Industrial enterprise typical fire scene construction system based on multi-source data fusion

Abstract

The application relates to the technical field of industrial safety monitoring, in particular to an industrial enterprise typical fire scene construction system based on multi-source data fusion. The application proposes the following scheme: a system acquires building information model, image monitoring, equipment state, personnel and vehicle trajectory and other multi-source data of a factory area, identifies and extracts a primitive set representing a temporary state, projects the primitive set to corresponding nodes and edges of a factory area reference reachability graph, and generates a restricted reachability graph. The system combines an emergency control target, adopts direction-sensitive morphological calculation to determine a passing and restricted area, filters an executable action sequence meeting time limit and safety constraints from the restricted reachability graph, constructs a control boundary of a fire scene, and realizes dynamic deduction and visualization of fire emergency response. The system can reflect changes of passing constraints in a complex industrial environment in real time, and improve the accuracy and reliability of emergency decision and fire disposal.

Description

Technical Field

[0001] This application relates to the field of industrial safety monitoring technology, and in particular to a system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion. Background Technology

[0002] Existing industrial enterprise fire scenario construction and disaster prevention decision-making systems largely rely on static plant structure models or single sensor data, lacking a comprehensive representation of temporary plant conditions and multi-source dynamic information. This leads to discrepancies between the generated fire scenarios and the actual emergency environment. Traditional fire simulations are mostly based on fixed channels and preset parameters, failing to reflect unstructured constraints such as construction access, equipment maintenance, and temporary closures, resulting in distorted emergency access paths. While existing monitoring systems can collect images and status data, they lack a unified spatial projection and accessibility mapping mechanism, making it difficult to translate real-time obstacles, smoke flow, or wind direction changes into firefighting action constraints. Furthermore, existing algorithms generally employ isotropic path search or simple topology filtering methods, failing to consider comprehensive factors such as traffic directionality, hydraulic limitations, and operational conditions. This results in a mismatch between fire emergency dispatch results and on-site executability, making it difficult to support dynamic fire response and decision-making in complex industrial plants.

[0003] To address the above issues, this application presents a system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion. Summary of the Invention

[0004] The technical problem this application aims to solve is to address the shortcomings of existing technologies by providing a system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion. The system acquires multi-source data from the factory area, including building information models, image monitoring, equipment status, and personnel and vehicle trajectories. It identifies and extracts a set of primitives representing temporary states and projects these primitives onto corresponding nodes and edges of the factory's baseline accessibility map to generate a restricted accessibility map. Combining emergency control objectives, the system uses direction-sensitive morphological calculations to determine passable and restricted areas. From the restricted accessibility map, it selects executable action sequences that meet timeliness and safety constraints to construct the control boundaries of the fire scenario, enabling dynamic simulation and visualization of fire emergency response.

[0005] To achieve the above objectives, this application provides the following technical solution: A system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion is applied to a disaster prevention and monitoring platform. The platform is used to update the control boundaries of fire scenarios and to construct and control typical fire scenarios in real time. The platform maintains a baseline reachability graph. The system includes: The multi-source data acquisition module is used to acquire multi-source data related to the factory area, including text data, image data, trajectory data, and monitoring status data. A temporary state identification module is used to identify and extract a set of primitives for characterizing temporary states based on the multi-source data. A restricted graph generation module is used to project the primitive set onto the corresponding nodes and edges of the baseline reachability graph to generate a restricted reachability graph; The control boundary reasoning module is used to combine preset emergency control objectives and filter executable action sequences from the restricted accessibility map to obtain the control boundary of the corresponding fire scenario.

[0006] The method for constructing the baseline reachability graph includes: Based on the building information model, fire protection system drawings, and equipment ledger data of the plant area, fire protection facilities, passages, doors, staircases, valves, shut-off devices, and hazard sources within the plant area are identified as nodes on the baseline accessibility map. According to the preset activity rules, establish the connection relationship between nodes and generate corresponding edges, where the edges are used to represent action paths; According to the preset response rules, constraint parameters are set for each edge. The constraint parameters include geometric constraints, hydraulic constraints, and operation method constraints. The geometric constraints are used to describe the clearance height, width, turning radius, slope, and passage distance of the channel. The hydraulic constraints are used to describe the pressure loss, flow capacity, minimum water supply pressure, and jet range of the fire water supply network. The operation method constraints are used to describe the number of personnel required for operation, operation sequence, and safety conditions. The nodes, edges, and constraint parameters are stored as a baseline reachability graph.

