A risk quantitative evaluation method considering leakage point uncertainty of hydrogen station operation

Abstract

The application discloses a risk quantitative evaluation method considering leakage point uncertainty of hydrogenation station operation, and the method comprises the following steps: identifying a systematic hazard source of a hydrogenation station; constructing an accident scene based on a Bayesian network and performing evolution analysis; performing leakage diffusion analysis and data collection; performing jet fire, flash fire and explosion accident consequence analysis; constructing a risk evaluation method considering hydrogenation station operation uncertainty; and performing comprehensive risk evaluation and visualization based on a physical information neural network.The method can comprehensively and accurately dynamically quantize and visually evaluate the operation risk of the hydrogenation station.

Description

Technical Field

[0001] This invention relates to the field of hydrogen energy safety and risk assessment technology. More specifically, this invention relates to a risk quantification assessment method that takes into account the uncertainty of leakage points during the operation of hydrogen refueling stations. Background Technology

[0002] With the rapid development of the hydrogen energy industry, the safe operation of hydrogen refueling stations, as a crucial infrastructure for hydrogen fuel cell vehicles, is paramount. Traditional risk assessment methods for hydrogen refueling stations are mostly based on deterministic leakage scenarios (such as setting specific equipment and fixed leak orifice diameters) for consequence simulation and risk calculation. However, in actual operation, hydrogen refueling station systems are complex, containing multiple potential leak points such as compressors, hydrogen storage tanks, hydrogen dispensers, valves, and pipeline connections. Furthermore, the specific location of a leak, the direction of hydrogen leakage, and its evolution are highly uncertain. In addition, factors such as ignition sources, personnel exposure, and weather conditions also exhibit significant randomness. Using existing deterministic assessment methods may lead to risk assessment results that deviate from reality, failing to provide accurate guidance for risk management and safety barrier optimization at hydrogen refueling stations.

[0003] Therefore, there is an urgent need for an evaluation method that can systematically quantify the risks caused by multiple uncertainties in the operation of hydrogen refueling stations, so as to more realistically and dynamically reflect the overall risk level of hydrogen refueling stations. Summary of the Invention

[0004] One objective of this invention is to provide a risk quantification assessment method that considers the uncertainty of leakage points in the operation of hydrogen refueling stations. This method integrates system hazard source identification, Bayesian network accident evolution analysis, leakage diffusion analysis and dataset analysis, jet, flash fire, and explosion accident consequence analysis, multi-source uncertainty probability quantification, and a surrogate model based on physical neural networks, ultimately achieving accurate superposition and visualization of the risks of the entire station.

[0005] To address the aforementioned technical problems, this invention provides a risk quantification assessment method considering the uncertainty of leakage points during hydrogen refueling station operation, comprising the following steps: S1. Based on the collected general layout plan, process flow diagram, equipment data and environmental information of the hydrogen refueling station, construct the geometric model of the hydrogen refueling station required for evaluation, and clarify the physical parameters of hydrogen in each piece of equipment; and systematically identify potential hydrogen leakage points in all equipment bodies and pipeline systems within the hydrogen refueling station. S2. Based on historical data, fault tree and event tree analysis results, construct a Bayesian network to describe the evolution of "leakage-diffusion-ignition-consequence", assign conditional probabilities to each node in the network, and deduce the probability of different leak points causing various accident scenarios under different conditions. S3. For the representative leak points identified in step S1 and various meteorological conditions, use computational fluid dynamics (CFD) software to perform transient simulation of hydrogen leakage and diffusion, obtain the spatiotemporal distribution of hydrogen concentration and key physical field data of the accident, and combine it with the established personnel exposure model to provide a data foundation for subsequent accident consequence analysis and model training. S4. Based on the CFD simulation results of step S3, quantitative analysis is performed on the thermal radiation field of jet fire accident, the combustible cloud range of flash fire accident and the shock wave overpressure field of vapor cloud explosion accident. Combined with the corresponding injury criteria, the injury range of various accidents to personnel and equipment is assessed. S5. The system identifies and selects key uncertainty factors that affect the risk assessment results, assigns leakage probabilities to each factor, and uses the Bayesian network model in S2 and various accident injury consequence models in S4 to calculate the probability of accident occurrence under specific scenarios at the leakage point and the probability distribution of personal and social risks at each leakage point, thereby quantifying the uncertainty of risk and forming a risk assessment framework. S6. Using the limited amount of CFD simulation data from step S3, train the physical information neural network PINN to construct a fast proxy prediction model for the accident consequence field; integrate this proxy model with the risk assessment framework from step S5 to quickly calculate and overlay the risks of all potential leakage points in the entire site under multiple uncertainties, generate a comprehensive risk heat map and present it visually.

[0006] Preferably, step S1 specifically includes: S1.1 Based on the general layout plan, process flow diagram, equipment data and surrounding environmental information of the hydrogen refueling station, construct a geometric model including the long tube trailer, hydrogen storage cylinder group, compressor, hydrogen dispenser, unloading column, sequence control panel, hydrogen storage valve group and process pipeline, and clarify the physical parameters of hydrogen in each device, including storage form, working pressure and working temperature, as boundary conditions for leakage source analysis; S1.2 Systematically identify hazards in the constructed geometric model, including specific potential leak points such as welds of hydrogen storage equipment, valves, compressor shaft seals, flanges, hydrogen refueling machine hoses, interfaces, and all process pipeline connection points.

[0007] Preferably, step S2 specifically includes: S2.1 Define the initial event as the leakage of a specific potential hydrogen leak point identified in step S1, intermediate events include leak mode, ventilation effect, detection system response, and ignition source appearance, and the terminal event is a fire or explosion accident at a hydrogen refueling station. S2.2 Construct a Bayesian network topology that connects the above events, and assign conditional probability tables to each node in the network based on historical data and expert knowledge; S2.3 Input specific leakage frequency and environmental conditions, and dynamically infer and output the conditional probability of occurrence of various terminal accident scenarios through Bayesian network inference.

