Simulation deduction and auxiliary decision system for multi-point concurrent leakage accident of hydrogen refueling station

Abstract

The present application relates to the technical field of fire accident monitoring and early warning and fire rescue, in particular to a simulation deduction and auxiliary decision system for multi-point concurrent leakage accidents of hydrogen stations, which comprises an accident scene dynamic risk mapping module, which is used for receiving and gridding a rescue area, collecting real-time monitored hydrogen concentration, heat radiation flux and explosion overpressure values for each grid. In the present application, by receiving and gridding the rescue area, collecting real-time monitored hydrogen concentration, heat radiation flux and explosion overpressure values, and combining the path interruption probability caused by hydrogen storage and hydrogenation structure collapse, a unified grid risk value is obtained, and a time-varying three-dimensional risk field data covering the entire accident scene is established, which converts discrete and multidimensional on-site monitoring data into an intuitive global danger situation picture, enabling fire rescue commanders to master the evolution law of risk in space and time.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fire accident monitoring and early warning and fire rescue technology, and particularly relates to a simulation deduction and auxiliary decision system for multi-point concurrent leakage accidents of hydrogen refueling stations. BACKGROUND

[0002] The goal of the field of fire accident monitoring and early warning and fire rescue technology is to monitor potential hazards in real time and continuously in industrial production, urban operation or natural environment.

[0003] In actual operation, the existing fire accident monitoring and early warning and fire rescue technology mainly relies on independent monitoring and threshold alarm of various physical parameters. This mode leads to limitations in recognizing complex accident scenes with concurrent coupling of multiple disasters. The temperature, pressure, gas concentration and other data collected by various sensors are often treated as isolated information, so that the early warning information remains at the point, and it is difficult for commanders to intuitively judge the real boundary of the dangerous area, the evolution direction and the superposition effect between different dangerous factors. For example, in a chemical leakage accident, it may report that the toxic gas in a certain area exceeds the standard and the temperature of the pressure container in another area rises, but it is difficult to prospectively reveal whether the diffusion path of the gas cloud will meet the heat source, thereby causing a more serious secondary explosion disaster, resulting in that the emergency decision seriously depends on the personal experience of the fire commanders. Therefore, improvement is needed. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art, and a simulation deduction and auxiliary decision system for multi-point concurrent leakage accidents of hydrogen refueling stations is proposed.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: the simulation deduction and auxiliary decision system for multi-point concurrent leakage accidents of hydrogen refueling stations comprises:

[0006] An accident scene dynamic risk mapping module is used to receive and grid the rescue area. For each grid, the hydrogen concentration, heat radiation flux and explosion overpressure values monitored in real time are collected, and the path interruption probability caused by the collapse of hydrogen storage and hydrogenation structure is combined to calculate the grid risk value, and a time-varying three-dimensional risk field data is established;

[0007] A rescue personnel space-time path planning module is used to calculate the metabolic equivalent according to the time-varying three-dimensional risk field data, combined with the weight equipment, moving speed and environmental temperature of the rescue personnel, to quantify the physical consumption rate, and then accumulate the dangerous dose and total travel time of the path, generate a multi-dimensional path comprehensive cost sequence, search for the path in the multi-dimensional path comprehensive cost sequence, and obtain a four-dimensional space-time safety corridor.

[0008] The command decision foresight deduction module is configured to input the decision options of the fire-fighting commanders, deduce the accident evolution process within a preset time period, obtain a multi-dimensional accident evolution increment set, extract a cascade failure probability increment, a dangerous area area increment, a safe evacuation window shortening time, and a resource consumption rate increment from the multi-dimensional accident evolution increment set, and generate a decision option multi-dimensional influence vector according to the time-varying three-dimensional risk field data.

[0009] The decision option threat quantification module is configured to normalize the cascade failure probability increment, the dangerous area area increment, the safe evacuation window shortening time, and the resource consumption rate increment according to the decision option multi-dimensional influence vector, obtain a normalized threat index matrix, and calculate each index in the normalized threat index matrix to generate a command decision option threat index according to a preset command principle weight value.

[0010] Preferably, the step of obtaining the time-varying three-dimensional risk field data comprises:

[0011] The rescue area boundary is received and gridded, the hydrogen concentration numerical value, the thermal radiation flux numerical value, the explosion overpressure numerical value, and the path interruption probability are synchronized in time sequence, the hazard probability is converted into a hazard probability according to a hazard threshold curve and time stamp alignment is completed, and the hydrogen hazard probability sequence, the thermal radiation hazard probability sequence, the explosion hazard probability sequence, and the path interruption probability sequence are obtained;

[0012] According to the hydrogen hazard probability sequence, the thermal radiation hazard probability sequence, the explosion hazard probability sequence, and the path interruption probability sequence, the risk value is calculated to obtain a grid risk numerical value sequence;

[0013] According to the grid risk numerical value sequence, each risk value is mapped to a corresponding three-dimensional voxel according to the time index, and the risk values of adjacent time slices are connected, the spatial resolution and the time step are unified, and the time-varying three-dimensional risk field data is formed.

[0014] Preferably, the step of obtaining the multi-dimensional path synthesis cost sequence comprises:

[0015] According to the time-varying three-dimensional risk field data, the hydrogen hazard probability time sequence, the thermal radiation hazard probability time sequence, and the explosion hazard probability time sequence are read voxel by voxel, the temperature field data corresponding to each grid is extracted, the metabolic equivalent per unit time is calculated in combination with the load-carrying equipment parameters and the moving speed parameters of the rescue personnel, and the metabolic equivalent per unit time is accumulated according to the time step to obtain the physical energy consumption rate result of each time period.

[0016] Based on the energy consumption rate results, the risk value of each path node at the corresponding time step is read sequentially along the candidate path. The dangerous dose encountered at each node on the path is calculated. The dangerous doses of each node are accumulated in the order of the path, and the corresponding travel time is calculated at the same time. The dangerous dose, energy consumption rate and travel time are combined to generate a multi-dimensional path comprehensive cost sequence.

