Fire emergency simulation method and system based on scene digital twinning

Abstract

The invention relates to the technical field of fire disaster simulation, and discloses a fire disaster emergency simulation method and system based on scene digital twinning. The method comprises the following steps: acquiring spatial topological data of a target building through a three-dimensional laser scanning technology, and constructing a geometric model and a thermodynamic attribute model of the digital twin of the building in combination with temperature gradient distribution data acquired by a thermal imaging sensor; according to fire source diffusion mode data in a historical fire case library, pyrolysis kinetic parameters of different materials are extracted, and a fire spread probability matrix is established; inputting real-time meteorological monitoring data into the thermodynamic attribute model, and calculating a dynamic coupling relationship between a temperature field and a smoke concentration field in the building; and generating a fire scene evolution path set based on the fire spreading probability matrix and the dynamic coupling relationship, and outputting a traffic capacity attenuation curve of the key escape node. According to the method, accurate simulation and dynamic evolution prediction of a fire scene are realized, and scientific support is provided for fire emergency decision and personnel evacuation planning.

Description

Technical Field

[0001] This invention relates to the field of fire simulation technology, specifically to a fire emergency simulation method and system based on scene digital twins. Background Technology

[0002] Fire, as one of the most destructive emergencies in building safety, poses a serious threat to human life and property due to its sudden outbreak, complex spread, and uncertain smoke diffusion. Effective fire emergency simulation can predict fire evolution trends and identify critical escape routes, making it an important means to improve the efficiency of fire emergency response and reduce disaster losses. However, current mainstream fire emergency simulation technologies still have many problems that urgently need to be solved.

[0003] In building modeling, traditional methods often rely on digitizing two-dimensional drawings or simple three-dimensional modeling, making it difficult to accurately reproduce the actual spatial topology of buildings, especially details such as the distribution of beams and columns and the routing of pipelines within complex buildings. Furthermore, existing models often focus only on geometric replication, lacking accurate representation of the thermodynamic properties of building materials. This fails to realistically reflect the temperature response characteristics of different areas during a fire, leading to significant discrepancies between subsequent fire simulations and actual scenarios. In addition, while some technologies attempt to incorporate sensor data, they often use single-type sensors, failing to effectively integrate spatial and thermodynamic data, further reducing the reliability of the models.

[0004] In the fire spread prediction stage, existing models typically rely on fixed diffusion formulas, neglecting the impact of differences in the pyrolysis kinetics of various building materials on fire propagation. The rich data on fire source diffusion patterns contained in historical fire cases have not been fully explored, resulting in insufficient probabilistic predictions of fire spread and an inability to cope with random changes during fire development. Furthermore, meteorological conditions, as a crucial external factor influencing fire spread, are often treated as static parameters in current simulation methods. These methods fail to integrate them into the real-time calculation of temperature and smoke concentration fields, failing to accurately reflect the dynamic impact of meteorological factors such as wind and humidity on fire evolution. This leads to a disconnect between the coupled calculation results of temperature and smoke concentration fields and actual conditions.

[0005] In terms of emergency escape guidance, most existing technologies can only provide fixed escape route suggestions, lacking an assessment of the dynamic changes in the accessibility of key escape nodes during the development of a fire. Because factors such as rising temperatures and smoke accumulation continuously weaken the accessibility of passageways after a fire breaks out, fixed escape routes may become ineffective in real fire scenarios due to passageway blockages, failing to provide timely and effective guidance for personnel evacuation, thus affecting the safety and efficiency of emergency evacuation. These problems mean that current fire emergency simulation technologies are insufficient to meet the actual needs of accurate emergency decision-making in complex building scenarios, necessitating a new method that can integrate multi-source data and achieve dynamic and accurate simulation of fire scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide a fire emergency simulation method and system based on scene digital twins to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a fire emergency simulation method based on scenario digital twins, the method comprising: Spatial topological data of the target building is obtained by using 3D laser scanning technology, and combined with temperature gradient distribution data collected by thermal imaging sensors, a geometric model and thermodynamic property model of the building's digital twin are generated. Based on fire source diffusion pattern data in historical fire case database, pyrolysis kinetic parameters of different materials are extracted to establish fire spread probability matrix. Real-time meteorological monitoring data is input into a thermodynamic property model to calculate the dynamic coupling relationship between the temperature field and the smoke concentration field inside the building. Based on the fire spread probability matrix and dynamic coupling relationship, a set of fire scene evolution paths is generated, and the passability decay curves of key escape nodes are output.

[0008] Preferably, the generation of the geometric model and thermodynamic property model of the building digital twin includes the following steps: Non-uniform rational B-spline surface reconstruction is performed on spatial topology data to eliminate topological faults caused by scanning blind zones; The temperature gradient distribution data is mapped to the surface mesh vertices of the geometric model using the Kriging interpolation algorithm; By matching the thermal resistance values ​​of building components with a material thermal conductivity database, the parameterization calibration of the thermodynamic property model is completed.

[0009] Preferably, establishing the fire spread probability matrix includes the following steps: Ignition time series of fire sources were extracted from historical fire case databases, and critical values ​​of heat release rates for different materials were fitted using a Weiber distribution. Based on the spatial adjacency relationship between building components, a directed propagation graph is constructed with thermal radiation flux as the weight. The ignition probability threshold of each node in the directed propagation graph is calculated iteratively using Monte Carlo simulation to generate a fire spread probability matrix.

[0010] Preferably, the calculation of the dynamic coupling relationship between the internal temperature field and the smoke concentration field of the building includes the following steps: The wind speed vector in real-time meteorological monitoring data is decomposed into the normal component and tangential component of the building facade. The enhancement factor of the normal component to the chimney effect of the ventilation opening is determined based on the large eddy simulation method. A particle system is used to track the tangential component-driven smoke diffusion trajectory, and a spatiotemporal coupling equation between the temperature field and the smoke concentration field is established.

