Hydrogen leakage accident holographic perception and disaster situation dynamic prediction system in tunnel scene

Abstract

The invention discloses a hydrogen leakage accident holographic perception and disaster situation dynamic prediction system in a tunnel scene, and belongs to the technical field of hydrogen energy traffic safety and intelligent risk management. According to the system, multi-modal information such as environmental parameters, hydrogen concentration, heat source temperature, leakage acoustic characteristics and vehicle states is collected in real time through a multi-modal sensing unit; abnormal detection, time synchronization and space registration are carried out through the information fusion module to generate fusion data in a unified format; the hazard source analysis module locates a leakage source, estimates a leakage rate, locates an ignition source and extracts disaster characteristics based on the fused data; the disaster dynamic prediction module predicts spatio-temporal evolution of diffusion, combustion and explosion by using an agent model and calculates disaster levels; the feedback early warning unit synchronously outputs the prediction information to the tunnel monitoring platform and issues early warning information; and the data storage and self-optimization module uniformly stores historical data and optimizes the prediction model through incremental learning. The system realizes real-time monitoring, rapid evaluation and intelligent disposal of tunnel hydrogen leakage accidents.

Description

Technical Field

[0001] This invention relates to the field of hydrogen energy transportation safety and intelligent risk management technology, and in particular to a holographic perception and dynamic disaster prediction system for hydrogen leak accidents in tunnel scenarios. Background Technology

[0002] Hydrogen energy, as a clean and efficient secondary energy source, has been widely used in fuel cell vehicles, industrial production, and energy storage and transportation. However, hydrogen has characteristics such as low density, rapid diffusion, low ignition point, and a wide explosive range. Once it leaks and forms a flammable mixture with air, it is highly likely to be ignited by a small ignition source, leading to fires or explosions with extremely serious hazards. Tunnels are semi-enclosed confined spaces with limited airflow, diverse ignition sources, and rapid accident spread. Furthermore, evacuation and rescue conditions are limited, making hydrogen leaks in such environments extremely difficult to control; improper handling can result in serious casualties and property damage.

[0003] However, current technologies have the following shortcomings and deficiencies: 1. Most technical solutions rely solely on hydrogen concentration sensors for monitoring, lacking the ability to simultaneously collect multi-dimensional information such as thermal anomalies, acoustic characteristics, and vehicle operating status, making it difficult to identify the source of an accident and its leakage-diffusion-explosion evolution trend in a timely and comprehensive manner; 2. It is difficult to balance prediction speed and accuracy. While computational fluid dynamics (CFD) methods can achieve high prediction accuracy, they involve large computational loads and are time-consuming, making it difficult to meet the second-level prediction requirements when an accident occurs. Although rapid estimation methods based on simplified assumptions or equivalent models (such as the TNT equivalent method and simplified shock wave model) are fast in computation, they often fail to accurately reflect the non-uniform diffusion, stratified accumulation, and explosive impact effects of hydrogen in confined spaces such as tunnels, resulting in significant deviations between prediction results and actual conditions; 3. There is a lack of dynamic prediction and update mechanisms for disaster development under different time and space conditions, leading to delays in disaster mitigation decisions; 4. Most disaster assessment models are built based on fixed historical data and expert experience, lacking mechanisms for parameter correction and adaptive updates using real-time monitoring data, making it difficult to cope with real-time changes in environmental conditions and operating status. Summary of the Invention

[0004] The purpose of this invention is to provide a holographic perception and dynamic disaster prediction system for hydrogen leak accidents in tunnel scenarios, which solves the shortcomings of existing technologies in terms of multi-dimensional information acquisition, prediction speed and accuracy, dynamic evaluation and adaptive model updating.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a holographic perception and dynamic disaster prediction system for hydrogen leak accidents in tunnel scenarios, which consists of six parts: a multimodal holographic perception unit, an information fusion module, a hazard source analysis module, a dynamic disaster prediction module, a feedback early warning unit, and a data storage and self-optimization module;

[0006] The multimodal holographic sensing unit is used to collect environmental parameters such as temperature, humidity, air pressure, wind speed, and wind direction in real time, as well as multidimensional information data such as hydrogen concentration distribution in the tunnel, changes in heat source temperature, acoustic characteristic signals generated by hydrogen leakage, vehicle position and driving speed.

[0007] The information fusion module is used to perform millisecond-level time synchronization and spatial registration processing on data from different sensors to generate multimodal fusion data in a unified format for subsequent use by the hazard source analysis module and disaster prediction model.