[0008] Preferably, step S3 specifically includes: S3.1. Combining the physical parameters and geometric model of S1, set leakage boundary conditions for each representative potential hydrogen leakage point; collect meteorological conditions of the hydrogen refueling station location, select representative combinations as calculation conditions, and assign annual occurrence frequency to each condition. S3.2 Obtain the distribution patterns of various personnel within the hydrogen refueling station and establish a personnel exposure model; S3.3 Perform transient simulation on the set leakage scenario, and output the following: spatiotemporal distribution cloud map of hydrogen concentration and key physical field data of the accident; The above data provides a data foundation for subsequent risk probability calculations and proxy prediction model training.

[0009] Preferably, step S5 specifically includes: S5.1 Identify and select key uncertainty factors, including: uncertainty of leakage frequency, uncertainty of leakage scenario, uncertainty of accident evolution path, uncertainty of environmental parameters and uncertainty of simulation model parameters; S5.2. Assign reasonable leakage probabilities to each key uncertainty factor; S5.3. The Bayesian network model of S2 is combined with the various accident injury consequence models of S4. The Bayesian network model is used to calculate the probability of accident occurrence in a specific scenario at the leak point. Risk post-processing software is used to calculate the probability distribution of personal risk and social risk at each leak point.

[0010] Preferably, step S6 specifically includes: S6.1. Using the leak point parameters, leak parameters, environmental parameters and time as inputs, and the spatial physical field data output by CFD simulation as labels, train the physical information neural network, and embed simplified fluid control equations as physical constraints in the loss function. S6.2 Integrate the trained physical information neural network proxy model into the risk assessment framework for direct simulation; S6.3. Based on the agent model, the rapid prediction results of the whole scenario are superimposed to calculate the comprehensive risk of the whole site, and visualized in the form of personal risk contour lines, social risk curves or probability cloud maps of specific injury thresholds.

[0011] Preferably, in step S4, jet fire analysis assesses the damage range using thermal radiation flux and damage models; flash fire analysis assesses the damage range by identifying clouds with hydrogen concentrations within the flammability limit obtained from CFD simulations; and vapor cloud explosion analysis assesses the damage range using shock wave overpressure and corresponding damage models.

[0012] The present invention has at least the following beneficial effects: The risk quantification and evaluation method of this invention introduces a Bayesian network to handle the uncertainty of accident evolution paths and combines a physical information neural network to achieve efficient calculation of large-scale risk scenarios. It can dynamically and accurately reflect the overall risk level of a hydrogen refueling station under the influence of multiple uncertain factors, and provide a comprehensive, accurate, dynamic quantification and visualization evaluation of the operational risks of hydrogen refueling stations.

[0013] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0014] Figure 1 This is an overall flowchart of the method of the present invention.

[0015] Figure 2 This is a plan view of the hydrogen refueling station in an embodiment of the present invention.

[0016] Figure 3 This is a process flow diagram of a hydrogen refueling station in an embodiment of the present invention.

[0017] Figure 4 This is a three-dimensional model of a hydrogen refueling station in an embodiment of the present invention.

[0018] Figure 5 This is a schematic diagram of the risk points of a hydrogen refueling station in an embodiment of the present invention.

[0019] Figure 6 This is a schematic diagram of an accident evolution analysis based on a Bayesian network in an embodiment of the present invention.

[0020] Figure 7 This is a spatiotemporal distribution cloud map of hydrogen concentration at different leakage orifice diameters in the hydrogen storage bottle valve of this invention.

[0021] Figure 8 This is a cloud map showing the temperature evolution of a large-scale leakage jet fire at the hydrogen storage bottle valve in an embodiment of the present invention.

[0022] Figure 9 This is a cloud map showing the temperature evolution of a large-scale leak and flashover at the hydrogen storage bottle valve in an embodiment of the present invention.

[0023] Figure 10 This is a cloud map illustrating the overpressure evolution of a large-scale leak and explosion at the hydrogen storage bottle valve in an embodiment of the present invention.

[0024] Figure 11 This is a risk standard diagram of acceptable risk in Chinese society, as described in this embodiment of the invention.

[0025] Figure 12 This is a schematic diagram of the training and application process of the proxy model based on the Physical Information Neural Network (PINN) in an embodiment of the present invention. Detailed Implementation

[0026] To better understand the purpose, structure, and function of this invention, the invention will be further described in detail below with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0027] It should be noted that, unless otherwise specified, the experimental methods described in the following embodiments are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified. In the description of this invention, the terms "lateral", "longitudinal", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0028] like Figures 1 to 12 As shown, this invention provides a risk quantification assessment method that considers the uncertainty of leakage points in the operation of hydrogen refueling stations. The method includes: identification of systemic hazards in hydrogen refueling stations; construction and evolution analysis of accident scenarios based on Bayesian networks; leakage diffusion analysis and data collection; consequence analysis of jet fire, flash fire and explosion accidents; construction of a risk assessment method considering the uncertainty of hydrogen refueling station operation; comprehensive risk assessment and visualization based on physical information neural networks; and comprehensive, accurate dynamic quantification and visualization evaluation of the operational risks of hydrogen refueling stations.

[0029] Specifically, the following steps are included: S1: Hydrogen Refueling Station Parameter Integration and System Hazard Identification. Based on the collected general layout plan, process flow diagram, and equipment data of the hydrogen refueling station, a geometric model of the hydrogen refueling station required for assessment is constructed; and potential hydrogen leak points in all equipment bodies and pipeline systems within the hydrogen refueling station are systematically identified.

