A Real-Time Prediction Method and System for Harmful Gas Diffusion in Tunnels Based on Digital Twins

Abstract

This invention proposes a real-time prediction method and system for hazardous gas diffusion in tunnels based on digital twins, belonging to the field of hazardous gas prediction technology. Addressing the problems of high computational load, long prediction cycles, and large prediction deviations due to neglecting the influence of dynamic construction factors in existing hazardous gas diffusion prediction technologies, this invention employs an adaptive grid fusion technology based on criticality assessment. By dividing the tunnel space into non-uniform grids, coarse grids are fused in low-risk areas while fine grids are maintained in high-risk areas, significantly reducing the total number of computational grids and improving computational efficiency. Simultaneously, by using multi-block overlapping grid technology to accurately simulate the disturbance of construction machinery, combined with the turbulent diffusion coefficient field assimilated from real-time data, accurate prediction of gas diffusion dynamics under the influence of mobile machinery is achieved, effectively improving the real-time performance and accuracy of the prediction results, and providing reliable technical support for gas safety management in tunnel construction.

Description

Technical Field

[0001] This invention relates to the field of tunnel hazardous gas prediction technology, and in particular to a method and system for real-time prediction of tunnel hazardous gas diffusion based on digital twins. Background Technology

[0002] With the rapid development of smart construction sites and intelligent construction technologies, digital twin technology has shown broad application prospects in tunnel construction safety management. A real-time prediction method for the diffusion of harmful gases in tunnels based on digital twins, by constructing a virtual model highly consistent with the physical tunnel, achieves visualized monitoring and predictive early warning of the diffusion process of harmful gases such as methane and hydrogen sulfide. It not only reflects the spatial distribution characteristics of gas concentrations in real time but also predicts short-term concentration trends through data-driven methods, providing forward-looking decision support for construction safety management. The introduction of digital twin technology transforms the gas diffusion process from invisible to transparent, shifting from passive management to proactive prevention, significantly improving the safety management level of tunnel construction and possessing significant promotional value and application potential in the future field of intelligent construction.

[0003] However, existing methods for predicting hazardous gases in tunnels still have significant shortcomings. First, traditional computational fluid dynamics-based prediction models typically employ uniform grid partitioning and fixed parameter settings, resulting in high computational resource consumption and long prediction cycles, making it difficult to meet the urgent need for real-time early warning during construction. Second, existing methods often treat the tunnel environment as a static system, ignoring the impact of dynamic factors such as the movement of construction machinery and personnel activities on the airflow field, leading to significant deviations between predicted results and actual diffusion conditions. Furthermore, most prediction systems lack sufficient optimization of computational load, failing to balance prediction accuracy with computational efficiency, thus limiting their practicality and scalability in engineering projects. These limitations make it difficult for existing prediction methods to provide accurate and timely safety warnings for tunnel construction.

[0004] Therefore, there is an urgent need to develop a new real-time prediction method and system for the diffusion of harmful gases in tunnels based on digital twins, which can both ensure prediction accuracy and have high-speed computing performance. Summary of the Invention

[0005] To address this, the present invention provides a method and system for real-time prediction of harmful gas diffusion in tunnels based on digital twins, which overcomes the problems of existing technologies such as high computational resource consumption, long prediction cycle, poor real-time performance, and neglect of the disturbance of the airflow field by dynamic factors such as construction machinery movement and personnel activities, resulting in a large deviation between the prediction results and the actual diffusion situation.

[0006] To achieve the above objectives, this invention provides a real-time prediction method for the diffusion of hazardous gases in tunnels based on digital twins, comprising: S1, real-time concentration measurements of various harmful gases are obtained through a sensor network deployed in the tunnel, and the location and intensity of the leak source are determined based on the time difference of arrival method; S2, input the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation to obtain the correspondence between the turbulent diffusion coefficient of harmful gas and different regions of the tunnel; S3, based on the correspondence between the turbulent diffusion coefficient of the harmful gas and different areas of the tunnel, as well as the location and intensity of the leakage source, the criticality assessment and non-uniform grid division of the tunnel space are carried out. S4, input the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel into a gas diffusion prediction model based on multiple overlapping grids to obtain the dynamic concentration distribution of the harmful gas in the tunnel; S5. Based on the dynamic concentration distribution and dangerous concentration threshold of the harmful gas in the tunnel, identify dangerous concentration areas and predict their spatiotemporal evolution to generate prediction information.

[0007] Furthermore, inputting the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation includes: S21, The tunnel space is divided into uniform grids to form a tunnel space grid; S22, based on tunnel geometric parameters and initial ventilation conditions, a static plume gas diffusion model is used to calculate the initial background distribution field of the turbulent diffusion coefficient in the tunnel space; S23, using the ensemble Kalman filter algorithm, the real-time concentration measurement value is fused with the initial background distribution field. By minimizing the difference between the observed value and the model prediction value, the calibrated effective turbulent diffusion coefficient spatial distribution field is obtained by inversion. S24. As ventilation conditions change, machinery moves, and construction activities proceed during tunnel construction, the data assimilation and calibration process is continuously repeated to form a dynamic parameter spectrum reflecting the gas diffusion capacity of the current tunnel environment.