[0008] The hazard source analysis module is used to locate the leakage source and estimate the leakage rate based on synchronized multimodal information, while locating potential ignition sources and extracting disaster characteristics.

[0009] The disaster dynamic prediction module is used to receive tunnel environmental parameters, hydrogen leak location, leak rate, concentration distribution and potential ignition source location output by the hazard source analysis module, and transmit the information as input features to the CFD-Net proxy model. The model is used to quickly predict and analyze the spatiotemporal evolution trend of accidents such as combustion and explosion that may be caused by hydrogen leaks, and calculate the disaster level distribution of accidents at different times and in different spatial locations.

[0010] The CFD-Net model is a disaster prediction proxy model trained on a large-scale CFD simulation prior dataset. The simulation dataset was obtained through CFD simulations of the hydrogen leakage-diffusion-ignition-detonation process under various parameter combinations, including different leakage locations, leakage orifice diameters, gas storage pressures, wind speed and direction, tunnel structures, vehicle distribution locations, ignition locations, and ignition delay times. Each simulation result outputs the corresponding thermal radiation distribution H(x, y, z,t) and the explosion overpressure field P. o The spatiotemporal evolution data of (x, y, z, t), where x, y, z are the three-dimensional coordinates of a certain location in the tunnel, and t is the time after the leak occurred;

[0011] The feedback and early warning unit, based on the tunnel digital twin model, maps the predicted information onto the three-dimensional scene, outputs it synchronously to the tunnel monitoring platform, and releases early warning information to vehicles and personnel inside the tunnel through variable information signs and tunnel broadcasting systems.

[0012] The data storage and self-optimization module is used to uniformly store and manage multimodal sensor data and real-time assessment data of historical accidents, and to iteratively update the disaster prediction model through incremental learning.

[0013] Furthermore, the multimodal holographic sensing unit includes an environmental monitoring module, a hydrogen sensor array, a distributed infrared thermal imager, a high-frequency acoustic array, and a millimeter-wave radar system;

[0014] The environmental monitoring module includes a temperature and humidity sensor, an air pressure sensor, and a wind speed and direction sensor, which are used to monitor environmental parameters such as temperature, humidity, air pressure, wind speed, and wind direction inside the tunnel in real time.

[0015] The hydrogen sensor array is arranged at intervals along the longitudinal direction and both sides of the tunnel. It adopts a catalytic combustion type sensor to collect the hydrogen concentration distribution at different locations in real time.

[0016] The distributed infrared thermal imager is deployed at key locations in the tunnel to monitor changes in heat source temperature and detect potential abnormal heating ignition sources.

[0017] The high-frequency acoustic array includes multiple acoustic sensors deployed on the tunnel wall and ceiling to collect characteristic acoustic signals generated during high-speed hydrogen leakage and to analyze the leakage status.

[0018] The millimeter-wave radar system is deployed at key locations in the tunnel to monitor the real-time location and speed of vehicles inside the tunnel, as well as information on obstacles inside the tunnel.

[0019] Furthermore, the information fusion module includes an anomaly detection module, a spatiotemporal synchronization module, and a formatting module;

[0020] The anomaly detection module automatically detects abnormal data such as sensor drift, disconnection, and sudden noise changes, and repairs or removes the data.

[0021] The spatiotemporal synchronization module adopts the IEEE-1588 precision time protocol to achieve millisecond-level time alignment of data from different types of sensors, and achieves spatial registration based on the mapping relationship between the sensor coordinate system and the tunnel global coordinate system established by the tunnel digital twin model.

[0022] The formatting module normalizes and formats the spatiotemporally aligned multimodal data to form formatted data that can be directly called by the hazard source analysis module.

[0023] Furthermore, the hazard source analysis module includes a leak source location module, a leak rate estimation module, an ignition source identification module, and a feature extraction module;

[0024] The working steps of the leakage source location module include:

[0025] a. Based on multi-point concentration distribution and leakage sound source signals, spatial interpolation and concentration gradient analysis are used to determine the suspected area of ​​high hydrogen concentration, and the sound source search range is narrowed by combining tunnel geometry and wind field prior information.

[0026] b. Bandpass filtering, noise reduction and event trigger detection are performed on the multi-channel signals acquired by the acoustic array. The beam power output of each candidate position is calculated by the beamforming method based on the maximum controllable response power, and redundant candidate points with the same arrival time difference are eliminated by searching the space cluster.