[0030] This step aims to establish the underlying physical and geometric model for the assessment and systematically identify all potential leakage points.

[0031] S1.1 Parameter Description and Model Construction: Collect and describe in detail the overall layout plan, process flow diagram, equipment data, and surrounding environmental information of the hydrogen refueling station. Based on the drawings, accurately determine the location, shape, and dimensions of key facilities including the long-tube trailer, hydrogen storage cylinder group, compressor, sequential control panel, heat exchanger, hydrogen dispenser, and station building, as well as process piping, and construct the two-dimensional and three-dimensional geometric models required for the evaluation. Simultaneously, clarify the physical parameters of hydrogen storage form, operating pressure, and operating temperature within each piece of equipment, serving as boundary conditions for the leak source analysis in step S3.

[0032] S1.2 Leakage Point System Identification: Based on the constructed model and equipment data, a systematic hazard identification is conducted across the entire station to identify all specific locations where hydrogen leaks may occur, including welds in hydrogen storage equipment, valves, compressor shaft seals, flanges, hydrogen dispenser hoses, interfaces, and all process piping connections. These are mainly divided into two categories: Leakage points in the equipment body include: welds, end caps, valves, return pipe flanges, safety valves, and pressure gauge joints of the hydrogen storage cylinder / tank; valves and interfaces on the long-tube trailer; compressor pump housing, shaft seals, and flange gaskets; hydrogen dispenser hoses, hose-to-nozzle connections, and pull-out valves, etc.

[0033] Piping system leak points: including welds, elbows, tees, pipe joints, various valves, instrument interfaces, etc. on all process pipelines.

[0034] S2: Accident Scenario Construction and Evolution Analysis Based on Bayesian Networks. Based on historical data, fault tree and event tree analysis results, a Bayesian network describing the evolution of "leakage-diffusion-ignition-consequence" is constructed. Conditional probabilities are assigned to network nodes to deduce the probability of different accident scenarios such as jet fire, flash fire and vapor cloud explosion caused by different leak points under different conditions.

[0035] This step introduces a Bayesian Network (BN) to dynamically describe the evolution path and probability of accidents after a leak, replacing the simple leak-ignition frequency model.

[0036] S2.1 Definition of Initial and Intermediate Events: An event that could potentially cause a leak at each specific leak point identified in S1 is considered a possible initial event. Intermediate events include leak patterns, ventilation effectiveness, detection system response, and the presence of ignition sources, such as: persistent small-hole leaks, large-hole / rupture leaks, success / failure of leak detection and isolation systems (e.g., gas detectors, emergency shut-off valves), the effectiveness of natural and forced ventilation, the formation and spread of flammable clouds, and the presence of different types of ignition sources (e.g., static electricity, electrical sparks, open flames, hot surfaces). The final event is a fire or explosion at a hydrogen refueling station.

[0037] S2.2 Bayesian Network Structure Learning and Parameter Learning: Based on historical accident data, Fault Tree Analysis (FTA), Event Tree Analysis (ETA) results, and expert knowledge, a Bayesian network topology describing the chain of "leakage → diffusion → ignition → accident consequences" is constructed. Through historical data, literature review (such as the "Purple Book" and industry databases), and expert judgment, a conditional probability table (CPT) is assigned to each node in the network to quantify the causal probability between events. For example, given the conditions of "large hole leakage" and "ventilation failure," the probability of "forming an explosive cloud" is calculated.

[0038] S2.3 Accident Scenario Probability Inference: By using the constructed Bayesian network and inputting the initial frequency and on-site conditions (such as equipment status and wind speed) of a specific leak point, the probability of occurrence of various terminal accident scenarios (such as non-ignition diffusion, jet fire, flash fire, and vapor cloud explosion) can be inferred. This provides dynamic and condition-dependent accident scenario probabilities for subsequent risk quantification, rather than fixed ignition probabilities.

[0039] S3: Leakage and diffusion analysis and data collection. For the representative leak points identified in step S1 and various meteorological conditions, computational fluid dynamics (CFD) software was used to perform transient simulations of hydrogen leakage and diffusion, obtain the spatiotemporal distribution of hydrogen concentration, and combine it with the established personnel exposure model to provide a data foundation for accident consequence analysis and model training.

[0040] This step utilizes FLACS software to perform high-precision accident consequence simulation, providing a data foundation for risk calculation and surrogate model training.

[0041] S3.1 Boundary Conditions and Operating Conditions: Combining the physical parameters and geometric model of S1, leakage boundary conditions (orifice size, pressure, direction) are set for each representative leakage point. Historical meteorological statistics of the hydrogen refueling station location are collected, and representative combinations of wind speed, wind direction, temperature, and humidity are selected as calculation operating conditions, with annual occurrence frequencies assigned to each condition.

[0042] S3.2 Personnel Exposure Model: Investigate the number of staff (fixed positions) and transient personnel (drivers, visitors) in hydrogen refueling stations, and their spatiotemporal distribution patterns within the stations, and establish a personnel exposure model.

[0043] S3.3 CFD Simulation Execution: A transient simulation of the defined leakage scenario is performed using professional CFD software (FLACS). The core output is: ① Spatiotemporal distribution cloud map of hydrogen concentration (used to determine the range of combustible cloud clusters).

[0044] ② Key physical field data of the accident, such as the spatiotemporal evolution data of temperature field, pressure field, thermal radiation flux field, and shock wave overpressure field. This high-fidelity data will serve as the basis for the S4 accident consequence analysis and as training labels for the neural network surrogate model in S6.