[0017] Preferably, the steps for obtaining the four-dimensional spatiotemporal safety corridor are as follows:

[0018] Based on the multidimensional path comprehensive cost sequence, a cumulative risk exposure threshold and a physical exertion rate threshold are set. Paths that simultaneously meet both threshold conditions are selected, and the travel time is used as the sorting criterion. The path with the shortest travel time is selected and connected sequentially according to spatial coordinates and time indices to form a continuous time series corridor, thus obtaining a four-dimensional spatiotemporal safety corridor.

[0019] Preferably, the steps for obtaining the multidimensional accident evolution increment set are as follows:

[0020] Based on the time-varying three-dimensional risk field data, a preset time period is defined and a unified timestamp sequence is established. The decision options of firefighters are analyzed as action type, location of action, and start and end time. The control quantity is modified at the corresponding location step by step, and the compressor status identifier and hydrogen dispenser status identifier are updated. The status transition timestamp is recorded. At the same time, a decision-free reference record is generated. The equipment cascade failure probability, dangerous area boundary and area, safe evacuation window start and end time, and resource consumption rate are written with timestamps to obtain a decision inference record set.

[0021] Based on the decision-making simulation record set, the decision-making records are aligned to the non-decision-making reference records according to the timestamp and location index. The differences in the probability of cascading equipment failure, the differences in the area of ​​dangerous areas, the differences in the time of shortening of the safety evacuation window, and the differences in the rate of resource consumption are calculated step by step. The results are arranged in chronological order and the source identifier is retained to generate a multi-dimensional accident evolution increment set.

[0022] Preferably, the steps for obtaining the multidimensional influence vector of the decision option are as follows:

[0023] Based on the multidimensional accident evolution increment set, the equipment cascading failure probability increment, dangerous area area increment, safe evacuation window shortening time and resource consumption rate increment are taken from the time stamp at the end of the preset time period and combined into four-dimensional entries in a fixed order to generate a multidimensional influence vector for decision options.

[0024] Preferably, the steps for obtaining the normalized threat index matrix are as follows:

[0025] Based on the multidimensional influence vector of the decision options, the increments of cascade failure probability, dangerous area area, safe evacuation window shortening time and resource consumption rate are extracted in sequence and arranged into column vectors according to the decision options. The maximum and minimum values ​​of each indicator are recorded to obtain the original indicator matrix and extreme value table.

[0026] Based on the original indicator matrix and extreme value table, the maximum and minimum values ​​of each indicator are normalized to unify the value range of the four indicators to the interval between 0 and 1. A normalized threat indicator matrix is ​​constructed according to the indicator number and decision option index, and the preset command principle weight values ​​are read to obtain the weight vector.

[0027] Preferably, the step of obtaining the threat index of the command and decision options is as follows:

[0028] The threat index of command and decision options is calculated based on the normalized threat index matrix and the weight vector.

[0029] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0030] In this invention, by receiving and gridding the rescue area, combining real-time monitored hydrogen concentration, thermal radiation flux, and explosion overpressure values, and considering the probability of path interruption caused by the collapse of hydrogen storage and refueling structures, a unified grid risk value is calculated. This establishes a time-varying three-dimensional risk field data covering the entire accident site, transforming discrete, multi-dimensional on-site monitoring data into an intuitive global hazard situation map. This allows commanders to grasp the spatial and temporal evolution of risks. Based on this dynamic risk field, the physical exertion of rescue personnel is further calculated by considering their load, speed, and ambient temperature, and the amount of hazardous agents along the path is accumulated. By measuring quantity and travel time, a four-dimensional spatiotemporal safety corridor can be searched and generated. This not only provides clear and feasible path guidance for rescue operations, but also improves the survival probability and mission success rate of rescuers by quantifying physical strength and risk exposure. At the same time, by using the decision options of firefighters as input conditions for simulation, it can proactively track changes in key indicators such as the probability of cascading equipment failure and the area of ​​dangerous areas caused by decisions, and generate multi-dimensional impact vectors of decision options. Furthermore, it can condense the consequences of decisions into scalars, enabling scientific weighing among various emergency plans and selecting the disposal strategy with the least overall harm and the highest benefit. Attached Figure Description

[0031] Figure 1 This is a system flowchart of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0033] Please see Figure 1 This invention provides a technical solution: a simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations, comprising:

[0034] The accident site dynamic risk mapping module is used to receive and grid the rescue area. For each grid, it combines the real-time monitored hydrogen concentration, thermal radiation flux and explosion overpressure values, and the probability of path interruption caused by the collapse of hydrogen storage and hydrogen refueling structures to calculate the grid risk value and establish time-varying three-dimensional risk field data.

[0035] The spatiotemporal path planning module for rescuers is used to calculate metabolic equivalents based on time-varying three-dimensional risk field data, combined with the rescuers' load-bearing equipment, movement speed and ambient temperature, to quantify the rate of physical energy consumption, and then accumulate the dangerous doses encountered along the path and the total travel time to generate a multi-dimensional path comprehensive cost sequence. The module searches for paths in the multi-dimensional path comprehensive cost sequence to obtain a four-dimensional spatiotemporal safety corridor.

[0036] The command and decision-making forward simulation module is used to simulate the accident evolution process within a preset time period based on time-varying three-dimensional risk field data and the decision options of firefighters as input conditions. It obtains a multi-dimensional accident evolution increment set, extracts the cascading failure probability increment, dangerous area area increment, safe evacuation window shortening time and resource consumption rate increment from the multi-dimensional accident evolution increment set, and generates a multi-dimensional influence vector of decision options.

[0037] The decision option threat quantification module is used to normalize the increments of cascading failure probability, dangerous area area, safe evacuation window shortening time, and resource consumption rate based on the multi-dimensional influence vector of decision options, and obtain a normalized threat index matrix. It then calls the preset command principle weight values ​​to calculate the threat index of each indicator in the normalized threat index matrix.

[0038] The steps for obtaining time-varying three-dimensional risk field data are as follows:

[0039] The rescue area boundary is received and gridded. The hydrogen concentration value, thermal radiation flux value, explosion overpressure value and path interruption probability are synchronized according to the time series. Based on the hazard threshold curve, the table is looked up grid by grid to convert into hazard probability and the timestamp is aligned to obtain the hydrogen hazard probability sequence, thermal radiation hazard probability sequence, explosion hazard probability sequence and path interruption probability sequence.