[0027] c. Within the convergence region, determine the location and three-dimensional coordinates of the leakage sound source based on the location of maximum power output;

[0028] Based on the location of the leakage source, the leakage rate estimation module analyzes the time series data of the hydrogen concentration sensor, extracts features such as the concentration change rate and diffusion trend, and inputs the features into the rate inversion model to output the leakage rate estimate.

[0029] The rate inversion model is trained based on a pre-built simulation dataset. The model analyzes the evolution characteristics of hydrogen concentration over time and space and maps it to the corresponding leakage rate. The simulation dataset is obtained by conducting CFD simulations of hydrogen diffusion under different leakage locations, leakage orifice diameters, gas storage pressures, wind speeds, wind directions, tunnel structural parameters, vehicle distributions, etc., covering a variety of diffusion modes and boundary conditions.

[0030] The ignition source identification module is based on the temperature distribution data collected by the infrared thermal imager. Combined with the operating status of the electrical equipment in the tunnel, it uses temperature threshold segmentation and temperature rise rate analysis methods to extract the location and characteristics of high-temperature heat sources. At the same time, it determines whether the electrical equipment is in a live, faulty, or high-load state in order to identify the possible types, locations, and possible ignition times of potential ignition sources.

[0031] The feature extraction module integrates and formats key elements such as temperature, humidity, air pressure, wind speed, wind direction, tunnel size, tunnel slope, leak location, leak rate, hydrogen concentration distribution, ignition location, and obstacle distribution to generate a standardized dataset for the disaster dynamic prediction module to use.

[0032] Furthermore, the disaster dynamic prediction module consists of a situation prediction module and a disaster assessment module, and they work together according to the following process:

[0033] The situation prediction module inputs features such as temperature, humidity, air pressure, wind speed, wind direction, tunnel size, tunnel slope, leak location, leak rate, hydrogen concentration distribution, ignition location, and obstacle distribution into the CFD-Net surrogate model to predict the thermal radiation distribution H(x, y, z, t) and the explosion overpressure field P of hydrogen diffusion and combustion. o The spatiotemporal evolution of (x, y, z, t) is output, showing the scenario in continuous time steps t0~t1. n Spatiotemporal evolution dataset within;

[0034] The disaster assessment module inputs the spatiotemporal distribution information predicted by the situation prediction module and combines it with the structural damage and personnel injury model to calculate the probability of tunnel damage, personnel injury, and personnel death at different locations and times, quantifying the consequences of the accident. This is then combined with a pre-set threshold {R}. EH , R H , R M , R L The disaster situation is classified, and the disaster level results are sent to the feedback early warning unit, where R... EH R H R M R L These correspond to extremely high risk, high risk, medium risk, and low risk, respectively.

[0035] Furthermore, the working steps of the data storage and self-optimization module include:

[0036] a. The preprocessed fusion data, accident mechanism simulation data, and prediction results are multi-dimensionally tagged and stored in the accident and operation database in the form of metadata tables. The fusion data includes multi-modal sensor data such as hydrogen concentration distribution, heat source temperature changes, acoustic characteristics, vehicle position and speed; the simulation data includes leak source location results, leak rate estimates, potential ignition source locations, and corresponding environmental condition parameters; and the prediction results include accident types and consequences at different times and locations.

[0037] b. Combine key parameters such as temperature, humidity, air pressure, wind speed, wind direction, tunnel size, slope, leak location, leak rate, hydrogen concentration distribution, ignition location and obstacle distribution into various working conditions, carry out CFD simulation to expand the original dataset, and retrain the CFD-Net proxy model based on the expanded dataset.

[0038] c. Evaluate the performance of the retrained model in different operating scenarios, and fine-tune the parameters based on the performance evaluation results to achieve continuous adaptive optimization of the model.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] 1. This invention integrates multiple sensors such as a hydrogen sensor array, a distributed infrared thermal imager, a high-frequency acoustic array, and a millimeter-wave radar to form a multimodal holographic perception network covering the longitudinal direction of the tunnel and key nodes. This enables real-time acquisition of multi-dimensional information such as concentration distribution, thermal anomalies, acoustic characteristics, and vehicle operating status. Furthermore, through a spatiotemporal synchronization mechanism and abnormal data management, the completeness and accuracy of data acquisition are improved.

