A Comprehensive Safety Analysis and Early Warning Method and System for Tunnels Based on Multi-Source Data Fusion

Abstract

This invention proposes a comprehensive tunnel safety analysis and early warning method and system based on multi-source data fusion, belonging to the field of tunnel safety early warning technology. Addressing the problems of insufficient early warning accuracy in existing technologies, lack of comprehensive consideration of secondary disasters related to the accumulation of harmful gases and decrease in oxygen concentration, and inability to support critical decision-making during time-sensitive emergency rescue operations, this invention employs a comprehensive approach that considers the spatiotemporal fusion of multi-source heterogeneous data, a risk prediction model integrating an attention mechanism, and dynamic modeling of the tunnel gas environment. This improves the accuracy and comprehensiveness of early warning based solely on structural risks. Furthermore, through a matching mechanism between construction activities and key driving forces, risk probability time series prediction, and dynamic path planning incorporating gas safety time, it achieves accurate judgment of construction safety windows and real-time optimization of personnel evacuation routes, improving the timeliness and reliability of early warning decisions.

Description

Technical Field

[0001] This invention relates to the field of tunnel safety early warning technology, and in particular to a tunnel comprehensive safety analysis and early warning method and system based on multi-source data fusion. Background Technology

[0002] In recent years, with the rapid development of sensing technology, the Internet of Things, and big data analytics, tunnel comprehensive safety analysis and early warning methods based on multi-source data fusion have become a hot research topic and an inevitable trend in the industry. These methods integrate heterogeneous data from sources such as ground-penetrating radar, stress sensors, displacement sensors, and video surveillance to achieve collaborative perception and comprehensive analysis of multi-dimensional safety conditions, including tunnel structural stability and abnormal construction activities. In particular, the introduction of artificial intelligence models such as machine learning and deep learning has made it possible to automatically mine potential risk patterns from massive, high-dimensional monitoring data and achieve early warnings, demonstrating enormous development potential and broad application prospects.

[0003] However, existing technologies lack sufficient early warning accuracy when facing maintenance and construction during tunnel operation or emergency repair and construction in case of emergencies. This results in a lack of attribution of the causes of safety risks during emergency repair and construction activities. In the confined space and time-sensitive maintenance or emergency environment, there is a lack of comprehensive consideration of secondary disasters related to the accumulation of harmful gases and the decrease in oxygen concentration. Due to complex factors such as equipment stacking and narrow working areas at the emergency site, existing technologies do not adequately consider the evacuation time dimension and cannot provide safety analysis and early warning to support critical decision-making in time-sensitive emergency rescue processes.

[0004] Therefore, developing a comprehensive tunnel safety analysis and early warning method and system based on multi-source data fusion with higher accuracy, more dimensions, and the ability to guide construction operations is of great significance to tunnel safety. Summary of the Invention

[0005] To address these issues, this invention provides a tunnel integrated safety analysis and early warning method and system based on multi-source data fusion. This system overcomes the problems of insufficient early warning accuracy, lack of comprehensive consideration of secondary disasters related to the accumulation of harmful gases and the decrease in oxygen concentration in existing technologies, insufficient consideration of the evacuation time dimension, and inability to provide safety analysis and early warning to support critical decision-making in time-sensitive emergency rescue processes.

[0006] To achieve the above objectives, this invention provides a tunnel integrated safety analysis and early warning method based on multi-source data fusion, comprising: S1, acquire multi-source heterogeneous data of the tunnel, and perform spatiotemporal registration and fusion processing on the multi-source heterogeneous data to generate a unified spatiotemporal reference dataset; S2, input the spatiotemporal reference dataset into the safety risk prediction model, and output the comprehensive safety risk probability and prediction confidence of the tunnel monitoring area; wherein, the safety risk prediction model integrates an attention mechanism to dynamically assign weights to each feature variable in multi-source heterogeneous data to generate a weighted feature vector. S3, based on construction activities, divides the tunnel monitoring area into several spatial sub-regions; S4. Based on the output of the security risk prediction model, determine the comprehensive key index of each feature variable, and determine the key driving force and its spatial sub-region based on the comprehensive key index. S5 matches the key driving forces corresponding to each spatial sub-region with construction activities. When the match is successful, based on the current progress and construction speed of the construction activity, a prediction spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined. S6. Based on the comprehensive safety risk probability of each time period, generate early warning information corresponding to each spatial sub-region, and establish a tunnel gas environment model based on the early warning information; S7. Based on the tunnel gas environment model, the early warning information is corrected, and the final early warning information is output.

[0007] Furthermore, in step S1, the spatiotemporal registration and fusion processing of multi-source heterogeneous data to generate a unified spatiotemporal reference dataset includes: S11, For multi-source data with different sampling frequencies, the time series of the highest frequency data is used as the benchmark, and the cubic spline interpolation algorithm is used to interpolate the lower frequency data series to unify all data into the same timestamp sequence. S12, based on tunnel geographic information, assigns unified three-dimensional spatial coordinates to all sensor data, aligns the physical locations of all sensors with coordinates, and establishes a mapping relationship between data and the physical space of the tunnel. S13 aggregates the spatiotemporally registered data according to a preset time window; for the data within each time window, extracts its statistical features to form a unified multidimensional feature vector; among which, the statistical features include mean, variance, peak value and trend.

[0008] Furthermore, in step S4, determining the comprehensive key index for each feature variable based on the output of the security risk prediction model includes: S41, For the current feature vector of the target spatial sub-region, random sampling is performed in the numerical neighborhood of the feature vector to generate a sample set containing multiple perturbation samples; S42, each of the disturbance samples is predicted using the security risk prediction model to obtain the corresponding comprehensive security risk probability prediction value; S43, Using the disturbance sample as input and its corresponding risk probability prediction value as output, a locally weighted linear regression model is established; S44, Extract the absolute values ​​of the regression coefficients of each feature variable in the linear regression model and use them as the contribution values ​​of each feature variable to the risk prediction results; S45, the contribution value is fused with the dynamic weights assigned by the attention mechanism, and a comprehensive key index for each feature variable is calculated by weighted summation.