[0008] Furthermore, the criticality assessment and non-uniform mesh generation of the tunnel space includes: S31. Establish a criticality evaluation index that integrates multiple factors, and use a weighted fusion algorithm to calculate the comprehensive criticality of each tunnel spatial grid. S32, based on the comprehensive criticality of each tunnel spatial grid, performs grid fusion to obtain a non-uniform grid.

[0009] Furthermore, the key evaluation index for establishing multi-factor fusion includes: S31a, calculate the Euclidean distance between each tunnel spatial grid and the tunnel spatial grid where the leak source is located based on the location of the leak source, and generate the leak source proximity factor; S31b, Based on the current measured concentration of harmful gases and the concentration growth rate of harmful gases, a real-time concentration distribution factor is generated; S31c, based on personnel density and activity intensity, generates a personnel activity intensity factor; S31d generates gas flow field characteristic factors based on real-time wind speed and real-time temperature data; S31e generates a sensor layout optimization factor based on the existing sensor deployment density.

[0010] Furthermore, the mesh fusion based on the comprehensive criticality of each tunnel spatial mesh includes: S32a treats each grid cell as a network node and establishes spatial adjacency relationships between grids; S32b calculates the multidimensional similarity between grid nodes, where multidimensional similarity includes comprehensive criticality value similarity, historical accident data similarity, geometric feature similarity, and environmental parameter correlation similarity. S32c constructs weighted connection edges for grid nodes based on multidimensional similarity, forming a grid similarity topology network; S32d calculates the local density and relative distance of each grid node and identifies density peak points as cluster centers; S32e, adaptively determines the number of clusters based on the local density distribution of grid nodes; S32f, based on a mesh similarity topology network, assigns non-peak mesh nodes to the cluster to which the nearest density peak belongs. For each cluster, it merges all the smaller original meshes within it into one or more larger new meshes.

[0011] Furthermore, in S4, the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel is input into a gas diffusion prediction model based on multiple overlapping grids to obtain the dynamic concentration distribution of the harmful gas in the tunnel, including: S41, establish multiple overlapping meshes including non-uniform meshes and component meshes; wherein, the component meshes are independent meshes that precisely fit the geometry of the construction machinery; S42 uses the momentum source term method to simulate the disturbance of airflow by construction machinery; S43, Based on the turbulent diffusion coefficient of the harmful gas and the real-time wind speed vector, construct the gas transport equation; S44, Solve the gas transport equations on multiple overlapping grids; S45, the solution results of each grid are merged to obtain the dynamic concentration distribution of harmful gases in the tunnel.

[0012] Furthermore, the simulation of the disturbance of airflow by construction machinery using the momentum source term method includes: S42a, identify grids occupied by construction machinery and mark them as solid areas; S42b applies a no-slip wall boundary condition to the solid region to simulate the obstruction of airflow by the mechanical body; S42c adds a momentum source term to the grid corresponding to the airflow generation part of the construction machinery to simulate the airflow disturbance generated by the machinery itself.

[0013] Furthermore, the process of identifying hazardous concentration areas and predicting their spatiotemporal evolution based on the dynamic concentration distribution and hazardous concentration threshold of the harmful gas within the tunnel, and generating prediction information, includes: S51, based on current data on the concentration distribution of hazardous gases, predicts the movement speed, diffusion rate, and concentration change trend of hazardous concentration areas; S52, in conjunction with the construction plan, predict the range of hazardous concentration areas during the construction period; S53 generates prediction information based on the prediction content.

[0014] Furthermore, the prediction information includes the location, risk level, impact range, and predicted evolution path of the hazardous concentration area.

