A method and system for predicting and preventing risks in sulfur-containing formation tunnel construction

Abstract

The present application belongs to the technical field of tunnel engineering construction safety and risk intelligent management and control, and specifically relates to a sulfur-containing stratum tunnel construction risk prediction and control method and system. The method comprises the following steps: S1, real-time synchronous acquisition of toxic gas concentration data in the tunnel space, geological feature data of the stratum to be excavated in front, and behavior and physiological state data of the personnel working in the tunnel; S2, establishment of a construction risk safety evaluation index system covering gas environment risk, geological condition risk and personnel behavior risk; S3, calculation of the comprehensive weight of each evaluation index in the construction risk safety evaluation index system; S4, calculation of the comprehensive relative closeness between the current construction state and the ideal safety state, and output of the real-time construction risk grade according to the result; S5, prediction of the change trend of each key risk index in the future preset period; and S6, triggering of a graded control response and execution of risk disposal based on the real-time construction risk grade and the predicted future risk state change trend. The present application improves the tunnel construction safety management and control capability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent management and control technology for tunnel construction safety and risks, specifically a method and system for predicting and controlling risks in tunnel construction in sulfur-containing strata. Background Technology

[0002] Sulfur-bearing strata tunnels are a typical high-risk engineering environment frequently encountered in oil and gas resource extraction, water conservancy and hydropower, and transportation tunnel construction projects in central and western my country. These strata not only have complex geological conditions and poor surrounding rock stability, but also contain large amounts of toxic and harmful gases such as hydrogen sulfide and sulfur dioxide. This leads to a long-term double risk of "sudden release of toxic gases" and "personnel gathering for work" during tunnel construction, easily causing catastrophic accidents such as mass poisoning, asphyxiation, and even gas explosions. Such accidents are highly sudden, spread rapidly, and have serious consequences, and traditional prevention and control methods are slow to respond. However, current methods for assessing and controlling the risks of sulfur-bearing strata tunnel construction still have significant shortcomings, making it difficult to meet the needs of modern intelligent tunnel construction and precise safety management. The main technical bottlenecks are: 1. Existing methods mostly rely on gas concentration threshold alarms at fixed locations, failing to effectively integrate the spatiotemporal diffusion patterns of gas, geological conditions, and construction dynamics. This results in insufficient predictability of hidden gas sources and abnormal outburst trends, making it difficult to achieve early identification and warning of risks before dangerous concentrations are formed.

[0003] 2. Gas monitoring data, advanced geological forecast information, and personnel behavior and location data in the construction environment are usually collected independently and processed separately. There is a lack of effective multi-source information fusion and correlation analysis mechanisms, which makes it impossible to accurately depict the dynamic coupling relationship between "strata-gas-personnel" and restricts the comprehensive assessment of the overall risk status.

[0004] 3. Existing assessment methods are mostly post-event alarms or static assessments at a single point in time, failing to continuously track risk status, predict trends, and provide graded early warnings based on time-series data, making it difficult to adapt to the dynamic evolution of risks during construction.

[0005] In summary, existing methods for assessing the safety of tunnel construction in sulfur-bearing strata suffer from technical problems such as one-sided risk perception, fragmented multi-source information, and disconnect between prediction and prevention, making it difficult to achieve real-time and accurate grasp of risks during construction and proactive intervention. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method and system for predicting and controlling construction risks in sulfur-containing strata tunnels.

[0007] This invention adopts the following technical solution: a method for predicting and controlling construction risks in sulfur-bearing strata tunnels, comprising: S1: Construct a multi-source real-time monitoring system for tunnel construction in sulfur-containing strata, and collect data on the concentration of toxic gases in the tunnel space, the geological characteristics of the strata to be excavated ahead, and the behavior and physiological state of the workers inside the tunnel in real time. S2: Based on the monitoring data collected in S1, establish a construction risk safety evaluation index system covering gas environment risk, geological condition risk and personnel behavior risk, and determine the evaluation index and risk level classification standards. S3: Calculate the comprehensive weight of each evaluation index in the construction risk safety evaluation index system; S4: After standardization and comprehensive weighting, the real-time monitoring data is input into the comprehensive evaluation model to calculate the comprehensive relative closeness between the current construction status and the ideal safety status, and output the real-time construction risk level accordingly. S5: Construct an LSTM-Transformer hybrid time series forecasting model to predict the changing trends of key risk indicators within a preset future time period; S6: Based on the real-time construction risk level output by S4 and the future risk status change trend predicted by S5, the hierarchical prevention and control response mechanism is automatically triggered to execute matching risk disposal measures.

[0008] In some embodiments, the multi-source real-time monitoring system includes: A gas concentration monitoring system, comprising: A multi-parameter gas sensor array is deployed at locations where tunnel gases tend to accumulate or at key airflow nodes to continuously monitor hydrogen sulfide and oxygen concentrations, forming a baseline monitoring network covering the entire tunnel. Individual gas detectors are provided, and each worker entering the tunnel is equipped with a portable one that continuously monitors and uploads data on the concentration of hydrogen sulfide and oxygen at the worker's breathing zone in real time. Stratigraphic feature acquisition system, the stratigraphic feature acquisition system comprising: Ground-penetrating radar (GPR) performs long-range scanning of the tunnel section to be excavated, macroscopically delineating strata rich in hydrogen sulfide toxic gases and abnormal areas. Advanced horizontal drilling is used for verification, core sampling and on-site gas testing to obtain key information on the lithology, fracture development and water-bearing capacity of the strata ahead, and to quantitatively analyze the sulfide mineral content of the core, the gas concentration and pressure parameters in the borehole. Personnel file management system, the personnel file management system including: An image recognition array is used to collect and record in real time the status of workers' safety equipment wearing and the compliance of their work behavior. Wearable physiological monitoring devices are used to collect real-time data on the physiological load and individual exposure gas concentrations of the wearer.

[0009] In some embodiments, S1 specifically includes: S11: Fixed multi-parameter gas sensor arrays are installed at the tunnel excavation face, cross passages and ventilation dead corners to continuously monitor the real-time concentration of hydrogen sulfide and oxygen in the tunnel; at the same time, portable personal gas detectors are provided to all workers in the tunnel to monitor and upload the concentration of toxic gases at the breathing zone of the workers in real time. S12: Integrating ground-penetrating radar and advanced horizontal drilling technology, it can explore the distribution of sulfur-containing strata, gas occurrence status and potential gas inrush channels in the section of the tunnel to be excavated, and obtain geological characteristic data of the strata ahead. S13: Through smart wearable devices and image recognition arrays, real-time data is collected on the safety equipment wearing status, work behavior compliance, physiological status indicators, and real-time location information of workers.

[0010] In some embodiments, S13 specifically includes: S131: Assign a unique digital identification code to each worker entering the tunnel, build a corresponding personal digital profile, and store and update the worker's historical behavior data, physiological status data, individual gas exposure data and violation records in real time; S132: Based on real-time collected data on the behavior and status of workers, their personal digital profiles are updated synchronously. The weight coefficients of the personnel behavior risk dimension indicators in the comprehensive evaluation model are dynamically adjusted according to the changing trends of the frequency of violations and the continuous changes in the standardization of protective equipment wearing. S133: Set a dynamic violation decay factor and introduce a time decay function to reduce the long-term impact of historical violation records on the current weight. The time decay function is as follows: ,in The time interval since the last violation. This is the attenuation coefficient, which is dynamically adjusted according to the effect. S134: Dynamic quantitative assessment and personalized weighting of individual worker risks based on adjusted weighting coefficients. The dynamic adjustment formula for weights is as follows: , For personnel i At any moment t The adjusted weights personnel i Based on the fundamental weights assigned by the combination at any given time. This refers to the recent frequency of violations. The average frequency of violations for the group. The sensitivity coefficient can be dynamically adjusted according to the risk level.

[0011] In some embodiments, the construction risk safety evaluation index system in S2 specifically includes: The gas environment risk characteristic evaluation indicators include: fixed monitoring network coverage, fixed-point hydrogen sulfide concentration, individual exposure to hydrogen sulfide concentration, gas concentration change trend, and oxygen concentration. Geological condition risk characteristic evaluation indicators include: key gas-bearing structure detection rate, rock mass gas-bearing grade, gas occurrence state parameters, peak concentration of hydrogen sulfide in borehole and gas inrush channel risk index. The personnel behavior risk characteristic evaluation indicators include: the rate of standardized wearing of respiratory protective equipment, the frequency of unauthorized entry into high-risk areas, and the real-time physiological fatigue index.

[0012] In some embodiments, the risk level classification criteria in S2 are as follows: Level I: Safety and controllability level; Level II: Low-risk concern level; Level III: Medium-risk warning level; Level IV: High-risk danger level.