[0009] Furthermore, when the key driving force corresponding to the spatial sub-region does not match the construction activities, and the overall safety risk probability is higher than the first preset threshold: Based on the physical type of the key driving force, determine the type of safety risk; Based on the security risk probability and the security risk type, an early warning message is generated; Based on the tunnel's overall risk heat map, combined with the tunnel's topology and the distribution of emergency facilities, an evacuation route is generated through a path planning algorithm, and the route and the early warning information are pushed to the mobile terminals of relevant personnel in real time. During the evacuation, the risk status of each spatial sub-area is continuously monitored. When a new risk area appears on the original route, the evacuation route is replanned and updated. Among them, the risk area is a spatial sub-region where the probability of safety risk is higher than the first preset threshold.

[0010] Furthermore, in step S5, the process of establishing a predictive spatiotemporal benchmark dataset for several time periods based on the current progress and speed of the construction activity, and determining the comprehensive safety risk probability for each time period, includes: S51, based on the current progress, construction speed and remaining workload of the construction activities, extrapolate the construction status after several consecutive periods in the future; wherein, the construction status includes the location of the excavation face, the parameters of the support structure and the range of surrounding rock disturbance; S52, For each future time period, based on the projected construction status and the changing patterns of the key driving forces, predict the corresponding multi-source monitoring data for that time period; S53, The multi-source monitoring data predicted for each future time period are used to construct a predicted spatiotemporal reference dataset for the corresponding time period according to the spatiotemporal registration and fusion processing method in step S1. S54, input the predicted spatiotemporal benchmark datasets for each time period into the security risk prediction model in sequence to obtain the comprehensive security risk probability corresponding to each future time period, forming a risk probability time series.

[0011] Furthermore, in S6, the step of generating early warning information for each spatial sub-region based on the comprehensive safety risk probability of each time period, and establishing a tunnel gas environment model based on the early warning information, includes: S61 integrates the current comprehensive safety risk probability and prediction confidence, key driving force type and spatial location, remaining safe construction time calculated based on risk probability time series, and evacuation path generated based on path planning algorithm to generate early warning information. S62a, based on the harmful gas emissions corresponding to the key driving force, simulates the accumulation process of harmful gases in the tunnel; S62b calculates the oxygen consumption rate of each spatial sub-region based on the number of construction personnel, equipment operating status, and possible combustion processes, and predicts the spatiotemporal distribution of oxygen concentration. S63, Based on the accumulation process of the harmful gases in the tunnel and the spatiotemporal distribution of the oxygen concentration, a tunnel gas environment model is established.

[0012] Furthermore, in S7, the correction of the early warning information based on the tunnel gas environment model includes: S71, based on the gas environment model, predicts the diffusion paths of harmful gases and the distribution of low-oxygen areas, and optimizes the evacuation route; S72, Based on the gas environment model, generate the gas safety time; S73 compares the gas safety time with the remaining safe construction time, and takes the smaller of the two as the corrected remaining safe construction time.

[0013] Furthermore, generating the gas safety time based on the gas environment model includes: S72a, Based on the current concentration and growth rate of the harmful gas, calculate the time required for the concentration of the harmful gas to reach the preset harmful gas threshold. S72b, based on the current oxygen concentration and consumption rate, calculates the time required for the oxygen concentration to reach the preset oxygen threshold, and compares it with the time required for the concentration of harmful gas to reach the preset harmful gas threshold, taking the smaller value of the two as the theoretical gas safety time of the spatial sub-region. S72c, based on the optimized evacuation path, calculates the time required to evacuate from each spatial sub-region to the safe area, and denots it as the evacuation time; S72d, subtract the corresponding evacuation time from the theoretical gas safety time of each spatial sub-region to obtain the gas safety time for each spatial sub-region.

[0014] Furthermore, based on the optimized evacuation path, the time required to evacuate from each spatial sub-region to the safe area is calculated and denoted as the evacuation time, including: S72c-1, based on the three-dimensional model of the tunnel, determines the actual distance of the evacuation path corresponding to each spatial sub-area; S72c-2, determining the dynamic velocity attenuation factor based on visibility level and obstacle density; S72c-3, based on the dynamic speed decay factor, the basic movement speed and the personnel density, establishes a personnel movement speed model and outputs the average value of the personnel movement speed; S72c-4, based on the average speed of personnel movement and the path distance of the evacuation path corresponding to each spatial sub-area, the evacuation time corresponding to each spatial sub-area is calculated.

[0015] This invention also proposes a tunnel comprehensive safety analysis and early warning system based on multi-source data fusion, which adopts the tunnel comprehensive safety analysis and early warning method based on multi-source data fusion as described in any of the preceding claims, and specifically includes the following modules: The spatiotemporal unification module is used to acquire multi-source heterogeneous data of the tunnel, and to perform spatiotemporal registration and fusion processing on the multi-source heterogeneous data to generate a unified spatiotemporal reference dataset. The risk prediction module is used to input the spatiotemporal reference dataset into the safety risk prediction model and output the comprehensive safety risk probability and prediction confidence of the tunnel monitoring area; wherein, the safety risk prediction model integrates an attention mechanism to dynamically assign weights to each feature variable in multi-source heterogeneous data to generate a weighted feature vector. The spatial division module is used to divide the tunnel monitoring area into several spatial sub-regions based on construction activities. The risk attribution module is used to determine the comprehensive key index of each feature variable based on the output of the safety risk prediction model, and to determine the key driving force and its spatial sub-region based on the comprehensive key index. The construction prediction module is used to match the key driving forces corresponding to each spatial sub-region with construction activities. When the match is successful, based on the current progress and construction speed of the construction activity, a prediction spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined. The early warning generation module is used to generate early warning information for each spatial sub-area based on the comprehensive safety risk probability of each time period, and to establish a tunnel gas environment model based on the early warning information. The early warning correction module is used to correct the early warning information based on the tunnel gas environment model and output the final early warning information.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: Firstly, this invention dynamically calculates the weights of each feature variable by embedding an attention mechanism in the prediction model; at the same time, it adopts the LIME locally interpretable model and quantifies the feature contribution through perturbation sampling and linear surrogate model, thus solving the problems of black box decision-making and difficulty in tracing the source of traditional models and improving the accuracy of tunnel safety early warning.

