is collected to establish period data. Mutual information is a measure of the correlation between two random variables, which is used to determine the amount of information in the period data and its contribution to the prediction. Specifically, the mutual information formula is used to calculate the amount of information between different variables in the period data. For two random variables X and Y, the mutual information calculation formula is as follows:

[0066] ;

[0067] in, is the joint probability distribution of X and Y, and is the marginal probability distribution of each. The mutual information calculation results are used to evaluate the correlation between variables.

[0068] The parent-child nodes are established according to the mutual information calculation results and the weight update formula as follows:

[0069] ;

[0070] The learning rate is a hyperparameter that controls the adjustment amplitude of the weights at each update. A larger learning rate will result in a larger weight update amplitude, while a smaller learning rate will result in a smaller update amplitude. The choice of learning rate will affect the convergence speed and stability of the model.

[0071] The weight update is achieved by subtracting the gradient and multiplying it by the learning rate. The direction of the gradient represents the steepest descent direction of the loss function at the current weight, so adjusting the weight along the gradient direction can effectively reduce the error. If the gradient of the loss function is large on a certain weight, then the update amplitude will also be large, so that the weight can be adjusted quickly; conversely, the update amplitude is small and the adjustment is slow. This formula optimizes the model's prediction ability by gradually updating the weights. By calculating the gradient of the loss function and updating the weights in combination with the learning rate, the prediction error can be gradually reduced, making the model's prediction results at the next moment more accurate.

[0072] Furthermore, the configuring dependency identification network according to the parent-child nodes further includes:

[0073] The dependency identification network is as follows:

[0074] ;

[0075] Represents the conditional probability, that is, given the previous time step Under the condition of The probability of Characterization time The state vector of all nodes under Characterization time The state vector of all nodes under Indicates time Next The state of the node, Characterization The state of the parent node set of a node, After characterizing the state of a given parent node, Nodes at time The conditional probability of .

[0076] The dependency identification network is as follows:

[0077] ;

[0078] Represents the conditional probability, that is, given the previous time step Under the condition of This conditional probability describes the probability distribution of the states of all nodes at the current moment after knowing the states of all nodes at the previous moment.

[0079] After characterizing the state of a given parent node, Nodes at time The conditional probability under describes how the state of the i-th node depends on the state of its parent node. This is a local conditional probability and is used to build a global model for the entire time series.

[0080] The dependency recognition network represents the node state at the current time step in the time series model. How does the probability of depend on the state at the previous time step? , through the conditional probability Based on the state of the parent node of the i-th node, its state at the current time step can be quantified to predict the state of the node in the future time step.

[0081] Furthermore, the method further comprises:

[0082] Obtain an alarm response level; if the alarm response level meets a preset level threshold, generate a closing response instruction; perform entrance vehicle detection at the tunnel entrance through the closing response instruction, and generate a closing strategy based on the entrance vehicle detection result; perform closing management on the tunnel entrance based on the closing strategy, and synchronously generate an acoustic and visual coordinated warning signal.

[0083] According to the severity of the warning results, the scope of impact and other factors, set the corresponding alarm response level, such as low, medium and high. The specific evaluation criteria are based on the system design and application requirements.

[0084] A preset level threshold is set according to actual conditions to trigger corresponding response measures. The obtained alarm response level is compared with the preset level threshold. If the alarm response level meets or exceeds the preset level threshold, a closing response instruction is automatically generated. This instruction is used to initiate the closure measures of the tunnel entrance to prevent further risk spread.

[0085] Based on the closing response instruction, the vehicles entering and leaving the tunnel are detected at the entrance to ensure that the vehicles can be safely guided out of the tunnel before the closing measures take effect. The detection can be achieved through sensors, cameras or other detection equipment. Based on the entrance vehicle detection results, a closing strategy is generated, which may include vehicle evacuation, closing time arrangement, warning sign setting, etc., to decide how to implement the closed management of the tunnel entrance. According to the closing strategy, actual closed management measures are implemented, including physical closing of the tunnel entrance, setting of closed signs, and control of traffic flow at the tunnel entrance. At the same time, sound and light coordinated early warning signals are generated to remind people inside and outside the tunnel about the implementation of the closing measures. These signals include alarm sounds, flashing lights, etc., which are used to improve the warning effect and ensure the safety of personnel.