[0011] Preferably, generating the set of fire scene evolution paths includes the following steps: Extract the component number sequence that exceeds the ignition probability threshold from the fire spread probability matrix; Predict the abrupt temperature field change points corresponding to the component number sequence based on the spatiotemporal coupling equation; Aligning the abrupt change time point with the inflection point of the traffic capacity decay curve generates a set of fire scenario evolution paths with timestamps.

[0012] Preferably, the output of the passability decay curve of the key escape node includes the following steps: Mark the topological center coordinates of the stairwell and safety exit in the geometric model; The shortest path distance from the topology center coordinates to the fire source location is calculated using Dijkstra's algorithm. The traffic resistance coefficient of the shortest path distance is dynamically corrected based on the spatiotemporal coupling equation of the smoke concentration field, and a traffic capacity attenuation curve is generated.

[0013] Preferably, the method further includes the following steps: When a new component number sequence is added to the fire spread probability matrix, the incremental update mechanism of the evolution path set is triggered. The measured and predicted data at the time points of temperature field abrupt change were fused using a Kalman filter to recalibrate the parameter errors of the thermodynamic property model.

[0014] Preferably, the execution process of the incremental update mechanism includes: Compare the spatial distribution similarity between the newly added component numbering sequence and the historical case library; If the similarity exceeds the preset threshold, the evolution path branch is directly generated by calling the pre-stored pyrolysis kinetic parameters; If the similarity is lower than a preset threshold, the local recalculation process of the thermodynamic property model will be initiated.

[0015] Preferably, the local recalculation process includes: Isolate the geometric model sub-regions corresponding to the newly added component number sequence; Adaptive mesh refinement technology is used to improve the solution accuracy of the heat conduction equation in sub-regions; The calculation results are then Gaussian smoothed and stitched together with the global temperature field to update the evolution path set.

[0016] Preferably, the present invention also includes a fire emergency simulation system based on scene digital twin, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the fire emergency simulation method based on scene digital twin as described above.

[0017] Compared with the prior art, the beneficial effects of the present invention are: At the model building level, by combining 3D laser scanning technology with thermal imaging sensor data, the spatial topological accuracy of the building digital twin geometric model is ensured, enabling it to completely replicate the internal structural details of complex buildings. Furthermore, by establishing a thermodynamic property model, the temperature response characteristics of different areas are accurately captured, allowing the digital twin to truly reflect the actual fire characteristics of buildings. This breaks the limitation of traditional models that only focus on geometric shape and neglect thermodynamic properties, making the simulation basis more closely aligned with the actual scenario.

[0018] In the fire prediction phase, the fire spread probability matrix constructed based on material pyrolysis kinetic parameters extracted from a historical fire case database fully considers the impact of different materials on fire propagation. It also incorporates the patterns of historical fire source diffusion, allowing fire spread prediction to move beyond fixed formula derivations and acquire probabilistic analytical capabilities, thus more comprehensively covering the various possibilities of fire development. Furthermore, the introduction of real-time meteorological monitoring data enables the calculation of the building's internal temperature field and smoke concentration field to dynamically respond to changes in the external environment, accurately capturing the coupling relationship between the two and avoiding simulation biases caused by static parameters. This makes the prediction of fire scenario evolution more timely and accurate.

[0019] At the emergency application level, the generated fire scenario evolution path set can provide fire commanders with a clear reference for fire development trends, helping them to plan firefighting strategies in advance, allocate resources rationally, and avoid delays in response due to insufficient judgment of fire development. Simultaneously, the capacity decay curves of key escape nodes can reflect the changes in the availability of escape routes in real time during a fire, providing precise basis for personnel evacuation planning, helping to formulate dynamic escape plans, guiding personnel to avoid dangerous areas and nodes with reduced capacity, and significantly improving the safety and efficiency of evacuation.

[0020] This method deeply integrates digital twin technology with fire emergency simulation, streamlining the entire process from data acquisition and model building to scenario prediction and emergency guidance. It can adapt to the fire simulation needs of different types of buildings, providing effective simulation guidance for complex commercial complexes, high-rise buildings, and densely populated public venues. The resulting simulation results provide multi-dimensional decision support for fire emergency management, not only improving the targeting and effectiveness of fire rescue and reducing property damage caused by fires, but also providing strong support for protecting human life. This drives the development of fire emergency simulation technology towards precision, dynamism, and practicality, meeting the high requirements of modern building safety emergency management. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the working principle of a fire emergency simulation method based on scene digital twins. Figure 2 A flowchart for establishing the fire spread probability matrix; Figure 3 The curve showing the dynamic coupling relationship between the temperature field and the smoke concentration field; Figure 4 A flowchart for outputting the passability decay curve of key escape nodes. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1This invention provides a fire emergency simulation method based on scene digital twins. The method includes: acquiring spatial topological data of a building using three-dimensional laser scanning technology, which includes the geometric dimensions, shape, and spatial relationships between components of the building structure; and collecting temperature gradient distribution data of the building surface through a network of thermal imaging sensors deployed in key areas. The spatial topological data and temperature gradient distribution data are fused to construct a digital twin of the building, containing both precise geometric information and initial thermal property information. This twin is composed of a geometric model and a thermodynamic property model. The method accesses a historical fire case database to analyze and extract fire source diffusion pattern data of different building materials under fire conditions, particularly the pyrolysis kinetic parameters of the materials. Based on the spatial adjacency relationships of components in the geometric model of the building digital twin, a fire spread probability matrix is ​​established to quantify the possibility of fire propagation between components. During the simulation operation phase, external meteorological monitoring data, including wind speed, wind direction, and ambient temperature, are acquired in real time and input as boundary conditions into the constructed thermodynamic property model. Through computational fluid dynamics and thermodynamic simulation, the dynamic coupling relationship between the internal temperature field and smoke concentration field of the building with time and space is solved. By combining the probability of fire spread paths provided by the fire spread probability matrix with the details of disaster evolution revealed by the dynamic coupling relationship, multiple sets of possible fire scenario evolution paths are generated. For key nodes on evacuation routes, such as stairwells and safety exits, the curves of their passage capacity decaying over time are calculated and output, providing quantitative basis for emergency decision-making.