[0013] In some embodiments, the grading of evaluation indicators corresponding to each level is as follows: Level I: Gas environment risk dimension, fixed monitoring network coverage ≥95%, fixed point hydrogen sulfide concentration <5ppm, individual exposure hydrogen sulfide concentration <5ppm, gas concentration trend is stable or decreasing, oxygen concentration is 20.0%~20.9%Vol; Geological condition risk dimension, key gas-bearing structure detection rate ≥90%, rock mass gas content level is low, gas occurrence state parameter <0.3, peak hydrogen sulfide concentration in borehole <10ppm, gas surge channel risk index <0.3; Personnel behavior risk dimension, respiratory protective equipment standard wearing rate ≥98%, frequency of unauthorized entry into high-risk areas is 0 times / hour, real-time physiological fatigue index <0.3; Level II: Gas environment risk dimension: Fixed monitoring network coverage is 85%~95%, fixed point hydrogen sulfide concentration is 5~10 ppm, individual exposure hydrogen sulfide concentration is 5~10 ppm, gas concentration trend is slight fluctuation, oxygen concentration is 19.5%~20.0% Vol; Geological condition risk dimension: Key gas-bearing structures are identified at 70%~90%, rock mass gas content is moderate, gas occurrence state parameter is 0.3~0.5, peak hydrogen sulfide concentration in boreholes is 10~30 ppm, gas surge channel risk index is 0.3~0.5; Personnel behavior risk dimension: Respiratory protective equipment standard wearing rate is 95%~98%, frequency of unauthorized entry into high-risk areas is 1 time / hour, real-time physiological fatigue index is 0.3~0.5; Level III: Gas environment risk dimension, fixed monitoring network coverage is 75%~85%, fixed point hydrogen sulfide concentration is 10~30ppm, individual exposure hydrogen sulfide concentration is 10~30ppm, gas concentration trend is a continuous and slow increase, oxygen concentration is 18.0%~19.5%Vol; Geological condition risk dimension, key gas-bearing structures are identified at 50%~70%, rock mass gas content is high, gas occurrence state parameter is 0.5~0.7, peak hydrogen sulfide concentration in boreholes is 30~100ppm, gas surge channel risk index is 0.5~0.7; Personnel behavior risk dimension, respiratory protective equipment standard wearing rate is 90%~95%, frequency of unauthorized entry into high-risk areas is 2~3 times / hour, real-time physiological fatigue index is 0.5~0.7; Level IV: Gas environment risk dimension: Fixed monitoring network coverage <75%, fixed point hydrogen sulfide concentration ≥30ppm, individual exposure hydrogen sulfide concentration ≥30ppm, gas concentration change trend is rapidly increasing, oxygen concentration <18.0%Vol; Geological condition risk dimension: Key gas-bearing structure detection rate <50%, rock mass gas content level is extremely high, gas occurrence state parameter ≥0.7, peak hydrogen sulfide concentration in borehole ≥100ppm, gas surge channel risk index ≥0.7; Personnel behavior risk dimension: Respiratory protective equipment standard wearing rate <90%, frequency of unauthorized entry into high-risk areas ≥4 times / hour, real-time physiological fatigue index ≥0.7.

[0014] In some embodiments, the calculation of the comprehensive weight in S3 specifically includes: S31: The fuzzy network hierarchical analysis method is used for subjective weighting. This method addresses the fuzziness in experts' judgments on the importance of indicators, while quantifying the networked correlation and mutual influence among the evaluation indicators, and calculating the subjective weight vector of each evaluation indicator. S32: An improved entropy weighting method is used for objective weighting. The real-time monitored data of each indicator are standardized to eliminate the influence of dimensions. Then, the dispersion and effective information of each indicator data are calculated and analyzed through information entropy to obtain the objective weight vector of each evaluation indicator. S33: Based on game theory, construct a combined weighting optimization model. With minimizing the deviation between subjective weights and objective weights as the optimization objective, solve for the optimal linear combination coefficients of subjective and objective weights. After normalizing the combination coefficients, calculate the optimal comprehensive weights of each evaluation index.

[0015] In some embodiments, S31 includes: S311: Based on a preset quantitative scale for the importance of indicators, all evaluation indicators in the construction risk and safety evaluation indicator system are scored pairwise for relative importance, and a triangular fuzzy evaluation matrix is ​​constructed. S312: Using the Grubbs test criterion, outlier removal is performed on the scoring evaluation index. Abnormal data with detection values ​​exceeding the critical value are removed to obtain the comprehensive triangular fuzzy number. S313: Perform consistency verification on the comprehensive triangular fuzzy matrix constructed by the comprehensive triangular fuzzy number, calculate the maximum eigenvalue of the matrix and the consistency ratio CR, and determine that the matrix is ​​consistent when CR < 0.1; S314: Convert the comprehensive fuzzy matrix that has passed the consistency verification into a fuzzy evaluation matrix, and calculate the adjustment judgment matrix by combining it with the median matrix composed of the median of the comprehensive fuzzy matrix. S315: Adjust the diagonal elements of the judgment matrix to 1, calculate the relative importance vector of each evaluation indicator using the square root method, standardize the relative importance vector, and obtain the subjective weight vector of each evaluation indicator.

[0016] In some embodiments, S32 includes: S321: Construct the original data matrix of real-time monitoring data for each evaluation indicator. For positive indicators with larger values ​​and higher safety levels, and negative indicators with smaller values ​​and higher safety levels, respectively, use the corresponding range standardization formula for standardization processing to eliminate the differences in the dimensions and orders of magnitude of different indicators. S322: Based on the standardized data, calculate the data weight of each evaluation indicator in each monitoring sample, and then calculate the information entropy value of each evaluation indicator according to the definition of information entropy. S323: Based on the information entropy values ​​of each evaluation index, an improved standardized weighting formula is used for calculation. The objective weight vectors corresponding to each evaluation index are calculated.

[0017] In some embodiments, S33 includes: S331: Construct a linear combination model of subjective weight vector and objective weight vector, set linear combination coefficients, and obtain the linear combination expression of comprehensive weight; S332: Based on game theory, an optimal model for combined weighting is constructed with the goal of minimizing the difference between subjective and objective weights. S333: Based on the properties of matrix differentiation, the optimization model is transformed into a system of linear equations that satisfy the condition that the first derivative is 0. Solving this system yields the optimal linear combination coefficients corresponding to the subjective weights and the objective weights. S334: Normalize the coefficients of the optimal linear combination obtained by the solution, and calculate the optimal comprehensive weight of each evaluation index by combining the subjective weight vector and the objective weight vector.

[0018] In some embodiments, the comprehensive evaluation model used in step S4 is the GRA-TOPSIS comprehensive evaluation model, and its specific implementation steps include: S41: Standardize the data of each evaluation indicator under real-time monitoring, and construct a weighted normalized evaluation matrix by combining the comprehensive weights of each evaluation indicator obtained in step S3. S42: Based on the weighted normalized evaluation matrix, select the optimal value for each evaluation index to construct a positive ideal solution, and select the worst value to construct a negative ideal solution. The positive ideal solution corresponds to the theoretically safest state, and the negative ideal solution corresponds to the theoretically most dangerous state. S43: Calculate the grey relational coefficients between the index sequence and the positive ideal solution and the negative ideal solution for each evaluation period, and obtain the corresponding grey relational degree by weighted average; S44: Calculate the Euclidean distance between the weighted index value vector and the positive ideal solution and the negative ideal solution for each time period to be evaluated; S45: After normalizing the grey relational degree and Euclidean distance respectively, linearly fuse them to calculate the comprehensive relative closeness between the construction state and the ideal safety state in each evaluation period; S46: Based on the preset range of the comprehensive relative proximity, determine and output the corresponding real-time construction risk level.

[0019] In some embodiments, in step S46, the correspondence between the overall relative proximity and the construction risk level is as follows: When the overall relative similarity is within the range of [0.85, 1], it is judged as Level I safety and controllability. When the overall relative proximity falls within the range of [0.7, 0.85), it is classified as a Level II low-risk concern level. When the overall relative proximity falls within the range of [0.5, 0.7), it is classified as a Level III medium-risk warning. When the overall relative proximity is in the range of [0, 0.5), it is judged as a level IV high-risk hazard.

[0020] In some embodiments, S5 includes: S51: Using historical multi-source monitoring data sequences as model input, the short-term and long-term time dependence characteristics of each risk indicator are captured through the LSTM layer; S52: Input the feature sequence output from the LSTM layer into the Transformer encoder layer, and fuse the spatiotemporal coupling relationship between signals from different risk sources through a multi-head self-attention mechanism; S53: Output the predicted value sequence of key risk indicators within a preset future time period, and input the predicted value sequence into the comprehensive evaluation model to achieve an advanced assessment of future risk status.

[0021] In some embodiments, in S6, the tiered prevention and control response mechanism includes five categories of measures: monitoring optimization, ventilation control, personnel management, emergency preparedness, and disaster relief. The corresponding level of measures is matched according to the real-time risk level and future risk trends. After the disposal measures are implemented, the multi-source real-time monitoring system collects the on-site data simultaneously and repeats S2 to S6 to achieve closed-loop iteration of risk control.