[0086] Furthermore, the independent point concentration warning analysis is performed based on the gas concentration feature extraction result to establish an independent position warning result, and further includes:

[0087] Establish concentration threshold warning and concentration change warning; trigger verification of gas concentration feature extraction results through the concentration threshold warning and concentration change warning, and establish independent position warning results based on the trigger verification results.

[0088] Concentration threshold warning is used to detect whether the gas concentration exceeds the predetermined safety limit. According to the safety standards and environmental characteristics of different gases in the tunnel, multiple thresholds are set. For example, low, medium and high concentration thresholds are set to mark different warning levels. Concentration change warning is used to detect the change rate or trend of gas concentration within a certain time window, identify abnormal change patterns, and determine the rate or amplitude threshold of concentration change to trigger an early warning. For example, when the concentration changes in a short period of time and exceeds the set change threshold, the concentration change warning is triggered.

[0089] According to the concentration threshold warning and concentration change warning, the gas concentration feature extraction results are triggered and verified, that is, to check whether the gas concentration exceeds the threshold or has undergone abnormal changes. Each monitoring point is judged separately. If it exceeds the threshold or the change rate is abnormal, it is judged whether there is an abnormal situation at that location. Based on the trigger verification results, an independent warning status is generated for each location, and an independent location warning result is established to help operators quickly identify and respond to areas with abnormal gas concentration.

[0090] In summary, the multi-point coordinated tunnel gas sensing alarm method provided by the embodiment of the present application has the following technical effects:

[0091] By laying distributed sensing optical fibers in the tunnel and distributing them evenly according to the preset monitoring accuracy, the coverage and accuracy of gas monitoring are ensured, and the comprehensiveness of the data is improved. By integrating traffic flow monitoring, construction activity monitoring and structural characteristics, the initial point distribution is dynamically adjusted so that the monitoring points can adapt to changes in actual conditions, thereby providing more reliable monitoring data. A dynamic pre-capture period is established to respond to real-time environmental changes, so that the system can be dynamically adjusted, enhancing the real-time and adaptability of the monitoring system. A multi-modal parent-child node is established based on the dynamic pre-capture period to update the point distribution results, which can process different types of data and improve the analysis ability and accuracy of complex environments. The long-term and short-term characteristics of gas concentration are extracted through multi-scale time series analysis, which can comprehensively capture the different time scales of gas concentration changes. The system can obtain detailed gas change trend data based on the degree of information, realize more accurate identification of gas concentration change patterns, and improve the accuracy and response speed of the early warning system; based on the gas concentration feature extraction results, the concentration warning analysis of independent points can be carried out, which can accurately warn at specific points, detect potential gas leaks or abnormal conditions early, and provide independent warning capabilities for each monitoring point; through the dependency identification network, the position concentration data is comprehensively analyzed, the dependency relationship between different points can be identified, and a comprehensive warning result can be generated, which effectively integrates the data of each point and enhances the accuracy of the overall warning; the gas sensing alarm processing is carried out based on the independent warning and dependent warning results, which provides a comprehensive alarm mechanism, enhances the response capability of the system, can effectively protect the safety in the tunnel, and reduce the possibility of accidents.

[0092] Embodiment 2, based on the same inventive concept as the multi-point coordinated tunnel gas sensing alarm method in the above embodiment, Figure 2 As shown, the embodiment of the present application provides a multi-point coordinated tunnel gas sensing alarm device, the device comprising:

[0093] A uniform point distribution module 10, the uniform point distribution module 10 is used to establish a distributed sensing optical fiber of the tunnel, and obtain a preset monitoring accuracy of the tunnel, perform uniform point distribution according to the distributed sensing optical fiber and the preset monitoring accuracy, and establish an initial point distribution result; a position compensation module 20, the position compensation module 20 is used to perform traffic flow monitoring and construction monitoring on the tunnel, establish traffic flow characteristics and construction activity characteristics, obtain the structural characteristics of the tunnel, establish point correction parameters based on the traffic flow characteristics, construction activity characteristics, and structural characteristics, perform position compensation on the initial point distribution result based on the point correction parameters, and establish an updated point distribution result; a parent-child node establishment module 30, the parent-child node establishment module 30 is used to establish a dynamic pre-capture period, and establish a parent-child node for updating the point distribution result based on the dynamic pre-capture period, and the parent-child node has a parent-child association weight and is a multi-modal parent-child Node; a long-short feature extraction module 40, the long-short feature extraction module 40 is used to call the distributed sensing optical fiber to monitor the gas concentration according to the updated point distribution result, establish the position concentration data, extract the long-short features of the gas concentration change in the position concentration data through a multi-scale time series, and establish the gas concentration feature extraction result; an independent position warning module 50, the independent position warning module 50 is used to perform independent point concentration warning analysis based on the gas concentration feature extraction result, and establish an independent position warning result; a dependent position warning module 60, the dependent position warning module 60 is used to configure a dependent identification network according to the parent-child nodes, input the position concentration data into the configured dependent identification network, and generate a dependent position warning result; an alarm processing module 70, the alarm processing module 70 is used to perform gas sensing alarm processing in the tunnel according to the independent position warning result and the dependent position warning result.

[0094] Furthermore, the device also includes a length feature extraction module to perform the following operation steps:

[0095] A scale perception model is established, wherein the scale perception model includes an equilibrium point extraction layer and a fluctuation perception layer; equilibrium value extraction of position concentration data is performed through the equilibrium point extraction layer, an equilibrium value extraction result is established, and the equilibrium value extraction result is synchronized to the fluctuation perception layer; a perception zero point of the fluctuation perception layer is established based on the equilibrium value extraction result, fluctuation perception of position concentration data is performed based on the perception zero point, and a time series scale set is output based on the fluctuation perception result; a multi-scale time series is established with the time series scale set to extract the long and short features of gas concentration changes in the position concentration data.

[0096] Specifically, the multi-scale time series is as follows:

[0097] ;

[0098] in, Characterize the original time series signal, Represents a point in time, is the wavelet function, is the time scale, is the displacement parameter, which is used to control the movement of the wavelet function on the time axis. is the wavelet coefficient, Characterization function With time series signal The convolution result is used to measure the time series scale Upward displacement The characteristic strength of is a function that characterizes scale and displacement.

[0099] Furthermore, the device further includes a parent-child node establishment module to perform the following operation steps:

[0100] Obtain the historical data change rate characteristics, the prediction error characteristics and the abnormal characteristics of the time series; determine the pre-capture period according to the historical data change rate characteristics, the prediction error characteristics and the abnormal characteristics; configure the period data with the pre-capture period, and calculate the mutual information based on the period data; establish the parent-child nodes according to the mutual information calculation result and the weight update formula, as follows:

[0101] ;

[0102] Representing the next moment Next, from the node To Node The updated weights of Representing the current moment Next, from the node To Node The weight of Represents the learning rate, Characterizing the loss function About weight The gradient of Representation loss function, used to calculate the actual observation value and the model prediction value of error.

[0103] Specifically, the dependency identification network is as follows:

[0104] ;

[0105] Represents the conditional probability, that is, given the previous time step Under the condition of The probability of Characterization time The state vector of all nodes under Characterization time The state vector of all nodes under Indicates time Next The state of the node, Characterization The state of the parent node set of a node, After characterizing the state of a given parent node, Nodes at time The conditional probability of .

[0106] Furthermore, the device also includes a collaborative early warning module to perform the following operation steps:

[0107] Obtain an alarm response level; if the alarm response level meets a preset level threshold, generate a closing response instruction; perform entrance vehicle detection at the tunnel entrance through the closing response instruction, and generate a closing strategy based on the entrance vehicle detection result; perform closing management on the tunnel entrance based on the closing strategy, and synchronously generate an acoustic and visual coordinated warning signal.