[0045] S4: Consequence Analysis of Jet Fire, Flash Fire, and Explosion Accidents. Based on the CFD simulation results of step S3, quantitative analyses are performed on the thermal radiation field of jet fire accidents, the combustible cloud range of flash fire accidents, and the shock wave overpressure field of vapor cloud explosion accidents. Combined with relevant injury criteria, the scope of damage to personnel and equipment caused by each type of accident is assessed.

[0046] Based on the CFD simulation results of S3, a quantitative analysis of the consequences of three major fire and explosion accidents is conducted.

[0047] S4.1 Jet Fire Analysis: This analysis addresses scenarios where pressurized leaks are immediately ignited. Flame morphology, length, and temperature field are extracted from CFD results, and the thermal radiation flux at different locations is calculated. Combining thermal radiation hazard criteria and the Probit hazard model, contour maps of different thermal radiation intensities (e.g., 4 kW / m², 12.5 kW / m², 37.5 kW / m²) are plotted to assess the extent of harm to personnel and equipment.

[0048] S4.2 Flash Fire Analysis: This analysis addresses scenarios where delayed ignition leads to the formation of a flammable cloud. Based on the spatiotemporal distribution of hydrogen concentration obtained from FLACS simulations (typically flammable within a 4%-75% volume concentration range), the extent of the flammable cloud is determined. It is assumed that ignition at any point within the cloud will result in rapid combustion (i.e., a flash fire) throughout the entire flammable volume. By analyzing the spatiotemporal relationship between personnel location and the extent of the flammable cloud, the probability of casualties due to a flash fire is assessed.

[0049] S4.3 Vapor Cloud Explosion (VCE) Analysis: This analysis addresses scenarios where a flammable cloud formed in a confined or semi-confined space (such as an equipment area or under a pipe gallery) is ignited by a strong ignition source, resulting in significant overpressure. The overpressure field of the shock wave generated by the explosion is extracted from FLACS results. Combining the overpressure injury criterion and the Probit injury model, contour maps of different overpressure levels (e.g., 0.03 bar, 0.1 bar, 0.3 bar) are plotted to assess the extent of structural damage and personnel injury.

[0050] S5: Construction of a Risk Assessment Method for Uncertainty in Hydrogen Refueling Station Operation. The system identifies and selects key uncertainty factors that affect the risk assessment results, assigns leakage probabilities to each factor, calculates the probability distributions of individual and social risks, quantifies the uncertainty of risks, and forms a risk assessment framework.

[0051] S5.1 Identification and Selection of Uncertainty Factors: Identify and select key uncertainty factors that affect the risk assessment results of hydrogen refueling stations, including: Uncertainty in leakage frequency: differences in the data sources for the reference leakage frequency of equipment / pipeline section, and fluctuations in the value of frequency F per unit length.

[0052] Uncertainty in leakage scenarios: the precise location of the leak point, the size of the leak hole, and the randomness of the leak direction.

[0053] Evolutionary path uncertainty: that is, the uncertainty of the conditional probability of each node in the Bayesian network in S2.

[0054] Uncertainty in environmental parameters: real-time changes in wind speed, wind direction, and atmospheric stability.

[0055] Uncertainty in simulation model parameters: the sensitivity of key parameter settings in CFD simulations, such as turbulence model and combustion model.

[0056] S5.2 Uncertainty Probability Assignment: Assign a reasonable probability to each uncertainty factor selected in S5.1: Risk Calculation Where L represents the severity of the consequences and P represents the probability of the accident occurring.

[0057] For the piping system, the annual leakage frequency per unit length is set as P. For any given section, the total leakage frequency is P for that section.

[0058] Distribute this frequency evenly to all specific risk points (such as valves and flanges) within the section.

[0059] , where P i Let n be the leakage frequency at each risk point, and n be the number of risk points per unit pipeline.

[0060] For other equipment outside the pipeline (such as compressors and inlet / outlet units of storage tanks), the total leakage frequency is also decomposed to each specific valve or connection point based on its structural complexity and the number of sealing points. Thus, each specific risk point in the entire station is assigned an annual leakage frequency that considers the uncertainty of spatial distribution. .

[0061] S5.3 Risk Quantification Assessment Considering Uncertainty: A combination of Bayesian network model and accident consequence model in S4 is adopted.

[0062] Calculate the probability of an accident occurring under a specific scenario at the leak point using a Bayesian network model; Calculate the risk (personal risk and social risk) at each leakage point using risk post-processing software.

[0063] S6: Comprehensive Risk Assessment and Visualization Based on Physical Information Neural Network. Using the limited amount of CFD simulation data from step S3, a Physical Information Neural Network (PINN) is trained to construct a rapid proxy prediction model for the accident consequence field. This proxy model is integrated with the risk assessment framework from step S5, and combined with the Probit injury model and population distribution data, the individual risk at each grid point is calculated. Through full-scenario probability weighted overlay, the risks of all potential leakage points across the entire site under multiple uncertainties are rapidly calculated and overlaid to generate a comprehensive risk heatmap, which is then visualized.

[0064] To efficiently achieve large-scale scene computation in S5.3, a Physics-Informed Neural Network (PINN) is introduced as a proxy model for CFD simulation.

[0065] S6.1 Proxy Model Training: A limited number of high-fidelity CFD simulation samples from S3 were used as the training set. The input layer of the neural network includes: the coordinates of the leak point location, leak parameters (flow rate, direction), environmental parameters (wind speed, wind direction), and time. The output layer represents the values ​​of the physical fields of interest (such as concentration, temperature, thermal radiation, and overpressure) at spatial grid points. Simplified physical equations (such as the Navier-Stokes equation and partial forms of the component transport equation) controlling hydrogen gas diffusion and combustion are embedded in the loss function as regularization terms, enabling the network to maintain physical plausibility even in sparse data regions and improving generalization ability.