[0022] In some embodiments, the corresponding handling measures for matching the real-time risk level with the future risk trend include: Level I security controllability: If future indicators remain stable, routine intelligent monitoring and continuous operation of the ventilation system will be implemented to ensure H2S concentration <6.6ppm, and sensor and alarm system functions will be inspected. If there are any transient minor fluctuations in the future, perform disturbance source analysis and labeling, increase the monitoring frequency of the affected area, and conduct on-site inspections to check the ventilation and equipment status; Level II Low Risk Attention Level: If future fluctuation indicators can fall back to a safe range, implement measures such as increasing the monitoring frequency of relevant points, increasing the frequency of manual inspections to once every 2 hours, strengthening ventilation in local areas, and ensuring that the wind speed at the working face is ≥0.25m / s; If there is a possibility of future upgrades, implement ventilation enhancement measures in advance, issue preliminary warnings to personnel in affected areas, and prepare grouting and sealing equipment and materials. Level III Medium Risk Warning: If future risks can be contained through intervention, implement regional control and enhanced ventilation procedures, have workers wear self-contained breathing apparatus and explosion-proof communication equipment, and have emergency response teams on standby. If there is a risk of rapid escalation in the future, implement a 30-60 minute risk window countdown reminder, stop non-emergency operations and evacuate personnel, move the emergency team to the standby position and put the grouting and drainage equipment in place; Level IV High Risk Rating: Immediately implement the orders for full evacuation, emergency ventilation, and energy isolation; activate life-saving channels and transmit disaster and personnel location data to the emergency command center; trigger the pollution isolation barrier and carry out emergency transfer of poisoned personnel and environmental emergency response.

[0023] A system for predicting and controlling construction risks in sulfur-bearing strata tunnels, used to implement methods for predicting and controlling construction risks in sulfur-bearing strata tunnels, includes: The monitoring and sensing unit is used to build and operate a multi-source real-time monitoring system, and simultaneously collect data on gas concentration, geological information and personnel status. The indicator construction unit is used to establish a construction risk and safety evaluation indicator system and determine the evaluation benchmarks and grade classification standards for each indicator. The weight calculation unit is used to integrate the fuzzy network hierarchical analysis method, the improved entropy weight method, and the game theory combined weighting method to calculate the comprehensive weight of each evaluation index. The dynamic weight adjustment unit is used to dynamically adjust the weight coefficients of personnel behavior indicators based on real-time behavioral and physiological data of workers and their personal digital profiles. The risk assessment unit is used to run the GRA-TOPSIS comprehensive evaluation model, calculate the comprehensive relative proximity, and output the real-time construction risk level. The trend prediction unit is used to run the LSTM-Transformer hybrid time series prediction model to predict the future trends of key risk indicators. The intelligent response unit is used to automatically trigger and execute a tiered prevention and control response mechanism based on real-time risk levels and future trend predictions.

[0024] Compared with the prior art, the present invention has the following beneficial effects: This invention enables dynamic monitoring of the tunnel construction environment, the geological features ahead, and the behavior of workers. It overcomes the limitations of traditional monitoring that relies solely on a single gas alarm, and can identify hidden gas sources and abnormal gas outburst trends in advance. At the same time, it takes into account personnel risks as well as environmental and geological risks, greatly improving the early perception and warning capabilities of construction risks.

[0025] This invention employs an LSTM-Transformer hybrid model for time-series risk prediction, combined with a GRA-TOPSIS dynamic comprehensive evaluation model, to uncover risk evolution patterns from time-series monitoring data. Simultaneously, it dynamically integrates subjective and objective weights through a game theory-based weighting method, ensuring that the risk assessment results conform to both the objective statistical patterns of the monitoring data and the actual conditions of the tunnel engineering site. The output risk level and trend prediction results are highly accurate and interpretable, enabling advanced assessment and graded early warning of construction risks.

[0026] This invention can automatically trigger matched graded response measures such as enhanced ventilation, personnel management, grouting and sealing, and emergency evacuation based on real-time risk levels and future risk trends. It changes the traditional passive alarm and post-event management mode, realizes proactive risk prediction and early intervention, and forms a self-adaptive closed-loop risk management system for the entire process of sulfur-containing strata tunnel construction, significantly improving the intelligence level and proactive prevention and control capabilities of tunnel construction safety management. Attached Figure Description

[0027] Figure 1 A flowchart illustrating a method for risk prediction and control in tunnel construction in sulfur-containing strata; Figure 2 This is a schematic diagram of the TF-LSTM prediction model architecture. Figure 3 A schematic diagram of the closed-loop feedback mechanism of the personnel file management system. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figure 1 As shown, a method for risk prediction and prevention in tunnel construction in sulfur-bearing strata includes: S1: Construct a multi-source real-time monitoring system for tunnel construction in sulfur-containing strata, and collect data on the concentration of toxic gases in the tunnel space, the geological characteristics of the strata to be excavated ahead, and the behavior and physiological state of the workers inside the tunnel in real time. S2: Based on the monitoring data collected in S1, establish a construction risk safety evaluation index system covering gas environment risk, geological condition risk and personnel behavior risk, and determine the evaluation index and risk level classification standards. S3: Calculate the comprehensive weight of each evaluation index in the construction risk safety evaluation index system; S4: After standardization and comprehensive weighting, the real-time monitoring data is input into the comprehensive evaluation model to calculate the comprehensive relative closeness between the current construction status and the ideal safety status, and output the real-time construction risk level accordingly. S5: Construct an LSTM-Transformer hybrid time series forecasting model to predict the changing trends of key risk indicators within a preset future time period; S6: Based on the real-time construction risk level output by S4 and the future risk status change trend predicted by S5, the hierarchical prevention and control response mechanism is automatically triggered to execute matching risk disposal measures.

[0030] In a specific embodiment, step S1 involves constructing a multi-source real-time monitoring system, including a gas concentration monitoring system, a formation feature acquisition system, and a personnel file management system.

[0031] The gas concentration monitoring system aims to accurately capture the spatiotemporal distribution and dynamic changes of toxic gases. Specifically, fixed multi-parameter gas sensor arrays are deployed at key airflow points where gases tend to accumulate, such as tunnel faces, inverts, cross passages, and ventilation dead zones. These arrays continuously monitor the concentrations of hydrogen sulfide (H2S) and oxygen (O2), forming a baseline monitoring network covering the entire tunnel. Simultaneously, portable individual gas detectors are provided to every worker entering the tunnel to continuously monitor and upload real-time data on the concentrations of these key gases in their surrounding breathing zone. The monitoring point deployment needs to cover different mileage sections along the tunnel's longitudinal direction, as well as different elevation locations on the cross-section, such as the arch crown, arch waist, and sidewalls. Emphasis is placed on strengthening the monitoring density in high-risk areas such as sulfur-bearing strata, structural fracture zones, and unsealed areas behind the tunnel face, thereby constructing a dynamic sensing network for the gas concentration field across the entire space and all time periods.

[0032] The stratigraphic feature acquisition system aims to proactively determine the spatial distribution of sulfur-bearing strata, the state of gas occurrence, and potential gas migration and accumulation structures. By integrating ground-penetrating radar (GPR) for long-range geophysical scanning sensitive to gas-bearing media, it macroscopically delineates stratigraphic segments and anomaly zones potentially rich in toxic gases such as hydrogen sulfide. Combined with advanced horizontal drilling for precise verification, core sampling, and on-site gas testing, it directly obtains key information such as lithology, fracture development, and water content of the strata ahead. It also quantitatively analyzes the sulfide mineral content in the core, the odor of fracture-filling materials, and the gas concentration and pressure parameters within the borehole. The implementation of this system requires precise focus on key geological units closely related to the generation, storage, and sudden release of toxic gases, such as known sulfur-bearing strata, sulfide mineral enrichment zones, closed bedding or fracture development zones, and karst development zones. The detection operation was carried out in sections along the tunnel axis and staggered on the left and right lines to achieve a progressive identification of gas hazard sources from "possible sections" to "precise locations" and from "existence state" to "emergence risk," providing direct geological feature basis for subsequent prevention and control plans.

[0033] The personnel file management system aims to assign a unique digital identification code to each worker, using this code as the core to achieve dynamic quantitative assessment of individual risks. The system uses IoT terminals such as image recognition arrays, wearable physiological monitoring devices, and personal portable gas detectors to collect and input key behavioral indicator data for each worker in real time. This data primarily includes: the proper wearing status of safety equipment, compliance of work behavior, real-time physiological load, and the concentration of individual exposed gases. This continuously fed-back dynamic data is not only used for real-time analysis but also drives an adaptive weight adjustment mechanism. This mechanism dynamically optimizes the contribution weight of each indicator in the risk assessment model based on changes in indicators and the frequency of violations. Finally, the weighted comprehensive indicator data is first input into a time-series prediction network model to analyze the evolution trend of risk behavior; then it is imported into a safety risk assessment model to achieve a graded dynamic assessment and precise early warning of personnel safety risk levels.

[0034] Specifically, S1 includes: S11: Fixed multi-parameter gas sensor arrays are installed at the tunnel excavation face, cross passages and ventilation dead corners to continuously monitor the real-time concentration of hydrogen sulfide and oxygen in the tunnel; at the same time, portable personal gas detectors are provided to all workers in the tunnel to monitor and upload the concentration of toxic gases at the breathing zone of the workers in real time. S12: Integrating ground-penetrating radar and advanced horizontal drilling technology, it can explore the distribution of sulfur-containing strata, gas occurrence status and potential gas inrush channels in the section of the tunnel to be excavated, and obtain geological characteristic data of the strata ahead. S13: Through smart wearable devices and image recognition arrays, real-time data is collected on the safety equipment wearing status, work behavior compliance, physiological status indicators, and real-time location information of workers.