[0108] Furthermore, the device also includes an independent location warning result establishment module to perform the following operation steps:

[0109] Establishing concentration threshold warning and concentration change warning; performing trigger verification of gas concentration feature extraction results through the concentration threshold warning and concentration change warning, and establishing independent position warning results based on the trigger verification results.

[0110] Through the above-mentioned detailed description of a multi-point coordinated tunnel gas sensing alarm method in this specification, technical personnel in this field can clearly understand a multi-point coordinated tunnel gas sensing alarm device in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0111] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-point coordinated tunnel gas sensing and alarm method, characterized in that: The method comprises: Establishing a distributed sensing fiber in the tunnel and obtaining a preset monitoring accuracy for the tunnel; performing a uniform point distribution based on the distributed sensing fiber and the preset monitoring accuracy; and establishing an initial point distribution result, wherein the uniform point distribution refers to distributing sensing points evenly in the longitudinal and transverse directions of the tunnel in a uniform manner; and the initial point distribution result includes the position of each sensing point and the corresponding monitoring accuracy; Carry out traffic flow monitoring and construction monitoring on the tunnel, establish traffic flow characteristics and construction activity characteristics, obtain the structural characteristics of the tunnel, establish point correction parameters based on the traffic flow characteristics, construction activity characteristics, and structural characteristics, perform position compensation on the initial point distribution result based on the point correction parameters, and establish an updated point distribution result, wherein the traffic flow characteristics include traffic volume distribution, heavy vehicle frequency, and average vehicle speed, and the construction activity characteristics include construction frequency, type of construction equipment, and type and concentration of gaseous pollutants generated by the construction equipment. Acquiring the structural characteristics of the tunnel refers to determining the geometry of the tunnel, ventilation system layout, tunnel slope, and other key structures in the tunnel by obtaining the structural design drawings and related data of the tunnel, and establishing a structural characteristic model of the tunnel according to the structural design. The point correction parameters include density adjustment of monitoring points and monitoring weight adjustment of specific areas. The position compensation refers to applying the correction parameters to the initial point distribution result to adjust the point position, including adding or reducing monitoring points and adjusting the position of monitoring points. Establish a dynamic pre-capture period, and establish a parent-child node based on the dynamic pre-capture period to update the point distribution result. The parent-child node has a parent-child association weight and is a multimodal parent-child node. The dynamic pre-capture period perceives the gas concentration anomaly in advance, extracts the change rate feature, prediction error feature and anomaly feature from the historical monitoring data, and some time periods in the historical data show a sharp change in gas concentration. These time periods need to be monitored in the dynamic pre-capture period. According to the matching of real-time data and historical features, the length and start time of the pre-capture period are dynamically adjusted to complete the establishment of the capture period. The parent-child node refers to treating each monitoring point as a node, and establishing the association between the parent node and the child node based on the spatial and data correlation between it and other points. The multimodal parent-child node refers to the fact that different types of data have different influences on the prediction of gas concentration changes. These data categories are divided into different modes for correlation construction to obtain multimodal parent-child nodes; Calling the distributed sensing optical fiber to monitor gas concentration according to the updated point distribution result, establishing position concentration data, extracting the long and short features of the gas concentration change in the position concentration data through a multi-scale time series, and establishing a gas concentration feature extraction result, wherein the position concentration data is the gas concentration value at different monitoring points in the tunnel, the long and short feature extraction includes extracting long-term trends and short-term fluctuation features, the long-term feature extraction is used to analyze the changing trend of the gas concentration on a longer time scale, and the short-time feature extraction is used to analyze the rapid changing characteristics of the gas concentration on a short time scale, and the gas concentration feature extraction result includes the average value, fluctuation amplitude and changing trend of the gas concentration; Based on the gas concentration feature extraction results, independent point concentration warning analysis is performed to establish independent position warning results. The independent point concentration warning analysis refers to setting a warning threshold. For each point, the current gas concentration feature is judged according to the warning threshold to determine whether there is an abnormality. The independent position warning result includes the warning level, warning time and location, and warning type. Configuring a dependency identification network according to the parent-child nodes, inputting the position concentration data into the configured dependency identification network to generate a dependency position warning result, wherein the dependency identification network identifies possible abnormal conditions by analyzing the dependency relationship between the parent node and the child node, and the dependency identification network performs a joint analysis on the input gas concentration data after combining the relationship between the parent and child nodes to generate a dependency position warning result; Gas sensing alarm processing is performed in the tunnel based on the independent position warning results and the dependent position warning results, including: integrating the independent position warning results with the dependent position warning results, using a weighted average or confidence scoring method to determine the final warning level and response plan, and performing gas sensing alarm processing in the tunnel.

