Building fire risk dynamic assessment method and system based on big data

Abstract

This invention relates to the field of smart fire protection technology, specifically to a method and system for dynamic assessment of building fire risks based on big data. It includes: S1, constructing a heterogeneous information network model, forming a heterogeneous information network graph of building fire protection based on multi-source data, and constructing a digital twin environment; S2, inputting the heterogeneous information network graph into a spatiotemporal graph convolutional network model, predicting the dynamic attribute weight changes of nodes and edges; S3, constructing a fire spread seepage model and a rescue failure seepage model based on evolutionary trends, and using a graph cut algorithm to identify critical seepage nodes / edges; S4, calculating fire resilience entropy based on the predicted network state; S5, extrapolating the runaway time window based on critical seepage nodes / edges and fire resilience entropy, generating risk assessment results, and mapping them to the digital twin environment for visualization. This invention can identify key weak points that lead to system collapse, predict the remaining time for fire spread to critical areas, and provide forward-looking decision support for fire rescue.

Description

Technical Field

[0001] This invention relates to the field of smart fire protection technology, and more specifically, to a method and system for dynamic assessment of building fire risk based on big data. Background Technology

[0002] Building fire safety risk assessment is a crucial means of protecting people's lives and property. Traditional building fire safety assessment methods mainly rely on regular manual inspections and static indicator systems. Risk levels are determined by checking indicators such as the integrity of fire protection facilities, the unobstructedness of evacuation routes, and the storage of combustible materials, combined with expert scoring or weighted summation. With the development of Internet of Things (IoT) technology, some existing technologies have begun to incorporate sensors to monitor data in real time, triggering alarms when thresholds are set, thus achieving a shift from static assessment to dynamic monitoring.

[0003] In recent years, some studies have attempted to combine big data analytics with fire safety assessments. For example, they have used historical fire data to train machine learning models to predict fire probabilities, or utilized building information models (BIM) for 3D visualization. These methods have improved the real-time nature and intuitiveness of the assessments to some extent.

[0004] The existing technology still has the following shortcomings, specifically:

[0005] 1. Existing technologies mainly rely on static indicator systems and threshold alarm mechanisms, using preset weighted scoring or parameter exceeding limits to assess risk. This "post-event alarm" mode can only respond passively after an anomaly occurs, and cannot proactively model the risk propagation path based on fire dynamics. It is even more difficult to issue early warnings in the early stages of risk when data fluctuates abnormally but has not yet exceeded the limits, thus missing the best opportunity for intervention.

[0006] 2. Existing methods only output a single risk score or level, failing to quantify the overall resilience of a building system against fire spread, and even less able to answer critical practical questions such as "how long can the system hold out?" and "which node failure would lead to overall loss of control." Two areas with the same risk score may have fire resistance times that differ by several times, but current assessments cannot distinguish this fundamental difference.

[0007] 3. The vast majority of existing technologies stop at risk detection, providing only alarm information without outputting solutions. After assessing high risks, they cannot automatically generate optimal emergency resource dispatch plans, personnel evacuation route planning, and other decision support, resulting in a gap between risk assessment and actual fire rescue operations, and making it difficult to translate assessment results into concrete action guidelines. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for dynamic assessment of building fire risk based on big data, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention aims to provide a method for dynamic assessment of building fire safety risks based on big data, comprising: S1, constructing a heterogeneous information network model: acquiring multi-source data of the target building, aggregating physical space units, fire protection facilities, and dynamic personnel within the building into heterogeneous nodes based on graph theory, constructing heterogeneous edges based on the physical connectivity, functional dependencies, and thermal conduction relationships between nodes, and assigning time-varying dynamic attribute weights to nodes and edges to form a building fire safety heterogeneous information network graph. ,in, For a set of nodes, Let be the set of edges. The weight matrix is ​​time-varying, and a digital twin environment for building fire protection that maps to the physical building is constructed based on this matrix.

[0010] S2. State Evolution Prediction: Input the heterogeneous information network graph into the spatiotemporal graph convolutional network model to predict the dynamic attribute weight changes of nodes and edges within a preset time window in the future, and generate the network state evolution trend.

[0011] S3. Critical seepage analysis: Based on the aforementioned evolution trend, a fire spread seepage model and a rescue failure seepage model are constructed. A graph cut algorithm is used to iteratively remove nodes or edges in the network and monitor changes in network connectivity components. When the removal of a few nodes or edges causes a sudden change in network connectivity, the critical point is identified and the corresponding node or edge is marked as a critical seepage node / edge.

[0012] S4. Resilience Entropy Calculation: Based on the predicted network state, the probability distribution of each fire compartment being in a safe state is defined, and the fire resilience entropy of the system is calculated to quantify the degree of orderliness of the system in resisting the spread of risks.

[0013] S5. Dynamic assessment: Based on the seepage critical node / edge and the fire resistance resilience entropy, the time window for the fire to spread to the preset key area is deduced, multi-dimensional dynamic risk assessment results are generated, and the dynamic risk assessment results are mapped to the building fire protection digital twin environment for visualization.

[0014] As a further improvement to this technical solution, the multi-source data of the target building includes building information model (BIM) data, real-time monitoring data from IoT sensors, video surveillance stream data, and static attribute data.

[0015] The BIM data includes at least the building structure dimensions, fire compartment division, fire protection facility layout, and geometric and physical information of material properties, which are parsed and extracted through the BIM interface.

[0016] IoT sensor data includes at least multiple types of data such as temperature, smoke, humidity, electrical, and pressure. Edge computing technology is used for real-time preprocessing to remove abnormal data and noise.

[0017] The video surveillance stream data uses a lightweight target detection algorithm to identify and extract features of people, flames, and smoke, reducing the burden of data transmission and processing.

[0018] Static attribute data should include at least the building's service life, fire protection design rating, and historical fire data, which are used to initialize the network model and calibrate parameters.

[0019] As a further improvement to this technical solution, the dynamic attribute weights include at least one of the following or a combination thereof: the dynamic coefficient of fire load calculated based on the temperature, smoke concentration and electrical parameters collected in real time by IoT sensors.

[0020] The channel blockage coefficient is calculated based on the flow field of personnel density, movement speed, and direction identified by the target detection algorithm in the video surveillance stream.