[0066] S6.2 Rapid Risk Calculation Across the Entire Site: The pre-trained PINN proxy model is integrated into the risk assessment framework. When calculating the leakage probability and consequences at each leak point, time-consuming CFD software is not required; instead, the pre-trained neural network instantly (in milliseconds) predicts the corresponding consequence field. This makes it possible to perform rapid and comprehensive risk calculations for tens of thousands of potential leak points across the entire site under various uncertain parameter combinations.

[0067] S6.3 Comprehensive Risk Visualization: Based on a large amount of results obtained from PINN accelerated calculations, the risk contributions of all leakage points and all uncertain scenarios are overlaid to generate a "comprehensive risk contour map" for the entire station. This map can be presented in the form of personal risk contour lines, social risk distribution curves, or probability cloud maps of specific damage thresholds (such as thermal radiation >12.5kW / m²), intuitively displaying high-risk areas within the hydrogen refueling station and providing precise spatial decision support for optimizing safety distances, setting up protective barriers, and formulating emergency plans. Example

[0068] This application provides a specific description of the application of hydrogen refueling stations in actual engineering projects at Yansteel.

[0069] A risk quantification assessment method considering the uncertainty of leakage points in the operation of hydrogen refueling stations includes the following steps: S1: Parameter Integration and System Hazard Identification of Yangang Hydrogen Refueling Station S1.1 Parameter Description and Model Construction: Collect the general layout plan, process flow diagram, equipment data, and surrounding environmental information of the Yansteel hydrogen refueling station. (See attached document) Figure 2 , 3 .

[0070] Key equipment modeling: Based on the drawings, accurately determine the location, shape, and dimensions of key facilities including the hydrogen long-tube trailer 1, 45MPa hydrogen storage cylinder group 2, 45MPa compressor group 3, compressor canopy 9, unloading column 6, hydrogen storage valve group 7, sequence control panel 8, heat exchanger / chiller unit 10, ammonia container 11, nitrogen container canopy 12, 35MPa hydrogen dispenser 13, hydrogen dispenser canopy 4, and station building 5. Construct the two-dimensional and three-dimensional geometric models required for the assessment. See [link / details]. Figure 4 Meanwhile, the storage form, operating pressure, operating temperature, and other physical parameters of hydrogen in each piece of equipment are clearly defined as boundary conditions for leak source analysis. Table 1 below is a list of the main equipment in a hydrogen refueling station.

[0071] Table 1. List of Main Equipment at Hydrogen Refueling Stations S1.2 System Identification of Leak Points Across the Entire Site: Based on the constructed model and equipment data, a systematic hazard identification was conducted across the entire station, identifying all specific locations where hydrogen leaks might occur. These are mainly divided into two categories, see [link to relevant documentation]. Figure 5 .

[0072] Leakage points in the equipment body include: welds, end caps, valves, return pipe flanges, safety valves, and pressure gauge joints of the hydrogen storage cylinder / tank; valves and interfaces on the long-tube trailer; compressor pump housing, shaft seals, and flange gaskets; hydrogen dispenser hoses, hose-to-nozzle connections, and pull-out valves, etc.

[0073] Piping system leak points include welds, elbows, tees, pipe joints, various valves (manual valves, pneumatic valves, solenoid valves, safety valves), and instrument interfaces (pressure gauges, thermometers, flow meters), etc., on all process pipelines.

[0074] S2: Accident Scenario Construction and Evolution Analysis Based on Bayesian Networks This step introduces a Bayesian Network (BN) to dynamically describe the evolution path and probability of post-leakage accidents, replacing the simple leak-ignition frequency model. (See...) Figure 6 .

[0075] S2.1 Definition of Initial and Intermediate Events: Each specific leak point identified in S1 is considered a possible initial event. Intermediate events include: persistent small hole leaks, large hole / rupture leaks, success / failure of leak detection and isolation systems (such as gas detectors, emergency shut-off valves), effectiveness of natural and forced ventilation, formation and spread of combustible clouds, and the appearance of different types of ignition sources (such as static electricity, electrical sparks, open flames, hot surfaces), etc.

[0076] Initial Event: The event that could cause each specific leak point in the S1.2 list to leak is taken as the initial node.

[0077] Intermediate events: Ignition sources. Potential ignition sources unique to the steel plant environment need to be considered, such as in-plant transport vehicles, maintenance arcs, and hot surfaces in specific areas, and their spatial proximity to the leak point should be assessed.

[0078] Terminal events: jet fire, flash fire, vapor cloud explosion.

[0079] S2.2 Bayesian Network Structure Learning and Parameter Learning: Based on historical accident data, Fault Tree Analysis (FTA) and Event Tree Analysis (ETA) results, and expert knowledge, a Bayesian network topology describing the chain of "leakage → diffusion → ignition → accident consequences" is constructed. Through historical data, literature review (such as the "Purple Book" and industry databases), and expert judgment, a conditional probability table (CPT) is assigned to each node in the network to quantify the causal probability between events.

[0080] S2.3 Accident Scenario Probability Inference: Using a pre-constructed Bayesian network, and inputting the initial frequency and site conditions (such as equipment status and wind speed) of a specific leak point, the probability of occurrence of various terminal accident scenarios (such as non-ignition diffusion, jet fire, flash fire, and vapor cloud explosion) can be inferred. This provides dynamic, condition-dependent accident scenario probabilities for subsequent risk quantification, rather than fixed ignition probabilities. Table 2 below shows the ignition probabilities under different hydrogen leakage rates.

[0081] Table 2 Ignition Probability S3: Leakage Propagation Analysis and Data Collection This step utilizes FLACS software to perform high-precision accident consequence simulation, providing a data foundation for risk calculation and surrogate model training.