2. A multi-point coordinated tunnel gas sensing and alarm method according to claim 1, characterized in that: The extracting of the length feature of the gas concentration change in the position concentration data by using the multi-scale time series to establish the gas concentration feature extraction result also includes: Establishing a scale perception model, wherein the scale perception model includes an equilibrium point extraction layer and a fluctuation perception layer; Extracting equilibrium values ​​of position concentration data through the equilibrium point extraction layer, establishing equilibrium value extraction results, and synchronizing the equilibrium value extraction results to the fluctuation perception layer; Establishing a perception zero point of the fluctuation perception layer based on the equilibrium value extraction result, performing fluctuation perception of position concentration data based on the perception zero point, and outputting a time series scale set based on the fluctuation perception result; A multi-scale time series is established with the time series scale set to extract the long and short features of the gas concentration changes in the position concentration data.

3. The multi-point coordinated tunnel gas sensing and alarm method according to claim 1, characterized in that: The establishment of the parent-child nodes for updating the point distribution result based on the dynamic front capture period also includes: Obtain historical data change rate characteristics, time series prediction error characteristics, and abnormal characteristics; Determining a pre-capture period based on the historical data change rate characteristics, prediction error characteristics, and abnormal characteristics; configuring period data with the pre-capture period, and calculating mutual information based on the period data; The parent-child nodes are established based on the mutual information calculation results and the weight update formula as follows: ； Representing the next moment Next, from the node To Node The updated weights of Representing the current moment Next, from the node To Node The weight of Represents the learning rate, Representation loss function About weight The gradient, Representation loss function, used to calculate the actual observation value and the model prediction value error.

4. The multi-point coordinated tunnel gas sensing and alarm method according to claim 3, characterized in that: The configuration of the dependency identification network according to the parent-child nodes further includes: The dependency identification network is as follows: ； Represents the conditional probability, that is, given the previous time step Under the condition of The probability of Representation time The state vectors of all nodes under Representation time The state vectors of all nodes under Indicates time Next The status of the node, Characterization The state of the parent node set of each node, After characterizing the state of a given parent node, the Nodes at time The conditional probability under .

5. The multi-point coordinated tunnel gas sensing and alarm method according to claim 1, characterized in that: The method further comprises: Get the alarm response level; If the alarm response level meets the preset level threshold, a shutdown response instruction is generated; The closing response instruction is used to detect the entrance vehicle of the tunnel, and a closing strategy is generated based on the entrance vehicle detection result. The tunnel entrance is closed and managed based on the closing strategy, and an acoustic and optical coordinated warning signal is generated synchronously.

6. The multi-point coordinated tunnel gas sensing and alarm method according to claim 1, characterized in that: The performing of independent point concentration warning analysis based on the gas concentration feature extraction result and establishing an independent position warning result further includes: Establish concentration threshold warning and concentration change warning; The trigger verification of the gas concentration feature extraction result is performed through the concentration threshold warning and concentration change warning, and an independent position warning result is established according to the trigger verification result.