[0021] The facility integrity coefficient is based on feedback from fire water system pressure and fire facility status sensors.

[0022] External environment correction coefficients are calculated based on wind speed, wind direction, temperature, and humidity data obtained through a meteorological interface.

[0023] As a further improvement to this technical solution, the probabilistic flux calculation method for the fire spread seepage model is as follows:

[0024]

[0025] in, express The fire was spreading from the node. Spread to adjacent nodes The probability, Represents a node Combustible load factor, Represents a node With nodes Oxygen supply coefficient between Indicates the barrier factor. , This indicates the adjustment parameter.

[0026] As a further improvement to this technical solution, the method for calculating the rescue flux attenuation coefficient of the rescue failure seepage model is as follows:

[0027]

[0028] in, Indicates the connection node and The channel at all times The channel blocking coefficient, Indicates the connection node and The channel at all times The integrity rating of fire protection facilities Indicates the connection node and The channel at all times The fire threat coefficient is obtained by accumulating the probability flux of the fire spread seepage model. Indicates the threat impact coefficient. This represents the fire threat normalization factor.

[0029] As a further improvement to this technical solution, the formula for calculating the fire resilience entropy is:

[0030]

[0031] in, Indicates the total number of fire compartments. Indicates the first Each partition The probability of maintaining a safe state at all times is obtained by normalizing the output layer of the spatiotemporal graph convolutional network model. An alert is triggered when the threshold is exceeded.

[0032] As a further improvement to this technical solution, the method for extrapolating the runaway time window includes: locking the key seepage path based on the seepage critical analysis results, combining the fire source location and the predicted network state, and using a reduced-order partial differential equation solver to calculate the shortest remaining time for the fire front to reach the boundary of the preset key area, which is used as the runaway time window.

[0033] As a further improvement to this technical solution, it also includes an intervention strategy generation step based on causal inference: introducing a structural causal model, when the fire resilience entropy increases or the runaway time window is lower than the safety threshold, tracing back to the root cause node or edge that leads to the deterioration of the risk state; combining reinforcement learning algorithm, with the goal of maximizing the delay of the runaway time window or reducing the system resilience entropy, outputting the optimal emergency intervention strategy.

[0034] As a further improvement to this technical solution, the optimal emergency intervention strategy includes: dynamically planned personnel evacuation routes that avoid predicted high-risk seepage areas in real time within the digital twin environment; and a fire resource dispatching scheme that prioritizes the seepage critical node / edge.

[0035] The second aspect of this invention provides a system for dynamic assessment of building fire risk based on big data, comprising: a module for constructing a heterogeneous information network model: acquiring multi-source data of the target building, aggregating physical space units, fire protection facilities, and dynamic personnel within the building into heterogeneous nodes based on graph theory, constructing heterogeneous edges based on the physical connectivity, functional dependencies, and thermal conduction relationships between nodes, and assigning time-varying dynamic attribute weights to nodes and edges to form a heterogeneous information network graph for building fire protection. ,in, For a set of nodes, Let be the set of edges. The weight matrix is ​​time-varying, and a digital twin environment for building fire protection that maps to the physical building is constructed based on this matrix.

[0036] State evolution prediction module: Input the heterogeneous information network graph into the spatiotemporal graph convolutional network model, predict the dynamic attribute weight changes of nodes and edges within a preset time window, and generate the network state evolution trend.

[0037] Critical seepage analysis module: Based on the evolution trend, a fire spread seepage model and a rescue failure seepage model are constructed. A graph cut algorithm is used to iteratively remove nodes or edges in the network and monitor the changes in the network connectivity components. When the removal of a few nodes or edges causes a sudden change in network connectivity, the critical point is identified and the corresponding node or edge is marked as a critical seepage node / edge.

[0038] Resilience Entropy Calculation Module: Based on the predicted network state, it defines the probability distribution of each fire compartment being in a safe state, calculates the fire resilience entropy of the system, and uses it to quantify the degree of orderliness of the system in resisting the spread of risks.

[0039] Dynamic assessment module: Based on the seepage critical node / edge and the fire resilience entropy, it extrapolates the time window for the fire to spread to the preset key area out of control, generates multi-dimensional dynamic risk assessment results, and maps the dynamic risk assessment results to the building fire protection digital twin environment for visualization.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. This invention introduces the seepage theory from statistical physics into fire safety assessment, constructs a dual seepage model for fire spread and rescue failure, and uses a graph cut algorithm to identify critical seepage nodes / edges that cause abrupt changes in network connectivity. For the first time, it achieves a fundamental shift from assessing "how high the risk" to assessing "when the system will collapse and where it will go out of control," enabling precise location of key weak points that, once failed, will trigger overall collapse.

[0042] 2. This invention innovatively proposes a fire resilience entropy index, which quantifies the degree of orderliness of the system in resisting the spread of risks based on the probability distribution of maintaining a safe state in each fire compartment. When the entropy value exceeds the limit, a system-level chaotic early warning is triggered. At the same time, a reduced-order partial differential equation solver accurately predicts the time window for the fire front to reach the critical area, providing a clear time benchmark and action window for emergency decision-making, enabling fire management to move from qualitative to quantitative.

[0043] 3. This invention employs a spatiotemporal graph convolutional network to achieve forward-looking prediction of risk evolution, introduces a structural causal model to enable interpretable risk root cause tracing, and combines reinforcement learning algorithms to dynamically plan optimal evacuation paths and resource scheduling schemes in a digital twin environment. This technology chain connects all stages of "assessment-early warning-decision-making," truly realizing a complete closed loop from problem discovery to problem solving. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the implementation steps of the method of the present invention.

[0046] Figure 2 This is a schematic diagram of the system structure connection of the present invention. Detailed Implementation

[0047] 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.

[0048] Example: Please refer to Figure 1 As shown, a method for dynamic assessment of building fire risk based on big data is provided, including: S1, constructing a heterogeneous information network model: acquiring multi-source data of the target building, aggregating physical space units, fire protection facilities, and dynamic personnel within the building into heterogeneous nodes based on graph theory, constructing heterogeneous edges based on the physical connectivity, functional dependencies, and thermal conduction relationships between nodes, and assigning time-varying dynamic attribute weights to nodes and edges to form a building fire heterogeneous information network graph. ,in, For a set of nodes, Let be the set of edges. The weight matrix is ​​time-varying, and a digital twin environment for building fire protection that maps to the physical building is constructed based on this matrix.