[0015] This invention also provides a real-time prediction system for the diffusion of hazardous gases in tunnels based on digital twins, which uses a real-time prediction method for the diffusion of hazardous gases in tunnels based on digital twins, and includes the following modules: The data acquisition module is used to acquire real-time concentration measurements of various harmful gases through a sensor network deployed in the tunnel, and to determine the location and intensity of the leak source based on the time difference of arrival method. The parameter assimilation module, connected to the data acquisition module, is used to input the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation, and obtain the correspondence between the turbulent diffusion coefficient of harmful gases and different regions of the tunnel. The mesh optimization module, connected to the parameter assimilation module, is used to perform criticality assessment and non-uniform mesh division of the tunnel space based on the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel, as well as the location and intensity of the leakage source. The diffusion simulation module, connected to the mesh optimization module, is used to input the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel into a gas diffusion prediction model based on multiple overlapping meshes, so as to obtain the dynamic concentration distribution of the harmful gas in the tunnel. The risk warning module, connected to the diffusion simulation module, is used to identify hazardous concentration areas and predict their spatiotemporal evolution based on the dynamic concentration distribution and hazardous concentration threshold of the harmful gas in the tunnel, and generate prediction information.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: Firstly, this invention significantly improves computational efficiency and real-time prediction by using an adaptive mesh fusion technology based on criticality assessment. Through multi-dimensional similarity calculation and density peak clustering, it intelligently identifies low-criticality areas within the tunnel and merges a large number of fine meshes in these areas into a small number of coarse meshes, directly reducing the total number of computational meshes by up to 70%. This fundamentally reduces the computational complexity of numerical simulation and significantly reduces the computation time for transient gas diffusion simulation in the entire tunnel, thereby enabling real-time prediction of gas concentration distribution and risk evolution, providing a critical time window for on-site emergency response. Secondly, this invention decouples the dynamic and static computational domains by employing a multi-block overlapping mesh technology, significantly reducing the mesh update overhead caused by the movement of construction machinery, thereby further improving computational efficiency and prediction real-time performance. The system separates the static tunnel background mesh from the dynamic mechanical component mesh, forming an independent overlapping mesh system. When the machinery moves, data exchange is only required in the overlapping area through an interpolation algorithm, without the need to reconstruct or re-divide the background mesh. This avoids the large amount of computational consumption caused by frequent mesh deformation or reconstruction in traditional dynamic mesh methods, ensuring the real-time performance of predictions in complex dynamic construction environments. Third, this invention introduces multi-block overlapping grid technology to accurately simulate the disturbance of construction machinery, which greatly improves the accuracy and reliability of the prediction results. The system establishes an independent body-fitting component grid for the mobile machinery and uses the momentum source term method to accurately simulate the obstruction of airflow by the machinery body and the additional disturbances generated by its engine and fan. This solves the prediction deviation problem caused by the traditional method of treating the tunnel environment as a static system and ignoring dynamic disturbances. This enables the model to accurately capture local complex flows such as the wake zone and the flow around the machinery, thereby more accurately predicting the accumulation trend and diffusion path of harmful gases in these areas. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for real-time prediction of harmful gas diffusion in tunnels based on digital twins, according to an embodiment of the present invention. Figure 2 This is a structural block diagram of a real-time prediction system for the diffusion of hazardous gases in tunnels based on digital twins, according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0020] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0021] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] Example 1 like Figure 1 As shown, this invention proposes a real-time prediction method for the diffusion of harmful gases in tunnels based on digital twins, which specifically includes the following steps: S1, real-time concentration measurements of various harmful gases are obtained through a sensor network deployed in the tunnel, and the location and intensity of the leak source are determined based on the time difference of arrival method; For example, a sensor network includes: Catalytic combustion sensor for detecting methane; Electrochemical sensor for detecting hydrogen sulfide; Wind speed and direction sensor, used to measure the three-dimensional wind speed vector at a specific point inside the tunnel; Temperature and humidity sensors; Visual sensors are used to locate the positions of construction workers and construction machinery.

[0023] In one possible implementation, multiple hazardous gas concentration sensors, such as methane sensors and hydrogen sulfide sensors, are arranged in a pre-defined topology within the tunnel construction area. These sensors are preferably installed on the tunnel sidewalls and arches, covering key areas such as the working face, escape routes, and equipment gathering areas. All sensors are connected to the central processor via wired or wireless communication networks to synchronously collect and upload gas concentration data and their corresponding timestamps at each monitoring point at a high frequency, such as once per second.

[0024] The central processing unit continuously monitors the concentration data of each sensor. When 10 sensors detect that the concentration of harmful gas exceeds the preset background threshold, a leak event is determined. The central processing unit then backtracks and analyzes the concentration time series of all sensors to identify the moment when the concentration of each sensor first exceeds the preset background threshold, which is the arrival time of that sensor.

[0025] Select a sensor as a reference, such as the sensor that first alarms, and combine it with other sensors to form multiple TDOA equations. Incorporate real-time measurements of the average wind speed vector within the tunnel for estimation and correction. By solving this overdetermined system of equations, using algorithms such as least squares or particle swarm optimization, the most probable location coordinates of the leak source can be determined.

[0026] To determine the location of the leak source, the intensity of the leak source is determined based on the earliest arrival time and the concentration of the hazardous gas at that time, combined with a simplified diffusion model, such as an instantaneous point source diffusion model that considers convection.

[0027] This invention analyzes the timing differences in gas arrival at different sensors, enabling rapid source location after a leak occurs. Compared to traditional manual investigation or on-site confirmation after a single-point alarm, this improves response speed. By calculating the intensity of the leak source, it provides a quantitative basis for assessing the severity of the leak and predicting the spread range, thus laying the foundation for improving prediction accuracy.

[0028] S2, input the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation to obtain the correspondence between the turbulent diffusion coefficient of harmful gas and different regions of the tunnel; The location and intensity of the leakage source are input into a turbulent diffusion coefficient field based on real-time data assimilation, including: S21, The tunnel space is divided into uniform grids to form a tunnel space grid; S22, based on tunnel geometric parameters and initial ventilation conditions, a static plume gas diffusion model is used to calculate the initial background distribution field of the turbulent diffusion coefficient in the tunnel space; S23, using the ensemble Kalman filter algorithm, the real-time concentration measurement value is fused with the initial background distribution field. By minimizing the difference between the observed value and the model prediction value, the calibrated effective turbulent diffusion coefficient spatial distribution field is obtained by inversion. S24. As ventilation conditions change, machinery moves, and construction activities proceed during tunnel construction, the data assimilation and calibration process is continuously repeated to form a dynamic parameter spectrum reflecting the gas diffusion capacity of the current tunnel environment.

[0029] In one possible implementation, the three-dimensional space of the tunnel is discretized into a regular grid. Based on the tunnel BIM model, the tunnel's geometric parameters, such as cross-sectional shape, size, and length, and initial ventilation conditions, such as design ventilation volume and fan location, are obtained. A Gaussian static plume gas diffusion model is used as the forward physical model to calculate the initial background distribution field of the turbulent diffusion coefficient in the tunnel space, which serves as the prior knowledge basis for data assimilation.