7. A multi-point coordinated tunnel gas sensing alarm device, characterized in that: A device for implementing a multi-point coordinated tunnel gas sensing and alarm method according to any one of claims 1 to 6, comprising: A uniform point distribution module, which is used to establish a distributed sensing fiber in the tunnel and obtain a preset monitoring accuracy for the tunnel. The module then performs uniform point distribution based on the distributed sensing fiber and the preset monitoring accuracy to establish an initial point distribution result. The uniform point distribution refers to evenly distributing sensing points in the longitudinal and transverse directions of the tunnel. The initial point distribution result includes the position of each sensing point and its corresponding monitoring accuracy. A position compensation module, the position compensation module is used to perform traffic flow monitoring and construction monitoring on the tunnel, establish traffic flow characteristics and construction activity characteristics, obtain the structural characteristics of the tunnel, establish point correction parameters based on the traffic flow characteristics, construction activity characteristics, and structural characteristics, perform position compensation on the initial point distribution result based on the point correction parameters, and establish an updated point distribution result, wherein the traffic flow characteristics include traffic flow distribution, heavy vehicle frequency, and average vehicle speed, and the construction activity characteristics include construction frequency, type of construction equipment, and type and concentration of gaseous pollutants generated by the construction equipment. Acquiring the structural characteristics of the tunnel refers to determining the geometry of the tunnel, ventilation system layout, tunnel slope, and other key structures in the tunnel by obtaining the structural design drawings and related data of the tunnel, and establishing a structural characteristic model of the tunnel according to the structural design. The point correction parameters include density adjustment of monitoring points and monitoring weight adjustment of specific areas. The position compensation refers to applying the correction parameters to the initial point distribution result to adjust the point positions, including adding or reducing monitoring points and adjusting the positions of monitoring points. A parent-child node establishment module, which is used to establish a dynamic pre-capture period, and establish a parent-child node based on the dynamic pre-capture period to update the point distribution result. The parent-child node has a parent-child association weight and is a multimodal parent-child node. The dynamic pre-capture period perceives the gas concentration anomaly in advance, extracts the change rate feature, prediction error feature and anomaly feature from the historical monitoring data, and some time periods in the historical data show a sharp change in gas concentration. These time periods need to be monitored in the dynamic pre-capture period. According to the matching of real-time data and historical features, the length and start time of the pre-capture period are dynamically adjusted to complete the establishment of the capture period. The parent-child node refers to treating each monitoring point as a node, and establishing the association between the parent node and the child node based on the spatial and data correlation between it and other points. The multimodal parent-child node refers to the fact that different types of data have different impacts on the prediction of gas concentration changes. These data categories are divided into different modes for correlation construction to obtain multimodal parent-child nodes; A long-short feature extraction module, which is used to call the distributed sensing optical fiber to monitor gas concentration according to the updated point distribution result, establish position concentration data, extract the long-short features of the gas concentration change in the position concentration data through a multi-scale time series, and establish a gas concentration feature extraction result, wherein the position concentration data is the gas concentration value at different monitoring points in the tunnel. The long-short feature extraction includes extracting long-term trends and short-term fluctuation features. The long-term feature extraction is used to analyze the changing trend of gas concentration on a longer time scale, and the short-time feature extraction is used to analyze the rapid changing characteristics of gas concentration on a short time scale. The gas concentration feature extraction result includes the average value, fluctuation amplitude and changing trend of gas concentration; An independent location warning module, which is used to perform independent point concentration warning analysis based on the gas concentration feature extraction results and establish an independent location warning result. The independent point concentration warning analysis refers to setting a warning threshold. For each point, the current gas concentration feature is judged according to the warning threshold to determine whether there is an abnormality. The independent location warning result includes the warning level, warning time and location, and warning type; A dependent position warning module, configured to configure a dependent identification network based on the parent-child nodes, input the position concentration data into the configured dependent identification network, and generate a dependent position warning result, wherein the dependent identification network identifies possible abnormal conditions by analyzing the dependency relationship between the parent node and the child node. The dependent identification network, after combining the relationship between the parent and child nodes, performs a joint analysis on the input gas concentration data to generate a dependent position warning result; An alarm processing module is used to perform gas sensing alarm processing in the tunnel based on the independent position warning results and the dependent position warning results, including: integrating the independent position warning results with the dependent position warning results, using a weighted average or confidence scoring method to determine the final warning level and response plan, and performing gas sensing alarm processing in the tunnel.

Images