[0049] In one specific embodiment, the multi-source data of the target building includes building information model (BIM) data, real-time monitoring data from IoT sensors, video surveillance stream data, and static attribute data.

[0050] The BIM data includes at least the building structure dimensions, fire compartment division, fire protection facility layout, and geometric and physical information of material properties, which are parsed and extracted through the BIM interface.

[0051] IoT sensor data includes at least multiple types of data such as temperature, smoke, humidity, electrical, and pressure. Edge computing technology is used for real-time preprocessing to remove abnormal data and noise.

[0052] The video surveillance stream data uses a lightweight target detection algorithm to identify and extract features of people, flames, and smoke, reducing the burden of data transmission and processing.

[0053] Static attribute data should include at least the building's service life, fire protection design rating, and historical fire data, which are used to initialize the network model and calibrate parameters.

[0054] In one specific embodiment, the dynamic attribute weights include at least one of the following or a combination thereof:

[0055] The dynamic coefficient of fire load is calculated based on real-time temperature, smoke concentration, and electrical parameters collected by IoT sensors.

[0056] Definition: Representation node The level of flammable hazard (e.g., in a room or area) at the current moment is determined by a comprehensive analysis of real-time data from IoT sensors.

[0057] Calculation formula:

[0058]

[0059] in, Represents a node At any moment The dynamic coefficient of fire load, Represents a node At any moment Real-time temperature sensor value (°C). Represents a node At any moment The real-time value of the smoke concentration sensor (%obs / m). Represents a node At any moment The normalized risk value of electrical parameters (such as residual current and cable temperature) is obtained by fusing multiple electrical features. , , These represent the baseline safety thresholds for temperature, smoke concentration, and electrical parameters, respectively. , , These represent the extreme danger thresholds for temperature, smoke concentration, and electrical parameters, respectively (which can be set to the sensor range or historical statistical extreme values). , , Represents the weighting coefficients, satisfying It can be determined through the analytic hierarchy process or by training with historical fire data; This represents the Sigmoid function, which compresses the output to the (0,1) interval. The larger the value, the higher the fire load.

[0060] Analysis process:

[0061] Real-time acquisition of sensor data, and outlier removal and normalization.

[0062] When a parameter exceeds the baseline threshold, the risk begins to increase significantly; when it approaches the extreme threshold, the risk approaches 1.

[0063] Electrical parameters need to integrate multiple features such as residual current, electric arc, and cable temperature, and a comprehensive risk value can be generated using principal component analysis or expert rules.

[0064] The channel blockage coefficient is calculated based on the flow field of personnel density, movement speed, and direction identified by the target detection algorithm in the video surveillance stream.

[0065] Definition: Represents a connected node and The passageway (such as corridor, staircase) at any time The degree of capacity attenuation is determined by video surveillance stream analysis.

[0066] Calculation formula:

[0067]

[0068] in, Indicates the connection node and The channel at all times The channel blocking coefficient, The real-time personnel density (people / m²) within the passage is obtained by dividing the number of people counted by the passage area using a video target detection algorithm. This indicates the critical population density (usually 2-3 people / m², exceeding which traffic almost comes to a standstill). The average speed (m / s) of people moving within the passageway is calculated using optical flow or trajectory tracking. This represents the free-flow velocity (usually taken as 1.2~1.5 m / s). This represents the speed-affect factor, adjusting for the contribution of speed deviation to congestion. This represents the disorder of the directional flow field, quantifying the randomness of the pedestrian flow direction (value range [0,1]), and is obtained by calculating the entropy or standard deviation of the velocity direction vector. This represents the influence coefficient of directional confusion.

[0069] Analysis process:

[0070] When the density exceeds the critical value, the blockage coefficient increases rapidly; deviations from the optimal value in speed or confusion in direction will also exacerbate the blockage.

[0071] This coefficient is used to correct edges. In the rescue failure seepage model, the flux decay is related to the flow flux. Proportional.

[0072] The facility integrity coefficient is based on feedback from fire water system pressure and fire facility status sensors.

[0073] Definition: Characterizing fire protection facility nodes (Such as fire hydrants, sprinkler pumps, fire doors) at all times The degree of functional integrity.

[0074] Calculation formula:

[0075]

[0076] in, Indicates fire protection facility nodes At any moment The facility's condition rating This indicates the number of key components or status parameters affecting the facility (such as water pressure, valve status, and power supply status). Indicates the first Each component at time An indicator function for whether the function is invalid (0 for normal, 1 for invalid). Indicates the first The weight of individual component failures on overall function (e.g., a power outage of the water pump has a weight of 1, and low water pressure has a weight of 0.5). This indicates the time (in days) that has elapsed since the facility was last maintained or inspected. Indicates the aging degradation coefficient (preset according to facility type).

[0077] Analysis process:

[0078] The facility integrity coefficient is a continuous value of [0,1], where 1 represents complete normal operation and 0 represents complete failure.

[0079] The system uses IoT sensors to monitor water pressure, power supply, and status feedback in real time, and updates these data dynamically based on maintenance records.

[0080] If the facility involves multiple redundant components, a parallel reliability model can be used, which is simplified here to a product form to reflect the "bottleneck effect".

[0081] External environment correction coefficients are calculated based on wind speed, wind direction, temperature, and humidity data obtained through a meteorological interface.

[0082] Definition: Global factor, used to correct the probabilistic flux in the fire spread model, reflecting the impact of meteorological conditions on fire development.

[0083] Calculation formula:

[0084]

[0085] in, Indicates at time External environment correction factor The wind speed (m / s) is displayed in real time and is obtained from the meteorological interface. This represents the baseline wind speed (e.g., 5 m / s) used for normalization. This indicates the real-time outdoor temperature (°C). Indicates the reference temperature (e.g., 20℃). Relative humidity (%) , , This represents the adjustment coefficient, set based on fire dynamics experience (e.g.) , , .

[0086] Analysis process:

[0087] The higher the wind speed, the faster the fire spreads (but excessively strong winds may inhibit the spread in certain directions, which is simplified to a positive correlation here).