[0030] Around the initial background distribution field, a set containing N members, such as 100, is generated by adding random perturbations. The location and intensity of the leakage source determined by S1 are used as the source term. Each member of the set is substituted into the gas diffusion model to predict the concentration field corresponding to each member. The predicted concentration value at each sensor location is extracted from the predicted concentration field by an interpolation operator. The actual concentration measurement values ​​of all sensors at the current moment are obtained, and the Kalman gain of each member of the set is calculated. The diffusion coefficient of each member is updated based on the Kalman gain. The updated set is used as the spatial distribution field of the effective turbulent diffusion coefficient after calibration at the current moment. This field accurately reflects the diffusion capacity of different regions.

[0031] As tunnel construction progresses, ventilation conditions, the layout and activity status of mechanical equipment constantly change, and the system continuously repeats the above data assimilation and calibration process. For example, the system sets a fixed assimilation cycle, such as every 5 minutes, or automatically triggers a new round of assimilation calibration when a significant change in the flow field is detected, such as the start / stop of the fan or a sudden change in wind speed.

[0032] This invention establishes an effective spatial distribution field of turbulent diffusion coefficients, breaking through the coarse characterization method of traditional approaches that use single empirical values, and precisely reflects the true differences in diffusion capacity in different regions within the tunnel. By using an ensemble Kalman filter algorithm, it provides information on the uncertainty of parameters and predictions, offering a more accurate basis for risk assessment.

[0033] S3, based on the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel, as well as the location and intensity of the leakage source, performs criticality assessment and non-uniform mesh generation of the tunnel space, including: S31. Establish a criticality evaluation index that integrates multiple factors, and use a weighted fusion algorithm to calculate the comprehensive criticality of each tunnel spatial grid. Establish key performance indicators that integrate multiple factors, including: S31a, calculate the Euclidean distance between each tunnel spatial grid and the tunnel spatial grid where the leak source is located based on the location of the leak source, and generate the leak source proximity factor; S31b, Based on the current measured concentration of harmful gases and the concentration growth rate of harmful gases, a real-time concentration distribution factor is generated; S31c, based on personnel density and activity intensity, generates a personnel activity intensity factor; S31d generates gas flow field characteristic factors based on real-time wind speed and real-time temperature data; S31e generates a sensor layout optimization factor based on the existing sensor deployment density; In one possible implementation, based on the location coordinates of the leak source determined by S1, the Euclidean distance from the center point of each grid cell within the tunnel to the leak source is calculated. Subsequently, a proximity factor to the leak source is generated using a distance decay function, for example, an exponential decay function. F a =e -k·d , of which F a Here, k is the proximity factor to the leak source, d is the attenuation coefficient, and d is the Euclidean distance from the center point of the grid cell to the leak source. The closer the distance to the leak source, the higher the proximity factor value (maximum 1), indicating a higher criticality. The Kriging interpolation algorithm transforms discrete sensor concentration measurements into a continuous concentration field. Higher concentration values ​​within a grid cell result in a higher concentration dimension score. The concentration growth rate within each grid cell is calculated; regions with faster growth rates have greater potential risk and higher growth dimension scores. The concentration dimension and growth dimension scores are then weighted and combined to obtain the real-time concentration distribution factor for each grid cell.

[0034] The number of personnel appearing in each grid cell within a certain time window is counted. Based on the construction plan, a higher foundation activity intensity is assigned to specific work areas, such as the excavation face and support work area. A personnel activity intensity factor is generated by combining real-time personnel density and foundation activity intensity. ， The higher the population density and the more frequent the work activities in the area, the higher the value of this factor.

[0035] Based on the turbulent diffusion coefficient field obtained by S2 assimilation and real-time wind speed data, grids with wind speeds below a critical value, such as 0.2 m / s, are identified where gas tends to accumulate. By combining the accumulation with the turbulent diffusion coefficient of harmful gases, a gas flow field characteristic factor is generated.

[0036] Calculate the distance from each grid to the nearest hazardous gas concentration sensor. The farther the grid is from the sensor, the more difficult it is to directly perceive its true concentration, and the higher the uncertainty. Therefore, it needs to rely more on model prediction, and its criticality should be appropriately increased. Generate sensor layout optimization factors based on the distance from the grid to the nearest hazardous gas concentration sensor.

[0037] Finally, a weighted fusion algorithm is used to combine the above five factors into the overall criticality of each grid cell. The weights can be dynamically adjusted according to different construction stages and safety strategies. For example, during normal tunneling, more attention may be paid to leakage sources and personnel activities; while after ventilation maintenance, more attention may be paid to monitoring dead zones in the flow field.

[0038] This invention directly identifies high-risk core areas surrounding a leak source by quantizing the spatial correlation between the grid and the leak source based on Euclidean distance, providing a fundamental spatial basis for subsequent grid resource allocation and avoiding risk omissions due to ambiguous distance judgments. By combining current concentration values ​​with concentration growth rates, it reflects both the current risk status and predicts short-term risk trends, compensating for the lag of relying solely on static concentrations. By incorporating personnel density and activity intensity into the evaluation indicators, the disturbance effect is transformed into a calculable factor index, enabling the diffusion prediction to specifically consider the physical process of personnel-induced airflow disturbance, avoiding prediction bias caused by ignoring this factor and improving the accuracy of tunnel hazardous gas diffusion prediction. If a factor exhibits abnormal data due to sensor malfunction, the weighting mechanism allows other normal factors to offset the impact of the abnormal data through weight allocation, preventing a single factor error from causing the entire grid criticality assessment to fail, thus enhancing the assessment's anti-interference capability and stability.