[0082] S3.1 Boundary Conditions and Operating Conditions: Combining the physical parameters and geometric model of S1, leakage boundary conditions (orifice size, pressure, direction) are set for each representative leakage point. Historical meteorological statistics of the hydrogen refueling station location are collected, and representative combinations of wind speed, wind direction, temperature, and humidity are selected as calculation operating conditions, with annual occurrence frequencies assigned to each condition.

[0083] Table 3. Log-normal distribution parameters of leakage frequency of individual components Table 3 presents five leakage orifice sizes, corresponding to 0.01%, 0.1%, 1%, 10%, and 100% of the connecting pipe diameter, respectively. Considering that the risk posed by excessively small leakage orifice sizes is negligible, three leakage orifice sizes of 1%, 10%, and 100% of the connecting pipe diameter are selected for analysis to reduce the number of accidents. Table 3 also presents the log-normal distribution parameters of the leakage frequency of a single component. For ease of calculation, its mathematical expectation is used as the leakage frequency, resulting in the results shown in Table 4.

[0084] If the random variable Y follows a log-normal distribution, that is, if it satisfies: The formula for the expected value (mean) of Y is: Table 4 Leakage frequency of individual components S3.2 Personnel Exposure Model: Investigate the number of staff (fixed positions) and transient personnel (drivers, visitors) in hydrogen refueling stations, and their spatiotemporal distribution patterns within the stations, and establish a personnel exposure model.

[0085] The activity trajectories of the operators (2 people / shift) at the Yangang hydrogen refueling station (mainly in the control room and inspection routes), as well as the exposure of hydrogen-powered heavy truck drivers (approximately 20 truck trips per day, with each truck taking about 15 minutes to refuel and waiting in the waiting area).

[0086] S3.3 CFD Simulation Execution: A transient simulation of the defined leakage scenario is performed using professional CFD software (FLACS). The core output is: ① Spatiotemporal distribution cloud map of hydrogen concentration, see Figure 7 .

[0087] ② Key physical field data of the accident, such as the spatiotemporal evolution data of the temperature field, pressure field, thermal radiation flux field, and shock wave overpressure field. This data will serve as the basis for the S4 accident consequence analysis and the training labels for the neural network surrogate model in S6.

[0088] S4: Consequence Analysis of Jet Fire, Flash Fire and Vapor Cloud Explosion Accidents Based on the CFD simulation results of S3, a quantitative analysis of the consequences of three major fire and explosion accidents is conducted. Table 5 shows the damage and losses caused by different thermal radiation intensities. Table 6 shows the damage to buildings caused by shock waves.

[0089] Table 5 Damage and losses caused by different thermal radiation intensities. Table 6. Damage to buildings caused by shock waves S4.1 Jet Fire Analysis: This analysis addresses scenarios where pressurized leaks are immediately ignited. Flame morphology, length, and temperature field are extracted from CFD results, and the thermal radiation flux at different locations is calculated. Combining thermal radiation hazard criteria, contour maps of different thermal radiation intensities (e.g., 4 kW / m², 12.5 kW / m², 37.5 kW / m²) are plotted to assess the extent of harm to personnel and equipment. (See attached diagram). Figure 8 .

[0090] S4.2 Flash Fire Analysis: This analysis addresses scenarios where delayed ignition leads to the formation of a combustible cloud. Based on the hydrogen concentration distribution obtained from FLACS simulations (typically combustible within a 4%-75% volume concentration range), the extent of the combustible cloud is determined. It is assumed that ignition at any point within the cloud will result in rapid combustion (i.e., flash fire) throughout the entire combustible volume. By analyzing the spatiotemporal relationship between personnel location and the extent of the combustible cloud, the probability of casualties due to flash fire is assessed. (See [link to relevant documentation]). Figure 9 .

[0091] S4.3 Vapor Cloud Explosion (VCE) Analysis: This analysis addresses scenarios where a flammable cloud formed in a confined or semi-confined space is ignited by a strong ignition source, generating significant overpressure. The overpressure field of the shock wave generated by the explosion is extracted from FLACS results. Using overpressure damage criteria, contour maps of different overpressure levels are plotted to assess the extent of structural damage and personal injury. (See [link to relevant documentation]). Figure 10 .

[0092] S5: Construction of a Risk Assessment Method for Uncertainty in the Operation of Hydrogen Refueling Stations S5.1 Identification and Selection of Uncertainty Factors: The system identifies and selects key uncertainty factors that affect the final risk assessment results, including: Uncertainty in leakage frequency: differences in the data sources for the reference leakage frequency of equipment / pipeline section, and fluctuations in the value of frequency F per unit length.

[0093] Uncertainty in leakage scenarios: the precise location of the leak point, the size of the leak hole, and the randomness of the leak direction.

[0094] Evolutionary path uncertainty: that is, the uncertainty of the conditional probability of each node in the Bayesian network in S2.

[0095] Environmental uncertainties: real-time changes in wind speed, wind direction, and atmospheric stability.

[0096] S5.2 Uncertainty Probability Assignment: Assign a reasonable probability to each uncertainty factor selected in S5.1: Risk Calculation Where L represents the severity of the consequences and P represents the probability of the accident occurring.

[0097] For the piping system, the annual leakage frequency per unit length is set as P. For any given section, the total leakage frequency is P for that section. This frequency is then evenly distributed to all specific risk points (such as valves and flanges) within that section.

[0098] , where P i Let n be the leakage frequency at each risk point, and n be the number of risk points per unit pipeline.

[0099] For other equipment outside the pipeline (such as compressors and inlet / outlet units of storage tanks), the total leakage frequency is also decomposed to each specific valve or connection point based on its structural complexity and the number of sealing points. Thus, each specific risk point in the entire station is assigned an annual leakage frequency that considers the uncertainty of spatial distribution. .