[0040] Based on the probability sequences of hydrogen hazard, thermal radiation hazard, explosion hazard, and path interruption, risk values ​​are calculated to obtain a grid risk numerical sequence. The calculation formula is as follows:

[0041] ;

[0042] in, For time Time Risk value of each grid For time indexing, For grid indexing, For time Time The first grid The probability of direct harm and The corresponding probability of harm caused by hydrogen gas. Corresponding probability of harm caused by thermal radiation. Corresponding probability of damage from an explosion. For time Time The probability of a grid path being interrupted. The direct harmful coupling smoothing parameter. The cascading damage amplification index;

[0043] Based on the grid risk value sequence, each risk value is mapped to the corresponding three-dimensional voxel by time index and the risk values ​​of adjacent time slices are connected to unify the spatial resolution and time step, forming time-varying three-dimensional risk field data.

[0044] Specifically, after receiving the geographical boundary information of the rescue area, the system first performs three-dimensional mesh partitioning, dividing the entire rescue area into multiple uniformly sized cubic meshes. For example, a 100m x 100m x 20m area is divided into 1 million 1m x 1m x 1m cubic meshes, and each mesh cell is assigned a unique three-dimensional spatial index. Next, it connects to various sensor networks deployed within the hydrogen refueling station, continuously collecting and synchronizing real-time environmental parameters at the center point of each mesh at preset 1-second intervals. This includes hydrogen concentration values ​​(volume percentage) obtained using electrochemical or thermal conductivity hydrogen sensors, and infrared thermal imaging... The data includes thermal radiation flux values ​​obtained from instruments or heat flow meters, measured in kilowatts per square meter (kW / m²), and explosion overpressure values ​​obtained from pressure sensors, measured in kilopascals (kPa). Simultaneously, combining the building structure model of the hydrogen refueling station with real-time monitoring of hydrogen storage tank pressure and temperature data, the probability of the hydrogen storage tank structure collapsing under its current state and causing disruption to surrounding paths is calculated using finite element analysis, yielding the path disruption probability. All collected data includes precise timestamps. Then, based on a pre-established hazard threshold database, the real-time monitoring data for each grid is transformed through a grid-by-grid lookup table. This database is based on the Probit model; for example, the transformation relationship for the hydrogen hazard probability is defined as... ,in It's the hydrogen concentration. It is the exposure time. , , These coefficients are determined based on toxicological experimental data. By substituting the real-time hydrogen concentration and combining it with the exposure time (starting from the leak), the probability of harm to personnel due to hypoxia or poisoning within the grid is calculated. Similarly, the probability of harm due to thermal radiation is calculated by looking up a table based on the thermal radiation flux and exposure time, the probability of harm due to explosion is calculated based on the peak overpressure of the explosion, and the probability of path interruption is directly obtained from the calculation results. All converted probability values ​​are aligned with the timestamp of the original data, ultimately generating a hydrogen harm probability sequence, a thermal radiation harm probability sequence, an explosion harm probability sequence, and a path interruption probability sequence that correspond one-to-one with the grid index and timestamp.

[0045] The LogSumExp function is used in the formula for calculating the risk value. The probability of direct harm from hydrogen, thermal radiation, and explosion is coupled using a smooth approximation of a maximum value function. Compared to simply taking the maximum value or summing probabilities, this approach highlights the most significant risk source while also considering the contributions of other risk sources, resulting in a more comprehensive and robust risk assessment. Furthermore, direct harm risk is non-linearly coupled with cascading harm risk (path disruption caused by an explosion), using a cascading harm amplification index. To characterize the exacerbating effect of explosions on structural damage, the model is more closely aligned with the physical reality of accident evolution, enabling it to capture the complex interactions between different disaster factors.

[0046] This is a time index, representing a discrete time step. For example, if the time step is 1 second... This represents the 10th second after the incident. This parameter is provided by the system's master clock and is used to synchronize all time-series data to ensure the dynamism and timeliness of risk calculation.

[0047] For mesh indexes, a unique identifier is used for each cubic mesh in 3D space, such as in a 100x100x20 meshed space. It can be a tuple containing (x, y, z) coordinates, such as This index is generated from the initial rescue area gridding step and is used to locate the calculated risk value to a specific physical location.

[0048] For time Time The first grid The probability of direct harm caused by a class of organisms is a parameter obtained in the previous step by consulting a database of harm threshold curves. The corresponding probability of harm caused by hydrogen gas. Corresponding probability of harm caused by thermal radiation. Corresponding to the probability of harm caused by the explosion, these probability values ​​are all dimensionless values, ranging from 0 to 1. They are derived from the conversion calculation between sensor monitoring data and the established injury model, and serve as the input basis for assessing the risk of direct harm to personnel.

[0049] For time Time The probability of path interruption for each grid, which is also provided in the previous step, is dynamically calculated by performing structural finite element analysis on key facilities such as hydrogen storage tanks, combined with real-time monitoring of pressure and temperature data. It reflects the possibility that rescue or evacuation routes will be blocked due to equipment collapse and is a dimensionless value between 0 and 1.

[0050] This is a direct causative coupling smoothing parameter. It adjusts how closely the LogSumExp function approximates the maximum value function, controlling the coupling patterns of different direct causative factors. The setup is based on a conservative risk assessment strategy. The setup process is as follows: First, define a risk discrimination index. , ,in It is a set of three probabilities of harm. LSE is the result of LogSumExp calculation, which measures the extent to which the coupled risk exceeds the maximum risk. Then, a target discrimination index is set. For example, setting it to 5% means that we want the coupling risk to be no more than 5% higher than the maximum risk, by solving the equation. To calculate in reverse The value of can be determined in practical applications by establishing a . A table showing the correspondence between values ​​and typical probability distributions for risk scenarios allows for quick lookup and setting. For example, for a given scenario, its typical probability distribution of harm is... If we want to differentiate risk levels Keep it below 5%, that is Calculation yields The value must be greater than or equal to 20; therefore, in this calculation, we take... .