[0007] The temporary state identification module includes: The text parsing unit is used to perform semantic analysis on the maintenance work orders, temporary work permits and lock-up records in the text data, identify semantic entities that match the terms "road occupation", "closure", "energy removal", "width restriction", "height restriction" and "displacement", extract the corresponding work area, facility identification, start time and end time, and generate text primitives that include spatial range and timeliness attributes. The image recognition unit is used to extract targets from the image data using a target detection algorithm, obtain obstacle targets, calculate the projection range and height information of the obstacle targets in the factory area spatial coordinate system, and convert them into geometric primitives representing temporary occupation and restricted passage status. The trajectory clustering unit, based on the vehicle positioning trajectory and personnel positioning trajectory in the trajectory data, identifies the corresponding dense areas through a clustering algorithm and generates trajectory primitives; The status event unit detects changes in the opening and closing status of facilities based on the opening and closing signals of access control, roller shutters, valves, and shut-off devices in the monitored status data, and generates status primitives.

[0008] The restricted graph generation module includes: The spatial mapping unit maps the primitive set into a dynamic mask layer based on the factory area spatial coordinate system to represent the distribution of road occupancy, enclosure, width restriction, height restriction and obstacles in the spatial and temporal dimensions; The passage core calculation unit is used to obtain the anisotropic passage core corresponding to each side and matched with the direction based on the corresponding action characteristics of personnel passage, equipment transportation and water hose laying. The shape and size of the anisotropic passage core are determined by the net width of the passage, the corner radius, the slope and the minimum passage radius. A morphological calculation unit is used to perform morphological calculations on the dynamic mask layer and the anisotropic passage kernel to obtain a passage window layer. The morphological calculation determines that the passage is not passable in the overlapping area of ​​the anisotropic passage kernel and the dynamic mask layer, and that it is passable in the non-overlapping area. The accessibility labeling unit is used to project the windowing results of the access windowing layer onto the nodes and edges of the baseline accessibility map, and to mark the corresponding nodes and edges as enabled, flow-limited, or disabled in combination with the availability index of the access windowing layer, so as to obtain a restricted accessibility map.

[0009] The restricted graph generation module further includes an update strategy, which is used to dynamically adjust the baseline reachability graph when the primitive set changes. The update strategy also includes mapping logic and decision logic, with the mapping logic configured in the spatial mapping unit and the decision logic configured in the morphological calculation unit.

[0010] The mapping logic is configured as follows: Perform coordinate transformation and spatial alignment on the spatial description information, location information and target coordinate information carried by each primitive in the primitive set; Based on the primitive type, corresponding graphic elements are generated, and based on the time-effect attribute, effective time and expiration time are assigned to each graphic element in the time dimension to obtain a spatiotemporal mask sequence. Primitives belonging to the road occupation and closure generate closed polygon regions, primitives belonging to the width restriction and height restriction generate semi-closed polygon regions, and primitives belonging to the obstacle generate volume regions containing height information. The spatiotemporal mask sequences are superimposed and fused according to their spatial positions to obtain a dynamic mask layer.

[0011] The determination logic is configured as follows: Based on the spatial coordinate system of the factory area, a direction field is constructed to guide morphological operations based on the anisotropic access kernel. The direction field is obtained by fusing the opening and closing status of access control and roller shutters, the operating status of ventilation and smoke exhaust equipment, wind speed and differential pressure monitoring data, real-time trajectories of vehicles and personnel, and smoke plume flow clues. Based on the direction field and the action type of each side, a corresponding convection structure element is generated. The convection structure element is stretched along the main direction of the direction field and its shape is corrected at the corner. Based on the convection structure element, erosion and expansion operations are performed on the dynamic mask layer along the streamlines of the directional field to obtain a passage window layer. When the barrier boundary of the dynamic mask layer is consistent with the main direction of the directional field, erosion operation is performed; when the barrier boundary of the dynamic mask layer is inconsistent with the main direction of the directional field, expansion operation is performed.