[0039] S32, based on the comprehensive criticality of each tunnel spatial grid, performs grid fusion to obtain a non-uniform grid, including: S32a treats each grid cell as a network node and establishes spatial adjacency relationships between grids; S32b calculates the multidimensional similarity between grid nodes, where multidimensional similarity includes comprehensive criticality value similarity, historical accident data similarity, geometric feature similarity, and environmental parameter correlation similarity. S32c constructs weighted connection edges for grid nodes based on multidimensional similarity, forming a grid similarity topology network; S32d calculates the local density and relative distance of each grid node and identifies density peak points as cluster centers; S32e, adaptively determines the number of clusters based on the local density distribution of grid nodes; S32f, based on a mesh similarity topology network, assigns non-peak mesh nodes to the cluster to which the nearest density peak belongs. For each cluster, it merges all the small original mesh cells within it into one or more larger new meshes. In one possible implementation, each uniform mesh cell is treated as an independent network node, and a complete adjacency network is established based on the relative positions of the meshes in three-dimensional space. All neighboring cells of each mesh cell are identified, including directly face-adjacent, edge-adjacent, and vertex-adjacent neighboring nodes, forming the basic mesh topology connection structure.

[0040] Calculate the key similarity features between adjacent grid nodes, specifically including: Keyness similarity is calculated by taking the absolute difference between the combined keyness values ​​of two grid nodes and then normalizing the result to obtain a similarity score. The smaller the difference, the higher the similarity. Historical accident data similarity analysis: Based on historical tunnel construction data, the similarity of accident frequency and type in different regions is analyzed. Based on historical gas accumulation event records, equipment failure rate, and distribution of abnormal construction events, regions with similar historical accident characteristics and frequency are identified as having high similarity. Geometric feature similarity is compared by examining the geometric characteristics of the grid cells in the tunnel space, including the direction of the cell normal vector, local curvature features, and relative positional relationship with the tunnel's central axis. Environmental parameter correlation similarity is analyzed based on temperature gradient, humidity distribution and harmful gas concentration distribution. The coupling relationship pattern of temperature-humidity-gas concentration in different regions is analyzed, and grids with similar environmental parameters and gas concentration response characteristics are identified as having high similarity. The four similarity measures mentioned above are weighted and fused according to preset weights to obtain a comprehensive similarity score between adjacent grid nodes. Weighted connection edges are established for all node pairs with similarity exceeding a preset threshold to form a complete grid similarity topology network structure.

[0041] The distribution density of a node is evaluated by counting the number of neighboring nodes within a given distance range. The minimum distance from a node to other nodes with higher density is calculated, and nodes with both high local density and large relative distance are selected as candidate cluster centers.

[0042] By analyzing the distribution patterns of local density and relative distance of all grid nodes, significant density peaks that deviate significantly from the main distribution area in the density-distance decision map are automatically identified. Based on the number of these significant peaks, the optimal number of clusters is adaptively determined.

[0043] By analyzing the distribution patterns of local density and relative distance of all grid nodes, significant density peaks that deviate significantly from the main distribution area in the density-distance decision map are automatically identified. Based on the number of these significant peaks, the optimal number of clusters is adaptively determined.

[0044] Non-peak grid nodes are progressively assigned to the clusters of the nearest density peaks according to the connection relationships in the grid similarity topology network. For each cluster, all the small original grids within it are merged into one or more larger new grids.

[0045] This invention accurately identifies areas where hazardous gas concentrations remain consistently zero or extremely low, and which, based on the current flow field and leak source location, are expected to remain safe for the foreseeable future, through comprehensive criticality assessment. It merges numerous small grids within these safe zones into a few large grids, directly eliminating a large amount of unnecessary computational load with minimal impact on prediction results, thus reducing computational complexity and significantly improving prediction speed and real-time performance while maintaining prediction accuracy. Compared to single-dimensional similarity assessment, this invention achieves a comprehensive characterization of grid attributes by integrating four dimensions: criticality value, historical accident data, geometric features, and environmental parameter correlations. Criticality value similarity ensures targeted clustering of high-risk or high-concern grids; historical accident data similarity incorporates past safety experience; geometric feature similarity adapts to complex tunnel cross-sectional shapes; and environmental parameter correlation similarity responds to the influence of environmental factors such as airflow and temperature on diffusion. The multi-dimensional fusion judgment method significantly improves the accuracy of grid similarity recognition, avoiding clustering bias caused by single-dimensional deviations. It allows grids with similar properties to be accurately classified into one category, improving the accuracy of fusion. The consideration of geometric feature similarity enables the clustered non-uniform grids to better fit the complex geometric shapes of tunnels such as irregular cross-sections, turning sections, and equipment layout areas, avoiding the grid distortion problem of uniform grids in complex areas. The integration of environmental parameter correlation similarity allows the grid division to respond to the spatial distribution differences of environmental parameters such as wind speed and temperature in the tunnel, making the grid structure match the environmental factors affecting gas diffusion, and improving the adaptability of this invention in the face of complex scenarios.