[0024] Example 1: See Figure 2Generating the geometric and thermodynamic property models of a building's digital twin is a complex, multi-step process. It begins with in-depth processing of the original spatial topological data. While point cloud data acquired through 3D laser scanning can accurately record the building's surface geometry, data gaps inevitably arise during actual scanning operations due to indoor obstacles, component occlusion, or scanning angle limitations. These blind spots can lead to voids or topological discontinuities in the generated mesh model, directly affecting the accuracy of subsequent simulations. To address the topological discontinuity problem caused by blind spots, a non-uniform rational B-spline surface reconstruction technique is needed to process the original point cloud data. This algorithm utilizes surrounding valid point cloud data to construct smooth, continuous surface patches using control points and weighting factors. These surface patches can seamlessly fill the data gaps, reconstructing a complete, watertight building geometric surface model. The specific process is as follows: First, based on the existing spatial topological point cloud data, the boundary points of the data gap areas are identified. Then, an interpolation algorithm is used to calculate the positions of control points within the gap areas, ensuring that the newly generated surface patches maintain geometric continuity with the surrounding existing surfaces. The weighting factors are set based on the density and curvature changes of the data points, assigning higher weights to areas with greater curvature changes to enhance the flexibility of surface fitting. The constructed surface patch is not only smooth and continuous, but can also achieve local shape optimization by adjusting control points and weights, thus seamlessly filling data gaps and forming a watertight geometric surface model. After obtaining a complete and accurate geometric model, the next step is to fuse the temperature gradient distribution data collected by the thermal imaging sensor with the geometric model. The thermal imaging data provides the temperature distribution of the building surface in its initial state, but this data usually exists in the form of discrete points or image pixels. The Kriging interpolation algorithm is used to map discrete temperature measurements onto the surface mesh vertices of the geometric model. The Kriging interpolation algorithm is an optimization interpolation method based on spatial statistics. It not only considers the distance relationship between the interpolation point and the known sample points, but also the spatial autocorrelation between sample points. By calculating the variogram function, it fits the structural characteristics of the spatial data, so that the distribution of temperature values ​​on the mesh surface maintains the spatial distribution law of the original data and meets the requirements of continuity and smoothness. Finally, a three-dimensional surface mesh model is generated, with each vertex having an initial temperature value.

[0025] After completing the temperature data mapping, the focus shifted to constructing a thermodynamic property model. The thermodynamic behavior of a building largely depends on the physical properties of its constituent materials. By querying a pre-generated database of material thermal conductivity coefficients, the various building components in the geometric model (such as concrete load-bearing walls, steel beams and columns, wooden doors and windows, etc.) were matched with the material types in the database to obtain key thermal parameters such as thermal conductivity coefficient, specific heat capacity, density, and emissivity for each material. Using the obtained thermal conductivity coefficients and component thicknesses, the thermal resistance of the components can be calculated. Thermal resistance characterizes the component's ability to resist heat transfer and is an important parameter for solving the heat conduction equation. Through this series of parameter matching and assignment, the parameterization and calibration of the thermodynamic property model were completed, enabling the digital twin to not only possess external geometric information but also the ability to simulate real thermal responses. Establishing a fire spread probability matrix is ​​the core step in quantifying the uncertainty of fire development. The reliability of the matrix is ​​based on the analysis of a large amount of historical fire case data. The ignition time series data extracted from the historical fire case database contains time information for different materials from heating to ignition under different thermal radiation intensities. These time series data are fitted and analyzed using the Weiber distribution, a continuous probability distribution widely used in reliability engineering and failure time analysis. Its shape and scale parameters can flexibly describe the statistical laws of the pyrolysis ignition process of materials. By fitting, the probability distribution function of the time when the heat release rate of various materials reaches the critical value can be obtained, thus providing a probabilistic basis for predicting the ignition time of individual components.

[0026] Based on the component-level ignition probability, it is necessary to analyze the fire spread path within the building's interior space at the system level. This involves analyzing the spatial adjacency relationships between building components using the spatial topological information contained in the building's digital twin geometric model. A directed graph model is constructed, with building components as nodes and potential fire spread directions between components as edges. In this directed propagation graph, each edge connecting adjacent components is assigned a weight value, the magnitude of which is determined by the thermal radiation flux caused by the source component to the target component during combustion. The calculation of thermal radiation flux needs to consider factors such as flame characteristics, distance between components, viewing angle coefficient, and smoke absorption. This weighted directed graph vividly depicts the possible fire propagation paths and their intensity. To combine the determined propagation paths with the uncertain ignition time probabilities, Monte Carlo simulation is used for extensive random iterative calculations. Each Monte Carlo simulation represents a possible fire scenario. In a single simulation, an ignition time is randomly generated for each combustible component based on the time probability distribution function fitted by the Weiber distribution. By combining the heat radiation flux weights defined in the directed propagation graph, the process of fire igniting adjacent components sequentially from the initial fire source in time is simulated, recording the time and order in which each component is ignited in each simulation. After thousands of Monte Carlo simulation iterations, the frequency of ignition of each component and the time interval distribution from the ignition of one component to the ignition of the next are statistically analyzed. When the frequency of ignition of a component exceeds a preset ignition probability threshold, that component is considered flammable under the corresponding fire conditions. Finally, the ignition probability relationships between all components are summarized to form a fire spread probability matrix. This matrix represents building components in rows and columns, and the values ​​of the matrix elements represent the conditional probability of fire spreading from a row component to a column component. This probability matrix provides quantitative, probability-based path support for predicting the dynamic evolution of fires.