[0012] Based on the convection structure element, erosion and expansion calculations are performed on the dynamic mask layer along the streamlines of the directional field to obtain a passage window layer, including: Guided by the streamlines of the directional field, a cluster of streamlines adjacent to each side is selected in the dynamic mask layer; The streamline cluster is slid by the convection structure element, and the overlap ratio between the convection structure element and the barrier boundary is calculated at each sliding position to obtain the local impedance value. The local impedance value is then used to determine whether the streamline cluster is downstream. If the streamline cluster is in the direction of flow, the sliding positions are processed by erosion operation and then connected to obtain the smooth flow channel; If the streamline cluster is countercurrent, the sliding position is processed by expansion calculation and buffering operation is performed to obtain the stagnation range; By traversing all streamline clusters, the smooth passageway and the obstruction range are merged to obtain the passage window layer.

[0013] The control boundary inference module includes: The target analysis unit analyzes the emergency control target based on the emergency response strategy library stored in the disaster prevention and monitoring platform, and determines the type, priority and time constraints of the target action, wherein the target action includes water outlet coverage, valve shut-off, isolation closure, sprinkler start and stop and evacuation passage. A path filtering unit is used to filter a set of candidate paths that are adapted to the type and priority in the restricted reachability graph based on the enable, rate limiting and disable flags of nodes and edges, wherein the set of candidate paths is used to represent the spatial and sequential combination of executable action sequences. The optimization and solution unit is used to search for the executable action sequence with the minimum comprehensive cost and that meets the time constraints in the candidate path set using a multi-objective optimization algorithm; The boundary calculation unit is used to calculate the parameters of the control boundary and generate a control boundary model for fire scenario construction based on the executable action sequence.

[0014] The parameters of the control boundary include the shortest executable time, minimum water supply flow rate, remaining pressure margin, covered spatial range, and corresponding inaccessible areas.

[0015] Compared with the prior art, the beneficial effects of this application are: This application achieves a unified representation of industrial enterprise fire scenarios from static topology to dynamic accessibility through a collaborative design involving multi-source data fusion, spatial constraint modeling, and direction-sensitive morphological computation. By establishing a baseline accessibility map and integrating building information models, image monitoring, equipment status, and trajectory data of the plant area, it can reflect unstructured changes such as temporary road occupancy, closures, width restrictions, and height restrictions in real time within the same coordinate system, constructing a restricted accessibility map that is updated synchronously with actual working conditions. Based on morphological computation guided by directional fields, the fire scenario possesses continuous and gradual accessibility assessment capabilities in airflow, personnel evacuation, and water supply directions, significantly improving the physical consistency and computational accuracy of scenario modeling. Attached Figure Description

[0016] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 An exemplary application scenario diagram provided for an embodiment of this application; Figure 2 A schematic diagram of the modules of a typical fire scenario construction system for industrial enterprises based on multi-source data fusion provided in this application embodiment; Figure 3 A flowchart illustrating the method for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion, as provided in this application embodiment; Figure 4 A flowchart illustrating the steps for generating a restricted reachability graph provided in this application embodiment. Detailed Implementation

[0017] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0018] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] Those skilled in the art will understand that the following common problems typically arise when a typical fire scenario in an industrial enterprise is implemented on-site: Production continues uninterrupted and operating conditions are constantly changing. Temporary conditions such as scaffolding, barriers, temporary stacks, vehicles in storage, valve locking tags, and ventilation switching frequently appear and disappear between shifts, directly altering the accessibility of people, equipment, and water, as well as the operability of valves and shut-off devices.

[0020] Traditional scenario construction often relies on fixed template libraries or static lists of available paths and facilities generated from drawings, combined with fire detection results for extrapolation. This approach works when operating conditions are stable, but once maintenance and repair windows, high-turnover periods in warehousing, or interlocking strategy switching phases begin, the assumptions of unobstructed passages and accessible facilities in the plan quickly deviate from the actual situation on site. The scenario extrapolation's judgment of achievability and controllability is systematically distorted, thereby affecting the reliability of decisions regarding the loss-of-control threshold and response sequence.