[0046] The introduction of historical accident data similarity enables grid division to be associated with the characteristics of areas with high incidence of past accidents, keeping the grid division of such areas at a fine level and providing a more accurate grid foundation for subsequent identification of dangerous areas and prediction of spatiotemporal evolution. The core role of the comprehensive criticality value ensures the grid accuracy of key areas such as densely populated areas and areas near leakage sources, enabling the prediction results to accurately focus on high-risk areas, thus meeting the real-time requirements of tunnel construction early warning while ensuring the accuracy of prediction results.

[0047] S4, input the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel into a gas diffusion prediction model based on multiple overlapping grids to obtain the dynamic concentration distribution of the harmful gas in the tunnel, including: S41, establish multiple overlapping meshes including non-uniform meshes and component meshes; wherein, the component meshes are independent meshes that precisely fit the geometry of the construction machinery; S42 uses the momentum source term method to simulate the disturbance of airflow by construction machinery, including: S42a, identify grids occupied by construction machinery and mark them as solid areas; S42b applies a no-slip wall boundary condition to the solid region to simulate the obstruction of airflow by the mechanical body; S42c adds a momentum source term to the grid corresponding to the airflow generation part of the construction machinery to simulate the airflow disturbance generated by the machinery itself.

[0048] S43, Based on the turbulent diffusion coefficient of the harmful gas and the real-time wind speed vector, construct the gas transport equation; S44, Solve the gas transport equations on multiple overlapping grids; S45, the solution results of each grid are merged to obtain the dynamic concentration distribution of harmful gases in the tunnel; In one possible implementation, a background computational domain covering the entire tunnel space is established based on the non-uniform grid generated by S3. This grid maintains fine resolution in key areas such as the vicinity of the leak source and the area where personnel are active, while the fused coarse grid is used in areas where the concentration of harmful gases is consistently zero or extremely low and, based on the current flow field and the location of the leak source, it is determined that the area will remain safe in the foreseeable future. The precise geometric shape of the construction machinery is obtained by 3D laser scanning. The structured mesh generation technology is used to generate independent meshes. The component meshes are spatially overlapped with the background meshes, but each maintains an independent mesh topology. In the background mesh, based on the real-time positioning data of the machinery, the meshes occupied by the machinery are identified and marked as solid regions. These meshes are excluded in the flow field calculation. No-slip wall boundary conditions are applied to the solid regions, and wall functions are set to handle the turbulence characteristics of the near-wall region, accurately simulating the obstruction effect of the machinery body on the airflow. Add momentum source terms to the grid cells corresponding to airflow generating parts such as mechanical engines and cooling fans, and quantify the magnitude and direction of the momentum source terms according to parameters such as engine power and fan speed. The gas transport equation is constructed as follows: Where C represents the concentration of the harmful gas. For wind speed vectors, Where S is the turbulent diffusion coefficient, and S is the leakage source term; Substitute the spatially distributed turbulent diffusion coefficient field obtained in S2 into the diffusion term, substitute the leakage source intensity determined in S1 into the source term into the equation, and substitute the real-time monitored wind speed data into the velocity field of the convection term into the equation. Numerically solve the gas transport equation on a multi-overlapping grid system. The solution results of the background mesh and the mesh of each component are fused together. Based on the interpolation relationship of the overlapping area, a seamless full-field concentration distribution is synthesized, eliminating non-physical jumps at the boundaries of different mesh blocks. The three-dimensional concentration distribution data that evolves over time is output, a concentration field time series is established, trend analysis and prediction are supported, and visualization results such as isoconcentration surfaces and concentration cloud maps are generated.

[0049] This invention improves the accuracy of gas diffusion path prediction by establishing a multi-overlapping grid system. It can accurately capture the gas distribution characteristics in complex flow modes such as the wake region and the flow around the machine, enhance the ability to identify the risk of local gas accumulation, and accurately predict gas accumulation areas that are difficult to detect by traditional methods, such as the vortex region behind the machine. It avoids the complex grid reconstruction process and achieves seamless simulation of mechanical motion through the independent movement of component grids. It improves the adaptability of the system in complex construction environments, can track changes in the position of the machine in real time and quickly update the prediction results, and realizes a technological leap from static environment prediction to accurate early warning of dynamic construction environment.

[0050] S5, based on the dynamic concentration distribution and hazardous concentration threshold of the harmful gas in the tunnel, identify hazardous concentration areas and predict their spatiotemporal evolution, generating prediction information including: S51, based on current data on the concentration distribution of hazardous gases, predicts the movement speed, diffusion rate, and concentration change trend of hazardous concentration areas; S52, in conjunction with the construction plan, predict the range of hazardous concentration areas during the construction period; S53 generates prediction information based on the prediction content.

[0051] Specifically, the forecast information includes the location of hazardous concentration areas, risk level, impact range, and predicted evolution path.

[0052] In one possible implementation, the movement trajectory of the centroid of the danger zone is calculated using optical flow or feature point tracking techniques from image recognition, based on concentration field data at continuous time steps. By combining real-time wind speed and direction data, a correlation model between movement speed and flow field is established to predict the movement direction and speed of the danger zone in the next 5-15 minutes, and output the movement direction angle, movement speed and the estimated time to reach the construction location. Calculate the rate of change of the volume of the hazardous area over time, quantify the diffusion rate, analyze the changes in the concentration gradient, identify the trend of diffusion acceleration or deceleration, and predict the changes in diffusion characteristics in different tunnel sections based on the turbulent diffusion coefficient field. Input the construction schedule, identify changes in hazardous concentration areas during the future construction period, assess the impact of hazardous gases on the construction area during the construction period, generate hazardous area boundary prediction maps at different time points, identify the expected impact range, pay special attention to the intersection of construction activities and gas diffusion paths, and conduct key risk assessments.