[0041] 2. This invention balances the accuracy of traditional CFD methods with the timeliness of engineering operation. It adopts a leakage rate inversion model and a disaster prediction proxy model trained on a CFD prior simulation dataset. After the leakage source is located, the hydrogen leakage rate can be estimated quickly and accurately. The features such as leakage location, leakage rate, ignition location and environmental parameters are input into the CFD-Net proxy model to achieve dynamic and continuous spatiotemporal prediction of accident processes such as combustion and explosion.

[0042] 3. This invention has adaptive optimization capabilities, which unifies the management and multidimensional labeling of historical accident data and real-time monitoring data, continuously expands the dataset and iteratively optimizes the prediction model based on the incremental retraining mechanism, thereby maintaining prediction accuracy and stability in long-term operation. Attached Figure Description

[0043] Figure 1 This is a system flowchart of the present invention.

[0044] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0046] To address the shortcomings in multidimensional information acquisition, prediction speed and accuracy, dynamic evaluation, and adaptive model updates, such as... Figure 1 , 2 As shown, the present invention provides the following preferred technical solutions:

[0047] The holographic perception and dynamic disaster prediction system for hydrogen leak accidents in tunnel scenarios consists of six parts: a multimodal holographic perception unit, an information fusion module, a hazard source analysis module, a dynamic disaster prediction module, a feedback early warning unit, and a data storage and self-optimization module.

[0048] Among them, the multimodal holographic sensing unit is used to collect environmental parameters such as temperature, humidity, air pressure, and wind speed in real time, as well as multidimensional information data such as hydrogen concentration distribution in the tunnel, changes in heat source temperature, acoustic characteristic signals generated by hydrogen leakage, vehicle position and driving speed.

[0049] The information fusion module is used to perform millisecond-level time synchronization and spatial registration processing on data from different sensors to generate multimodal fusion data in a unified format for subsequent use by the hazard source analysis module and disaster prediction model.

[0050] The hazard source analysis module is used to locate the leak source and estimate the leak rate based on synchronized multimodal information, while locating potential ignition sources and extracting disaster characteristics.

[0051] The disaster dynamic prediction module is used to receive tunnel environmental parameters, hydrogen leak location, leak rate, concentration distribution and potential ignition source location output by the hazard source analysis module, and transmit the information as input features to the CFD-Net proxy model. The model is used to quickly predict and analyze the spatiotemporal evolution trend of accidents such as combustion and explosion that may be caused by hydrogen leaks, and calculate the distribution of accident disaster level at different times and in different spatial locations.

[0052] The feedback and early warning unit is based on the tunnel digital twin model, which maps the predicted information to the three-dimensional scene, outputs it synchronously to the tunnel monitoring platform, and releases early warning information to vehicles and personnel inside the tunnel through variable information signs and tunnel broadcasting systems.

[0053] The data storage and self-optimization module is used to uniformly store and manage multimodal sensor data and real-time assessment data of historical accidents, and to iteratively update the disaster prediction model through incremental learning.

[0054] The multimodal holographic sensing unit includes an environmental monitoring module, a hydrogen sensor array, a distributed infrared thermal imager, a high-frequency acoustic array, and a millimeter-wave radar system.

[0055] The environmental monitoring module consists of temperature, humidity, air pressure, and wind speed and direction sensors, which are deployed every 100 m along the longitudinal direction of the tunnel, and are densely deployed at the shaft and entrance sections. The sampling period is 5 s, and the wind speed range is 0~20 m / s.

[0056] The hydrogen sensor array uses catalytic combustion type sensors arranged every 50 m along both sides of the wall, with one additional sensor every 50 m at the top of the arch to monitor high-altitude stratification and accumulation, with a range of 0~4% volume fraction.

[0057] Distributed infrared thermal imagers are deployed every 100 m in the electromechanical equipment area, shaft entrance, and traffic bottleneck section, with a thermal sensitivity of no more than 50 mK, to identify abnormal temperature rises and potential ignition sources; the high-frequency acoustic array consists of an array unit composed of 48 MEMS microphones, installed on the wall and the vault, with an array spacing of 80 m and a sampling rate of no less than 48 kHz, to capture the noise and whistling characteristics generated by high-speed hydrogen leakage.

[0058] The millimeter-wave radar system is deployed every 100 m along the longitudinal direction of the tunnel, using a 77 / 79 GHz FMCW system to cover adjacent lanes, with a refresh rate of no less than 10 Hz. It is used to acquire vehicle position, speed and local obstacle information, and can be cross-checked with existing video detection.