[0021] The core logic of this application is based on the aforementioned problem. Instead of changing the template or adding detection types, it transforms the problem of temporary states affecting control boundary offsets into a computable process.

[0022] Specifically, this embodiment is aimed at a disaster prevention monitoring platform with multi-source operational data and video surveillance. It abstracts heterogeneous data such as work orders, tickets, access control and valve position status, vehicle and personnel trajectories, and video recognition results into primitives that can be processed uniformly. These primitives carry spatial range and time-sensitivity attributes and can be directly projected onto the benchmark accessibility map of the plant area to form spatiotemporal constraints that evolve over time.

[0023] To avoid simplifying the problem to a static process of edge deletion or weighting, the processing logic in this embodiment also includes a computational approach consistent with the on-site physics: The directional field is formed by integrating elements such as ventilation and smoke exhaust operation, access control opening and closing, human and material flow organization, and smoke plume direction. Then, an anisotropic passage kernel that matches the action type is used to perform morphological calculations on this directional field, so that the judgment of "passable / unpassable" and "hydraulic arrival / inability to arrive" is spatially directionally related and temporally hysteretic and stable for temporary states.

[0024] Furthermore, the platform only filters action sequences that meet the time and operation method constraints on the restricted accessibility graph, and provides quantitative results of control boundaries such as the shortest executable time, minimum water supply and remaining pressure margin, coverage area and inaccessibility certificate, and uses them as boundary conditions for typical fire scenarios for instantiation and online updating.

[0025] To facilitate understanding of the field adaptability of this application, several typical scenarios are described below.

[0026] For example, when scaffolding is erected for maintenance and partial valve tagging is carried out in a petrochemical tank area at night, the process constraint of unblocking the passage before passage becomes a decisive factor, even though the passage geometry remains unchanged. In this embodiment, the primitives are used to perform morphological processing on the projection and fitting direction field to provide an executable action sequence and corresponding time window for moving obstacles first and then passing through, so that the control boundary of the overflow pool fire scene is correctly tightened during construction.

[0027] When operating in high-level aisles in automated warehouses, temporary pallet stacking restricts width and turning angles. At the same time, the pump pressure drops at night, reducing hydraulic margin. This embodiment reflects the difference between seemingly passable but insufficient range on both geometric and hydraulic levels, thereby constraining and screening the high-level fire templates on the shelves, retaining only the solutions that can be executed in the current shift.

[0028] Within the battery module assembly line, the dense passage of AGVs and the linkage with the door curtain can easily cause local wind field reversal and flue gas backflow. In this embodiment, directional field guidance calculation is used to erode and expand the accessibility of the forward and reverse flow sides to different degrees, so that the propagation direction and critical time determination of the thermal runaway scenario are close to reality.

[0029] It is understood that this embodiment does not require a specific brand of sensor to be used in a particular factory area, nor is it based on fixed spatial partitioning or static cavity division. The key point is: The temporary state of the site is abstracted into primitives with spatiotemporal boundaries and operational meanings. A morphological process is guided by a direction field consistent with the physical state of the site to compile it into a computable accessibility change. Then, based on this, the control boundary is derived and the scene instantiation is driven.

[0030] Based on this, a typical fire scenario is no longer just about identifying what kind of fire is happening, but rather a scenario that, at this time and place, under these temporary constraints, is indeed possible and for how long can it be controlled; this transformation from seeing to doing is the source of the problem and the basis of the principle that this embodiment aims to solve.

[0031] refer to Figure 1 , Figure 1 This is an exemplary application scenario diagram provided for an embodiment of this application.

[0032] Figure 1 The overall application framework of this application in the disaster prevention and monitoring system is shown. As shown in the figure, sensors 1, 2 to N deployed in the factory area are used to collect multi-source data related to disaster prevention and monitoring, including but not limited to environmental monitoring data, equipment operation data, video image data, access control status information, and personnel and vehicle positioning information.