[0017] Secondly, this invention constructs a matching mechanism between construction activities and key driving forces, establishing a dynamic correlation model between construction progress, speed, and risk evolution. By extrapolating future construction states and predicting changes in monitoring data, it forms a risk probability time series. This can identify expected and abnormal risks caused by construction, and by determining safe construction time, it provides a clear time window for emergency rescue operations.

[0018] Thirdly, this invention establishes a dynamic gas environment model by simulating the diffusion process of harmful gases and predicting oxygen consumption based on personnel and equipment data. Simultaneously, it incorporates gas risk into the path planning cost function, dynamically generating evacuation routes to avoid hazardous areas. This allows for early warning of the coupling effect between structural and gas risks, ensuring safe evacuation of personnel in complex disaster environments through real-time updated path planning, and further improving the accuracy and reliability of safe construction time by calculating the time required for evacuation routes. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the method of the tunnel integrated safety analysis and early warning method based on multi-source data fusion according to an embodiment of the present invention. Figure 2 This is a structural block diagram of a tunnel integrated safety analysis and early warning system based on multi-source data fusion, according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0023] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0024] like Figure 1 As shown, a tunnel integrated safety analysis and early warning method based on multi-source data fusion includes: S1. Acquire multi-source heterogeneous data of the tunnel, and perform spatiotemporal registration and fusion processing on the multi-source heterogeneous data to generate a unified spatiotemporal reference dataset, including: S11, For multi-source data with different sampling frequencies, the time series of the highest frequency data is used as the benchmark, and the cubic spline interpolation algorithm is used to interpolate the lower frequency data series to unify all data into the same timestamp sequence. S12, based on tunnel geographic information, assigns unified three-dimensional spatial coordinates to all sensor data, aligns the physical locations of all sensors with coordinates, and establishes a mapping relationship between data and the physical space of the tunnel. S13 aggregates the spatiotemporally registered data according to a preset time window; for the data within each time window, extracts its statistical features to form a unified multidimensional feature vector; among which, the statistical features include mean, variance, peak value and trend.

[0025] Specifically, multi-source heterogeneous data is acquired in real time from various sensors deployed on the tunnel cross section. For example, the stress of the lining concrete is obtained from stress sensors, the settlement of the arch is obtained from displacement sensors, the area of ​​the cavity behind the lining is obtained from ground-penetrating radar, and the status of construction activities is obtained from video monitoring.

[0026] For example, among all data sources, the stress and displacement sensor was identified as having the highest sampling frequency of 1Hz, and therefore its time series was used as the reference time axis. For a data point acquired by the ground-penetrating radar every 10 seconds, a cubic spline interpolation algorithm is used to fit a continuous and smooth curve based on the data points before and after it. On this curve, interpolation calculations are performed for each timestamp of the reference time axis to generate a radar data sequence synchronized with high-frequency data; Specifically, tunnel geographic information includes the tunnel’s building information model, which contains the precise three-dimensional coordinates of all sensors.

[0027] For example, a sliding time window with a preset time window of 60 seconds is used to slide the window in steps of one second; The mean, variance, peak value, and trend of each sensor were calculated within a 60-second time window, reflecting the average load level, fluctuation intensity, maximum value, and development trend within the 60-second time window. All extracted statistical features are concatenated in a fixed order to form a unified multidimensional feature vector, which is the spatiotemporal benchmark dataset.

[0028] S2, input the spatiotemporal reference dataset into the safety risk prediction model, and output the comprehensive safety risk probability and prediction confidence of the tunnel monitoring area; wherein, the safety risk prediction model integrates an attention mechanism, which is used to dynamically assign weights to each feature variable in multi-source heterogeneous data to generate a weighted feature vector. Specifically, the security risk prediction model has an integrated architecture, with its core consisting of an attention mechanism and a Bayesian neural network connected together.

[0029] For example, the feature vector is first passed through a fully connected neural network layer to calculate the attention score of each feature. The attention scores of all features are then normalized into weights using the Softmax function, so that the sum of all weights is 1. The original feature vector is then multiplied element-wise with the corresponding attention weights to obtain the weighted feature vector.

[0030] For example, the output layer of the boundary network node is designed as a binary classification, using the Sigmoid activation function to directly output a value between 0 and 1, representing the overall security risk probability; To quantify uncertainty, multiple slightly different outputs are obtained through Monte Carlo sampling. The mean of the multiple sampling results is taken as the final probability of safety risk. The standard deviation or variance of the multiple sampling results is calculated as a measure of the uncertainty of the prediction.

[0031] This embodiment achieves dynamic focusing on risk-driving factors through an attention mechanism, enabling the model to adapt to changes in tunnel conditions. It provides prediction results with uncertainty metrics through a Bayesian neural network, making early warning decisions no longer a simple yes or no, but incorporating credibility information. When the risk probability is high and the confidence level is also high, the system can decisively trigger an alarm; when the risk probability is high but the confidence level is low, the system can prompt for manual review or increased monitoring, thereby greatly improving the scientific rigor and reliability of the early warning system.