[0053] Based on the above analysis results, risk levels are generated, specifically including: Level 1 warning: The danger zone is rapidly moving towards densely populated areas, or the concentration is rising sharply and approaching the lower explosive limit. Level II alert: The danger zone remains stable and the concentration is slowly increasing, affecting the main work access routes; Level 3 warning: The danger zone is limited to non-operational areas, and the concentration remains stable or shows a downward trend.

[0054] Generate diverse predictive information, specifically including: Predictive basic information includes the type of hazardous gas, the current maximum concentration, and the specific location of the hazardous concentration area; Risk level; The scope of impact includes the affected work areas, the number of personnel at risk, and the expected duration; Predict the evolution path, including a schematic diagram of the movement path, the key areas expected to be affected, and the time when the peak concentration is reached.

[0055] Generate multiple versions of early warning information suitable for different terminals, visualize the historical trajectory and predicted path of dangerous areas in the 3D tunnel model, and release early warning information through multiple channels such as sound and light alarms, SMS push, and broadcast system.

[0056] This invention quantifies and predicts the movement speed, diffusion rate, and concentration change trend of hazardous concentration areas, solving the problem that traditional monitoring systems only issue static alarms when concentrations exceed standards and cannot predict risk development trends. This advances the warning time and increases the time window. By combining construction plans to predict the scope of hazardous areas, it addresses the lack of forward-looking assessment of dynamic risks during construction in existing technologies. This enables early identification of the intersection risks between construction activities and gas diffusion paths, allowing for proactive intervention and providing risk prediction data for construction scheduling, supporting the optimized scheduling of high-risk operations. By generating structured prediction information including location, risk level, impact range, and evolution path, it solves the problem of traditional alarm information being singular and lacking specific spatial location and impact range descriptions. By quantifying the impact range by area area and involved objects, it enables managers to clearly grasp the scale of resources affected by the risk and rationally allocate emergency resources.

[0057] The structured predictive information of this invention significantly reduces the information integration costs for managers, enabling them to quickly grasp core risks and make critical decisions such as whether to suspend construction, evacuate personnel, or adjust ventilation plans. This improves the efficiency and accuracy of emergency decision-making and minimizes the risk of safety accidents caused by the spread of harmful gases.

[0058] Example 2 like Figure 2 As shown, the present invention also provides a real-time prediction system for the diffusion of hazardous gases in tunnels based on digital twins, which uses the real-time prediction method for the diffusion of hazardous gases in tunnels based on digital twins described in any one of Embodiment 1, and includes the following modules: The data acquisition module is used to acquire real-time concentration measurements of various harmful gases through a sensor network deployed in the tunnel, and to determine the location and intensity of the leak source based on the time difference of arrival method. The parameter assimilation module, connected to the data acquisition module, is used to input the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation, and obtain the correspondence between the turbulent diffusion coefficient of harmful gases and different regions of the tunnel. The mesh optimization module, connected to the parameter assimilation module, is used to perform criticality assessment and non-uniform mesh division of the tunnel space based on the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel, as well as the location and intensity of the leakage source. The diffusion simulation module, connected to the mesh optimization module, is used to input the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel into a gas diffusion prediction model based on multiple overlapping meshes, so as to obtain the dynamic concentration distribution of the harmful gas in the tunnel. The risk warning module, connected to the diffusion simulation module, is used to identify hazardous concentration areas and predict their spatiotemporal evolution based on the dynamic concentration distribution and hazardous concentration threshold of the harmful gas in the tunnel, and generate prediction information.

[0059] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for real-time prediction of hazardous gas diffusion in tunnels based on digital twins, characterized in that, include: S1, real-time concentration measurements of various harmful gases are obtained through a sensor network deployed in the tunnel, and the location and intensity of the leak source are determined based on the time difference of arrival method; S2, input the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation to obtain the correspondence between the turbulent diffusion coefficient of harmful gas and different regions of the tunnel; S3, based on the correspondence between the turbulent diffusion coefficient of the harmful gas and different areas of the tunnel, as well as the location and intensity of the leakage source, the criticality assessment and non-uniform grid division of the tunnel space are carried out. S4, input the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel into a gas diffusion prediction model based on multiple overlapping grids to obtain the dynamic concentration distribution of the harmful gas in the tunnel; S5. Based on the dynamic concentration distribution and dangerous concentration threshold of the harmful gas in the tunnel, identify dangerous concentration areas and predict their spatiotemporal evolution to generate prediction information.

2. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 1, characterized in that, The step of inputting the location and intensity of the leakage source into a turbulent diffusion coefficient field based on real-time data assimilation includes: S21, The tunnel space is divided into uniform grids to form a tunnel space grid; S22, based on tunnel geometric parameters and initial ventilation conditions, a static plume gas diffusion model is used to calculate the initial background distribution field of the turbulent diffusion coefficient in the tunnel space; S23, using the ensemble Kalman filter algorithm, the real-time concentration measurement value is fused with the initial background distribution field. By minimizing the difference between the observed value and the model prediction value, the calibrated effective turbulent diffusion coefficient spatial distribution field is obtained by inversion. S24. As ventilation conditions change, machinery moves, and construction activities proceed during tunnel construction, the data assimilation and calibration process is continuously repeated to form a dynamic parameter spectrum reflecting the gas diffusion capacity of the current tunnel environment.