[0027] Example 2: The calculation of the dynamic coupling relationship between the internal temperature field and smoke concentration field of a building is the core physical process of fire emergency simulation. The accuracy of the calculation directly depends on the accurate introduction of real-time external meteorological conditions. Real-time meteorological monitoring data provides key information about the external wind field. Decomposing the wind speed vector in the real-time meteorological monitoring data is the first step in simulating the impact of the external wind field on the fire environment inside the building. Wind speed vector decomposition requires decomposing the wind speed vector into a normal component perpendicular to the building's exterior surface and a tangential component parallel to the building's exterior surface, based on the geometric characteristics of the building's digital twin geometric model. The normal component directly affects the pressure distribution and air infiltration at the building's openings, while the tangential component dominates the external flow around the building and surface shear force. The normal component of the wind speed vector has a significant impact on the development of fire inside the building, especially the movement of smoke. The enhancement coefficient of the normal component to the chimney effect of the building's ventilation openings is solved based on the Large Eddy Simulation (LES) method. LES is a computational fluid dynamics technique that efficiently and relatively accurately simulates turbulent flow by directly simulating the motion of large-scale eddies and modeling the influence of small-scale eddies. When applying the large eddy simulation method, building ventilation openings (such as open doors, windows, and ventilation ducts) are set as boundary conditions. A dynamic pressure field calculated from the normal component of wind speed is applied. By solving the unsteady Navier-Stokes equations, the change in airflow caused by the external wind pressure difference can be calculated, thereby quantifying the degree to which the wind field enhances or weakens the buoyancy-driven chimney effect. This enhancement coefficient will be integrated as a key parameter into the control equations for indoor fire smoke movement. The tangential component of the wind speed vector has a significant effect on smoke transport behavior near the building's exterior walls. Using a particle system to track the smoke diffusion trajectory driven by the tangential component is an effective means of simulating smoke transport. The particle system treats smoke as a large collection of discrete, tiny particles with mass, momentum, energy, and species properties. During the simulation, each smoke particle moves under the influence of drag generated by the tangential component of the outer velocity, internal buoyancy, and turbulent diffusion. The position, velocity, and physical quantities carried by each particle at each time step are updated by numerical integration, thereby depicting the detailed trajectory of smoke in the interior space of the building over time. This method can intuitively demonstrate the diffusion, stratification, and aggregation of smoke in complex spaces.

[0028] The spatiotemporal coupling relationship between the temperature field and the smoke concentration field needs to be established by solving a set of interrelated partial differential equations. These equations include the continuity equation describing mass conservation, the Navier-Stokes equations describing momentum conservation, the fluid energy equation describing energy conservation, and the species transport equation describing the transport of smoke components. These equations are coupled through physical properties such as fluid density, viscosity, thermal conductivity, and species diffusion coefficient. In particular, changes in the temperature field directly lead to changes in air density, generating a buoyancy effect that drives the macroscopic movement of the smoke. The distribution of smoke concentration affects the absorption and scattering of thermal radiation in the air, thus influencing the distribution of the local temperature field. Numerical solving of these coupled partial differential equations is a crucial step in obtaining details of the dynamic evolution of the temperature and smoke concentration fields. Typically, the finite volume method is used to spatially discretize the computational domain, dividing the building's interior space into numerous control volume units. The governing equations are then integrated over each control volume, transforming the partial differential equations into a system of algebraic equations. The solution process requires the introduction of turbulence models (such as subgrid-scale models in large eddy simulation), combustion models (simplified chemical reaction models), and radiative heat transfer models to close the equation set and improve the physical realism of the calculation. The solution process is time-progressive; within each time step, the velocity, pressure, temperature, and concentration fields need to be solved simultaneously. Iterative algorithms ensure the convergence of the solution, ultimately outputting the temperature and smoke concentration values ​​at any location inside the building at different times, fully characterizing the dynamic coupling relationship between the temperature and smoke concentration fields. The interaction between the external wind field and the internal fire environment is a complex fluid dynamics problem, requiring dynamic coupling between the simulation of the external wind field and the internal fire simulation. This coupling can be achieved by setting dynamic boundary conditions. At the building opening boundaries, the inflow and outflow velocities and pressure conditions at the openings are dynamically updated based on real-time wind speed and direction data and the local pressure coefficients calculated by large eddy simulation. The heat plume and smoke generated by the internal fire simulation will be discharged to the outside through the openings, while the external wind field will replenish the indoor air with fresh air or change the flow field structure near the openings through the openings. This two-way coupling effect needs to be reflected in real time during the simulation through data exchange to ensure the consistency of the internal and external flow field simulations.

[0029] The results of the particle system tracking the smoke diffusion trajectory need to be fused with the temperature and concentration field calculations based on the Eulerian field. The specific fusion process includes: first, establishing an Eulerian computational grid within the building's interior space, with each grid cell storing temperature and smoke concentration values; then, statistically analyzing the particle positions at each time step in the particle system simulation, and accumulating the smoke mass or energy contribution carried by the particles based on their grid cell location. For example, the fusion of the smoke concentration field is achieved by calculating the average smoke mass of all particles within each grid cell; the fusion of the temperature field is achieved by weighted averaging of the heat values ​​carried by the particles. The fusion algorithm employs kernel function interpolation methods, such as the Gaussian kernel function, to smoothly distribute the contribution to surrounding grid cells centered on the particle position, avoiding abrupt numerical changes. The smoke trajectory information from a Lagrange perspective provided by the particle system can be used to verify the accuracy of the Eulerian field concentration calculations or as a post-processing tool to visualize the smoke flow path. The physical quantities carried by particles (such as smoke and dust concentration) can be inverted onto the Eulerian grid using statistical averaging methods, providing higher resolution field data for visibility calculation and toxicity assessment, thereby enriching the output information of dynamic coupling relationships and providing more detailed basis for subsequent safety assessment of escape routes.