[0032] S3, based on construction activities, divides the tunnel monitoring area into several spatial sub-regions; S4, based on the output of the safety risk prediction model, determines the comprehensive key index for each feature variable, and based on the comprehensive key index, determines the key drivers and their spatial sub-regions, including: S41, For the current feature vector of the target space sub-region, perform random sampling in the numerical neighborhood of the feature vector to generate a sample set containing multiple perturbation samples; S42, predict each disturbance sample using the security risk prediction model to obtain the corresponding comprehensive security risk probability prediction value; S43, taking the perturbation sample as input and its corresponding risk probability prediction value as output, establishes a locally weighted linear regression model; S44, Extract the absolute values ​​of the regression coefficients of each feature variable in the linear regression model and use them as the contribution values ​​of each feature variable to the risk prediction results; S45 integrates the contribution value with the dynamic weights assigned by the attention mechanism, and calculates the comprehensive key index of each feature variable by weighted summation.

[0033] For example, the current feature vector of the target space sub-region is [mean stress = 18.2, displacement variance = 4.5, void tendency = +0.08], the risk probability output by the safety risk prediction model is 0.92, and the feature weights assigned by the attention mechanism are [0.7, 0.2, 0.1]; Centered on the current feature vector, random sampling is performed within the numerical neighborhood of each feature dimension according to a normal distribution. For example, for the mean stress feature, sampling is performed within the range of [18.2 ± 2.0]. Repeat this process 5000 times to generate a sample set containing 5000 perturbation samples, each sample representing a possible but slightly different state of the spatial sub-region; 5000 disturbance samples are sequentially input into the safety risk prediction model in S2, and the comprehensive safety risk probability output by the model for each sample is recorded to obtain a vector containing 5000 probability values. Centered on the original feature vector, the Gaussian kernel function is used to calculate the similarity between each perturbed sample and it. The closer the sample is to the original vector, the greater its weight in the fitting process. With 5000 perturbed samples as input features and their corresponding 5000 risk probabilities as output targets, the calculated similarity is used as the sample weight to train a weighted linear regression model and obtain a well-fitted local linear model containing the regression coefficients of each feature. Extract the regression coefficients obtained from S43, and take the absolute value of each coefficient. The larger the absolute value of the coefficient, the greater the influence of the feature on the change of risk probability in the local range. Normalize these absolute values ​​so that their sum is 1 to obtain the LIME contribution value of each feature. The LIME contribution value is weighted and summed with the dynamic weights assigned by the attention mechanism in S2 to obtain the comprehensive key index of each feature variable. The key driving force is determined based on the comprehensive key index, which is the key reason for the occurrence of security risks.

[0034] This embodiment uses the LIME method to explore the decision-making basis of complex risk prediction models, quantifies the local contribution of each feature in specific risk cases, and then integrates it with the attention weights within the model to obtain an interpretable and traceable key driver report. This enables construction personnel to not only know the numerical value of the probability of safety risks, but also to clearly understand the specific causes of the risks, thus providing direct and powerful decision support for taking precise reinforcement or evacuation measures.

[0035] S5 matches the key driving forces corresponding to each spatial sub-region with construction activities. When a match is successful, based on the current progress and construction speed of the construction activity, a predictive spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined, including: S51, based on the current progress, construction speed and remaining workload of the construction activities, extrapolate the construction status after several consecutive periods in the future; the construction status includes the location of the excavation face, the parameters of the support structure and the range of surrounding rock disturbance; S52, for each future period, based on the projected construction status and the changing patterns of key driving forces, predicts the corresponding multi-source monitoring data for that period; S53, The multi-source monitoring data predicted for each future time period are used to construct a predicted spatiotemporal reference dataset for the corresponding time period according to the spatiotemporal registration and fusion processing method in step S1. S54. The predicted spatiotemporal benchmark datasets for each time period are sequentially input into the security risk prediction model to obtain the comprehensive security risk probability for each future time period, forming a risk probability time series.

[0036] In one possible implementation, a pre-established structured construction activity-key driving force matching table serves as the knowledge base for matching decisions.

[0037] Specifically, the system obtains the status of construction activities and the identification results of key driving forces. Based on the matching table, it logically determines whether the current construction type matches the key driving force. If the match is successful, it indicates that the current construction activity may be an emergency rescue operation targeting safety risks inside the tunnel. More accurate early warning information for the emergency rescue operation should be provided through longer-term inference simulation, rather than simple logical judgment.

[0038] Furthermore, the matching between construction type and key driving force can be verified through time correlation, spatial consistency, strength matching degree, and historical patterns.

[0039] This invention, through steps S51 and S52, extends the time dimension forward, proactively predicting future construction status and monitoring data based on the evolution of current construction progress, speed, and key driving forces. This enables safety early warnings to no longer be limited to the current moment, but to foresee the trajectory of risk evolution over a future period, achieving advanced early warning in the time dimension. This provides safety risk assessment criteria for emergency construction activities and improves the accuracy and reliability of tunnel safety early warnings.

[0040] When the key driving force corresponding to the spatial sub-region does not match the construction activities, and the overall safety risk probability is higher than the first preset threshold: Determine the type of safety risk based on the physical type of the key driving force; Early warning information is generated based on the probability and type of security risks. Based on the tunnel's overall risk heat map, combined with the tunnel's topology and the distribution of emergency facilities, an evacuation route is generated through a path planning algorithm, and the route and early warning information are pushed to the mobile terminals of relevant personnel in real time. During the evacuation, the risk status of each spatial sub-area is continuously monitored. When a new risk area appears on the original route, the evacuation route is replanned and updated. Among them, the risk area is a spatial sub-region where the probability of safety risk is higher than the first preset threshold.