[0059] The sensor terminal and the information fusion module are interconnected via gigabit Ethernet. Data is transmitted to the information fusion module after being collected and timestamped.

[0060] The information fusion module includes an anomaly detection module, a spatiotemporal synchronization module, and a formatting module.

[0061] The anomaly detection module is used to identify and process abnormal data such as sensor drift, disconnection, and noise mutation. For slow drift, it uses multi-segment linear regression based on a fixed time window combined with sliding median filtering for correction. For disconnection data, it performs local interpolation. For sudden noise, it uses a robust Z-score method and interquartile range method to remove it with dual thresholds.

[0062] The time-space synchronization module completes the time alignment of multi-source data based on the IEEE-1588 precision time protocol, with a maximum allowable deviation of no more than 5 ms, and establishes a mapping relationship between the sensor local coordinate system and the tunnel global coordinate system to achieve spatial registration.

[0063] The formatting module converts the aligned multimodal data into standardized fields including timestamps, three-dimensional location coordinates, hydrogen concentration vectors, wind field vectors, temperature, humidity and air pressure parameters, infrared temperature image indexes, acoustic array signal identifiers and vehicle status vectors, and performs normalization processing for direct use by the hazard source analysis module.

[0064] The hazard source analysis module includes a leak source location module, a leak rate estimation module, an ignition source identification module, and a feature extraction module.

[0065] The leak source location module is used to determine the spatial location of a hydrogen leak, and its steps include:

[0066] a. Based on the multi-point hydrogen concentration distribution, construct longitudinal and transverse profiles, calculate the spatial gradient and temporal growth rate of concentration, and combine real-time wind direction, wind speed and tunnel geometric constraints to generate suspected leakage source areas;

[0067] b. Bandpass filtering, noise reduction, and event-triggered detection are performed on the multi-channel signals acquired by the acoustic array. The arrival time difference between each sensor pair is obtained by using a signal delay estimation algorithm. The beam power output of each candidate position is calculated by using the maximum controllable response power beamforming method. Redundant candidate points equivalent to the arrival time difference are eliminated by searching the space cluster.

[0068] c. Perform a fine search within the intersection of the suspected leakage area and the acoustic clustering area, take the point with the maximum beam power as the three-dimensional coordinates of the leakage sound source, and perform a secondary verification based on the consistency of the concentration field gradient direction.

[0069] Given the leak location, the leak rate estimation module extracts features such as near-field concentration growth rate, concentration spatial gradient norm, wind speed, acoustic characteristics, and vehicle-induced disturbances. These features are then input into a rate inversion model trained on a CFD prior dataset to output the leak rate LR (kg / s). The rate inversion model is trained on a pre-built simulation dataset. The model analyzes the evolution of hydrogen concentration over time and space and maps it to the corresponding leak rate. The training dataset is obtained through CFD simulations of hydrogen diffusion under different leak locations, leak orifice diameters, gas storage pressures, wind speeds, wind directions, tunnel structural parameters, and vehicle distribution conditions, covering various diffusion modes and boundary conditions.

[0070] The ignition source identification module is based on the temperature field of the infrared thermal imager and the operating status of the electromechanical equipment. It sets temperature thresholds according to the equipment type, combines the temperature rise rate to identify the location of potential ignition sources and possible ignition time windows, and determines whether the equipment is in a powered, faulty, or high-load state.

[0071] The feature extraction module extracts wind speed, wind direction, tunnel size, tunnel slope, leak location (x, y, z), leak rate (LR), and potential ignition location (x). p , y p , z p The environmental vector E={Temp, RH, Wind, P}, vehicle obstacle distribution, and potential ignition locations are integrated into a fixed-format feature vector, which serves as the input to the disaster dynamic prediction module.

[0072] The disaster dynamic prediction module consists of two parts: situation prediction and disaster assessment.

[0073] The situation prediction module inputs the above features into the CFD-Net surrogate model and rapidly outputs the thermal radiation distribution H(x, y, z, t) and the explosion overpressure field P within a time step t=0~60 s and a step size ∆t=0.5 s. oThe spatiotemporal evolution of (x, y, z, t) is described. The CFD-Net surrogate model is a deep learning-based multi-output spatiotemporal prediction network, trained on a large-scale prior CFD simulation dataset. This dataset was obtained through three-dimensional unsteady-state CFD simulations under various parameter combinations, including different leak locations, leak orifice diameters, gas storage pressures, wind speeds and directions, tunnel structures, vehicle distribution locations, ignition locations, and ignition delay times. It covers various hydrogen diffusion, combustion, and explosion modes, including non-uniform diffusion, stratified accumulation, jet fire, flashover, and explosion. Each simulation result outputs the corresponding thermal radiation distribution H(x, y, z, t) and the explosion overpressure field P under the corresponding operating conditions. o Spatiotemporal evolution data of (x, y, z, t).