[0030] See Figure 3 This graph, with time on the horizontal axis, temperature on the left vertical axis, and smoke concentration on the right vertical axis, visually presents the dynamic coupling evolution of temperature and smoke concentration over time in a fire scenario. Both the temperature and smoke concentration curves show an upward trend over time. Smoke concentration gradually approaches saturation in the later stages of growth, while temperature continues to rise, reflecting the physical laws governing their mutual influence and coordinated evolution during fire development. From a technical perspective, this coupling relationship is calculated by decomposing real-time meteorological wind speed vectors, combining this with large eddy simulation to solve for the chimney effect enhancement coefficient at ventilation openings, and then using a particle system to track the smoke diffusion trajectory, ultimately establishing a spatiotemporal coupling equation between the temperature field and the smoke concentration field. The data in this graph is a visual representation of this equation, quantifying the dynamic interaction between heat and smoke in a fire. It provides crucial physical field data for fire spread prediction, fire scenario evolution path generation, and analysis of the attenuation of accessibility at key escape nodes, serving as an important basis for achieving accurate dynamic fire simulation based on scenario digital twins.

[0031] Example 3: See Figure 4The generation of the fire scenario evolution path set and the output of the passability decay curve of key escape nodes constitute the core of dynamic risk assessment. Generating the fire scenario evolution path set begins with a deep analysis of the fire spread probability matrix, which quantifies the possibility of fire propagation between building components. From the fire spread probability matrix, building component numbers whose ignition probability values ​​exceed a preset threshold are extracted; these high-probability components identify the areas most likely to be affected by the fire. Based on the spatial topological relationships defined by the building's digital twin geometric model, the extracted component numbers are sorted and combined according to their spatial adjacency and ignition probability, forming several potential component number sequences. Each component number sequence represents a possible main fire spread path, depicting the spatial logical order of fire expansion from its origin. After determining the potential component number sequences, each sequence needs to be assigned a clear temporal attribute. Combining the spatiotemporal coupling equations of the temperature field and the smoke concentration field, the predicted time point when the ambient temperature around each component in the component number sequence will change abruptly is predicted. The abrupt change in temperature field corresponds to the critical moment when a component is ignited or subjected to intense thermal radiation shock, causing a rapid increase in its surface temperature. This is a significant event marker in the fire evolution process. The prediction process is based on the numerical solution of the heat conduction, convection, and radiation heat transfer equations, taking into account the influence of component material properties, initial temperature distribution, and real-time meteorological conditions.

[0032] Aligning the predicted abrupt temperature field changes with characteristic inflection points on the accessibility decay curves of key escape nodes is a crucial step in connecting fire development with personnel evacuation safety. Inflection points on the accessibility decay curves represent the critical moments when the difficulty of traversing a key escape node (such as a stairwell or corridor turn) begins to increase significantly due to factors such as high temperatures, toxic smoke accumulation, or a sharp decrease in visibility. Through time alignment, the spatial path of fire spread is correlated with the time window for personnel evacuation safety. This ensures that each component number sequence not only represents the spatial direction of spread but also includes a clear timeline indicating the approximate time when the fire reaches each key location, ultimately generating a time-stamped set of fire scenario evolution paths containing multiple possible development scenarios. Outputting the accessibility decay curves of key escape nodes requires precise spatial positioning and dynamic path assessment. In the geometric model of the building's digital twin, accurately marking the topological center coordinates of all key escape nodes is fundamental to outputting the accessibility decay curves. Key escape nodes include major evacuation stairwells, safety exits, refuge rooms, and key decision points in corridors. The marking process requires analyzing the building's spatial topology, determining the central location of these nodes in the evacuation network, and recording their three-dimensional coordinates to provide accurate start or end point information for subsequent shortest path calculations. Dijkstra's algorithm is used to calculate the shortest path distance from the topological center coordinates of each critical escape node to the initial fire source location. Dijkstra's algorithm is a classic algorithm for finding single-source shortest paths in a weighted graph. It abstracts the building space into a graph structure, where nodes represent spatial locations or regions, edges represent connecting paths, and the edge weights can be initialized with the physical length of the path or standard travel time. By executing Dijkstra's algorithm, the theoretical shortest path distance from the fire source to various safe nodes can be obtained in the initial stage of a fire, before the environment deteriorates. This distance reflects the baseline level of evacuation difficulty.

[0033] The development of a fire significantly alters the passability of evacuation routes. Therefore, it is necessary to dynamically adjust the passability resistance coefficient along the shortest path distance based on smoke concentration and temperature field data calculated using spatiotemporal coupling equations. Smoke concentration field data is used to assess the impact of reduced visibility and toxic gas accumulation on personnel movement speed and psychological stress; high concentration areas correspond to high passability resistance. Temperature field data is used to determine the tolerance limits of high-temperature environments and the hazards of heat radiation; areas exceeding the human tolerance range are considered impassable. The passability resistance coefficient is a dimensionless correction factor that translates environmental hazards into a reduction in path passability. The dynamically corrected effective passability distance or time more realistically reflects the actual usability of evacuation routes in a fire environment. The passability decay curve of critical escape nodes is a visual representation of how passability changes over time. The curve is generated based on continuous calculations of the effective passability distance or time from the fire source to the escape node at different times. As the fire progresses for time T, smoke and high temperatures continuously erode the original evacuation routes, leading to a continuous increase in effective passability distance or a sharp extension of effective passability time. Traffic capacity decay curves are typically plotted with time (T) on the horizontal axis and traffic capacity indicators (such as effective traffic speed and remaining safe passage time) on the vertical axis. Inflection points on the curve correspond to the aforementioned abrupt changes in temperature or critical smoke concentration points, marking the deterioration of traffic conditions. A typical traffic capacity decay curve can describe the process of traffic capacity gradually decreasing until complete failure as the fire progresses.