[0041] For example, based on a preset risk type knowledge base, the physical type of the key driving force is matched with it to determine the type of safety risk. For example, if the key driving force is a sudden change in the stress of the surrounding rock, it is judged as a sudden risk; if the key driving force is a long-term stress relaxation of the steel bars, it is judged as a gradual risk. Optionally, the first preset threshold can be implemented in the range of [0.70, 0.75], and preferably, the preferred embodiment of the first preset threshold is 0.73.

[0042] Optionally, the optional implementation range of the second preset threshold is [0.80, 0.90], and preferably, the preferred embodiment of the second preset threshold is 0.85.

[0043] When the probability of a safety risk exceeds the second preset threshold, an early warning message containing an immediate evacuation signal is generated. When the probability of a safety risk is greater than the first preset threshold, less than the second preset threshold, and the type of safety risk is a sudden risk, an early warning message containing an immediate evacuation signal is generated. When the probability of a safety risk is greater than the first preset threshold, less than the second preset threshold, and the type of safety risk is a progressive risk, an early warning message containing a time-limited evacuation signal is generated. When the probability of safety risk is less than the first preset threshold, an early warning message containing a safety construction signal is generated.

[0044] In one possible implementation, the system polls the real-time risk probabilities of all spatial sub-regions, classifies the tunnel's 3D model into safety risk levels, and uses the A* algorithm, taking risk probability as one of the main factors in path cost, to find the path with the lowest cumulative risk exposure as the evacuation path. This evacuation path is integrated into the early warning information and pushed to the construction personnel's smartphones or dedicated handheld terminals. During the evacuation process, the system continues to execute the S1 to S2 process, updating the global risk heatmap in real time. When it detects that the original path has been blocked by a new high-risk area, the path planning algorithm is triggered again. Based on the latest heatmap, a new evacuation path is calculated and resent.

[0045] This invention achieves in-depth risk identification by using the mismatch between key driving forces and construction activities as a judgment criterion. It can identify potential and more dangerous abnormal risks not directly caused by construction, reducing false alarms caused by routine factors such as construction disturbances, improving the pertinence and reliability of safety warnings, and ensuring the personal safety of emergency rescue personnel. By generating warning information for different types of safety risks, it improves the balance between emergency rescue and safety risks. By generating evacuation routes based on a tunnel-wide risk heat map and continuously monitoring and dynamically replanning evacuation routes, it combines warning information with construction safety, improving the practicality of safety risk warnings. It also constructs a precise information direct channel from the central system to individual personnel. This allows on-site personnel to receive complete action instructions—what happened, what to do, and where to go—in the first instance, eliminating confusion and delays caused by information transmission delays or misunderstandings. Command personnel at the rear can also use the system to globally control the risk situation and personnel movement trajectories, achieving more efficient and scientific emergency command.

[0046] S6 generates early warning information for each spatial sub-region based on the comprehensive safety risk probability of each time period, and establishes a tunnel gas environment model based on the early warning information, including: S61 integrates the current comprehensive safety risk probability and prediction confidence, key driving force type and spatial location, remaining safe construction time calculated based on risk probability time series, and evacuation path generated based on path planning algorithm to generate early warning information. S62a, based on the harmful gas emissions corresponding to the key driving force, simulates the accumulation process of harmful gases in the tunnel; S62b calculates the oxygen consumption rate of each spatial sub-region based on the number of construction personnel, equipment operating status, and possible combustion processes, and predicts the spatiotemporal distribution of oxygen concentration. S63. Based on the accumulation process of harmful gases in the tunnel and the spatiotemporal distribution of oxygen concentration, a tunnel gas environment model is established.

[0047] In one possible implementation, the remaining time for safe construction is obtained by calculating the risk probability time series using a trend extrapolation forecasting method. For example, the generated warning information includes: Current security risk probability and confidence level; Key drivers and specific locations that lead to security risks; Remaining time for safe construction; Evacuation route.

[0048] In one possible implementation, the area corresponding to the key driving force is identified as undergoing emergency concrete pouring construction, and two high-power diesel generators are used to power the vibrating equipment. In this case, the overall safety risk probability is 0.62. Based on the exhaust emissions of this type of diesel generator at rated power, determine the type and emission rate of harmful gases; The real-time emission rate of harmful gases is calculated by combining the generator's real-time load rate and operating time. Predict oxygen consumption under the combined effects of diesel combustion and human respiration; Taking into account both oxygen consumption and oxygen supply from the ventilation system, the spatiotemporal changes in oxygen concentration are predicted. The predicted concentration fields of sulfur dioxide and carbon monoxide generated by the diesel generator are superimposed with the oxygen concentration field in the tunnel BIM model. The final output is a dynamic and visualized comprehensive gas environment risk map of the tunnel, which intuitively displays the comprehensive gas risk level of different regions and time points in the form of a heat map.

[0049] This invention incorporates gaseous environmental risks into the core assessment framework by accurately simulating the accumulation of harmful gases and oxygen consumption caused by construction activities, such as emissions from diesel equipment, or geological factors, such as the release of harmful gases. This expands the dimensions of consideration for tunnel safety early warning, improves the accuracy and reliability of tunnel safety early warning, and enhances robustness in dealing with complex coupled risks.

[0050] S7 corrects the early warning information based on the tunnel gas environment model and outputs the final early warning information.

[0051] The early warning information is corrected based on the tunnel gas environment model, including: S71 optimizes evacuation routes based on the predicted diffusion paths of harmful gases and the distribution of low-oxygen areas using a gas environment model.