3. The method for real-time prediction of tunnel hazardous gas diffusion based on digital twins according to claim 1, characterized in that, The criticality assessment and non-uniform grid division of the tunnel space include: S31. Establish a criticality evaluation index that integrates multiple factors, and use a weighted fusion algorithm to calculate the comprehensive criticality of each tunnel spatial grid. S32, based on the comprehensive criticality of each tunnel spatial grid, performs grid fusion to obtain a non-uniform grid.

4. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 3, characterized in that, The key evaluation indicators for establishing multi-factor fusion include: S31a, calculate the Euclidean distance between each tunnel spatial grid and the tunnel spatial grid where the leak source is located based on the location of the leak source, and generate the leak source proximity factor; S31b, Based on the current measured concentration of harmful gases and the concentration growth rate of harmful gases, a real-time concentration distribution factor is generated; S31c, based on personnel density and activity intensity, generates a personnel activity intensity factor; S31d generates gas flow field characteristic factors based on real-time wind speed and real-time temperature data; S31e generates a sensor layout optimization factor based on the existing sensor deployment density.

5. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 3, characterized in that, The mesh fusion based on the comprehensive criticality of each tunnel spatial grid includes: S32a treats each grid cell as a network node and establishes spatial adjacency relationships between grids; S32b calculates the multidimensional similarity between grid nodes, where multidimensional similarity includes comprehensive criticality value similarity, historical accident data similarity, geometric feature similarity, and environmental parameter correlation similarity. S32c constructs weighted connection edges for grid nodes based on multidimensional similarity, forming a grid similarity topology network; S32d calculates the local density and relative distance of each grid node and identifies density peak points as cluster centers; S32e, adaptively determines the number of clusters based on the local density distribution of grid nodes; S32f, based on a mesh similarity topology network, assigns non-peak mesh nodes to the cluster to which the nearest density peak belongs. For each cluster, it merges all the smaller original meshes within it into one or more larger new meshes.

6. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 1, characterized in that, In S4, the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel is input into a gas diffusion prediction model based on multiple overlapping grids to obtain the dynamic concentration distribution of the harmful gas in the tunnel, including: S41, establish multiple overlapping meshes including non-uniform meshes and component meshes; wherein, the component meshes are independent meshes that precisely fit the geometry of the construction machinery; S42 uses the momentum source term method to simulate the disturbance of airflow by construction machinery; S43, Based on the turbulent diffusion coefficient of the harmful gas and the real-time wind speed vector, construct the gas transport equation; S44, Solve the gas transport equations on multiple overlapping grids; S45, the solution results of each grid are merged to obtain the dynamic concentration distribution of harmful gases in the tunnel.

7. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 6, characterized in that, The method of simulating the disturbance of airflow by construction machinery using the momentum source term method includes: S42a, identify grids occupied by construction machinery and mark them as solid areas; S42b applies a no-slip wall boundary condition to the solid region to simulate the obstruction of airflow by the mechanical body; S42c adds a momentum source term to the grid corresponding to the airflow generation part of the construction machinery to simulate the airflow disturbance generated by the machinery itself.

8. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 1, characterized in that, The process of identifying hazardous concentration areas and predicting their spatiotemporal evolution based on the dynamic concentration distribution and dangerous concentration threshold of the harmful gas within the tunnel, and generating prediction information, includes: S51, based on current data on the concentration distribution of hazardous gases, predicts the movement speed, diffusion rate, and concentration change trend of hazardous concentration areas; S52, in conjunction with the construction plan, predict the range of hazardous concentration areas during the construction period; S53 generates prediction information based on the prediction content.

9. The method for real-time prediction of harmful gas diffusion in tunnels based on digital twins according to claim 1, characterized in that, The predicted information includes the location, risk level, impact range, and predicted evolution path of the hazardous concentration area.

10. A real-time prediction system for the diffusion of harmful gases in tunnels based on digital twins, characterized in that, The system employs the real-time prediction method for the diffusion of hazardous gases in tunnels based on digital twins as described in any one of claims 1 to 9, and specifically includes the following modules: The data acquisition module is used to acquire real-time concentration measurements of various harmful gases through a sensor network deployed in the tunnel, and to determine the location and intensity of the leak source based on the time difference of arrival method. The parameter assimilation module, connected to the data acquisition module, is used to input the location and intensity of the leakage source into the turbulent diffusion coefficient field based on real-time data assimilation, and obtain the correspondence between the turbulent diffusion coefficient of harmful gases and different regions of the tunnel. The mesh optimization module, connected to the parameter assimilation module, is used to perform criticality assessment and non-uniform mesh division of the tunnel space based on the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel, as well as the location and intensity of the leakage source. The diffusion simulation module, connected to the mesh optimization module, is used to input the correspondence between the turbulent diffusion coefficient of the harmful gas and different regions of the tunnel into a gas diffusion prediction model based on multiple overlapping meshes, so as to obtain the dynamic concentration distribution of the harmful gas in the tunnel. The risk warning module, connected to the diffusion simulation module, is used to identify hazardous concentration areas and predict their spatiotemporal evolution based on the dynamic concentration distribution and hazardous concentration threshold of the harmful gas in the tunnel, and generate prediction information.

Images