[0074] The disaster assessment module will combine H and P o Inputting a structural and personnel injury model, the system calculates the probability of structural damage, personnel injury, and personnel death at different locations (d) and times (t). It then calculates the disaster risk value using accident scenario probabilities and combines this value with a pre-set threshold {R}. EH , R H , R M , R L The system classifies disaster situations, outputs extremely high / high / medium / low disaster levels, and forms a level layer that evolves over time and space, which is then sent to the feedback early warning unit.

[0075] The formula for calculating the disaster risk value is as follows:

[0076] ;

[0077] In the formula, Accident scenario The probability of occurrence For the context The consequences at position d at time t;

[0078] The probability of the accident scenario is calculated by combining the hydrogen leakage rate (LR), the ignition probability model, and the accident consequence path. The calculation formula for the ignition probability model is as follows:

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] In the formula, ISP is the detection isolation probability, IP is the instantaneous ignition probability, DP is the delayed ignition probability, DEP is the probability of delayed ignition leading to explosion, the ignition probability is not greater than 1, and LR is the hydrogen release rate, in kg / s.

[0084] The path probability of the accident consequence is further calculated by combining the accident consequence path, and the calculation formula is as follows:

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] In the formula, To increase the probability of cutting off the gas supply in time before ignition. To determine the probability of flashover, The probability of jet fire occurring. This represents the probability of an explosion.

[0090] The quantification of consequences is exemplified by mortality probability models, including the TNO thermal radiation mortality probability unit model and the TNO overpressure mortality probability unit model, which are used to assess the degree of harm caused to personnel by thermal radiation and blast shock waves, respectively. The formulas are as follows:

[0091]

[0092]

[0093]

[0094]

[0095] In the formula, H is the thermal radiation dose, and the unit is 1000 kJ / m². ;P o P is atmospheric pressure in Pa. s The peak overpressure of the explosion is expressed in Pa, and i is the explosion shock wave pulse, in units of... m is the mass of a person, measured in kilograms.

[0096]

[0097]

[0098]

[0099] In the formula, P1 and P2 are the probabilities of death for a person exposed to a certain dose of heat radiation or overpressure, respectively. c This represents the overall probability of death from heat.

[0100] The feedback and early warning unit couples the received prediction data with the tunnel's digital twin model, mapping the predicted disaster consequences (such as the probability of death) and disaster level layers in three-dimensional space within the model. The digital twin model encompasses information such as tunnel geometry, sensor placement locations, ventilation shaft locations, lane layout, and real-time traffic conditions, ensuring an accurate correspondence between the visualization results and the actual tunnel space.

[0101] The mapped 3D visualization data stream is transmitted in real time to the tunnel monitoring platform via gigabit Ethernet. Upon receiving the data stream, the monitoring platform can simultaneously output and display dynamic 3D renderings of the thermal radiation field distribution, explosion overpressure field, and risk level distribution on multiple screen interfaces. Simultaneously, the feedback early warning unit automatically generates warning content on variable message signs at key locations within the tunnel entrance and interior.

[0102] The working steps of the data storage and self-optimization module include:

[0103] a. The preprocessed fusion data, accident mechanism simulation data, and prediction results are multi-dimensionally tagged and stored in the accident and operation database in the form of metadata tables. The fusion data includes multi-modal sensor data such as hydrogen concentration distribution, heat source temperature changes, acoustic characteristics, vehicle position and speed; the simulation data includes leak source location results, leak rate estimates, potential ignition source locations, and corresponding environmental condition parameters; and the prediction results include accident types and consequences at different times and locations.

[0104] b. Combine key parameters such as temperature, humidity, air pressure, wind speed, wind direction, tunnel size, slope, leak location, leak rate, hydrogen concentration distribution, ignition location and obstacle distribution into various working conditions, carry out CFD simulation to expand the original dataset, and retrain the CFD-Net proxy model based on the expanded dataset.