[0052] S72, based on a gas environment model, generates a gas safety time; S73 compares the gas safety time with the remaining safe construction time, and takes the smaller of the two as the corrected remaining safe construction time. In one possible implementation, the tunnel gas environment model predicts that after 2 hours, a high-concentration area of ​​carbon monoxide will appear in the tunnel due to the accumulation of carbon monoxide; after 1.5 hours, a low-concentration area of ​​oxygen will appear in the tunnel due to the consumption of oxygen. If the original evacuation route passes through both the high-concentration area of ​​carbon monoxide and the low-concentration area of ​​oxygen, the gas risk cost will be added to the path planning algorithm, and the final evacuation route will be regenerated.

[0053] Based on the gas environment model, the gas safety time is generated, including: S72a, Based on the current concentration and growth rate of the harmful gas, calculate the time required for the concentration of the harmful gas to reach the preset harmful gas threshold. S72b, based on the current oxygen concentration and consumption rate, calculates the time required for the oxygen concentration to reach the preset oxygen threshold, and compares it with the time required for the concentration of harmful gas to reach the preset harmful gas threshold, taking the smaller value of the two as the theoretical gas safety time of the spatial sub-region. S72c, based on the optimized evacuation path, calculates the time required to evacuate from each spatial sub-region to the safe area, and denots it as the evacuation time; S72d, subtract the corresponding evacuation time from the theoretical gas safety time of each spatial sub-region to obtain the gas safety time for each spatial sub-region.

[0054] In one possible implementation, the preset carbon monoxide threshold is 30 ppm, and the preset oxygen threshold is 19.5%. First, combine key driving forces to identify the core gas and determine the type of target harmful gas; collect the real-time gas concentration in the current space sub-region, and obtain the critical time of harmful gas by fitting the concentration growth rate through sensor data. The oxygen critical time is calculated by taking into account the number of construction workers, the oxygen consumption of equipment, and whether there is a combustion reaction. The critical time of harmful gases is compared with the critical time of oxygen, and the smaller value is taken as the theoretical gas safety time of the space sub-region to ensure coverage of the most dangerous gas risk scenarios.

[0055] Subtract the evacuation time from the theoretical gas safety time to ensure that personnel can reach the safe area before the gas environment deteriorates; if the calculation result is negative, it indicates that the current environment no longer meets the conditions for safe evacuation, and the gas safety time is directly set to 0 to trigger an emergency evacuation order.

[0056] This invention accurately captures the "weakest link" effect among different risk factors, and the output theoretical safety time is a precise reflection of the most urgent and fatal constraints, avoiding prediction bias caused by focusing on one-sided indicators and improving the accuracy of early warning. The gas safety event generated by the difference between the theoretical gas safety time and the evacuation time injects critical time redundancy into the early warning, ensuring that the published safety time is a reliable bottom-line time that can guarantee the safety of personnel in any unexpected situation, thus improving the reliability of safety early warning.

[0057] Based on the optimized evacuation path, the time required to evacuate from each spatial sub-region to the safe zone is calculated and denoted as the evacuation time, including: S72c-1, based on the three-dimensional model of the tunnel, determines the path distance of the evacuation path corresponding to each spatial sub-area; S72c-2, determining the dynamic velocity attenuation factor based on visibility level and obstacle density; S72c-3 establishes a personnel movement speed model based on dynamic speed decay factor, base movement speed, and personnel density, and outputs the average value of personnel movement speed; S72c-4 calculates the evacuation time for each spatial sub-area based on the average speed of personnel movement and the path distance of the evacuation path corresponding to each spatial sub-area.

[0058] In one possible implementation, a tunnel BIM model or a 3D laser scanning model is invoked. The model needs to include geographic information such as the tunnel's inner wall outline, exit location, and fixed facilities. The coordinate system adopts a unified 3D coordinate system for tunnel construction, such as the geodetic coordinate system.

[0059] Determine the coordinates of the center point of each spatial sub-region. For example, the coordinates of the center point of the 100m segment of the left arch waist of the sub-region are X=2568.2m, Y=1892.5m, Z=32.8m, and the center point of the exit is X=2800.0m, Y=1892.5m, Z=33.0m. These coordinates will be used as the start and end points for path calculation.

[0060] The evacuation route is segmented for sampling, that is, a sampling point is taken every 0.5m along the route. The spatial distance between adjacent sampling points is calculated using a three-dimensional coordinate system. All segment distances are summed to obtain the total evacuation route distance corresponding to the sub-area.

[0061] Real-time visibility is obtained by visibility sensors deployed inside the tunnel, and visibility values ​​are generated; the number of obstacles on the evacuation path is counted by AI recognition of video surveillance or on-site inspection records, and obstacle density values ​​are generated; the dynamic velocity attenuation factor is calculated using a weighted product method.

[0062] For example, the base moving speed is set to 1.4 m / s, when the population density is less than or equal to 0.5 people / m².2 When the population density is greater than 0.5 people / m², the population density coefficient is taken as 1.0; when the population density is greater than 0.5 people / m², the population density coefficient is taken as 1.0. 2 And less than or equal to 1.0 person / m 2 When the population density is greater than 1.0 people / m², the population density coefficient is taken as 0.9; when the population density is greater than 1.0 people / m², the population density coefficient is taken as 0.9. 2 And less than or equal to 2.0 people / m 2 When the population density is 0.7, the population density coefficient is taken as 0.7; when the population density is greater than 2 people / m² 2 When the personnel density coefficient is taken as 0.5, the personnel density coefficient represents the efficiency of personnel movement speed under the current personnel density; the personnel movement speed is the product of the basic movement speed, the personnel density coefficient, and the dynamic speed decay factor.