[0035] In a specific embodiment, to achieve accurate and dynamic assessment of individual risks, this application designs a closed-loop feedback mechanism ( Figure 3 This mechanism, based on combined weighting, links real-time monitored personnel behavior data with personal files and weighting factors, forming a continuously optimizing cycle. Specifically, it collects real-time work behavior and status information for each worker. This dynamic data is not only used to trigger immediate alerts, but more importantly, at the end of each assessment cycle, the system automatically updates the digital files of the corresponding personnel based on the latest behavioral records (such as whether protective equipment is worn correctly, whether high-risk areas are entered without authorization, etc.) and changes in their physiological state, and dynamically fine-tunes the weighting coefficients of their personal risk indicators in the evaluation model accordingly. This design allows personnel risk assessment to be dynamically adjusted according to their actual performance, ensuring the timeliness and relevance of the evaluation results, and providing a continuously optimized decision-making basis for subsequent implementation of differentiated and precise control.

[0036] The updated profile data is then directly used to drive the dynamic adjustment of weighting factors. Based on preset dynamic adjustment rules, the weighting algorithm can update each employee's individual weighting factors in the comprehensive evaluation model in real time, according to their recent behavioral patterns, such as the rising or falling trend of violation frequency or continuous changes in protective equipment usage habits. For example, if an employee with a good track record commits multiple violations recently, the weight of their behavioral indicator will be increased, making their overall risk level more sensitive to violations; conversely, for an employee whose behavior has continuously improved after targeted training, the weighting effect of their historical negative records will decrease.

[0037] Specifically, S13 includes: S131: Assign a unique digital identification code to each worker entering the tunnel, build a corresponding personal digital profile, and store and update the worker's historical behavior data, physiological status data, individual gas exposure data and violation records in real time; S132: Based on real-time collected data on the behavior and status of workers, their personal digital profiles are updated synchronously. The weight coefficients of the personnel behavior risk dimension indicators in the comprehensive evaluation model are dynamically adjusted according to the changing trends of the frequency of violations and the continuous changes in the standardization of protective equipment wearing. S133: Set a dynamic violation decay factor and introduce a time decay function to reduce the long-term impact of historical violation records on the current weight. The time decay function is as follows: ,in The time interval since the last violation. This is the attenuation coefficient, which is dynamically adjusted according to the effect. S134: Dynamic quantitative assessment and personalized weighting of individual worker risks based on adjusted weighting coefficients. The dynamic adjustment formula for weights is as follows: , For personnel i At any moment t The adjusted weights personnel i Based on the fundamental weights assigned by the combination at any given time. This refers to the recent frequency of violations. The average frequency of violations for the group. To adjust the sensitivity coefficient ( (This can change dynamically with the risk level.)

[0038] This mechanism enables risk assessment to continuously track and respond to an individual's latest behavioral performance, achieving feedback adjustment. It dynamically adjusts the weighting of the assessment based on recent actual data, allowing the risk assessment results to more timely and accurately reflect an individual's current safety status. This shifts safety management from relying on fixed records to real-time responses to dynamic behaviors, improving the targeting and accuracy of subsequent risk assessments.

[0039] In a specific embodiment, step S2 involves establishing a safety evaluation index system for tunnel construction risks in sulfur-bearing strata. This system comprises three dimensions: gas environment risk characteristics, geological condition risk characteristics, and personnel behavior risk characteristics. Gas environment risk characteristics include the coverage rate of a fixed monitoring network, gas concentration thresholds, and gas concentration change trends. Geological condition risk characteristics include the detection rate of key gas-bearing structures, the gas content level of the rock mass, gas occurrence state parameters, and the risk index of gas inrush channels. Personnel behavior risks include the rate of proper wearing of respiratory protective equipment, operational compliance index (such as whether high-risk areas have been entered, and whether operations comply with procedures), and physiological state level (such as a real-time fatigue index based on heart rate).

[0040] The risk level classification criteria are as follows: Level I, which is safe and controllable; Level II, which is low-risk and requires attention; Level III, which is a medium-risk warning level; and Level IV, which is a high-risk and dangerous level. The risk levels and their determinations are shown in the table below. Based on the safety risk levels of sulfur-containing tunnel construction and in conjunction with its indicator system, a benchmark system for sulfur-containing tunnel construction risk was established through literature review, on-site monitoring, and engineering analogy. The classification of evaluation indicators for each level is shown in the table below: The detailed descriptions of the evaluation indicators are shown in the table below: In a specific embodiment, step S3 calculates the safety assessment results for tunnel construction risks in sulfur-containing strata, which is the foundation for accurate risk prediction and dynamic prevention and control. Selecting a scientifically sound and reasonable weighting method for the indicators is crucial for accurate comprehensive evaluation of construction risks and safety. This step utilizes a subjective weighting method based on fuzzy analytic hierarchy process (FAHP), an objective weighting method based on improved entropy weighting, and a game theory-based optimal combination weighting method to obtain reasonable and accurate safety assessment indicator weights.

[0041] S3 specifically includes: S31: The fuzzy network hierarchical analysis method is used for subjective weighting. This method addresses the fuzziness in experts' judgments on the importance of indicators, while quantifying the networked correlation and mutual influence among the evaluation indicators, and calculating the subjective weight vector of each evaluation indicator.

[0042] In the risk and safety assessment system for tunnel construction in sulfur-bearing strata constructed in this application, the risk factors are not only diverse in type and nature, but more importantly, there are complex interdependencies and feedback relationships among the three major categories of indicators: gas, geology, and personnel. For example, geological conditions directly affect the risk of gas escape, while gas concentration dynamically restricts personnel operation standards and protection requirements. This non-hierarchical network relationship makes it difficult for traditional weighting methods to accurately characterize its intrinsic mechanism. The introduction of Fuzzy Network Hierarchical Analysis (FAHP) is precisely to address this challenge. By integrating fuzzy mathematics theory and network analysis structure, this method can effectively handle the inherent cognitive ambiguity of experts when judging the relative importance of indicators, allowing the use of natural language such as "slightly important" and "significantly important" for evaluation, which is more in line with the actual thinking mode of engineering decision-making. On the other hand, its network model can systematically describe and quantify the mutual influence relationships between different categories and indicators within the same category, thereby calculating weights that are more in line with the essence of complex risk systems. Therefore, adopting FAHP for subjective weighting lays a key methodological foundation for achieving a scientific and reasonable comprehensive risk assessment in this application.

[0043] S31 specifically includes: S311: Based on a preset quantitative scale for the importance of indicators, all evaluation indicators in the construction risk and safety evaluation indicator system are scored pairwise for relative importance, and a triangular fuzzy evaluation matrix is ​​constructed. Specifically, based on the factor importance judgment table, domain experts were invited to quantitatively score the 15 risk indicators in the 3 major categories, and to determine the triangular fuzzy evaluation matrix. : In the formula, N This represents the number of risk assessment indicators. N =15, The scale representing the relative importance of the k-th expert to the i-th and j-th risk indicators is a quantitative scale based on... The interval is the median value.

[0044] S312: Using the Grubbs test criterion, outliers are removed from the scoring evaluation indicators. Abnormal data with detection values ​​exceeding the critical value are removed to obtain the comprehensive triangular fuzzy number.

[0045] S313: Perform consistency verification on the comprehensive triangular fuzzy matrix constructed by the comprehensive triangular fuzzy number, calculate the maximum eigenvalue of the matrix and the consistency ratio CR, and determine that the matrix is ​​consistent when CR < 0.1; Based on the Grubbs test criterion, obviously unreasonable data information is removed to obtain the comprehensive triangular fuzzy number. And perform consistency verification: In the formula, L represent The three elements in it include , for K The average of the experts, s Standard deviation; critical value Gcrit Set as confidence threshold α =0.05, if the detected value G>Gcrit If so, then remove that value; λmax To synthesize the triangular fuzzy matrix The largest eigenvalue, RI As a random consistency indicator, CR For the consistency ratio, if CR If the value is less than 0.1, then the consistency is passed.

[0046] S314: Convert the comprehensive fuzzy matrix that has passed the consistency verification into a fuzzy evaluation matrix, and calculate the adjustment judgment matrix by combining it with the median matrix composed of the median of the comprehensive fuzzy matrix. The comprehensive fuzzy matrix B obtained after Grubbs detection is transformed to obtain the fuzzy evaluation matrix C; In the formula , which is the standard deviation rate, reflecting the degree of ambiguity in the expert review opinions; The median matrix V is derived from Median composition: ; Multiplying the median matrix by the fuzzy evaluation matrix yields the adjustment judgment matrix Q; .

[0047] S315: Adjust the diagonal elements of the judgment matrix to 1, calculate the relative importance vector of each evaluation indicator using the square root method, standardize the relative importance vector, and obtain the subjective weight vector of each evaluation indicator. The specific process is as follows: convert all the diagonal values ​​of the adjustment judgment matrix to 1 to obtain matrix Q′, and use the square root method to obtain the relative importance vector P; The relative importance matrix P is standardized to obtain the weight vector W. F This is the weight sought by FAHP.

[0048] Weight matrix W F Each column in This corresponds to the subjective weight given by each expert to each indicator.