[0088] The higher the temperature, the easier it is for combustible materials to ignite;

[0089] The lower the humidity, the drier the air, which helps combustion.

[0090] This coefficient can be applied to the external environment correction term in the fire spread probability flux formula, for example, by multiplying the probability flux by... Or it can be used as an adjustment parameter.

[0091] S2. State Evolution Prediction: Input the heterogeneous information network graph into the spatiotemporal graph convolutional network model to predict the dynamic attribute weight changes of nodes and edges within a preset time window in the future, and generate the network state evolution trend.

[0092] The spatiotemporal graph convolutional network model adopts an encoder-decoder architecture. The encoder encodes historical observation sequences into hidden states through stacked spatiotemporal convolutional blocks, and the decoder predicts the dynamic attribute weights of nodes and edges at future time points in an autoregressive manner based on the hidden states. The spatiotemporal convolutional block includes a spatial graph convolutional layer and a temporal convolutional layer. The spatial graph convolutional layer is used to aggregate neighborhood node information to capture spatial dependencies, and the temporal convolutional layer uses dilated causal convolution combined with a gating mechanism to capture temporal evolution patterns.

[0093] Specifically, the node attributes (such as temperature, smoke concentration, and personnel density) and edge attributes (such as channel blockage coefficient) of the heterogeneous information network graph at multiple historical time steps are first organized into a three-dimensional tensor input model. The encoder consists of stacked spatiotemporal convolutional blocks. Each spatiotemporal convolutional block first aggregates the features of neighboring nodes through a spatial graph convolutional layer to capture spatial dependencies, then captures the temporal evolution law through an expanded causal convolutional layer, and introduces a gating mechanism to enhance the modeling capability. Finally, the historical observation sequence is encoded into a hidden state. Based on this hidden state, the decoder gradually predicts the node and edge attributes at each moment within a preset future time window in an autoregressive manner, and outputs the future network state evolution trend. The entire model is trained end-to-end under supervision by minimizing the mean square error between the predicted value and the true value, thereby achieving forward-looking prediction of dynamic parameters related to building fire risk.

[0094] S3. Critical seepage analysis: Based on the aforementioned evolution trend, a fire spread seepage model and a rescue failure seepage model are constructed. A graph cut algorithm is used to iteratively remove nodes or edges in the network and monitor changes in network connectivity components. When the removal of a few nodes or edges causes a sudden change in network connectivity, the critical point is identified and the corresponding node or edge is marked as a critical seepage node / edge.

[0095] The fire spread seepage model is used to simulate the propagation path and probability flux of fire in the network. Using a heterogeneous information network diagram of building fire protection as a carrier, heterogeneous nodes such as physical spaces, fire protection facilities, and personnel gathering areas are regarded as seepage grids. The heterogeneous edges formed by thermal conduction and physical connection between nodes are regarded as fire propagation channels. Dynamic attributes such as node combustible load, oxygen supply between nodes, and fire barrier effectiveness are used as core parameters to construct a calculation formula for the fire propagation probability flux that evolves over time. By setting the node ignition probability, edge conduction probability, and fire transmission intensity, the spatial spread process of fire in the building is transformed into a seepage phase change process, thereby simulating the path, intensity, and probability distribution of fire spreading from the ignition node to the surrounding area.

[0096] The rescue failure seepage model is used to simulate the flux attenuation of the fire resource network due to node failure or edge blockage. Taking fire resource supply, facility function dependence, and channel connectivity guarantee as the logical main line, the model takes key nodes and functional edges of fire water sources, sprinkler systems, evacuation routes, and fire protection units as the seepage subjects of the rescue network. Using facility integrity coefficient, channel blockage coefficient, and resource transmission efficiency as dynamic weights, the model constructs a seepage transmission mechanism that characterizes the attenuation of rescue capability. By simulating failure modes such as node failure, edge blockage, and resource flux attenuation, the model abstracts the process of the fire system losing its overall rescue capability due to local damage as the seepage breakdown process of the rescue network, thereby identifying the key weak links that lead to the paralysis of the rescue system.

[0097] In one specific embodiment, the probabilistic flux calculation method for the fire spread seepage model is as follows:

[0098]

[0099] in, express The fire was spreading from the node. Spread to adjacent nodes The probability, Represents a node Combustible load factor, Represents a node With nodes Oxygen supply coefficient between Indicates the barrier factor. , This indicates the adjustment parameter.

[0100] Combustible material load factor definition: characterizing the node The combined danger level of the quantity and combustion characteristics of combustibles at the current moment is the material basis for whether a fire can continue to spread.

[0101] Calculation formula:

[0102]

[0103] in, Represents a node The real-time combustible material mass (kg) within the building is obtained through one of the following methods: Static data: based on the preset values ​​of furniture and decoration materials in the building BIM model;

[0104] Dynamic correction: Identify newly added temporary flammable materials (such as stacked cardboard boxes) or IoT weighing sensors (such as warehouse shelves) through video surveillance.

[0105] This represents the critical mass of combustible material (kg), which is the minimum amount of combustible material required to sustain continuous combustion. It is set based on fire dynamics experience (for example, 50 kg / m² for a typical office area). This represents the weighted coefficient of the calorific value of combustibles, reflecting the combustion characteristics of different types of combustibles:

[0106]

[0107] in, Indicate the type of combustible material (such as wood, plastic, chemicals). Indicates the first The mass percentage of combustible materials, It is expressed as its calorific value per unit mass (MJ / kg);

[0108] This is expressed as the preset maximum calorific value coefficient (e.g., a typical value from a chemical warehouse). This represents the calorific value influence factor, which adjusts the contribution intensity of calorific value to the combustible load (usually taken as 0.2~0.5).

[0109] This represents the correction factor for the moisture content of combustible materials, estimated based on humidity sensors or meteorological data.

[0110]

[0111] in, Represents a node Relative humidity (%) This represents the moisture content decay coefficient (set based on material experimental data; for example, 0.02 can be used for wood).

[0112] Analysis process:

[0113] When the mass of combustible material Below the critical value hour, It is directly proportional to the mass; when the critical value is exceeded, the coefficient saturates to 1, indicating that there is sufficient combustible material and the spread of fire is no longer limited by the material.