[0051] The cascade damage amplification index is a parameter used to quantify the nonlinear amplification effect of an explosion on the probability of path disruption; its value is greater than 1. The setup needs to be based on the structural vulnerability assessment results of the buildings within the hydrogen refueling station. The assessment process is as follows: First, the structural engineer assesses the vulnerability of each critical structural component (such as load-bearing walls and columns) based on the building's design drawings, building materials (such as reinforced concrete and steel structures), and blast resistance rating. The rating ranges from 1 to 10, with higher scores indicating greater vulnerability. Then, the distance between the structural component and a potential explosion source is considered. Calculate a distance influence factor Finally, the weighted average fragility of all structural components is used to calculate... The calculation formula is: ,in It is a structural component index. This refers to the total number of components. For example, if a support pillar next to the critical path of a hydrogen refueling station has a vulnerability score of 8 and is 10 meters away from an explosion source, then its distance impact factor is... If this is the only key structure in the region, then Take this place .

[0052] Calculations based on parameters:

[0053] In a specific scenario, set a time At that time, the grid The values ​​of each parameter are as follows:

[0054] Probability of harm from hydrogen .

[0055] Probability of harm from heat radiation .

[0056] Probability of damage from an explosion .

[0057] Path interruption probability .

[0058] Directly harmful coupling smoothing parameters .

[0059] Cascade Damage Amplification Index .

[0060] First, calculate the coupling probability component that directly causes harm, and let it be... :

[0061] ;

[0062] ;

[0063] Next, the non-disruptive probability of cascading harm is calculated, and let it be... :

[0064] ;

[0065] ;

[0066] Finally, calculate the total risk value. :

[0067] ;

[0068] ;

[0069] ;

[0070] This result indicates that at the current time , No. The overall risk value of each grid is 0.2085. 0.2085 is in the low-risk range (0-0.3 is low risk, 0.3-0.7 is medium risk, and above 0.7 is high risk). This means that when rescuers stay or pass through this area, the probability of being injured or trapped is relatively low, but it is not absolutely safe. This grid risk value is a data point that constitutes the grid risk value sequence. After calculating the risk values ​​of all grids at all time steps, the complete grid risk value sequence can be obtained.

[0071] Based on the calculated sequence of grid risk values, which contains the risk value of each two-dimensional grid at consecutive time points, a spatial mapping operation is first performed to map the risk values ​​of each two-dimensional grid. A three-dimensional voxel is assigned to each point. Specifically, a two-dimensional grid is used as the base of the voxel, and the risk value is extended vertically (Z-axis) upwards to a preset rescuer activity height, such as 2 meters, forming a risk voxel with volume. All points within this voxel are assigned the same risk value. This expands the original two-dimensional risk map into a three-dimensional risk layer with thickness. Next, the time dimension data is processed, separating adjacent time slices (e.g., time...) and The risk voxel data are connected to form a continuous risk field data stream on the time axis, constituting a four-dimensional dataset (X, Y, Z, T). To ensure data consistency and ease of subsequent processing, the spatiotemporal resolution of the entire dataset is unified. The spatial resolution is set to 1 meter by 1 meter by 1 meter, which means that each voxel is a 1-cubic-meter cube. If the original grid resolution is different, the risk values ​​are resampled using trilinear interpolation to match the target resolution. The time step is unified to 1 second. If the time step of the original data is inconsistent, new risk data points are inserted on the time axis using linear interpolation to ensure that there is a complete risk field snapshot every second. Through the above processing, a dynamic, continuous, time-varying three-dimensional risk field data consisting of a large number of voxels with risk attributes is finally formed in a unified spatiotemporal coordinate system.

[0072] The steps to obtain the multidimensional path comprehensive cost sequence are as follows:

[0073] Based on the time-varying three-dimensional risk field data, the time series of hydrogen hazard probability, thermal radiation hazard probability, and explosion hazard probability were read voxel by voxel. The temperature field data corresponding to each grid was extracted. Combined with the parameters of the rescuers' load-bearing equipment and movement speed, the metabolic equivalent per unit time was calculated. Then, the metabolic equivalent was accumulated according to the time step to obtain the results of the energy consumption rate for each time period.

[0074] Based on the results of the energy consumption rate, the risk value of each path node at the corresponding time step is read sequentially along the candidate path. The dangerous dose encountered at each node on the path is calculated, and the dangerous doses of each node are accumulated in the order of the path. At the same time, the corresponding travel time is calculated. The dangerous dose, energy consumption rate and travel time are combined to generate a multi-dimensional path comprehensive cost sequence.

[0075] Specifically, based on time-varying three-dimensional risk field data, voxels in each candidate path are traversed. At each time step, not only are the time series of hydrogen hazard probability, thermal radiation hazard probability, and explosion hazard probability read, but also the environmental temperature field data precisely corresponding to the voxel, generated by thermal sensor arrays or fluid dynamics simulations, is extracted simultaneously. Then, preset physiological and equipment parameters of the rescue personnel are retrieved, including the rescuer's weight (e.g., 80 kg) and the total weight of their load-bearing equipment. This total weight is obtained by adding the weights of each individual piece of equipment, such as a respirator (15 kg), fireproof suit (8 kg), and tool bag (12 kg), totaling 35 kg. Simultaneously, their movement speed is set, for example, 1 meter per second in a flat, unobstructed area. Then, the metabolic equivalent (MET) per unit time (1 second) is calculated using the following formula: ,in, The load capacity is 35 kg. It weighs 80 kilograms. The movement speed is 1 meter per second. This refers to the terrain slope, which is 0 here. This is the ambient temperature correction factor, which is set so that when the ambient temperature is below 25 degrees Celsius... When the temperature is above 25 degrees Celsius, ,in It is the ambient temperature extracted from temperature field data. For example, if the voxel temperature is 40 degrees Celsius at a certain moment, then... Substitute these parameters into the formula to calculate the metabolic equivalent per second. Finally, based on a 1-second time step, convert the metabolic equivalent value calculated at each time step into a specific energy consumption rate, and the conversion formula is the energy consumption rate (kcal per minute). The calculation results of all time steps along this path are arranged in chronological order to obtain the energy consumption rate results for each time period.