[0049] S32: An improved entropy weighting method is used for objective weighting. The real-time monitored data of each indicator are standardized to eliminate the influence of dimensions. Then, the dispersion and effective information of each indicator data are calculated and analyzed through information entropy to obtain the objective weight vector of each evaluation indicator.

[0050] In this application, gas concentration, geological parameters, and personnel behavior indicators were all collected at high frequency and continuously using sensors, forming a massive real-time monitoring data sequence. Traditional objective weighting methods (such as the standard entropy weighting method), while able to allocate weights based on the data's inherent dispersion, are insufficient to fully meet the specific needs of risk assessment in this scenario: First, the dimensions and orders of magnitude of different indicators differ significantly, requiring more scientific standardization to eliminate scale effects; second, the inherent engineering risk value of certain key indicators (such as "gas concentration change trends") may not be fully reflected by their static variability; third, the data distribution at different construction stages may dynamically change, requiring the weights to possess a certain degree of robustness and adaptability. Therefore, this application employs an improved entropy weighting method for objective weighting. This method, while retaining the core idea of ​​information entropy, introduces a more reasonable standardization mechanism, a sensitivity enhancement factor for key risk indicators, and a dynamic adjustment strategy considering data timeliness. This ensures that weight allocation is not only based on the statistical characteristics of the data but also closely aligns with the engineering reality of the evolution of risks in sulfur-containing tunnel construction, thus providing a solid and reliable data-driven foundation for the subsequent scientific integration of subjective and objective weights.

[0051] In a specific embodiment, S32 includes: S321: Construct the original data matrix of real-time monitoring data for each evaluation indicator. For positive indicators with larger values ​​and higher safety levels, and negative indicators with smaller values ​​and higher safety levels, respectively, use the corresponding range standardization formula for standardization processing to eliminate the differences in the dimensions and orders of magnitude of different indicators. Specifically, an original data matrix is ​​constructed, and the data for each indicator are standardized using 0-1 to eliminate the influence of dimensions. For positive indicators, the larger the value, the better the indicator. The standardized calculation formula for the indicator data is: For contrarian indicators, the smaller the value, the better the indicator. The standardized calculation formula for the indicator data is: In the formula, , These are the first and second parts before and after standardization. i The first object j Evaluation indicators; , The first j The maximum and minimum values ​​of each evaluation indicator.

[0052] S322: Based on the standardized data, calculate the data weight of each evaluation indicator in each monitoring sample, and then calculate the information entropy value of each evaluation indicator according to the definition of information entropy. Specifically, the first j If we consider the security status of each evaluation indicator as a system, then we can derive the result of the first evaluation indicator based on the definition of entropy. j Information entropy value of each evaluation indicator e j : in, For the first i The sample at the th j The proportion of each indicator .

[0053] S323: Based on the information entropy values ​​of each evaluation index, an improved standardized weighting formula is used for calculation. The objective weight vectors corresponding to each evaluation index are calculated.

[0054] S33: Based on game theory, construct a combined weighting optimization model. With minimizing the deviation between subjective weights and objective weights as the optimization objective, solve for the optimal linear combination coefficients of subjective and objective weights. After normalizing the combination coefficients, calculate the optimal comprehensive weights of each evaluation index.

[0055] Based on the established FAHP subjective weights and the improved entropy weight method objective weights, effectively combining the two is crucial to ensuring the reliability of risk assessment results. Simply using a weighted average is insufficient to reconcile potential discrepancies between expert experience and data patterns, and it fails to reflect the complex interrelationships between gas, geological, and personnel risks during sulfur-containing tunnel construction. Therefore, this patent employs a combined weighting method based on game theory, treating the determination of subjective and objective weights as a multi-party collaborative optimization process. By establishing a game equilibrium model, it seeks a comprehensive set of weights that maximizes consensus and minimizes discrepancies between the two. This method respects the domain experts' profound understanding of risk mechanisms while fully incorporating the statistical patterns revealed by real-time monitoring data, resulting in a weight allocation outcome that is more closely aligned with engineering realities and exhibits greater stability, providing a basis for subsequent accurate risk assessment.

[0056] In a specific embodiment, the following are the specific calculation steps of this combined weighting method: set up Given the linear combination coefficients, the linear combination of the two weight vectors is: In the formula: w F The set of weight vectors for the analytic hierarchy process; w E To improve the weight vector set of the entropy weight method; w F w E All n 3D feature vectors ( n This represents the total number of evaluation indicators for the construction risks of tunnels in sulfur-bearing strata. n =15); These are their coefficients.

[0057] Based on the idea of ​​game theory aggregation model, this paper optimizes the coefficients of two linear combinations with the objective of minimizing deviation to obtain the optimal weight W. The optimization problem is established as follows: In the formula: These are the weighting coefficients; For the l-th weight vector set ( l =1,2,…, k ; k To determine the number of weighting methods, this patent employs two methods: fuzzy analytic hierarchy process (AHP) and improved entropy weighting method. k =2.

[0058] Based on the properties of matrix differentiation, the above equation can be equivalently transformed into a system of linear equations with optimal first-order derivative conditions: Calculate the optimal combination coefficients , And after normalization, we obtain , The optimal overall weight of the indicators can be obtained as follows: .

[0059] In a specific embodiment, the comprehensive evaluation model used in S4 is the GRA-TOPSIS comprehensive evaluation model.

[0060] Based on the evaluation levels of safety assessment indicators obtained from multi-source monitoring data including gas, geology, and personnel, and the determination of the comprehensive weight of each indicator using game theory-based weighting, this application adopts the GRA-TOPSIS (Grey Relational Analysis-Approximation Ideal Solution Ranking Method) model as the core algorithm for comprehensive risk state assessment in order to ultimately achieve dynamic and accurate evaluation and ranking of the entire tunnel construction risk. This method integrates the advantages of both grey relational analysis and the ideal point method. On the one hand, it uses grey relational analysis (GRA) to assess the similarity in changing trends between each state to be evaluated (such as different times or different cross-sections) and the ideal risk sequence, effectively handling incomplete information and uncertainty in the data. On the other hand, it uses the TOPSIS method to calculate the spatial proximity of each state to the "safest ideal point" and the "most dangerous negative ideal point," thereby accurately determining its risk ranking and level under the constraint of indicator weights. This model is suitable for multi-dimensional risk assessment scenarios such as sulfur-containing tunnel construction risks, which involve mixed indicators, fluctuating data, and require clear classification, and can output intuitive and reliable comprehensive risk ranking results.

[0061] Its specific implementation steps include: S41: Standardize the data of each evaluation indicator under real-time monitoring, and construct a weighted normalized evaluation matrix by combining the comprehensive weights of each evaluation indicator obtained in step S3. The gas monitoring system, geological forecasting system, and personnel file management system acquire real-time observations of all safety evaluation indicators for the period or section to be evaluated, forming an initial data sequence.

[0062] For benefit-oriented (lower values ​​indicate greater safety, such as gas concentration) and cost-oriented (higher values ​​indicate greater safety, such as equipment wear rate) indicators, appropriate standardization formulas (such as the range method) are used to uniformly transform all indicator values ​​to the [0, 1] interval, eliminating the influence of dimensions. The standardization method is consistent with the objective weight determination method in step 3.

[0063] The combined weights of the indicators determined by the game theory combinatorial weighting method Multiply by the corresponding normalized index value respectively The weighted normalized evaluation matrix is ​​obtained. U This matrix reflects both the actual observation level of each indicator and its relative importance in the overall risk.

[0064] S42: Based on the weighted normalized evaluation matrix, select the optimal value for each evaluation index to construct a positive ideal solution, and select the worst value to construct a negative ideal solution. The positive ideal solution corresponds to the theoretically safest state, and the negative ideal solution corresponds to the theoretically most dangerous state. Determining the ideal solution: From the weighted normalized matrix, for each indicator, select the optimal value from all evaluated objects (e.g., different times or different cross-sections) to form a virtual optimal sequence. For benefit-type indicators, take the maximum value. For cost-related indicators, take the minimum value. This sequence represents the theoretically "safest" ideal state.

[0065] Determining the negative ideal solution: Similarly, select the worst-case values ​​for each indicator to form a worst-case sequence. For benefit-related indicators, take the minimum value. For cost-related indicators, take the maximum value. This sequence represents the theoretically "most dangerous" state.

[0066] .

[0067] S43: Calculate the grey relational coefficients between the index sequence and the positive ideal solution and the negative ideal solution for each evaluation period, and obtain the corresponding grey relational degree by weighted average; Specifically, the index sequence and the positive ideal solution sequence for each object to be evaluated are calculated separately. Negative ideal solution sequence Correlation coefficients at each corresponding indicator point , The correlation coefficient reflects how similar the trends of two sequences are in terms of a single indicator.

[0068] in, It is a two-level minor difference. It consists of two levels of small to large differences; The resolution coefficient is typically set to 0.5.

[0069] Calculating grey relational degree: Associating a given object with the ideal solution. The grey relational degree between the object and the "safest" state is obtained by taking a weighted average of the correlation coefficients across all indicators. Similarly, calculate its difference from the negative ideal solution. Grey relational degree The greater the correlation, the more similar the trend of the object is to the reference sequence.

[0070] .