[0114] Combustible materials with higher calorific value (such as chemicals) will have a higher load factor, reflecting their characteristic of releasing huge amounts of energy once they are burned.

[0115] The higher the humidity, the less likely flammable materials are to ignite; this physical law is reflected in the form of exponential decay.

[0116] Oxygen supply coefficient characterizes fire intensity from node Spread to nodes During the process, the adequacy of oxygen supply reflects the ability of ventilation conditions to support combustion.

[0117] Calculation formula:

[0118]

[0119] in, The effective ventilation area (m²) of the opening connecting nodes i and j is dynamically calculated based on the following data:

[0120] Static data: Door and window dimensions (from BIM model);

[0121] Dynamic correction: The door and window opening and closing status (open / closed / half-open) is obtained through the door magnetic sensor, and the opening angle is obtained through the smart window sensor;

[0122] The calculation formula is:

[0123] in, Indicates the first The total open area of ​​each opening. This indicates the opening coefficient (1 for fully open, 0 for fully closed, and 0.5 for half open).

[0124] This represents the minimum ventilation area (m²) required to maintain complete combustion, estimated based on the fire load and space volume (e.g., taking 1 / 20 of the space volume). This represents the critical population density that leads to significant oxygen consumption (e.g., 3 people / m²). 1 represents the oxygen consumption impact coefficient of personnel, reflecting the local oxygen consumption by high-density populations (usually taken as 0.3~0.6). This represents the real-time air volume (m³ / s) of the mechanical ventilation system, obtained from the building automation system. This indicates the design air volume (m³ / s) of the mechanical ventilation system. This represents the mechanical ventilation enhancement factor, which adjusts the contribution of the ventilation system to the oxygen supply (usually taken as 0.2~0.4).

[0125] Analysis process:

[0126] Oxygen supply depends primarily on the effective ventilation area of ​​physical openings; the larger the area, the more abundant the oxygen supply.

[0127] When the density of people in the passage is too high, human respiration will consume local oxygen. This effect is simulated by a linear decay term (but people are usually evacuated in a fire, so this effect may be short-lived).

[0128] If the mechanical ventilation system is operating normally, it can supplement the oxygen supply, but it is necessary to consider whether it is operating in fire mode (such as smoke exhaust mode, which may actually exhaust oxygen, so the sign needs to be adjusted according to the system mode, which is simplified to positive correlation here).

[0129] The upper limit of this coefficient is 1, which indicates that the oxygen supply is sufficient and the fire spreads without being restricted by oxygen; the lower limit is 0, which indicates that there is a complete lack of oxygen and the fire cannot spread.

[0130] The barrier factor calculation integrates the real-time effectiveness of three dimensions: passive fire prevention, active fire prevention, and human intervention.

[0131] Passive fire barriers are based on the fire resistance rating of building components and dynamically decrease with the duration of the fire and the state of damage to the components.

[0132] Active fire barriers quantify their suppression effectiveness based on the real-time operating status and malfunctions of sprinkler and smoke exhaust systems.

[0133] Manual intervention and obstruction take into account the impact of the number of on-site emergency personnel, response time, and accessibility on firefighting operations.

[0134] In one specific embodiment, the method for calculating the rescue flux attenuation coefficient of the rescue failure seepage model is as follows:

[0135]

[0136] in, Indicates the connection node and The channel at all times The channel blocking coefficient, Indicates the connection node and The channel at all times The integrity rating of fire protection facilities Indicates the connection node and The channel at all times The fire threat coefficient is obtained by accumulating the probability flux of the fire spread seepage model. Indicates the threat impact coefficient. This represents the fire threat normalization factor.

[0137]

[0138] in, Indicates from node To the node The set of all possible fire propagation paths In the fire spread and seepage model, the nodes represent the nodes. To the node The probability flux, Represents a node With nodes Graph distance (shortest path steps) in a network. This represents the spatial attenuation coefficient, with a value range of (0,1), typically between 0.5 and 0.8. Represents a node To the node The fireproof shielding effectiveness coefficient;

[0139]

[0140] in, Indicates the barrier factor. This indicates the threshold for complete blockage (a blockage factor exceeding this value is considered as fire being unable to pass).

[0141] S4. Resilience Entropy Calculation: Based on the predicted network state, the probability distribution of each fire compartment being in a safe state is defined, and the fire resilience entropy of the system is calculated to quantify the degree of orderliness of the system in resisting the spread of risks.

[0142] In one specific embodiment, the formula for calculating the fire resilience entropy is:

[0143]

[0144] in, Indicates the total number of fire compartments. Indicates the first Each partition The probability of maintaining a safe state at all times is obtained by normalizing the output layer of the spatiotemporal graph convolutional network model. An alert is triggered when the threshold is exceeded.

[0145] S5. Dynamic assessment: Based on the seepage critical node / edge and the fire resistance resilience entropy, the time window for the fire to spread to the preset key area is deduced, multi-dimensional dynamic risk assessment results are generated, and the dynamic risk assessment results are mapped to the building fire protection digital twin environment for visualization.

[0146] In one specific embodiment, the method for extrapolating the runaway time window includes: locking the key seepage path based on the seepage critical analysis results, combining the fire source location and the predicted network state, and using a reduced-order partial differential equation solver to calculate the shortest remaining time for the fire front to reach the boundary of the preset key area, which is used as the runaway time window.

[0147] Based on the critical seepage nodes / edges identified by S3, the key seepage paths are locked, combined with the preset ignition point in S5 or the most unfavorable ignition point determined by the risk identification algorithm, and the network state evolution trend predicted by S2.

[0148] In one specific embodiment, the method further includes an intervention strategy generation step based on causal inference: introducing a structural causal model, when the fire resilience entropy increases or the runaway time window is lower than the safety threshold, tracing back the root cause node or edge that leads to the deterioration of the risk state; combining reinforcement learning algorithms, with the goal of maximizing the delay of the runaway time window or reducing the system resilience entropy, outputting the optimal emergency intervention strategy.

[0149] In one specific embodiment, the optimal emergency intervention strategy includes: dynamically planned personnel evacuation routes that avoid predicted high-risk seepage areas in real time within a digital twin environment; and a fire resource dispatching scheme that prioritizes the seepage critical nodes / edges.