[0034] The mathematical expression of the capacity decay curve can be described using a function that includes time T and path environment parameters. For example, a simplified model can consider the effect of smoke concentration S(T) and temperature θ(T) on the base traffic speed. Corrective effect:

[0035] in: Indicates after the fire At any given moment, the effective speed at which people travel on a specific path segment. This indicates the normal walking speed of a person in a fire-free and harmless environment. Indicates in At any given time, the smoke concentration on that path segment is determined by the solution of the spatiotemporal coupling equation of the smoke concentration field. This indicates the critical smoke concentration threshold that would render the area completely impassable to people. Indicates in At any given time, the ambient temperature along this path segment is determined by the solution of the spatiotemporal coupling equation of the temperature field. This refers to the reference value for ambient temperature that the human body feels comfortable in. This represents the critical environmental temperature threshold that the human body cannot withstand. This function indicates the effective passage speed. With smoke concentration The rise and ambient temperature The increase and decrease, when achieve or achieve hour, A value approaching zero indicates that the path is completely invalid. This is equivalent to the basic traffic speed. The correction factor for dynamic adjustment is the aforementioned traffic resistance coefficient. By integrating or iteratively calculating this function over time along different path segments, a curve reflecting the decay of the entire path's traffic capacity over time can be generated.

[0036] The final output of the critical escape node accessibility decay curves provides quantified time pressure information for emergency evacuation decisions. Decision-makers can intuitively see the expected accessibility of each safety exit at different points in time after a fire, identify which exits have longer safety windows and which exits will fail prematurely, thereby developing more targeted evacuation plans and dynamic evacuation guidelines to maximize the protection of people's lives. The combination of the fire scenario evolution path set and the accessibility decay curves enables an integrated assessment of dynamic fire risks and personnel evacuation safety.

[0037] Example 4: The dynamic update and calibration mechanism of the fire emergency simulation system ensures that the simulation results can respond to real-time changes in the fire scenario. When the system detects a new component number sequence in the fire spread probability matrix through the deployed sensor network or real-time data analysis, it means that a new building component has reached the critical condition for ignition. The simulation system will immediately trigger the incremental update mechanism of the evolution path set. The incremental update mechanism is designed to avoid starting a computationally intensive full-scenario resimulation due to local fire changes. Instead, it adopts an efficient and targeted update strategy, recalculating and fusing data only on the relevant parts that have changed, thereby significantly improving the system response speed and meeting the real-time requirements of emergency decision-making. The core execution logic of the incremental update mechanism lies in assessing the degree of matching between the newly emerging fire and historical experience. The system compares the spatial distribution characteristics of the fire points represented by the newly added component number sequence with numerous fire development patterns stored in the historical fire case database. The comparison process extracts multiple feature dimensions for calculation, including the material type of the new component, the spatial relative position relationship between the new component and the already burned component, distance, azimuth angle, and the distribution density of surrounding combustibles, etc. These feature vectors are compared with the feature vectors of each record in the historical case database to calculate their similarity. This is typically done using metrics such as cosine similarity or the reciprocal of Euclidean distance, resulting in a comprehensive similarity score between zero and one. This similarity score is then compared to a preset threshold, an optimized value determined through machine learning training on a large amount of historical data, designed to balance accuracy and computational efficiency.

[0038] If the calculated overall similarity score exceeds a preset threshold, the system determines that the newly added fire point spread pattern belongs to a "normal" pattern with high similarity in the historical case library. In this case, the system directly calls the parameter set associated with the historical case with the highest matching degree from the pre-stored pyrolysis kinetic parameter library. These parameter sets contain key data such as the heat release rate curve and ignition temperature of typical materials under this spread pattern. Using these parameters verified by historical data, the system can quickly deduce the most likely development path and timeline of the newly added fire point, generating a new evolutionary path branch. This newly generated path branch will be marked with its referenced historical case source and integrated into the existing set of evolutionary paths in the form of timestamps and probability weights, realizing rapid expansion of the set. If the overall similarity score is lower than the preset threshold, it indicates that the newly added fire point presents an "abnormal" or "special" spread pattern that lacks high similarity records in historical cases. This anomalousness may stem from the special characteristics of the building structure, the combustion characteristics of new materials, or rare combinations of ventilation conditions. In this case, the system will initiate a local recalculation process of the thermodynamic property model. The local recalculation process does not involve rerunning the entire building's full-site simulation. Instead, it focuses computational resources on a localized area affected by the new fire source. The first step is to dynamically define the boundary of a sub-region requiring detailed recalculation within the overall digital twin geometric model of the building, based on the spatial coordinates of the new component's number sequence and their adjacency relationships. This sub-region typically includes the new fire source itself, its directly adjacent components, and the surrounding area that may be significantly affected by thermal radiation, forming a local computational domain.

[0039] Within a defined local sub-region, to more accurately capture potentially complex physical phenomena, such as drastic temperature gradient changes or unique flue gas flow patterns, an adaptive mesh refinement technique is employed to improve the solution accuracy of the governing equations. This technique dynamically adjusts the density of the computational mesh based on the gradient information of the currently calculated physical fields (such as temperature and velocity fields). In regions with drastic changes in physical quantities, such as near the flame front or the interface between high-temperature flue gas and cold air, the program automatically refines the mesh, using smaller mesh cells to distinguish subtle changes; while in regions with gentle physical quantity distribution, a relatively sparse mesh is maintained to save computational resources. This dynamic mesh optimization method based on physical field characteristics can significantly improve the computational accuracy of critical regions while controlling the total computational load. After completing the refined calculation of the local sub-region, the generated new physical field data needs to be seamlessly integrated with the unaffected global background field. Direct stitching may cause data jumps or discontinuities at the sub-region boundaries, leading to numerical instability. A Gaussian smoothing stitching algorithm is used to handle the boundary between the sub-region and the global region. The Gaussian smoothing function exhibits a bell-shaped weight distribution characteristic. The algorithm delineates a transition zone on both sides of the boundary. Within this transition zone, the final physical quantity value of each grid point is obtained by Gaussian weighted averaging of the results calculated in the sub-region and the original global field. The closer to the interior of the sub-region, the higher the weight of the sub-region calculation result; the closer to the global region, the higher the weight of the original global field value. This smooth stitching effectively eliminates numerical abrupt changes at the boundary, ensuring the continuity and smoothness of the physical field throughout the entire computational domain, thereby achieving reliable updates to the fire scene evolution path set.