[0071] S44: Calculate the Euclidean distance between the weighted index value vector and the positive ideal solution and the negative ideal solution for each time period to be evaluated; The weighted index value vector of each object to be evaluated to the positive ideal solution vector With negative ideal solution vector Euclidean distance , The smaller the distance value, the closer they are in space.

[0072] .

[0073] S45: After normalizing the grey relational degree and Euclidean distance respectively, linearly fuse them to calculate the comprehensive relative closeness between the construction state and the ideal safety state in each evaluation period; Dimensionless processing: Normalize the two grey relational degree sequences obtained in step three and the two Euclidean distance sequences obtained in step four to make them comparable.

[0074] Linear fusion: For each object to be evaluated, calculate its fusion proximity to the positive ideal solution. and the degree of fusion with the negative ideal solution. The fusion process sets a balance coefficient. This is to adjust the weighting of trend similarity and spatial proximity in the evaluation. Generally, This indicates that the contribution proportions are equal.

[0075] Calculate the overall proximity score: based on the proximity score after fusion. and distance Calculate the overall relative closeness of each evaluated object to the ideal state. , This value ranges from 0 to 1. The larger the value, the closer the overall risk status of the object is to the "safest" ideal solution, and the lower the risk; the smaller the value, the closer it is to the "most dangerous" state.

[0076] S46: Based on the preset range of the comprehensive relative proximity, determine and output the corresponding real-time construction risk level.

[0077] In a specific embodiment, in step S5, Building upon the static comprehensive evaluation of construction risks based on the GRA-TOPSIS model, this application introduces an LSTM-Transformer hybrid time-series prediction model to deeply mine and predict the complex time dependencies and dynamic coupling relationships in multi-source monitoring data, aiming to construct a deeper and more forward-looking dynamic early warning capability. The formation and evolution of construction risks in sulfur-bearing strata tunnels are not only reflected in the continuous changes of single indicators on the time axis (such as gas concentration accumulation), but more importantly, in the nonlinear interaction and resonance of multi-source heterogeneous signals (such as the frequency of personnel violations and the rate of proper wearing of respiratory protective equipment) under specific spatiotemporal conditions. While a single LSTM model can effectively capture the short-term memory and progressive patterns of sequences, it has limitations in processing ultra-long sequences and mining global spatiotemporal features; while the Transformer's self-attention mechanism can accurately quantify the global dependencies between data points at any given time, it is less sensitive to the strict order of the sequence. The LSTM-Transformer hybrid architecture constructed in this patent first utilizes the LSTM layer to perform preliminary feature extraction and context encoding on the historical sequences of various indicators, capturing their basic temporal evolution patterns. Then, the feature sequences rich in temporal information are input into the Transformer encoder layer. Through its multi-head attention mechanism, the mutual influence weights of different risk source signals at different time steps are dynamically calculated and fused, thereby achieving multi-step, high-precision prediction of the future state of sulfur-containing tunnel construction risks. By inputting the predicted "future indicator trajectory" into the safety evaluation system in advance, the evolution direction and potential coupling effects can be preemptively assessed before the risks manifest, greatly enhancing the initiative, reliability, and intelligence level of risk prediction. The following are the specific calculation steps for time-series prediction and trend analysis using this hybrid model.

[0078] The construction risk trend prediction model constructed in this application adopts an LSTM-Transformer fusion structure. The input is a 24-hour time series, and each time step contains six core features: fixed-point H2S concentration, individual exposure H2S concentration, oxygen concentration, gas concentration change trend, respiratory protective equipment standard wearing rate, and real-time physiological fatigue index. These features collectively characterize the key driving factors affecting the dynamic changes of construction risks, including the gas environment state and the real-time behavior and state of personnel. The model output is a sequence of predicted values ​​for each key risk indicator for the next hour. The training set, validation set, and test set are divided in a time-series ratio of 8:1:1.

[0079] The sample data is constructed into a feature matrix and input into the LSTM layer, which serves as the primary temporal feature extractor. Gating mechanisms (forget gate, input gate, and output gate) capture the short-term and long-term dependencies of each indicator over time, forming a hidden state sequence rich in temporal context information. The gating mechanism formula for LSTM with multi-source monitoring data is shown below: The forgetting gate determines which historical information in a cell's state needs to be discarded or attenuated. Its output is a value between 0 and 1, used to control the degree to which historical memory is retained. The formula for calculating the forgetting gate is as follows: The input gate determines how much of the current input information needs to be updated in the cell state. It consists of a sigmoid layer and a tanh layer for generating candidate values. The formula for calculating the input gate is as follows: The cell state update process first involves selectively forgetting the old state based on the output of the forget gate, and then selectively adding new candidate information under the control of the input gate, thus completing the memory update. The calculation formula for cell state update is as follows: The output gate determines the hidden state of the final output based on the updated cell state and the current input. This hidden state contains the temporal information most relevant to the current task, filtered by the network, and is passed to the next time step or the output layer. The formula for calculating the output gate is as follows: in, x t For time step t The input vector, h t For time step t Hidden state output, C t For time step t cellular state; f t The output vector of the forget gate controls the degree to which historical information is retained; i t The input gate output vector controls the degree to which new information is added. o t The output vector of the output gate controls the output level of the current state. This represents the candidate cell state vector; W f , W i , W C , W o This is the corresponding weight matrix; b f , b i , b C , bo This is the corresponding bias vector. for sigmoid function, tanh is the hyperbolic tangent function. .

[0080] After format adjustment in the feature transformation stage, the temporal feature sequences captured by the LSTM layer are fed into the Transformer layer. In this application, the Transformer layer, as the core module of risk temporal fusion, is responsible for deep modeling of the temporal features initially encoded by LSTM, in order to construct global dynamic features that can characterize the complex nonlinear correlations between multi-source risk indicators. This layer first combines the input features with positional encodings reflecting the construction time sequence pattern, thereby incorporating temporal sequence information. Subsequently, a multi-head self-attention mechanism is used to perform global interactive modeling of the sequence, fully exploring the dependencies between different times and different risk features while ensuring the legitimacy of the time sequence prediction. The multi-head attention mechanism can extract information from multiple feature subspaces in parallel, enhancing the model's representational ability. Finally, feature integration and transformation are performed using a feedforward neural network and residual connections, extracting high-level fusion features while retaining important temporal patterns, providing reliable predictions for the final risk trend forecast. The key calculation formulas of the Transformer layer are as follows: In the formula, H Input the feature sequence to the LSTM. Q , K and V These represent the query vector, key vector, and value vector, respectively. W Q , W K , W V and W O It is a trainable convolutional kernel; h i Indicates the first i One attention point, among which i =1,2,..., h , h It's about the number of heads; Indicates the dimension of the key vector; M It is a causal mask matrix to ensure that the model only focuses on historical and current information; softmax represents the activation function; Concat represents the concatenation function; Attention Represents the attention function; MultiHead This represents a multi-head self-attention function.

[0081] To improve the prediction accuracy of the LSTM-Transformer model, mean squared error is used during the model training process. MSE The network parameters are optimized as a core component of the loss function. The adaptive optimizer Adam iteratively updates the network parameters until the model can accurately capture the multidimensional risk evolution patterns based on gas concentration and human behavior. In the forward prediction stage, the model output is a normalized sequence of predicted values, which needs to be restored to their original dimensions through an inverse normalization operation to finally obtain the specific numerical predictions of each risk indicator for future periods.

[0082] Finally, to comprehensively evaluate the performance of the LSTM-Transformer hybrid model in predicting the construction risk trends of sulfur-containing tunnels, this patent uses the average relative error. MRE With root mean square error RMSE As a core indicator for evaluating prediction accuracy, the mean relative error (MRE) directly reflects the average deviation between predicted and observed values, demonstrating the overall relative accuracy of the model's predictions. The root mean square error (RMSE), by squaring the error, is more sensitive to larger deviations in the prediction, thus measuring the stability and reliability of the prediction results. The calculation formula for the indicators is as follows: In the formula, y i These are measured values. is the predicted value, and n is the total number of samples.

[0083] In a specific embodiment, in step S6, based on the forward-looking risk prediction of the LSTM-Transformer hybrid model and the dynamic comprehensive evaluation results of GRA-TOPSIS, combined with the "Technical Specification for Design and Construction of Hydrogen Sulfide Tunnels on Highways DB63T2384-2024", this application constructs the following hierarchical prevention and control response mechanism and implements matching risk disposal measures.

[0084] After the response measures are implemented, a new round of multi-source real-time monitoring is automatically initiated to continuously track changes in gas concentration, geological response, and personnel behavior. The feedback data is then input into the aforementioned prediction and evaluation model to verify and re-evaluate the risk control effect in real time, thereby driving the dynamic optimization of the management and control strategy and forming a precise, proactive, and adaptive construction risk management and control closed loop.