[0150] See Figure 2 As shown, a system for dynamic assessment of building fire risk based on big data is provided, including: a module for constructing a heterogeneous information network model: acquiring multi-source data of the target building, aggregating physical space units, fire protection facilities, and dynamic personnel within the building into heterogeneous nodes based on graph theory, constructing heterogeneous edges based on the physical connectivity, functional dependencies, and thermal conduction relationships between nodes, and assigning time-varying dynamic attribute weights to nodes and edges to form a building fire protection heterogeneous information network graph. ,in, For a set of nodes, Let be the set of edges. The weight matrix is ​​time-varying, and a digital twin environment for building fire protection that maps to the physical building is constructed based on this matrix.

[0151] State evolution prediction module: Input the heterogeneous information network graph into the spatiotemporal graph convolutional network model, predict the dynamic attribute weight changes of nodes and edges within a preset time window, and generate the network state evolution trend.

[0152] Critical seepage analysis module: Based on the evolution trend, a fire spread seepage model and a rescue failure seepage model are constructed. A graph cut algorithm is used to iteratively remove nodes or edges in the network and monitor the changes in the network connectivity components. When the removal of a few nodes or edges causes a sudden change in network connectivity, the critical point is identified and the corresponding node or edge is marked as a critical seepage node / edge.

[0153] Resilience Entropy Calculation Module: Based on the predicted network state, it defines the probability distribution of each fire compartment being in a safe state, calculates the fire resilience entropy of the system, and uses it to quantify the degree of orderliness of the system in resisting the spread of risks.

[0154] Dynamic assessment module: Based on the seepage critical node / edge and the fire resilience entropy, it extrapolates the time window for the fire to spread to the preset key area out of control, generates multi-dimensional dynamic risk assessment results, and maps the dynamic risk assessment results to the building fire protection digital twin environment for visualization.

[0155] In another specific embodiment:

[0156] Multi-source data acquisition and preprocessing

[0157] Deploy the following IoT sensors and data acquisition devices within the building:

[0158] Environmental sensors: Temperature sensors, smoke detectors, and humidity sensors are installed in each corridor, shop interior, and equipment room, totaling 320 sensors, with a sampling frequency of 1 time per minute.

[0159] Electrical fire monitoring: Residual current transformers and temperature sensors are installed in each distribution box and main power circuit, totaling 86.

[0160] Fire protection facility status monitoring: Pressure sensors (48) are installed in the fire hydrant network, water flow indicators (32) are installed in the sprinkler system, and door magnetic sensors (120) are installed in the fire doors.

[0161] Video surveillance: Utilizing existing security cameras (150 in total), deploy edge computing units to run a lightweight YOLOv5 target detection algorithm to identify personnel density, movement speed, directional flow field, and anomalies such as flames and smoke in real time.

[0162] Meteorological data interface: Connect to the local meteorological bureau's API to obtain real-time wind speed, wind direction, temperature, and humidity.

[0163] BIM Model: Import the building's as-built BIM model and extract static information such as floor plan layout, fire compartment division, location of fire protection facilities, fire resistance rating of materials, and door and window dimensions.

[0164] All sensor data is aggregated to an edge server via an IoT gateway for outlier removal, time alignment, and normalization. Video analytics results are stored as structured data (area number, timestamp, personnel density, average speed, etc.). BIM data is parsed into building topology via API. All data is stored uniformly in a time-series database, providing input for subsequent models.

[0165] Constructing a heterogeneous information network model

[0166] Based on BIM data and real-time dynamic information, a heterogeneous information network diagram is constructed.

[0167] The node set includes all physical space units (approximately 450 nodes per shop, corridor, stairwell, and equipment room), fire protection facility nodes (approximately 200 nodes for fire hydrants, sprinkler pumps, fire extinguishers, etc.), and dynamic personnel nodes (approximately 30 virtual personnel nodes aggregated by area on each floor). Each node has type attributes (room / corridor / facilities / personnel) and static attributes (area, fire resistance rating, etc.).

[0168] Edge set: Edges are constructed based on the connectivity of buildings, including:

[0169] Physical connectivity edges: Rooms are connected by doors and corridors. Connectivity relationships are extracted based on the BIM model, and each edge records the connectivity type (door / wall / window), opening size, etc.

[0170] Functional dependency edges: The dependencies between fire hydrants and water supply networks, and between sprinkler heads and sprinkler pumps, are extracted from fire protection design drawings.

[0171] Thermal conduction edge: The thermal relationship between adjacent rooms through the wall, established based on the thermal parameters of the materials.

[0172] Dynamic attribute weights: Time-varying weights are assigned to each node and edge, and the calculation method is as described in the foregoing dependent claims.

[0173] Meanwhile, based on the three-dimensional geometric information and texture data of the BIM model, a digital twin environment that is mapped 1:1 to the physical building is constructed, and a heterogeneous network diagram is embedded in it as the underlying data model.

[0174] Spatiotemporal Graph Convolutional Network Model Training and State Evolution Prediction

[0175] Collect sensor data, video analysis results, and actual event records (including false fire alarms, small-scale fires, drill data, etc.) from the past three months to construct a training dataset.

[0176] Model architecture: A spatiotemporal graph convolutional network employing an encoder-decoder structure. The encoder consists of four stacked spatiotemporal convolutional blocks, each containing:

[0177] Spatial graph convolutional layer (first-order ChebNet, K=3), input feature dimension F=16, output dimension 64;

[0178] Temporal convolutional layers (dilated causal convolutions with dilation rates of 1, 2, 4, and 8), kernel size 3;

[0179] Gated linear units and residual connections.

[0180] The decoder has a symmetrical structure and a fully connected output layer. It outputs the predicted values ​​of key attributes (temperature, smoke concentration, personnel density, facility integrity, etc.) of each node for the next 6 time steps (each time step is 5 minutes, i.e., the next 30 minutes) and the channel blockage coefficient of each edge.

[0181] Training: Data from the past 12 time steps (1 hour) is used as input, and the next 6 time steps are used as the prediction target. The loss function is mean squared error, the optimizer is Adam, the learning rate is 0.001, the batch size is 32, and the training is conducted for 50 epochs. The early stopping mechanism is based on the validation set loss. After training, the model is deployed on an edge server to receive the latest data in real time and predict the state evolution trend for the next 30 minutes every 5 minutes.