[0100] S5.3 Risk Quantification Assessment Considering Uncertainty: A combination of Bayesian network model and accident consequence model in S4 is adopted.

[0101] The probability of an accident occurring under a specific scenario at a leak point is calculated using a Bayesian network model; risk post-processing software is used to calculate the risk (personal risk and social risk) at each leak point.

[0102] The steps for calculating individual risk are as follows: a) Select a leakage scenario (LOC) and determine the frequency of LOC occurrence. f s ; b) Select a weather class M and a wind direction under that weather class. φ Provide weather level M and wind direction. φ The joint probability of simultaneous occurrence ; c) If it is a release of flammable material, select an ignition event. i And determine the ignition probability P i If the toxicity of the substance is taken into account, then ignition events are not considered; d) Calculate at a specific LOC, weather level M, and wind direction φ and ignition incident i (Combustible material) probability of death on a grid cell P 个体风险 The reference height in the calculation is taken as 1 m; e) Calculate (LOC, M, ...) φ , iContribution to individual grid cell risk under the following conditions; .

[0103] f) For all ignition events, repeat steps c)–e) to obtain the results. For all weather levels and wind directions, repeat steps b)–e) to obtain the results. and For all LOCs, repeat steps a)–e) to obtain... Then the individual risk at the grid point is calculated by the following formula; .

[0104] The steps for calculating social risk are as follows: a) First, determine the following conditions: 1) Determine the LOC and its frequency of occurrence. f s ; 2) Select the weather level M The probability is P M ; 3) Select the weather level M The next wind direction φ The probability is P φ ; 4) For combustible materials, the conditional probability of selection is: P i ignition incident i .

[0105] b) Select a grid cell and determine the number of people within that grid cell. N cell ; c) Calculate at a specific LOC, M , φ and i Below, the percentage of population deaths within a grid cell. P 社会风险 The reference height in the calculation is taken as 1 m.

[0106] d) Calculate at a specific LOC, M, φ and i The number of deaths in the grid cells below ; .

[0107] e) For all grid cells, repeat the calculations in steps b)–d), for LOC , M , φ and iCalculate the total number of deaths ; .

[0108] f) Calculation LOC , M , φ and i joint frequency ; .

[0109] For all LOC ( f S ), M , φ and i Repeat steps a)–f) to calculate the cumulative total number of deaths. Frequency of all accidents Construct the FN curve; .

[0110] Potential loss of life PLL The calculation should be performed using the following formula: ; In the formula: PLL —Potential loss of life; f i --event i The frequency of the results is expressed in per year. N i ——No. i The number of deaths in each incident.

[0111] Table 8. Standard Values ​​of Acceptable Risk for Individuals in my country Table 8. Standard Values ​​of Acceptable Risk for Individuals in my country Figure 11 This is a risk standard map acceptable to Chinese society. For hydrogen refueling stations, we consider them low-density locations and emerging infrastructure; therefore, a more conservative risk assessment should be made, and we selected 10. -5 / year of personal risk is used as an acceptable standard for personal risk.

[0112] S6: Comprehensive Risk Assessment and Visualization Based on Physical Information Neural Networks To efficiently perform large-scale scene calculations in S5.3, a Physics-Informed Neural Network (PINN) is introduced as a proxy model for CFD simulations, such as... Figure 12 As shown.

[0113] S6.1 Proxy Model Training: A limited number of high-fidelity CFD simulation samples from S3 were used as the training set. The input layer of the neural network includes: the coordinates of the leak point location, leak parameters (flow rate, direction), environmental parameters (wind speed, wind direction), and time. The output layer represents the values ​​of the physical fields of interest (such as concentration, temperature, thermal radiation, and overpressure) at spatial grid points. Simplified physical equations (such as the Navier-Stokes equation and partial forms of the component transport equation) controlling hydrogen gas diffusion and combustion are embedded in the loss function as regularization terms, enabling the network to maintain physical plausibility even in sparse data regions and improving generalization ability.

[0114] S6.2 Rapid Risk Calculation Across the Entire Site: The trained PINN proxy model is integrated into the risk assessment framework of step S5. When calculating the leakage probability and consequences at each leak point, time-consuming CFD software is no longer required; instead, the trained neural network instantly (in milliseconds) predicts the corresponding consequence field. This makes it possible to perform rapid and comprehensive risk calculations for tens of thousands of potential leak points across the entire site under various uncertain parameter combinations.

[0115] S6.3 Comprehensive Risk Visualization: Based on a large amount of results obtained from PINN accelerated calculations, the risk contributions of all leakage points and all uncertain scenarios are overlaid to generate a "comprehensive risk contour map" for the entire station. This map can be presented in the form of personal risk contour lines, social risk distribution, or probability cloud maps of specific injury thresholds (such as thermal radiation >12.5kW / m²), intuitively displaying high-risk areas within the hydrogen refueling station and providing precise spatial decision support for optimizing safety distances, setting up protective barriers, and formulating emergency plans.