[0105] c. Evaluate the performance of the retrained model in different operating scenarios, and fine-tune the parameters based on the performance evaluation results to achieve continuous adaptive optimization of the model.

[0106] The above embodiments are only used to illustrate the present invention. Any equivalent transformations and improvements made on the basis of the technical solutions of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A holographic perception and dynamic disaster prediction system for hydrogen leak accidents in tunnel scenarios, characterized in that, The system includes a multimodal holographic perception unit, an information fusion module, a hazard source analysis module, a disaster dynamic prediction module, a feedback early warning unit, and a data storage and self-optimization module; The multimodal holographic sensing unit is used to collect environmental parameters including temperature, humidity, air pressure, wind speed, and wind direction in real time, as well as multidimensional information data such as hydrogen concentration distribution in the tunnel, changes in heat source temperature, acoustic characteristic signals generated by hydrogen leakage, vehicle position, and driving speed. The information fusion module is used to perform millisecond-level time synchronization and spatial registration processing on data from different sensors to generate multimodal fusion data in a unified format for subsequent use by the hazard source analysis module and disaster prediction model. The hazard source analysis module is used to locate the leakage source and estimate the leakage rate based on synchronized multimodal information, while locating potential ignition sources and extracting disaster characteristics. The disaster dynamic prediction module is used to receive tunnel environmental parameters, hydrogen leak location, leak rate, concentration distribution and potential ignition source location output by the hazard source analysis module, and transmit the information as input features to the CFD-Net proxy model. The model is used to quickly predict and analyze the spatiotemporal evolution trend of accidents such as combustion and explosion that may be caused by hydrogen leaks, and calculate the disaster level distribution of accidents at different times and in different spatial locations. The CFD-Net model is a disaster prediction proxy model trained on a prior CFD simulation dataset. The simulation dataset is obtained by conducting CFD simulations of the hydrogen leakage-diffusion-ignition-detonation process using multiple parameter combinations. Each simulation result outputs the thermal radiation distribution H(x, y, z, t) and the explosion overpressure field P under the corresponding operating conditions. o The spatiotemporal evolution data of (x, y, z, t), where x, y, z are the three-dimensional coordinates of a certain location in the tunnel, and t is the time after the leak occurred; The feedback and early warning unit maps the predicted information onto the three-dimensional scene based on the tunnel digital twin model, outputs it synchronously to the tunnel monitoring platform, and releases early warning information to vehicles and personnel inside the tunnel through variable information signs and tunnel broadcasting systems. The data storage and self-optimization module is used to uniformly store and manage multimodal sensor data and real-time assessment data of historical accidents, and to iteratively update the disaster prediction model through incremental learning.

2. The system according to claim 1, characterized in that, The multimodal holographic sensing unit includes an environmental monitoring module, a hydrogen sensor array, a distributed infrared thermal imager, a high-frequency acoustic array, and a millimeter-wave radar system. The environmental monitoring module includes a temperature and humidity sensor, an air pressure sensor, and a wind speed and direction sensor, which are used to monitor environmental parameters such as temperature, humidity, air pressure, wind speed, and wind direction inside the tunnel in real time. The hydrogen sensor array is arranged at intervals along the longitudinal direction and both sides of the tunnel. It adopts a catalytic combustion type sensor to collect the hydrogen concentration distribution at different locations in real time. The distributed infrared thermal imager is deployed at key locations in the tunnel to monitor changes in heat source temperature and detect potential abnormal heating ignition sources. The high-frequency acoustic array includes multiple acoustic sensors deployed on the tunnel wall and ceiling to collect characteristic acoustic signals generated during high-speed hydrogen leakage and to analyze the leakage status. The millimeter-wave radar system is deployed at key locations in the tunnel to monitor the real-time location and speed of vehicles inside the tunnel, as well as information on obstacles inside the tunnel.

3. The system according to claim 2, characterized in that, The information fusion module includes an anomaly detection module, a spatiotemporal synchronization module, and a formatting module; The anomaly detection module automatically detects abnormal data and repairs or removes it; abnormal data includes sensor drift, disconnection, and sudden noise changes. The spatiotemporal synchronization module adopts the IEEE-1588 precision time protocol to achieve millisecond-level time alignment of data from different types of sensors, and achieves spatial registration based on the mapping relationship between the sensor coordinate system and the tunnel global coordinate system established by the tunnel digital twin model. The formatting module normalizes and formats the spatiotemporally aligned multimodal data to form formatted data that can be directly called by the hazard source analysis module.