[0040] In addition to incremental updates to the evolution path, the system also integrates a model parameter calibration mechanism. A Kalman filter is used to fuse measured and predicted data at temperature abrupt change points, recalibrating the parameter errors of the thermodynamic property model. When temperature sensors installed inside the building transmit actual temperature abrupt change time data, the Kalman filter compares this measured value with the model's predicted value. Based on the difference between the two and the preset statistical characteristics of model and observation noise, the relevant parameters in the thermodynamic property model are adjusted in reverse, allowing the model's subsequent prediction output to better approximate the actual physical process. This achieves online self-calibration of model parameters and continuously improves the prediction accuracy of the simulation system. Table 1 shows some of the feature vectors that may be involved in the system's similarity assessment of new fire points and historical cases during the incremental update process.

[0041] Table 1: Feature Vector Table for Similarity Assessment between New Fire Points and Historical Cases

[0042] Example 5: The execution of the incremental update mechanism relies heavily on the quantitative similarity assessment results obtained by comparing newly added fire points with recorded patterns in the historical case database. This assessment result directly determines the subsequent update strategy adopted by the system. When a new component number sequence is detected in the fire spread probability matrix, the system immediately initiates the spatial distribution similarity calculation process. The calculation process first extracts multi-dimensional feature vectors from the geometric entities represented by the newly added component number sequence. The feature vectors include the component's material combustion rating, the component's absolute coordinates in three-dimensional space, the three-dimensional Euclidean distance between the component and each member in the current set of burned components, the relative azimuth angle between the component and the burned components, and the distribution density index of all combustible components within a certain radius of the surrounding sphere. The extracted feature vectors are compared one by one with the feature vectors of each archived case in the historical fire case database. The historical fire case database is a knowledge base trained with a large amount of real fire data and simulation data. Each case not only records the development process of the fire but also annotates the corresponding architectural spatial structural features and fire dynamic parameters. The comparison process employs a weighted cosine similarity algorithm. This algorithm assigns different weight coefficients to different dimensions of the feature vector. These weight coefficients are determined based on historical data analysis and reflect the varying influences of different feature dimensions on fire spread patterns. For example, the weight of the combustion rating of a component material may be higher than that of its absolute coordinates, because material properties have a more direct impact on ignition rate. The calculated similarity score is a scalar value between zero and one.

[0043] This similarity score needs to be compared with a preset threshold. This threshold is not a fixed constant, but a dynamically adjusted parameter. Its initial value comes from the optimal value obtained after cross-validation statistical analysis of the historical case database, and it is fine-tuned during system operation based on the consistency between model predictions and measured data. The preset threshold serves to delineate the boundary between "normal spread patterns" and "abnormal spread patterns." If the similarity score exceeds the preset threshold, the system determines that the spread pattern of the newly added fire point is highly similar to a known pattern in the historical case database. The system then switches to a fast track to generate evolution path branches by calling pre-stored parameters. The system retrieves the parameter set associated with the historical case with the highest matching degree from the pre-stored pyrolysis kinetic parameter database. These parameter sets contain key data such as material pyrolysis reaction kinetic constants, heat release rate curve fitting parameters, and flue gas generation rate, which have been validated under similar scenarios. Using these pre-stored parameters, the system does not need to perform complex physical field recalculation. Instead, it can quickly deduce the subsequent spread path, speed, and key event time points of newly added fire points through parameter interpolation and extrapolation, generate a new evolution path branch, and integrate this branch with the set of evolution paths. The branch will be marked with the ID of its reference case and similarity score for subsequent analysis and tracing.

[0044] If the similarity score is below a preset threshold, it indicates that the characteristics of the newly added fire point differ significantly from historical experience, suggesting a potential encounter with an unrecognized fire scenario. The system then initiates a local recalculation process for the thermodynamic property model. The core idea of ​​this local recalculation process is to precisely focus computational resources on the affected local area, avoiding the enormous computational overhead of global recalculation. The first step is sub-region isolation. Based on the spatial location of the newly added component number sequence and its connectivity within the building topology network, a sub-region boundary requiring further detailed simulation is dynamically defined within the overall digital twin geometric model. The sub-region definition is not a simple geometric bounding box but rather based on an analysis of the potential thermal impact range. It typically includes the newly added fire point, all adjacent components with direct thermal radiation or convection paths, and surrounding spaces that may be rapidly affected due to structural connectivity.