[0076] Based on the energy consumption rate results obtained in the previous step for each time period, and the pre-planned multiple candidate paths, the comprehensive cost of each path is calculated. For any candidate path, the path consists of a series of nodes (voxels) arranged in chronological order. First, starting from the first node of the path, the entry time of that node is read (e.g., ...). The grid risk value (in seconds) is obtained directly from the time-varying 3D risk field data, for example, 0.25. Then, the hazardous dose that rescuers would be exposed to during a time step (1 second) at that node is calculated. The calculation method is that the hazardous dose equals the risk value multiplied by the time of stay, i.e. Next, move to the next node in the path, at which point the time is updated. Every second, repeatedly read the risk value (e.g., 0.3) of the new node and calculate its dangerous dose. The current node's hazard dose is added to the hazard doses of all previous nodes to obtain the cumulative hazard dose reached by the path. Meanwhile, the total travel time of the path is recorded synchronously. For each node passed, the total travel time increases by one time step (1 second). This process is carried out sequentially along all nodes of the path until the end of the path. Finally, the total cumulative dangerous dose and total travel time of the path are obtained. Finally, the total cumulative dangerous dose, total travel time of the path, and the complete physical exertion rate time series obtained from the input data are combined into a data entry. The above calculation and processing process is repeated for all candidate paths to generate a multi-dimensional path comprehensive cost sequence containing cost information of all candidate paths.

[0077] The steps to obtain the four-dimensional spacetime safety corridor are as follows:

[0078] Based on the multidimensional path comprehensive cost sequence, a cumulative risk exposure threshold and a physical exertion rate threshold are set. Paths that simultaneously meet both threshold conditions are screened, and the travel time is used as the sorting criterion. The path with the shortest travel time is selected and connected sequentially according to spatial coordinates and time indices to form a continuous time series corridor, thus obtaining a four-dimensional spatiotemporal safety corridor.

[0079] Specifically, based on the multidimensional path comprehensive cost sequence containing all path cost information, paths are screened and optimized. First, a cumulative risk exposure threshold and a physical exertion rate threshold are set. The cumulative risk exposure threshold is set with reference to the prescribed exposure limit and adjusted according to the urgency of the task. For example, for routine inspection tasks, a conservative threshold based on long-term health effects is set. This threshold is calculated by referencing the probability of harm corresponding to the short-term exposure limit (STEL) of a low-hazard chemical (such as ammonia), converting it into a dimensionless cumulative risk dose, for example, set to 15. For emergency rescue tasks, this threshold can be appropriately relaxed to 30. The physical exertion rate threshold is set based on research on maximum oxygen uptake (VO2max) in exercise physiology. For well-trained firefighters, their sustainable maximum working intensity is approximately 80% of their VO2max, which corresponds to a specific metabolic equivalent value. For example, if a firefighter's VO2max is 50 ml / kg / min, the metabolic equivalent corresponding to their sustainable working threshold is... Therefore, the physical exertion rate threshold is set to 11.4 METs. After setting the threshold, each path in the multidimensional path comprehensive cost sequence is traversed to check whether its total cumulative dangerous dose is less than or equal to the cumulative risk exposure threshold (e.g., 15) and whether the maximum value in its physical exertion rate sequence is less than or equal to the physical exertion rate threshold (11.4 METs). Only paths that meet both conditions are determined to be valid paths and are retained. Then, all valid paths that pass the screening are sorted in ascending order based solely on their total travel time. The path ranked first in the list, i.e., the path with the shortest total travel time, is selected as the final optimal path. Finally, the spatial coordinate sequence contained in this optimal path is sequentially connected with the corresponding time index sequence to form a continuous set of path points in four-dimensional spacetime from the starting point to the end point, thus obtaining the four-dimensional spacetime safety corridor.

[0080] The steps for obtaining the multidimensional accident evolution increment set are as follows:

[0081] Based on time-varying three-dimensional risk field data, a preset time period is defined and a unified timestamp sequence is established. The decision options of firefighters are analyzed as action type, location of action, and start and end time. The control variables are modified at the corresponding locations step by step, and the compressor status identifier and hydrogen dispenser status identifier are updated. The status transition timestamp is recorded. At the same time, a decision-free reference record is generated. The equipment cascade failure probability, dangerous area boundary and area, safe evacuation window start and end time, and resource consumption rate are written with timestamps to obtain a decision inference record set.

[0082] Based on the decision-making simulation record set, the decision-making records are aligned to the non-decision-making reference records according to the timestamp and location index. The differences in the probability of cascading equipment failure, the differences in the area of ​​dangerous areas, the differences in the time of shortening of the safety evacuation window, and the differences in the rate of resource consumption are calculated step by step. The results are arranged in chronological order and the source identifier is retained to generate a multi-dimensional accident evolution increment set.

[0083] Specifically, based on time-varying three-dimensional risk field data, a preset projection time period is first defined, such as the next 300 seconds from the current moment, and a unified timestamp sequence is established at 1-second intervals. Next, the decision options input by firefighters are analyzed and decomposed into three core elements: action type (e.g., "close valve," "start sprinkler," or "cut off power"), and location of action (i.e., the coordinates of the action in the three-dimensional model of the hydrogen refueling station, such as the location of valve V-101 (x1, ...). The system sets the time stamps for y1, z1, and the start and end times, i.e., the timestamps of the start and end of the action, such as from the 60th to the 65th second. Then, it starts the simulation engine and simulates the execution process of the decision step by step (1 second). During the start and end time of the decision, it modifies the corresponding physical model control variables at the specified action position. For example, for "closing the valve", it modifies the flow coefficient of the corresponding pipe in the fluid dynamics model to 0. For "starting the spray", it adds a negative heat source term and a water evaporation term in the corresponding area. These modifications will dynamically affect the accident evolution model, thereby changing the calculation results of subsequent parameters such as hydrogen concentration and thermal radiation. At the same time, according to the set equipment status logic, it updates the status flags of related equipment. For example, after closing valve V-101, the status flag of the compressor C-01 associated with it is updated from "running". The system is set to "standby" and the timestamp of this state transition is recorded. While conducting decision-making simulations, a baseline simulation without decision-making intervention is launched in parallel, representing the natural development of the accident without any human intervention. Both simulations simultaneously record a series of key performance indicators, including the probability of cascading equipment failure calculated using a fault tree analysis model, the area of ​​the danger zone obtained by identifying the boundaries of areas where the risk value exceeds a preset safety threshold (e.g., 0.5) and calculating the area enclosed by them, the start and end times of the safe evacuation window calculated based on the predicted time point when the safe evacuation route will be completely blocked according to the danger zone expansion rate, and the consumption rate of rescue resources such as water, electricity, and foam called up during the simulation. All these records are marked with precise timestamps and are finally compiled into a decision-making simulation record set.