[0063] Based on the early warning information of different spatial sub-zones, after the personnel gather on the evacuation route, the personnel density coefficient is recalculated, and the average personnel movement speed is calculated according to the movement speed of different road sections. Based on the actual distance of the evacuation path and the average speed of personnel movement for each spatial sub-zone, the evacuation time for each spatial sub-zone is calculated, the early warning information is corrected, and the final early warning information is output.

[0064] The final warning information includes: Overall security risk probability and prediction confidence level; Key driving force types and spatial locations; The revised remaining time for safe construction, and this time is marked as the latest recommended time to start evacuation; Revised evacuation route and visualization map.

[0065] This invention abandons the simplified method of traditional straight-line distance estimation. Relying on tunnel BIM models or 3D laser scanning models, it accurately extracts the actual path, including turns, slopes, and avoidance of fixed facilities. Segmented sampling calculations lay a precise spatial foundation for subsequent time calculations, improving the accuracy of tunnel safety early warning. Furthermore, it dynamically corrects movement speed based on visibility and obstacle density, and further adjusts it based on personnel density. Ultimately, the accuracy of evacuation time calculation is improved to within ±0.5 minutes, enhancing the accuracy and reliability of tunnel safety early warning.

[0066] Example 2 like Figure 2 As shown, the present invention also proposes a tunnel comprehensive safety analysis and early warning system based on multi-source data fusion, using the tunnel comprehensive safety analysis and early warning method based on multi-source data fusion as described in any of Embodiment 1, including the following modules: The spatiotemporal unification module is used to acquire multi-source heterogeneous data of the tunnel, and to perform spatiotemporal registration and fusion processing on the multi-source heterogeneous data to generate a unified spatiotemporal reference dataset. The risk prediction module is used to input the spatiotemporal benchmark dataset into the safety risk prediction model and output the comprehensive safety risk probability and prediction confidence of the tunnel monitoring area. The safety risk prediction model integrates an attention mechanism to dynamically assign weights to each feature variable in multi-source heterogeneous data to generate a weighted feature vector. The spatial division module is used to divide the tunnel monitoring area into several spatial sub-regions based on construction activities. The risk attribution module is used to determine the comprehensive key index of each feature variable based on the output of the safety risk prediction model, and to determine the key drivers and their spatial sub-regions based on the comprehensive key index. The construction prediction module is used to match the key driving forces corresponding to each spatial sub-region with construction activities. When the match is successful, based on the current progress and construction speed of the construction activity, a prediction spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined. The early warning generation module is used to generate early warning information for each spatial sub-area based on the comprehensive safety risk probability of each time period, and to establish a tunnel gas environment model based on the early warning information. The early warning correction module is used to correct the early warning information based on the tunnel gas environment model and output the final early warning information.

[0067] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A tunnel integrated safety analysis and early warning method based on multi-source data fusion, characterized in that, include: S1, acquire multi-source heterogeneous data of the tunnel, and perform spatiotemporal registration and fusion processing on the multi-source heterogeneous data to generate a unified spatiotemporal reference dataset; S2, input the spatiotemporal reference dataset into the safety risk prediction model, and output the comprehensive safety risk probability and prediction confidence of the tunnel monitoring area; wherein, the safety risk prediction model integrates an attention mechanism to dynamically assign weights to each feature variable in multi-source heterogeneous data to generate a weighted feature vector. S3, based on construction activities, divides the tunnel monitoring area into several spatial sub-regions; S4. Based on the output of the security risk prediction model, determine the comprehensive key index of each feature variable, and determine the key driving force and its spatial sub-region based on the comprehensive key index. S5 matches the key driving forces corresponding to each spatial sub-region with construction activities. When the match is successful, based on the current progress and construction speed of the construction activity, a prediction spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined. S6. Based on the comprehensive safety risk probability of each time period, generate early warning information corresponding to each spatial sub-region, and establish a tunnel gas environment model based on the early warning information; S7. Based on the tunnel gas environment model, the early warning information is corrected, and the final early warning information is output.

2. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 1, characterized in that, In S1, the spatiotemporal registration and fusion processing of multi-source heterogeneous data to generate a unified spatiotemporal reference dataset includes: S11, For multi-source data with different sampling frequencies, the time series of the highest frequency data is used as the benchmark, and the cubic spline interpolation algorithm is used to interpolate the lower frequency data series to unify all data into the same timestamp sequence. S12, based on tunnel geographic information, assigns unified three-dimensional spatial coordinates to all sensor data, aligns the physical locations of all sensors with coordinates, and establishes a mapping relationship between data and the physical space of the tunnel. S13 aggregates the spatiotemporally registered data according to a preset time window; for the data within each time window, extracts its statistical features to form a unified multidimensional feature vector; among which, the statistical features include mean, variance, peak value and trend.

3. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 1, characterized in that, In S4, the determination of the comprehensive key index for each feature variable based on the output of the security risk prediction model includes: S41, For the current feature vector of the target spatial sub-region, random sampling is performed in the numerical neighborhood of the feature vector to generate a sample set containing multiple perturbation samples; S42, each of the disturbance samples is predicted using the security risk prediction model to obtain the corresponding comprehensive security risk probability prediction value; S43, Using the disturbance sample as input and its corresponding risk probability prediction value as output, a locally weighted linear regression model is established; S44, Extract the absolute values ​​of the regression coefficients of each feature variable in the linear regression model and use them as the contribution values ​​of each feature variable to the risk prediction results; S45, the contribution value is fused with the dynamic weights assigned by the attention mechanism, and a comprehensive key index for each feature variable is calculated by weighted summation.

4. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 1, characterized in that, When the key driving force corresponding to the spatial sub-region does not match the construction activities, and the overall safety risk probability is higher than the first preset threshold: Based on the physical type of the key driving force, determine the type of safety risk; Based on the security risk probability and the security risk type, an early warning message is generated; Based on the tunnel's overall risk heat map, combined with the tunnel's topology and the distribution of emergency facilities, an evacuation route is generated through a path planning algorithm, and the route and the early warning information are pushed to the mobile terminals of relevant personnel in real time. During the evacuation, the risk status of each spatial sub-area is continuously monitored. When a new risk area appears on the original route, the evacuation route is replanned and updated. Among them, the risk area is a spatial sub-region where the probability of safety risk is higher than the first preset threshold.

5. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 1, characterized in that, In S5, based on the current progress and construction speed of the construction activity, a predictive spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined, including: S51, based on the current progress, construction speed and remaining workload of the construction activities, extrapolate the construction status after several consecutive periods in the future; wherein, the construction status includes the location of the excavation face, the parameters of the support structure and the range of surrounding rock disturbance; S52, For each future time period, based on the projected construction status and the changing patterns of the key driving forces, predict the corresponding multi-source monitoring data for that time period; S53, The multi-source monitoring data predicted for each future time period are used to construct a predicted spatiotemporal reference dataset for the corresponding time period according to the spatiotemporal registration and fusion processing method in step S1. S54, input the predicted spatiotemporal benchmark datasets for each time period into the security risk prediction model in sequence to obtain the comprehensive security risk probability corresponding to each future time period, forming a risk probability time series.

6. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 1, characterized in that, In S6, the step of generating early warning information for each spatial sub-region based on the comprehensive safety risk probability of each time period, and establishing a tunnel gas environment model based on the early warning information, includes: S61 integrates the current comprehensive safety risk probability and prediction confidence, key driving forces and their spatial sub-regions, the remaining safe construction time calculated based on the risk probability time series, and the evacuation path generated based on the path planning algorithm to generate early warning information. S62a, based on the harmful gas emissions corresponding to the key driving force, simulates the accumulation process of harmful gases in the tunnel; S62b calculates the oxygen consumption rate of each spatial sub-region based on the number of construction personnel, equipment operating status, and possible combustion processes, and predicts the spatiotemporal distribution of oxygen concentration. S63, Based on the accumulation process of the harmful gases in the tunnel and the spatiotemporal distribution of the oxygen concentration, a tunnel gas environment model is established.

7. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 6, characterized in that, In S7, the correction of the early warning information based on the tunnel gas environment model includes: S71, based on the predicted diffusion paths of harmful gases and the distribution of low-oxygen areas using the tunnel gas environment model, optimize the evacuation route; S72, Based on the tunnel gas environment model, generate the gas safety time; S73 compares the gas safety time with the remaining safe construction time, and takes the smaller of the two as the corrected remaining safe construction time.

8. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 7, characterized in that, The generation of the gas safety time based on the tunnel gas environment model includes: S72a, Based on the current concentration and growth rate of the harmful gas, calculate the time required for the concentration of the harmful gas to reach the preset harmful gas threshold. S72b, based on the current oxygen concentration and consumption rate, calculates the time required for the oxygen concentration to reach the preset oxygen threshold, and compares it with the time required for the concentration of harmful gas to reach the preset harmful gas threshold, taking the smaller value of the two as the theoretical gas safety time of the spatial sub-region. S72c, based on the optimized evacuation path, calculates the time required to evacuate from each spatial sub-region to the safe area, and denots it as the evacuation time; S72d, subtract the corresponding evacuation time from the theoretical gas safety time of each spatial sub-region to obtain the gas safety time for each spatial sub-region.

9. The tunnel integrated safety analysis and early warning method based on multi-source data fusion according to claim 8, characterized in that, Based on the optimized evacuation path, the time required to evacuate from each spatial sub-region to the safe zone is calculated and denoted as the evacuation time, including: S72c-1, based on the three-dimensional model of the tunnel, determines the actual distance of the evacuation path corresponding to each spatial sub-area; S72c-2, determining the dynamic velocity attenuation factor based on visibility level and obstacle density; S72c-3, based on the dynamic speed decay factor, the basic movement speed and the personnel density, establishes a personnel movement speed model and outputs the average value of the personnel movement speed; S72c-4, based on the average speed of personnel movement and the path distance of the evacuation path corresponding to each spatial sub-area, the evacuation time corresponding to each spatial sub-area is calculated.

10. A tunnel integrated safety analysis and early warning system based on multi-source data fusion, characterized in that, The system employs the tunnel comprehensive safety analysis and early warning method based on multi-source data fusion as described in any one of claims 1 to 9, specifically including the following modules: The spatiotemporal unification module is used to acquire multi-source heterogeneous data of the tunnel, and to perform spatiotemporal registration and fusion processing on the multi-source heterogeneous data to generate a unified spatiotemporal reference dataset. The risk prediction module is used to input the spatiotemporal reference dataset into the safety risk prediction model and output the comprehensive safety risk probability and prediction confidence of the tunnel monitoring area; wherein, the safety risk prediction model integrates an attention mechanism to dynamically assign weights to each feature variable in multi-source heterogeneous data to generate a weighted feature vector. The spatial division module is used to divide the tunnel monitoring area into several spatial sub-regions based on construction activities. The risk attribution module is used to determine the comprehensive key index of each feature variable based on the output of the safety risk prediction model, and to determine the key driving force and its spatial sub-region based on the comprehensive key index. The construction prediction module is used to match the key driving forces corresponding to each spatial sub-region with construction activities. When the match is successful, based on the current progress and construction speed of the construction activity, a prediction spatiotemporal benchmark dataset for several time periods is established, and the comprehensive safety risk probability for each time period is determined. The early warning generation module is used to generate early warning information for each spatial sub-area based on the comprehensive safety risk probability of each time period, and to establish a tunnel gas environment model based on the early warning information. The early warning correction module is used to correct the early warning information based on the tunnel gas environment model and output the final early warning information.

Images