[0045] Within the isolated geometric model sub-regions, to capture potentially complex physical phenomena, such as turbulent vortices caused by special spatial structures or abnormal combustion behavior due to abrupt changes in material properties, it is necessary to improve the spatiotemporal resolution of the calculations. Adaptive mesh refinement technology is employed to achieve on-demand improvement in computational accuracy. This technology does not generate an extremely fine mesh all at once, but rather dynamically optimizes based on the initial calculation results. The calculation program monitors the gradient changes of physical field variables within the sub-region in real time, such as temperature gradient, velocity gradient, and smoke concentration gradient. In areas where the gradient values ​​of field variables exceed a set threshold, the system automatically subdivides the background mesh, generating smaller mesh cells, thereby achieving higher resolution and computational accuracy in these critical areas. In areas with a gentle distribution of field variables, a relatively coarse mesh is maintained to conserve computational resources. This dynamic mesh optimization strategy based on physical field characteristics ensures that computational resources are efficiently allocated to the areas most requiring detailed simulation. After completing the adaptive mesh refinement and fine calculation within the sub-region, high-precision updated physical field data is obtained within that sub-region. The next step is to seamlessly integrate this locally updated data with the unchanged global background field. Direct data replacement can introduce discontinuous jumps at sub-region boundaries, leading to numerical instability and physical distortion. A Gaussian smoothing stitching algorithm is employed to address this boundary transition problem. This algorithm defines a transition band of a certain width on both sides of the sub-region boundary, typically proportional to the local grid size. At each grid point within the transition band, the final physical field value is no longer simply derived from either the sub-region calculation result or the global field result, but rather obtained by a weighted average of both using a Gaussian weighting function. This Gaussian weighting function ensures that the sub-region calculation result has a higher weight closer to the sub-region, the original global field value has a higher weight closer to the global region, and the weights are evenly distributed at the center of the transition band. This smooth transition effectively eliminates numerical discontinuities at sub-region boundaries, ensuring the smoothness and continuity of the physical field throughout the computational domain, thus making the updated evolution path set physically self-consistent.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fire emergency simulation method based on scene digital twins, characterized in that, Includes the following steps: Spatial topological data of the target building is obtained by using 3D laser scanning technology, and combined with temperature gradient distribution data collected by thermal imaging sensors, a geometric model and thermodynamic property model of the building's digital twin are generated. Based on fire source diffusion pattern data in historical fire case database, pyrolysis kinetic parameters of different materials are extracted to establish fire spread probability matrix. Real-time meteorological monitoring data is input into a thermodynamic property model to calculate the dynamic coupling relationship between the temperature field and the smoke concentration field inside the building. Based on the fire spread probability matrix and dynamic coupling relationship, a set of fire scene evolution paths is generated, and the passability decay curves of key escape nodes are output.

2. The fire emergency simulation method based on scene digital twin as described in claim 1, characterized in that, The process of generating the geometric and thermodynamic property models of the building digital twin includes the following steps: Non-uniform rational B-spline surface reconstruction is performed on spatial topology data to eliminate topological faults caused by scanning blind zones; The temperature gradient distribution data is mapped to the surface mesh vertices of the geometric model using the Kriging interpolation algorithm; By matching the thermal resistance values ​​of building components with a material thermal conductivity database, the parameterization calibration of the thermodynamic property model is completed.

3. The fire emergency simulation method based on scene digital twin according to claim 2, characterized in that, The establishment of the fire spread probability matrix includes the following steps: Ignition time series of fire sources were extracted from historical fire case databases, and critical values ​​of heat release rates for different materials were fitted using a Weiber distribution. Based on the spatial adjacency relationship between building components, a directed propagation graph is constructed with thermal radiation flux as the weight. The ignition probability threshold of each node in the directed propagation graph is calculated iteratively using Monte Carlo simulation to generate a fire spread probability matrix.

4. The fire emergency simulation method based on scene digital twin according to claim 3, characterized in that, The calculation of the dynamic coupling relationship between the internal temperature field and the smoke concentration field of the building includes the following steps: The wind speed vector in real-time meteorological monitoring data is decomposed into the normal component and tangential component of the building facade. The enhancement factor of the normal component to the chimney effect of the ventilation opening is determined based on the large eddy simulation method. A particle system is used to track the tangential component-driven smoke diffusion trajectory, and a spatiotemporal coupling equation between the temperature field and the smoke concentration field is established.

5. The fire emergency simulation method based on scene digital twin according to claim 4, characterized in that, The process of generating a set of fire scenario evolution paths includes the following steps: Extract the component number sequence that exceeds the ignition probability threshold from the fire spread probability matrix; Predict the abrupt temperature field change points corresponding to the component number sequence based on the spatiotemporal coupling equation; Aligning the abrupt change time point with the inflection point of the traffic capacity decay curve generates a set of fire scenario evolution paths with timestamps.

6. The fire emergency simulation method based on scene digital twin according to claim 5, characterized in that, The output of the passability decay curve of the key escape node includes the following steps: Mark the topological center coordinates of the stairwell and safety exit in the geometric model; The shortest path distance from the topology center coordinates to the fire source location is calculated using Dijkstra's algorithm. The traffic resistance coefficient of the shortest path distance is dynamically corrected based on the spatiotemporal coupling equation of the smoke concentration field, and a traffic capacity attenuation curve is generated.

7. The fire emergency simulation method based on scene digital twin according to claim 6, characterized in that, It also includes the following steps: When a new component number sequence is added to the fire spread probability matrix, the incremental update mechanism of the evolution path set is triggered. The measured and predicted data at the time points of temperature field abrupt change were fused using a Kalman filter to recalibrate the parameter errors of the thermodynamic property model.

8. The fire emergency simulation method based on scene digital twin according to claim 7, characterized in that, The execution process of the incremental update mechanism includes: Compare the spatial distribution similarity between the newly added component numbering sequence and the historical case library; If the similarity exceeds the preset threshold, the evolution path branch is directly generated by calling the pre-stored pyrolysis kinetic parameters; If the similarity is lower than a preset threshold, the local recalculation process of the thermodynamic property model will be initiated.

9. The fire emergency simulation method based on scene digital twin according to claim 8, characterized in that, The local recalculation process includes: Isolate the geometric model sub-regions corresponding to the newly added component number sequence; Adaptive mesh refinement technology is used to improve the solution accuracy of the heat conduction equation in sub-regions; The calculation results are then Gaussian smoothed and stitched together with the global temperature field to update the evolution path set.

10. A fire emergency simulation system based on scene digital twin, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the fire emergency simulation method based on scene digital twin as described in any one of claims 1 to 9.

Images