[0182] Construction of a dual-flow model and critical flow analysis

[0183] Based on the predicted network state evolution trend, a fire spread seepage model and a rescue failure seepage model were constructed respectively.

[0184] Critical seepage analysis: A graph cut algorithm is used, taking the current potential ignition point (such as a node in a shop where the temperature rises abnormally) as the fire source. Edges with a probability flux higher than the threshold (0.3) (simulating fire spread) and edges with a rescue flux attenuation coefficient higher than the threshold (0.7) (simulating rescue channel failure) are iteratively removed from the network, and changes in the network connectivity components are monitored. When the network connectivity components change abruptly after removing one or more edges (for example, the fire network changes from local spread to overall connectivity, or the rescue network changes from connectivity to complete interruption), these edges are marked as critical seepage edges, and the corresponding nodes may also be marked as critical seepage nodes.

[0185] For example, one analysis found that when the edge corresponding to the fire door between fire compartment 3 and fire compartment 4 on the third floor (which has a high probability flux prediction value) was removed, the fire spread rapidly to the entire third floor. In this case, the fire door was identified as a critical edge for seepage. At the same time, if a fire hydrant node causes a sharp increase in the attenuation coefficient of the rescue flux of multiple surrounding edges due to low predicted water pressure, the fire hydrant node is marked as a critical node for seepage.

[0186] Fire resistance resilience entropy calculation

[0187] Calculate the fire resilience entropy of the entire building system based on the predicted network state.

[0188] Out-of-control time window simulation

[0189] Suppose that an abnormal temperature rise is detected in a shop on the third floor, triggering a fire alarm confirmation. The system uses this as the ignition point and, combined with the predicted network state evolution trend and seepage criticality analysis results, identifies key seepage paths (i.e., the main spread paths with high probability flux starting from the ignition point). Using a reduced-order partial differential equation solver (based on a fast prediction model simplified from the fire dynamics FDS model), and inputting the ignition point location, current combustible material distribution, ventilation conditions, and barrier factors, the system quickly calculates the shortest remaining time for the fire front to reach preset critical areas (such as the fourth-floor children's activity area, the basement electrical room, and the fire control room).

[0190] For example, the system calculates that the fire will spread eastward along the third-floor corridor, ascend vertically through the stairwell, and is expected to reach the boundary of the fourth-floor children's activity area in 8 minutes and 30 seconds; it will also spread westward along another path, and is expected to reach the hazardous materials warehouse on the other side of the third floor in 5 minutes and 20 seconds. The system outputs these two times as out-of-control time windows and updates them dynamically.

[0191] Digital twin visualization

[0192] All dynamic assessment results are mapped onto the digital twin environment in real time: each area is rendered in a different color (green-yellow-red gradient) based on the fire resilience entropy value, forming a risk heat map;

[0193] Critical seepage nodes / edges are highlighted in flashing red and labeled with their type (e.g., "3F-Fire Door-12" indicates a critical seepage point).

[0194] The predicted fire spread path is represented by dynamically flowing arrows, with the direction of the arrows indicating the direction of spread;

[0195] The time window for the loss of control is attached above the key area in the form of a countdown label;

[0196] The current distribution of people is displayed in the form of a particle cloud, with areas of higher density being darker in color.

[0197] The fire control room displays the 3D situation map in real time on a large screen, and managers can interactively view specific forecast data for any area.

[0198] Intervention strategy generation based on causal inference

[0199] When the time window for loss of control is lower than the safety threshold (e.g., less than 10 minutes to reach the children's activity area) or the fire resilience entropy continues to rise, the system activates the intervention strategy generation module.

[0200] Causal attribution: A structural causal model is introduced to calculate the contribution of each factor to the deterioration of risk based on historical data and the current state. For example, system analysis shows that the main reasons for the shortened time window of loss of control are "the failure of the No. 3 fire door on the third floor to close properly due to the accumulation of debris" (high passage blockage coefficient) and "the failure of the smoke exhaust fan on the fourth floor" (low facility integrity coefficient). These two root cause nodes / edges are marked and pushed to the management personnel.

[0201] Reinforcement Learning Optimization: In a digital twin environment, deep reinforcement learning algorithms (such as PPO) are used for rapid simulation. The agent aims to maximize the delay of the runaway time window and reduce the system entropy, trying different combinations of intervention actions (such as "closing the fire door," "starting the backup smoke exhaust fan," and "evacuating people on the east side of the third floor"). After thousands of rapid simulations, the system outputs the optimal strategy:

[0202] Evacuation routes: Three main evacuation routes are dynamically planned and marked with green arrows in the digital twin environment. The routes avoid predicted high-risk seepage areas in real time (e.g., avoid corridors that are about to be spread by fire).

[0203] Firefighting resource dispatch plan: Prioritize dispatching firefighters to the critical seepage point (fire door No. 3 on the third floor and smoke exhaust fan room on the fourth floor) for handling; at the same time, instruct the mini fire station to carry fire extinguishing equipment to stand by near the fire point.

[0204] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A dynamic assessment method for building fire safety risks based on big data, characterized in that, include: S1. Constructing a heterogeneous information network model: Acquire multi-source data of the target building, aggregate physical space units, fire protection facilities, and dynamic personnel within the building into heterogeneous nodes based on graph theory, construct heterogeneous edges based on the physical connectivity, functional dependencies, and thermal conduction relationships between nodes, and assign time-varying dynamic attribute weights to nodes and edges to form a heterogeneous information network graph for building fire protection. ,in, For a set of nodes, Let be the set of edges. A time-varying weight matrix is ​​used as the basis for constructing a digital twin environment for building fire protection that maps to the physical building. S2. State Evolution Prediction: Input the heterogeneous information network graph into the spatiotemporal graph convolutional network model to predict the dynamic attribute weight changes of nodes and edges within a preset time window in the future, and generate the network state evolution trend. S3. Critical seepage analysis: Based on the aforementioned evolution trend, a fire spread seepage model and a rescue failure seepage model are constructed. A graph cut algorithm is used to iteratively remove nodes or edges in the network and monitor changes in the network connectivity components. When the removal of a few nodes or edges causes a sudden change in network connectivity, the critical point is identified and the corresponding node or edge is marked as a critical seepage node / edge. S4. Resilience Entropy Calculation: Based on the predicted network state, the probability distribution of each fire compartment being in a safe state is defined, and the fire resilience entropy of the system is calculated to quantify the degree of orderliness of the system in resisting the spread of risks. S5. Dynamic assessment: Based on the seepage critical node / edge and the fire resistance resilience entropy, the time window for the fire to spread to the preset key area is deduced, multi-dimensional dynamic risk assessment results are generated, and the dynamic risk assessment results are mapped to the building fire protection digital twin environment for visualization.