[0085] A risk prediction and control system for tunnel construction in sulfur-bearing strata is provided for implementing risk prediction and control methods for tunnel construction in sulfur-bearing strata. The system includes: a monitoring and sensing unit for constructing and operating a multi-source real-time monitoring system to simultaneously collect gas concentration, geological information, and personnel status data; an indicator construction unit for establishing a construction risk safety evaluation indicator system and determining the evaluation benchmarks and level classification standards for each indicator; a weight calculation unit for integrating fuzzy network hierarchical analysis, improved entropy weighting, and game theory combined weighting methods to calculate the comprehensive weight of each evaluation indicator; a dynamic weight adjustment unit for dynamically adjusting the weight coefficients of personnel behavior indicators based on real-time behavior and physiological data of workers, combined with personal digital profiles; a risk assessment unit for running a GRA-TOPSIS comprehensive evaluation model to calculate the comprehensive relative proximity and output the real-time construction risk level; a trend prediction unit for running an LSTM-Transformer hybrid time series prediction model to predict the future changing trends of key risk indicators; and an intelligent response unit for automatically triggering and executing a graded prevention and control response mechanism based on the real-time risk level and future trend prediction results.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for risk prediction and prevention in tunnel construction in sulfur-bearing strata, characterized in that, include: S1: Construct a multi-source real-time monitoring system for tunnel construction in sulfur-containing strata, and collect data on the concentration of toxic gases in the tunnel space, the geological characteristics of the strata to be excavated ahead, and the behavior and physiological state of the workers inside the tunnel in real time. The multi-source real-time monitoring system includes: A gas concentration monitoring system, comprising: A multi-parameter gas sensor array is deployed at locations where tunnel gases tend to accumulate or at key airflow nodes to continuously monitor hydrogen sulfide and oxygen concentrations, forming a baseline monitoring network covering the entire tunnel. Individual gas detectors are provided, and each worker entering the tunnel is equipped with a portable one that continuously monitors and uploads data on the concentration of hydrogen sulfide and oxygen at the worker's breathing zone in real time. Stratigraphic feature acquisition system, the stratigraphic feature acquisition system comprising: Ground-penetrating radar (GPR) performs long-range scanning of the tunnel section to be excavated, macroscopically delineating strata rich in hydrogen sulfide toxic gases and abnormal areas. Advanced horizontal drilling is used for verification, core sampling and on-site gas testing to obtain key information on the lithology, fracture development and water-bearing capacity of the strata ahead, and to quantitatively analyze the sulfide mineral content of the core, the gas concentration and pressure parameters in the borehole. Personnel file management system, the personnel file management system including: An image recognition array is used to collect and record in real time the status of workers' safety equipment wearing and the compliance of their work behavior. Wearable physiological monitoring devices are used to collect real-time data on the physiological load and individual exposure gas concentrations of the wearer; S1 specifically includes: S11: Fixed multi-parameter gas sensor arrays are installed at the tunnel excavation face, cross passages and ventilation dead corners to continuously monitor the real-time concentration of hydrogen sulfide and oxygen in the tunnel; at the same time, portable personal gas detectors are provided to all workers in the tunnel to monitor and upload the concentration of toxic gases at the breathing zone of the workers in real time. S12: Integrating ground-penetrating radar and advanced horizontal drilling technology, it can explore the distribution of sulfur-containing strata, gas occurrence status and potential gas inrush channels in the section of the tunnel to be excavated, and obtain geological characteristic data of the strata ahead. S13: Through smart wearable devices and image recognition arrays, real-time data is collected on the safety equipment wearing status, work behavior compliance, physiological status indicators, and real-time location information of workers. S2: Based on the monitoring data collected in S1, establish a construction risk safety evaluation index system covering gas environment risk, geological condition risk and personnel behavior risk, and determine the evaluation index and risk level classification standards. In S2, the construction risk safety evaluation index system specifically includes: The gas environment risk characteristic evaluation indicators include: fixed monitoring network coverage, fixed-point hydrogen sulfide concentration, individual exposure to hydrogen sulfide concentration, gas concentration change trend, and oxygen concentration. Geological condition risk characteristic evaluation indicators include: key gas-bearing structure detection rate, rock mass gas-bearing grade, gas occurrence state parameters, peak concentration of hydrogen sulfide in borehole and gas inrush channel risk index. Personnel behavioral risk characteristic evaluation indicators include: the rate of standardized wearing of respiratory protective equipment, the frequency of unauthorized entry into high-risk areas, and the real-time physiological fatigue index. S3: Calculate the comprehensive weight of each evaluation index in the construction risk safety evaluation index system; The calculation of the comprehensive weight in S3 specifically includes: S31: The fuzzy network hierarchical analysis method is used for subjective weighting. This method addresses the fuzziness in experts' judgments on the importance of indicators, while quantifying the networked correlation and mutual influence among the evaluation indicators, and calculating the subjective weight vector of each evaluation indicator. S32: An improved entropy weighting method is used for objective weighting. The real-time monitored data of each indicator are standardized to eliminate the influence of dimensions. Then, the dispersion and effective information of each indicator data are calculated and analyzed through information entropy to obtain the objective weight vector of each evaluation indicator. S33: Based on game theory, construct a combined weighting optimization model, take minimizing the deviation between subjective weights and objective weights as the optimization objective, solve for the optimal linear combination coefficients of subjective and objective weights, and calculate the optimal comprehensive weights of each evaluation index after normalizing the combination coefficients. S4: After standardization and comprehensive weighting, the real-time monitoring data is input into the comprehensive evaluation model to calculate the comprehensive relative closeness between the current construction status and the ideal safety status, and output the real-time construction risk level accordingly. In step S4, the comprehensive evaluation model used is the GRA-TOPSIS comprehensive evaluation model, and its specific implementation steps include: S41: Standardize the data of each evaluation indicator under real-time monitoring, and construct a weighted normalized evaluation matrix by combining the comprehensive weights of each evaluation indicator obtained in step S3. S42: Based on the weighted normalized evaluation matrix, select the optimal value for each evaluation index to construct a positive ideal solution, and select the worst value to construct a negative ideal solution. The positive ideal solution corresponds to the theoretically safest state, and the negative ideal solution corresponds to the theoretically most dangerous state. S43: Calculate the grey relational coefficients between the index sequence and the positive ideal solution and the negative ideal solution for each evaluation period, and obtain the corresponding grey relational degree by weighted average; S44: Calculate the Euclidean distance between the weighted index value vector and the positive ideal solution and the negative ideal solution for each time period to be evaluated; S45: After normalizing the grey relational degree and Euclidean distance respectively, linearly fuse them to calculate the comprehensive relative closeness between the construction state and the ideal safety state in each evaluation period; S46: Determine and output the corresponding real-time construction risk level based on the preset range of the overall relative proximity; S5: Construct an LSTM-Transformer hybrid time series forecasting model to predict the changing trends of key risk indicators within a preset future time period; The S5 includes: S51: Using historical multi-source monitoring data sequences as model input, the short-term and long-term time dependence characteristics of each risk indicator are captured through the LSTM layer; S52: Input the feature sequence output from the LSTM layer into the Transformer encoder layer, and fuse the spatiotemporal coupling relationship between signals from different risk sources through a multi-head self-attention mechanism; S53: Output the predicted value sequence of key risk indicators within a preset future time period, and input the predicted sequence into the comprehensive evaluation model to achieve an advanced assessment of future risk status; S6: Based on the real-time construction risk level output by S4 and the future risk status change trend predicted by S5, the hierarchical prevention and control response mechanism is automatically triggered to execute matching risk disposal measures. In S6, the tiered prevention and control response mechanism includes five categories of measures: monitoring optimization, ventilation control, personnel management, emergency preparedness, and emergency response. The corresponding level of response measures is matched according to the real-time risk level and future risk trends. After the disposal measures are implemented, the multi-source real-time monitoring system collects the on-site data simultaneously and repeats S2 to S6 to achieve closed-loop iteration of risk control.

2. The method for risk prediction and prevention in tunnel construction in sulfur-bearing strata according to claim 1, characterized in that, S13 specifically includes: S131: Assign a unique digital identification code to each worker entering the tunnel, build a corresponding personal digital profile, and store and update the worker's historical behavior data, physiological status data, individual gas exposure data and violation records in real time; S132: Based on real-time collected data on the behavior and status of workers, their personal digital profiles are updated synchronously. The weight coefficients of the personnel behavior risk dimension indicators in the comprehensive evaluation model are dynamically adjusted according to the changing trends of the frequency of violations and the continuous changes in the standardization of protective equipment wearing. S133: Set a dynamic violation decay factor and introduce a time decay function to reduce the long-term impact of historical violation records on the current weight. The time decay function is as follows: ,in The time interval since the last violation. This is the attenuation coefficient, which is dynamically adjusted according to the effect. S134: Dynamic quantitative assessment and personalized weighting of individual worker risks based on adjusted weighting coefficients. The dynamic adjustment formula for weights is as follows: , For personnel i At any moment t The adjusted weights personnel i Based on the fundamental weights assigned by the combination at any given time. This refers to the recent frequency of violations. The average frequency of violations for the group. The sensitivity coefficient can be dynamically adjusted according to the risk level.

3. The method for risk prediction and prevention in sulfur-bearing strata tunnel construction according to claim 1, characterized in that, In S2, the risk level classification standard is as follows: Level I, Level I is a safe and controllable level; Level II, Level II is a low-risk level of concern; Level III, which is a medium-risk warning level; Level IV, which is a high-risk hazard level.