[0084] Based on the decision-making simulation record set generated in the previous step, data alignment and difference calculation are performed. First, using timestamps and spatial location indices as dual keys, simulation records with "decision-making" data are matched and aligned with those without "decision-making reference" data. This ensures that there are two sets of data for comparison at each time step and each spatial location. Next, after alignment, difference calculation is performed step-by-step. For the probability of cascading failure, the probability value in the "decision-making" record is subtracted from the corresponding probability value in the "no-decision-making" record to obtain the difference in cascading failure probability. A positive difference indicates that the decision increased the risk of cascading failure, while a negative difference indicates a decrease in risk. Similarly, the areas of the hazardous areas are subtracted to obtain the difference in hazardous area area. A positive difference indicates that the decision led to an expansion of the hazardous area, while a negative difference indicates a decrease in risk. This indicates that the danger zone has been effectively controlled or reduced. For the safe evacuation window, the difference in its shortening time is calculated. Specifically, the window closing time in the "no decision reference" record is subtracted from the window closing time in the "with decision" record to obtain the difference in the shortening time of the safe evacuation window. A positive value indicates that the decision-making behavior shortened the evacuation time, increasing the risk of personnel evacuation. Finally, the resource consumption rate is subtracted to obtain the resource consumption rate difference. A positive difference indicates that the decision requires more resources, while a negative difference indicates that resources are saved. All calculated differences are arranged in chronological order and their source identifiers are retained, i.e., which specific decision option generated the difference. All difference data generated by all decision options throughout the entire extrapolation period are summarized to generate a multidimensional accident evolution increment set.

[0085] The steps to obtain the multidimensional influence vector of decision options are as follows:

[0086] Based on the multidimensional accident evolution increment set, the increments of the probability of cascading failure of equipment, the area of ​​dangerous areas, the time of shortening of the safe evacuation window and the rate of resource consumption at the end of the preset time period are taken and combined into four-dimensional items in a fixed order to generate a multidimensional influence vector of decision options.

[0087] Specifically, based on the multidimensional accident evolution increment set, four key increment values ​​corresponding to the end timestamp (e.g., the 300th second) of a preset time period are extracted from the dataset for each decision option. These four values ​​are the increment of equipment cascade failure probability, the increment of hazardous area area, the increment of safe evacuation window shortening time, and the increment of resource consumption rate. The end timestamp value is selected because it best represents the final impact of the decision throughout the entire simulation period. The extracted four values ​​are combined in a fixed, predefined order. For example, the order is fixed as follows: the first value is the increment of equipment cascade failure probability, the second value is the increment of hazardous area area, the third value is the increment of safe evacuation window shortening time, and the fourth value is the increment of resource consumption rate. These four values ​​are combined into a four-dimensional entry, that is, a vector containing four elements. This vector represents the multidimensional impact of a single decision option. This operation is repeated for each decision option being simulated. Finally, the four-dimensional entries corresponding to all decision options are collected together to generate a multidimensional impact vector of the decision option.

[0088] The steps to obtain the normalized threat index matrix are as follows:

[0089] Based on the multidimensional influence vector of the decision options, the increments of cascade failure probability, dangerous area area, safe evacuation window shortening time and resource consumption rate are extracted in sequence and arranged into column vectors according to the decision options. The maximum and minimum values ​​of each indicator are recorded to obtain the original indicator matrix and extreme value table.

[0090] Based on the original indicator matrix and extreme value table, the maximum and minimum values ​​of each indicator are normalized to unify the value range of the four indicators to the interval between 0 and 1. A normalized threat indicator matrix is ​​constructed according to the indicator number and decision option index, and the preset command principle weight values ​​are read to obtain the weight vector.

[0091] Specifically, based on the multi-dimensional influence vector of decision options generated in the previous step, the original indicator matrix is ​​constructed. First, each four-dimensional entry in the multi-dimensional influence vector of decision options is traversed, and the values ​​of the four indicators—cascading failure probability increment, hazardous area area increment, safe evacuation window shortening time, and resource consumption rate increment—are extracted sequentially. Then, the values ​​of the same indicator for all decision options are grouped together to form a column vector. For example, the cascading failure probability increment values ​​of all decision options are arranged into the first column vector, the hazardous area area increment values ​​are arranged into the second column vector, and so on, forming a total of four column vectors. These four column vectors are then arranged according to a preset indicator order (cascading failure probability, hazardous area area increment, hazardous area area increment, etc.). The area, the shortened safe evacuation window, and the resource consumption rate are horizontally pieced together to form a matrix, where each row represents a decision option and each column represents a threat indicator. This matrix is ​​the original indicator matrix. While constructing the matrix, each column vector (i.e., each indicator) is traversed to find the maximum and minimum values, and these two extreme values ​​are recorded and associated with the corresponding indicator names. For example, the maximum value of "cascading failure probability increment" is recorded as 0.08 and the minimum value as -0.02. The maximum and minimum values ​​of all four indicators are recorded to form a list or table containing four pairs of extreme values, and finally, the complete original indicator matrix and extreme value table are obtained.