2. The method for dynamic assessment of building fire risk based on big data according to claim 1, characterized in that: The multi-source data of the target building includes building information model (BIM) data, real-time monitoring data from IoT sensors, video surveillance stream data, and static attribute data; The BIM data includes at least the building structure dimensions, fire compartment division, fire protection facility layout, and geometric and physical information of material properties, which are parsed and extracted through the BIM interface. IoT sensor data includes at least multiple types of data such as temperature, smoke, humidity, electrical, and pressure. Edge computing technology is used for real-time preprocessing to remove abnormal data and noise. The video surveillance stream data uses a lightweight target detection algorithm to identify and extract features of people, flames, and smoke, reducing the burden of data transmission and processing. Static attribute data should include at least the building's service life, fire protection design rating, and historical fire data, which are used to initialize the network model and calibrate parameters.

3. The method for dynamic assessment of building fire safety risks based on big data according to claim 2, characterized in that: The dynamic attribute weights include at least one of the following or a combination thereof: Fire load dynamic coefficient calculated based on real-time temperature, smoke concentration, and electrical parameters collected by IoT sensors; The channel blockage coefficient is calculated based on the flow field of personnel density, movement speed, and direction identified by the target detection algorithm in the video surveillance stream; The facility integrity coefficient is based on feedback from fire water system pressure and fire facility status sensors. External environment correction coefficients are calculated based on wind speed, wind direction, temperature, and humidity data obtained through a meteorological interface.

4. The method for dynamic assessment of building fire safety risks based on big data according to claim 3, characterized in that: The probabilistic flux calculation method for the fire spread seepage model is as follows: in, express The fire was spreading from the node. Spread to adjacent nodes The probability, Represents a node Combustible load factor, Represents a node With nodes Oxygen supply coefficient between Indicates the barrier factor. , This indicates the adjustment parameter.

5. The method for dynamic assessment of building fire safety risks based on big data according to claim 4, characterized in that: The method for calculating the rescue flux attenuation coefficient of the rescue failure seepage model is as follows: in, Indicates the connection node and The channel at all times The channel blocking coefficient, Indicates the connection node and The channel at all times The integrity rating of fire protection facilities Indicates the connection node and The channel at all times The fire threat coefficient is obtained by accumulating the probability flux of the fire spread seepage model. Indicates the threat impact coefficient. This represents the fire threat normalization factor.

6. The method for dynamic assessment of building fire safety risks based on big data according to claim 5, characterized in that: The formula for calculating the fire resistance resilience entropy is: in, Indicates the total number of fire compartments. Indicates the first Each partition The probability of maintaining a safe state at all times is obtained by normalizing the output layer of the spatiotemporal graph convolutional network model. An alert is triggered when the threshold is exceeded.

7. The method for dynamic assessment of building fire risk based on big data according to claim 6, characterized in that: The method for extrapolating the out-of-control time window includes: Based on the critical seepage analysis results, the key seepage path is identified. Combining the fire source location and the predicted network state, a reduced-order partial differential equation solver is used to calculate the shortest remaining time for the fire front to reach the boundary of the preset key area, which is used as the out-of-control time window.

8. The method for dynamic assessment of building fire risk based on big data according to claim 7, characterized in that: It also includes an intervention strategy generation step based on causal inference: introducing a structural causal model, when the fire resilience entropy increases or the runaway time window is lower than the safety threshold, tracing back to the root cause node or edge that leads to the deterioration of the risk state; combining reinforcement learning algorithms, with the goal of maximizing the delay of the runaway time window or reducing the system resilience entropy, outputting the optimal emergency intervention strategy.

9. The method for dynamic assessment of building fire risk based on big data according to claim 8, characterized in that: The optimal emergency intervention strategy includes: dynamically planned personnel evacuation routes that avoid predicted high-risk seepage areas in real time within the digital twin environment; and a fire resource dispatching scheme that prioritizes the seepage critical nodes / edges.

10. A system for implementing the big data-based dynamic assessment method for building fire risk as described in any one of claims 1-9, characterized in that: include: The module for constructing a heterogeneous information network model involves acquiring multi-source data of the target building, aggregating physical space units, fire protection facilities, and dynamic personnel within the building into heterogeneous nodes based on graph theory, constructing heterogeneous edges based on the physical connectivity, functional dependencies, and thermal conduction relationships between nodes, and assigning time-varying dynamic attribute weights to nodes and edges to form a heterogeneous information network graph for building fire protection. ,in, For a set of nodes, Let be the set of edges. A time-varying weight matrix is ​​used as the basis for constructing a digital twin environment for building fire protection that maps to the physical building. State evolution prediction module: Input the heterogeneous information network graph into the spatiotemporal graph convolutional network model, predict the dynamic attribute weight changes of nodes and edges within a preset time window in the future, and generate the network state evolution trend; Seepage Critical Analysis Module: Based on the aforementioned evolution trend, a fire spread seepage model and a rescue failure seepage model are constructed. A graph cut algorithm is used to iteratively remove nodes or edges in the network and monitor changes in network connectivity components. When the removal of a few nodes or edges causes a sudden change in network connectivity, the critical point is identified and the corresponding node or edge is marked as a seepage critical node / edge. Resilience Entropy Calculation Module: Based on the predicted network state, it defines the probability distribution of each fire compartment being in a safe state, calculates the fire resilience entropy of the system, and uses it to quantify the degree of orderliness of the system in resisting the spread of risks. Dynamic assessment module: Based on the seepage critical node / edge and the fire resilience entropy, it extrapolates the time window for the fire to spread to the preset key area out of control, generates multi-dimensional dynamic risk assessment results, and maps the dynamic risk assessment results to the building fire protection digital twin environment for visualization.

Images