[0116] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention, and other modifications can be easily implemented by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A risk quantification assessment method considering the uncertainty of leakage points in the operation of hydrogen refueling stations, characterized in that, Includes the following steps: S1. Based on the collected general layout plan, process flow diagram, equipment data and environmental information of the hydrogen refueling station, construct the geometric model of the hydrogen refueling station required for evaluation, and clarify the physical parameters of hydrogen in each piece of equipment; and systematically identify potential hydrogen leakage points in all equipment bodies and pipeline systems within the hydrogen refueling station. S2. Based on historical data, fault tree and event tree analysis results, construct a Bayesian network to describe the evolution of "leakage-diffusion-ignition-consequence", assign conditional probabilities to each node in the network, and deduce the probability of different leak points causing various accident scenarios under different conditions. S3. For the representative leak points identified in step S1 and various meteorological conditions, use computational fluid dynamics (CFD) software to perform transient simulation of hydrogen leakage and diffusion, obtain the spatiotemporal distribution of hydrogen concentration and key physical field data of the accident, and combine it with the established personnel exposure model to provide a data foundation for subsequent accident consequence analysis and model training. S4. Based on the CFD simulation results of step S3, quantitative analysis is performed on the thermal radiation field of jet fire accident, the combustible cloud range of flash fire accident and the shock wave overpressure field of vapor cloud explosion accident. Combined with the corresponding injury criteria, the injury range of various accidents to personnel and equipment is assessed. S5. The system identifies and selects key uncertainty factors that affect the risk assessment results, assigns leakage probabilities to each factor, and uses the Bayesian network model in S2 and various accident injury consequence models in S4 to calculate the probability of accident occurrence under specific scenarios at the leakage point and the probability distribution of personal and social risks at each leakage point, thereby quantifying the uncertainty of risk and forming a risk assessment framework. S6. Using the limited amount of CFD simulation data from step S3, train the physical information neural network PINN to construct a fast proxy prediction model for the accident consequence field; integrate this proxy model with the risk assessment framework from step S5 to quickly calculate and overlay the risks of all potential leakage points in the entire site under multiple uncertainties, generate a comprehensive risk heat map and present it visually.

2. The risk quantification assessment method considering the uncertainty of leakage points in the operation of hydrogen refueling stations as described in claim 1, characterized in that, Step S1 specifically includes: S1.1 Based on the general layout plan, process flow diagram, equipment data and surrounding environmental information of the hydrogen refueling station, construct a geometric model including the long tube trailer, hydrogen storage cylinder group, compressor, hydrogen dispenser, unloading column, sequence control panel, hydrogen storage valve group and process pipeline, and clarify the physical parameters of hydrogen in each device, including storage form, working pressure and working temperature, as boundary conditions for leakage source analysis; S1.2 Systematically identify hazards in the constructed geometric model, including specific potential leak points such as welds of hydrogen storage equipment, valves, compressor shaft seals, flanges, hydrogen refueling machine hoses, interfaces, and all process pipeline connection points.

3. The risk quantification assessment method considering the uncertainty of leakage points in hydrogen refueling station operation as described in claim 1, characterized in that, Step S2 specifically includes: S2.1 Define the initial event as the leakage of a specific potential hydrogen leak point identified in step S1, intermediate events include leak mode, ventilation effect, detection system response, and ignition source appearance, and the terminal event is a fire or explosion accident at a hydrogen refueling station. S2.2 Construct a Bayesian network topology that connects the above events, and assign conditional probability tables to each node in the network based on historical data and expert knowledge; S2.3 Input specific leakage frequency and environmental conditions, and dynamically infer and output the conditional probability of occurrence of various terminal accident scenarios through Bayesian network inference.

4. The risk quantification assessment method considering the uncertainty of leakage points in hydrogen refueling station operation as described in claim 1, characterized in that, Step S3 specifically includes: S3.

1. Combining the physical parameters and geometric model of S1, set leakage boundary conditions for each representative potential hydrogen leakage point; collect meteorological conditions of the hydrogen refueling station location, select representative combinations as calculation conditions, and assign annual occurrence frequency to each condition. S3.2 Obtain the distribution patterns of various personnel within the hydrogen refueling station and establish a personnel exposure model; S3.3 Perform transient simulation on the set leakage scenario, and output the following: spatiotemporal distribution cloud map of hydrogen concentration and key physical field data of the accident; The above data provides a data foundation for subsequent risk probability calculations and proxy prediction model training.

5. The risk quantification assessment method considering the uncertainty of leakage points in the operation of hydrogen refueling stations as described in claim 1, characterized in that, Step S5 specifically includes: S5.1 Identify and select key uncertainty factors, including: uncertainty of leakage frequency, uncertainty of leakage scenario, uncertainty of accident evolution path, uncertainty of environmental parameters and uncertainty of simulation model parameters; S5.

2. Assign reasonable leakage probabilities to each key uncertainty factor; S5.

3. The Bayesian network model of S2 is combined with the various accident injury consequence models of S4. The Bayesian network model is used to calculate the probability of accident occurrence in a specific scenario at the leak point. Risk post-processing software is used to calculate the probability distribution of personal risk and social risk at each leak point.

6. The risk quantification assessment method considering the uncertainty of leakage points in the operation of hydrogen refueling stations as described in claim 1, characterized in that, Step S6 specifically includes: S6.

1. Using the leak point parameters, leak parameters, environmental parameters and time as inputs, and the spatial physical field data output by CFD simulation as labels, train the physical information neural network, and embed simplified fluid control equations as physical constraints in the loss function. S6.2 Integrate the trained physical information neural network proxy model into the risk assessment framework for direct simulation; S6.

3. Based on the agent model, the rapid prediction results of the whole scenario are superimposed to calculate the comprehensive risk of the whole site, and visualized in the form of personal risk contour lines, social risk curves or probability cloud maps of specific injury thresholds.

7. A method for quantitatively evaluating the operational risks of hydrogen refueling stations considering uncertainties, as described in any one of claims 1-6, characterized in that, In step S4, jet fire analysis assesses the damage range using thermal radiation flux and damage models; flash fire analysis assesses the damage range by identifying clouds with hydrogen concentrations within the flammability limit obtained from CFD simulations; and vapor cloud explosion analysis assesses the damage range using shock wave overpressure and corresponding damage models.

Images