4. The system according to claim 3, characterized in that, The hazard source analysis module includes a leak source location module, a leak rate estimation module, an ignition source identification module, and a feature extraction module.

5. The system according to claim 4, characterized in that, The working steps of the leakage source location module include: a. Based on multi-point concentration distribution and leakage sound source signals, spatial interpolation and concentration gradient analysis are used to determine the suspected area of ​​high hydrogen concentration, and the sound source search range is narrowed by combining tunnel geometry and wind field prior information. b. Bandpass filtering, noise reduction and event trigger detection are performed on the multi-channel signals acquired by the acoustic array. The beam power output of each candidate position is calculated by the beamforming method based on the maximum controllable response power, and redundant candidate points with the same arrival time difference are eliminated by searching the space cluster. c. Within the convergence region, determine the location and three-dimensional coordinates of the leakage sound source based on the location of the maximum power output.

6. The system according to claim 4, characterized in that, The working steps of the leakage source location module include: Based on the location of the leakage source, the leakage rate estimation module analyzes the time series data of the hydrogen concentration sensor, extracts the concentration change rate and diffusion trend characteristics, and inputs the characteristics into the rate inversion model to output the leakage rate estimate. The rate inversion model is trained based on a pre-built simulation dataset. The model analyzes the evolution characteristics of hydrogen concentration over time and space and maps it to the corresponding leakage rate. The simulation dataset is obtained by conducting CFD simulations of hydrogen diffusion under different leakage locations, leakage orifice diameters, gas storage pressures, wind speeds, wind directions, tunnel structural parameters, and vehicle distribution conditions, covering a variety of diffusion modes and boundary conditions. The ignition source identification module is based on temperature distribution data collected by an infrared thermal imager. Combined with the operating status of electrical equipment in the tunnel, it uses temperature threshold segmentation and temperature rise rate analysis methods to extract the location and characteristics of high-temperature heat sources. At the same time, it determines whether the electrical equipment is in a live, faulty, or high-load state in order to identify the possible types and locations of potential ignition sources. The feature extraction module integrates and formats key elements to generate a standardized dataset for the disaster dynamic prediction module to call; the key elements are temperature, humidity, air pressure, wind speed, wind direction, tunnel size, tunnel slope, leak location, leak rate, hydrogen concentration distribution, ignition location, and obstacle distribution.

7. The system according to claim 6, characterized in that, The disaster dynamic prediction module consists of a situation prediction module and a disaster assessment module, and they work together according to the following process: The situation prediction module inputs feature information into the CFD-Net surrogate model to predict the thermal radiation distribution H(x, y, z, t) and the explosion overpressure field P of hydrogen diffusion combustion. o The spatiotemporal evolution of (x, y, z, t) is output, showing the scenario in continuous time steps t0~t1. n The spatiotemporal evolution dataset within the tunnel; the feature information includes temperature, humidity, air pressure, wind speed, wind direction, tunnel size, tunnel slope, leak location, leak rate, hydrogen concentration distribution, ignition location, and obstacle distribution; The disaster assessment module inputs the spatiotemporal distribution information predicted by the situation prediction module and combines it with the structural damage and personnel injury model to calculate the probability of tunnel damage, personnel injury, and personnel death at different locations and times, quantifying the consequences of the accident. It then classifies the disaster based on pre-set thresholds and sends the disaster level results to the feedback early warning unit.

8. The system according to claim 7, characterized in that, The working steps of the data storage and self-optimization module include: a. The preprocessed fusion data, accident mechanism simulation data and prediction results are tagged in multiple dimensions and stored in the accident and operation database in the form of a metadata table; The fused data includes modal sensor acquisition results such as hydrogen concentration distribution, heat source temperature changes, acoustic characteristics, and vehicle position and speed. The simulation data includes the leak source location results, estimated leak rate, potential ignition source locations, and corresponding environmental condition parameters; The prediction results include the types of accidents and their consequences in different times and locations; b. Combine key parameters such as temperature, humidity, air pressure, wind speed, wind direction, tunnel size, slope, leak location, leak rate, hydrogen concentration distribution, ignition location, and obstacle distribution into multiple working conditions, conduct CFD simulations to expand the original dataset, and retrain the CFD-Net proxy model based on the expanded dataset. c. Evaluate the performance of the retrained model in different operating scenarios, and perform parameter tuning based on the performance evaluation results to continuously and adaptively optimize the model.

Images