[0092] Based on the original index matrix and extreme value table, perform maximum and minimum value normalization on each element of the matrix. For any element in the matrix... Its normalized value Through formula The calculation shows that, among which and These are the minimum and maximum values ​​of the column containing the element (i.e., the corresponding indicator). These values ​​are obtained from the extreme value table. For example, for the column "Increment of Cascade Failure Probability," if the original value of a decision option is 0.03, and the maximum value of this indicator is 0.08 and the minimum value is -0.02, then its normalized value is... By performing this operation on all elements of the original indicator matrix, the numerical ranges of all four indicators are uniformly mapped to a closed interval of 0 and 1, where 0 represents the best performance (minimum threat) for that decision option and 1 represents the worst performance (maximum threat). Then, according to the index of the original indicator and the index of the decision option, these normalized values ​​are reorganized into a new matrix, which is the normalized threat indicator matrix. After that, the preset command principle weight values ​​are read from the system configuration file. These weight values ​​are preset by the emergency command expert group according to the overall goal of the rescue operation and the priority of the current accident stage, using the analytic hierarchy process (AHP). For example, in the early stage of an accident, personnel safety is the primary goal, so the weight related to "shortening the safe evacuation window" will be set higher. After the accident is under control, the weights of protecting property and reducing resource consumption may be increased. These weight values ​​constitute a four-dimensional weight vector, the sum of its components of which is 1, and finally the normalized threat indicator matrix and weight vector are obtained.

[0093] The steps for obtaining the threat index of command and decision-making options are as follows:

[0094] Based on the normalized threat index matrix and weight vector, the threat index of command and decision options is calculated using the following formula:

[0095] ;

[0096] in, For the first Threat index of command decision options for each decision-making option For the first The decision options are in the first... Normalized values ​​for each indicator and The corresponding increment of the cascade failure probability, The corresponding increase in the area of ​​the danger zone. The corresponding safe evacuation window time is shortened. Corresponding to the increase in resource consumption rate, For the first The command principle weight value of each indicator This is a risk preference adjustment factor, with a value range of [0,1], used to control the balance between average risk and extreme risk in the model. This is a smoothing factor.

[0097] Specifically, the formula for calculating the threat index of command and decision-making options uses a risk preference adjustment factor. To flexibly control the ratio of the two, when When the formula degenerates into a standard weighted average, reflecting the average threat level of the decision, when At this point, the formula becomes a smoothed maximum function, highlighting the penalty for the worst-performing decision option on any single indicator. This design allows firefighters to adjust the calculation of the threat index according to their own risk preferences (risk-averse, risk-neutral, or risk-seeking), thereby obtaining ranking results that better match their decision-making style. and Two adjustable parameters give the decision-making model greater flexibility and adaptability, enabling it to meet the diverse decision-making needs of different firefighters and in different emergency scenarios.

[0098] For the first The decision options are in the first... The normalized value for each indicator is calculated by normalizing the maximum and minimum values ​​from the previous step, and ranges from 0 to 1. These are the normalized values ​​corresponding to the increments of cascade failure probability, hazardous area area, shortened safe evacuation window time, and resource consumption rate, respectively.

[0099] For the first The command principle weight values ​​for each indicator are determined by experts using the Analytic Hierarchy Process (AHP). This weight vector reflects the relative importance of different threat indicators under specific emergency command principles. The steps are as follows: First, a judgment matrix is ​​constructed. Five emergency management experts compare the importance of the four indicators pairwise using a 1-9 scale. For example, if an expert considers "shortening the safe evacuation window" slightly more important than "increase in resource consumption rate," the score is 3; otherwise, it is 1 / 3. Then, the geometric mean of the judgment matrices from the five experts is taken to obtain the comprehensive judgment matrix. Next, the largest eigenvalue and its corresponding eigenvector of this matrix are calculated, and the eigenvector is normalized to obtain the weight vector. For example, the calculated weight vector is... ,in , , , ,and .

[0100] This is a risk preference adjustment factor, with a value range of [0,1]. This factor is set by firefighters based on their risk attitude and is used to balance average risk and extreme risk. If a firefighter is risk-averse and tends to avoid any extreme situation in any indicator, a larger adjustment factor will be set. A value (e.g., 0.8) would be set if the firefighters are risk-neutral and focused on overall average performance. The value (e.g., 0.2) can also be used to quantify the risk preferences of firefighters through questionnaires. For example, a questionnaire with multiple risk scenario choices can be designed, and their risk aversion coefficient can be statistically analyzed based on their choices, then mapped to... In this calculation, firefighters are assumed to be of a low-risk-averse type, and the value is taken as... .

[0101] As a smoothing factor, this parameter controls how closely the LogSumExp function approximates the maximum value function, and its function is similar to that described earlier. parameter, The larger the value, the closer the result of LogSumExp is to the expected value. Therefore, the greater the penalty for extreme values, the more sensitive the penalty becomes to extreme risks. The setting process is to define an extreme risk sensitivity index. ,in It is a set of weighted threat values, setting a target sensitivity. For example, 10%, can be calculated by solving equations. The value, and The process of obtaining these values ​​is similar. Here, based on experience, to achieve better smoothing and discrimination of extreme values ​​in numerical calculations, we take... .

[0102] Calculations based on parameters:

[0103] Set a decision option Its normalized values ​​across the four indicators are:

[0104] (Incremental probability of cascade failure).

[0105] (Increase in the area of ​​the danger zone)

[0106] (The safe evacuation window is shortened.)

[0107] (Increment in the rate of resource consumption).

[0108] Command principle weight value .

[0109] Risk preference adjustment factor .

[0110] Smoothing factor .

[0111] First, calculate the weighted average:

[0112] ;

[0113] ;

[0114] ;

[0115] Next, calculate the LogSumExp part:

[0116] ;

[0117] ;

[0118] Finally, the threat index of command and decision-making options is calculated. :

[0119] ;

[0120] ;

[0121] This result indicates that decision options The threat index of the command decision option is 0.38428. By calculating the threat index of all candidate decision options, a threat index list can be obtained. Firefighters can sort the options according to this list and select the decision option with the lowest threat index as the current optimal emergency response plan. For example, if the threat index of another decision option is 0.45, then the current decision option (0.38428) is better than that option. Conversely, if the threat index of another option is 0.30, then the current option is worse.