4. The method for risk prediction and prevention in sulfur-bearing strata tunnel construction according to claim 3, characterized in that, The grading of evaluation indicators for each level is as follows: Level I: Gas environment risk dimension, fixed monitoring network coverage ≥95%, fixed point hydrogen sulfide concentration <5ppm, individual exposure hydrogen sulfide concentration <5ppm, gas concentration trend is stable or decreasing, oxygen concentration is 20.0%~20.9%Vol; Geological condition risk dimension, key gas-bearing structure detection rate ≥90%, rock mass gas content level is low, gas occurrence state parameter <0.3, peak hydrogen sulfide concentration in borehole <10ppm, gas surge channel risk index <0.3; Personnel behavior risk dimension, respiratory protective equipment standard wearing rate ≥98%, frequency of unauthorized entry into high-risk areas is 0 times / hour, real-time physiological fatigue index <0.3; Level II: Gas environment risk dimension: Fixed monitoring network coverage is 85%~95%, fixed point hydrogen sulfide concentration is 5~10 ppm, individual exposure hydrogen sulfide concentration is 5~10 ppm, gas concentration trend is slight fluctuation, oxygen concentration is 19.5%~20.0% Vol; Geological condition risk dimension: Key gas-bearing structures are identified at 70%~90%, rock mass gas content is moderate, gas occurrence state parameter is 0.3~0.5, peak hydrogen sulfide concentration in boreholes is 10~30 ppm, gas surge channel risk index is 0.3~0.5; Personnel behavior risk dimension: Respiratory protective equipment standard wearing rate is 95%~98%, frequency of unauthorized entry into high-risk areas is 1 time / hour, real-time physiological fatigue index is 0.3~0.5; Level III: Gas environment risk dimension, fixed monitoring network coverage is 75%~85%, fixed point hydrogen sulfide concentration is 10~30ppm, individual exposure hydrogen sulfide concentration is 10~30ppm, gas concentration trend is a continuous and slow increase, oxygen concentration is 18.0%~19.5%Vol; Geological condition risk dimension, key gas-bearing structures are identified at 50%~70%, rock mass gas content is high, gas occurrence state parameter is 0.5~0.7, peak hydrogen sulfide concentration in boreholes is 30~100ppm, gas surge channel risk index is 0.5~0.7; Personnel behavior risk dimension, respiratory protective equipment standard wearing rate is 90%~95%, frequency of unauthorized entry into high-risk areas is 2~3 times / hour, real-time physiological fatigue index is 0.5~0.7; Level IV: Gas environment risk dimension: Fixed monitoring network coverage <75%, fixed point hydrogen sulfide concentration ≥30ppm, individual exposure hydrogen sulfide concentration ≥30ppm, gas concentration change trend is rapidly increasing, oxygen concentration <18.0%Vol; Geological condition risk dimension: Key gas-bearing structure detection rate <50%, rock mass gas content level is extremely high, gas occurrence state parameter ≥0.7, peak hydrogen sulfide concentration in borehole ≥100ppm, gas surge channel risk index ≥0.7; Personnel behavior risk dimension: Respiratory protective equipment standard wearing rate <90%, frequency of unauthorized entry into high-risk areas ≥4 times / hour, real-time physiological fatigue index ≥0.

7.

5. The method for risk prediction and prevention in tunnel construction in sulfur-bearing strata according to claim 1, characterized in that, S31 includes: S311: Based on a preset quantitative scale for the importance of indicators, all evaluation indicators in the construction risk and safety evaluation indicator system are scored pairwise for relative importance, and a triangular fuzzy evaluation matrix is ​​constructed. S312: Using the Grubbs test criterion, outlier removal is performed on the scoring evaluation index. Abnormal data with detection values ​​exceeding the critical value are removed to obtain the comprehensive triangular fuzzy number. S313: Perform consistency verification on the comprehensive triangular fuzzy matrix constructed by the comprehensive triangular fuzzy number, calculate the maximum eigenvalue of the matrix and the consistency ratio CR, and determine that the matrix is ​​consistent when CR < 0.1; S314: Convert the comprehensive fuzzy matrix that has passed the consistency verification into a fuzzy evaluation matrix, and calculate the adjustment judgment matrix by combining it with the median matrix composed of the median of the comprehensive fuzzy matrix. S315: Adjust the diagonal elements of the judgment matrix to 1, calculate the relative importance vector of each evaluation indicator using the square root method, standardize the relative importance vector, and obtain the subjective weight vector of each evaluation indicator.

6. The method for risk prediction and prevention in tunnel construction in sulfur-bearing strata according to claim 1, characterized in that, S32 includes: S321: Construct the original data matrix of real-time monitoring data for each evaluation indicator. For positive indicators with larger values ​​and higher safety levels, and negative indicators with smaller values ​​and higher safety levels, respectively, use the corresponding range standardization formula for standardization processing to eliminate the differences in the dimensions and orders of magnitude of different indicators. S322: Based on the standardized data, calculate the data weight of each evaluation indicator in each monitoring sample, and then calculate the information entropy value of each evaluation indicator according to the definition of information entropy. S323: Based on the information entropy values ​​of each evaluation index, an improved standardized weighting formula is used for calculation. The objective weight vectors corresponding to each evaluation index are calculated.

7. The method for risk prediction and prevention in tunnel construction in sulfur-bearing strata according to claim 1, characterized in that, S33 includes: S331: Construct a linear combination model of subjective weight vector and objective weight vector, set linear combination coefficients, and obtain the linear combination expression of comprehensive weight; S332: Based on game theory, an optimal model for combined weighting is constructed with the goal of minimizing the difference between subjective and objective weights. S333: Based on the properties of matrix differentiation, the optimization model is transformed into a system of linear equations that satisfy the condition that the first derivative is 0. Solving this system yields the optimal linear combination coefficients corresponding to the subjective weights and the objective weights. S334: Normalize the coefficients of the optimal linear combination obtained by the solution, and calculate the optimal comprehensive weight of each evaluation index by combining the subjective weight vector and the objective weight vector.

8. The method for risk prediction and prevention in tunnel construction in sulfur-bearing strata according to claim 1, characterized in that, In S46, the correspondence between the overall relative proximity and the construction risk level is as follows: When the overall relative similarity is within the range of [0.85, 1], it is judged as Level I safety and controllability. When the overall relative proximity falls within the range of [0.7, 0.85), it is classified as a Level II low-risk concern. When the overall relative proximity falls within the range of [0.5, 0.7), it is classified as a Level III medium-risk warning. When the overall relative proximity is in the range of [0, 0.5), it is judged as a level IV high-risk hazard.

9. The method for risk prediction and prevention in tunnel construction in sulfur-bearing strata according to claim 1, characterized in that, The corresponding handling measures for matching the real-time risk level with the future risk trend include: Level I security controllability: If future indicators remain stable, routine intelligent monitoring and continuous operation of the ventilation system will be implemented to ensure H2S concentration <6.6ppm, and sensor and alarm system functions will be inspected. If there are any transient minor fluctuations in the future, perform disturbance source analysis and labeling, increase the monitoring frequency of the affected area, and conduct on-site inspections to check the ventilation and equipment status; Level II Low Risk Attention Level: If future fluctuation indicators can fall back to a safe range, implement measures such as increasing the monitoring frequency of relevant points, increasing the frequency of manual inspections to once every 2 hours, strengthening ventilation in local areas, and ensuring that the wind speed at the working face is ≥0.25m / s; If there is a possibility of future upgrades, implement ventilation enhancement measures in advance, issue preliminary warnings to personnel in affected areas, and prepare grouting and sealing equipment and materials. Level III Medium Risk Warning: If future risks can be contained through intervention, implement regional control and enhanced ventilation procedures, have workers wear self-contained breathing apparatus and explosion-proof communication equipment, and have emergency response teams on standby. If there is a risk of rapid escalation in the future, implement a 30-60 minute risk window countdown reminder, stop non-emergency operations and evacuate personnel, move the emergency team to the standby position and put the grouting and drainage equipment in place; Level IV High Risk Rating: Immediately implement the orders for full evacuation, emergency ventilation, and energy isolation; activate life-saving channels and transmit disaster and personnel location data to the emergency command center; trigger the pollution isolation barrier and carry out emergency transfer of poisoned personnel and environmental emergency response.

10. A system for predicting and controlling construction risks in sulfur-bearing strata tunnels, characterized in that, The method for risk prediction and prevention in sulfur-bearing strata tunnel construction as described in any one of claims 1-9 includes: The monitoring and sensing unit is used to build and operate a multi-source real-time monitoring system, and simultaneously collect data on gas concentration, geological information and personnel status. The indicator construction unit is used to establish a construction risk and safety evaluation indicator system and determine the evaluation benchmarks and grade classification standards for each indicator. The weight calculation unit is used to integrate the fuzzy network hierarchical analysis method, the improved entropy weight method, and the game theory combined weighting method to calculate the comprehensive weight of each evaluation index. The dynamic weight adjustment unit is used to dynamically adjust the weight coefficients of personnel behavior indicators based on real-time behavioral and physiological data of workers and their personal digital profiles. The risk assessment unit is used to run the GRA-TOPSIS comprehensive evaluation model, calculate the comprehensive relative proximity, and output the real-time construction risk level. The trend prediction unit is used to run the LSTM-Transformer hybrid time series prediction model to predict the future trends of key risk indicators. The intelligent response unit is used to automatically trigger and execute a tiered prevention and control response mechanism based on real-time risk levels and future trend predictions.

Images