Claims

1. A simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations, characterized in that, The system includes: The accident site dynamic risk mapping module is used to receive and grid the rescue area. For each grid, it combines the real-time monitored hydrogen concentration, thermal radiation flux and explosion overpressure values, and the probability of path interruption caused by the collapse of hydrogen storage and hydrogen refueling structures to calculate the grid risk value and establish time-varying three-dimensional risk field data. The rescuer spatiotemporal path planning module is used to calculate metabolic equivalents based on the time-varying three-dimensional risk field data, combined with the rescuer's load-bearing equipment, movement speed and ambient temperature, quantify the rate of physical energy consumption, and then accumulate the dangerous doses encountered on the path and the total travel time to generate a multi-dimensional path comprehensive cost sequence. The module searches for paths in the multi-dimensional path comprehensive cost sequence to obtain a four-dimensional spatiotemporal safety corridor. The command and decision-making forward simulation module is used to simulate the accident evolution process within a preset time period based on the time-varying three-dimensional risk field data and the decision options of firefighters as input conditions. It obtains a multi-dimensional accident evolution increment set, extracts the cascading failure probability increment, the dangerous area area increment, the safe evacuation window shortening time and resource consumption rate increment from the multi-dimensional accident evolution increment set, and generates a multi-dimensional influence vector of decision options. The decision option threat quantification module is used to normalize the increments of cascading failure probability, dangerous area area, safe evacuation window shortening time, and resource consumption rate based on the multidimensional influence vector of the decision option, obtain a normalized threat index matrix, call the preset command principle weight values, calculate each index in the normalized threat index matrix, and generate a command decision option threat index.

2. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 1, characterized in that, The steps for obtaining the time-varying three-dimensional risk field data are as follows: The rescue area boundary is received and gridded. The hydrogen concentration value, thermal radiation flux value, explosion overpressure value and path interruption probability are synchronized according to the time series. Based on the hazard threshold curve, the table is looked up grid by grid to convert into hazard probability and the timestamp is aligned to obtain the hydrogen hazard probability sequence, thermal radiation hazard probability sequence, explosion hazard probability sequence and path interruption probability sequence. Based on the hydrogen hazard probability sequence, thermal radiation hazard probability sequence, explosion hazard probability sequence and path interruption probability sequence, the risk value is calculated to obtain the grid risk value sequence. Based on the grid risk value sequence, each risk value is mapped to the corresponding three-dimensional voxel by time index and the risk values ​​of adjacent time slices are connected to unify the spatial resolution and time step, forming time-varying three-dimensional risk field data.

3. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 1, characterized in that, The steps for obtaining the multidimensional path comprehensive cost sequence are as follows: Based on the time-varying three-dimensional risk field data, the time series of hydrogen hazard probability, thermal radiation hazard probability, and explosion hazard probability are read voxel by voxel. The temperature field data corresponding to each grid is extracted. Combined with the weighted equipment parameters and movement speed parameters of the rescuers, the metabolic equivalent per unit time is calculated. Then, the metabolic equivalent is accumulated according to the time step to obtain the physical energy consumption rate results for each time period. Based on the energy consumption rate results, the risk value of each path node at the corresponding time step is read sequentially along the candidate path. The dangerous dose encountered at each node on the path is calculated. The dangerous doses of each node are accumulated in the order of the path, and the corresponding travel time is calculated at the same time. The dangerous dose, energy consumption rate and travel time are combined to generate a multi-dimensional path comprehensive cost sequence.

4. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 1, characterized in that, The steps for obtaining the four-dimensional spatiotemporal safety corridor are as follows: Based on the multidimensional path comprehensive cost sequence, a cumulative risk exposure threshold and a physical exertion rate threshold are set. Paths that simultaneously meet both threshold conditions are selected, and the travel time is used as the sorting criterion. The path with the shortest travel time is selected and connected sequentially according to spatial coordinates and time indices to form a continuous time series corridor, thus obtaining a four-dimensional spatiotemporal safety corridor.

5. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 1, characterized in that, The steps for obtaining the multidimensional accident evolution increment set are as follows: Based on the time-varying three-dimensional risk field data, a preset time period is defined and a unified timestamp sequence is established. The decision options of firefighters are analyzed as action type, location of action, and start and end time. The control quantity is modified at the corresponding location step by step, and the compressor status identifier and hydrogen dispenser status identifier are updated. The status transition timestamp is recorded. At the same time, a decision-free reference record is generated. The equipment cascade failure probability, dangerous area boundary and area, safe evacuation window start and end time and resource consumption rate are written with timestamps to obtain a decision inference record set. Based on the decision-making simulation record set, the decision-making records are aligned to the non-decision-making reference records according to the timestamp and location index. The differences in the probability of cascading equipment failure, the differences in the area of ​​dangerous areas, the differences in the time of shortening of the safety evacuation window, and the differences in the rate of resource consumption are calculated step by step. The results are arranged in chronological order and the source identifier is retained to generate a multi-dimensional accident evolution increment set.

6. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 1, characterized in that, The steps for obtaining the multidimensional influence vector of the decision option are as follows: Based on the multidimensional accident evolution increment set, the equipment cascading failure probability increment, dangerous area area increment, safe evacuation window shortening time and resource consumption rate increment are taken from the time stamp at the end of the preset time period and combined into four-dimensional entries in a fixed order to generate a multidimensional influence vector for decision options.

7. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 1, characterized in that, The steps for obtaining the normalized threat index matrix are as follows: Based on the multidimensional influence vector of the decision options, the increments of cascade failure probability, dangerous area area, safe evacuation window shortening time and resource consumption rate are extracted in sequence and arranged into column vectors according to the decision options. The maximum and minimum values ​​of each indicator are recorded to obtain the original indicator matrix and extreme value table. Based on the original indicator matrix and extreme value table, the maximum and minimum values ​​of each indicator are normalized to unify the value range of the four indicators to the interval between 0 and 1. A normalized threat indicator matrix is ​​constructed according to the indicator number and decision option index, and the preset command principle weight values ​​are read to obtain the weight vector.

8. The simulation and decision support system for multi-point concurrent leakage accidents at hydrogen refueling stations according to claim 7, characterized in that, The steps for obtaining the threat index of the command and decision-making options are as follows: The threat index of command and decision options is calculated based on the normalized threat index matrix and the weight vector.

Images