[0033] Various sensors transmit the collected data to the disaster prevention and monitoring platform in real time. The disaster prevention and monitoring platform is used to perform fusion processing, primitive recognition and projection calculation on the multi-source data, and maintain the dynamic update of the baseline accessibility map and the restricted accessibility map.

[0034] The disaster prevention monitoring platform is connected to a processor, which executes the algorithm logic of this application, including steps such as dynamic mask layer generation, anisotropic passage kernel construction, morphological calculation, and control boundary solution. Based on the calculation results, the processor generates control boundary models for typical fire scenarios and transmits corresponding control commands or early warning information to the controller.

[0035] Based on the output of the disaster prevention and monitoring platform, the controller performs coordinated control of on-site fire-fighting facilities, valve shut-off devices, sprinkler systems, smoke exhaust systems, and emergency evacuation instructions to achieve automatic response and safety decision-making in fire scenarios.

[0036] It is understood that the processor in this application can be understood as an electronic processing unit used to execute the algorithm instructions, data fusion logic, and scene construction program configured in the disaster prevention and monitoring platform. Specific implementations may include a central processing unit (CPU), graphics processing unit (GPU), digital signal processor (DSP), field-programmable gate array (FPGA), or other embedded processing modules with computing power support. These modules can perform functions such as parsing multi-source data, feature extraction, morphological operations, and control boundary calculations through software, firmware, or hardware circuitry.

[0037] Similarly, the controller in this application can be understood as an execution control unit used to execute control instructions and coordinate with field devices based on the calculation results of the processor. Specific implementations may include programmable logic controllers (PLCs), industrial control computers (IPCs), distributed control systems (DCS), control modules, smart gateways, or other devices with output control capabilities. These devices can interact with fire protection facilities, sprinkler systems, smoke exhaust systems, valve shut-off devices, and evacuation indicator equipment via fieldbus, Ethernet, or wireless communication for signal exchange and status control.

[0038] It should be noted that the specific form of the processor and controller can be selected and combined according to the configuration and application requirements of the factory disaster prevention system, and this application does not limit it; in different implementations, the processor and controller can be independent hardware units, or they can be integrated into the same industrial control platform or edge computing node, which will not be elaborated here.

[0039] refer to Figure 2 , Figure 2 A schematic diagram of the modules of a typical fire scenario construction system for industrial enterprises based on multi-source data fusion provided in this application embodiment.

[0040] In one example, this application embodiment provides a system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion, applied to a disaster prevention and monitoring platform. The disaster prevention and monitoring platform is used to refresh the control boundaries of fire scenarios and to construct and control typical fire scenarios in real time. The disaster prevention and monitoring platform maintains a baseline reachability graph. The system includes: The multi-source data acquisition module is used to acquire multi-source data related to the factory area, including text data, image data, trajectory data, and monitoring status data. A temporary state identification module is used to identify and extract a set of primitives for characterizing temporary states based on the multi-source data. A restricted graph generation module is used to project the primitive set onto the corresponding nodes and edges of the baseline reachability graph to generate a restricted reachability graph; The control boundary reasoning module is used to combine preset emergency control objectives and filter executable action sequences from the restricted accessibility map to obtain the control boundary of the corresponding fire scenario.

[0041] The temporary state identification module includes: The text parsing unit is used to perform semantic analysis on the maintenance work orders, temporary work permits and lock-up records in the text data, identify semantic entities that match the terms "road occupation", "closure", "energy removal", "width restriction", "height restriction" and "displacement", extract the corresponding work area, facility identification, start time and end time, and generate text primitives that include spatial range and timeliness attributes. The image recognition unit is used to extract targets from the image data using a target detection algorithm, obtain obstacle targets, calculate the projection range and height information of the obstacle targets in the factory area spatial coordinate system, and convert them into geometric primitives representing temporary occupation and restricted passage status. The trajectory clustering unit, based on the vehicle positioning trajectory and personnel positioning trajectory in the trajectory data, identifies the corresponding dense areas through a clustering algorithm and generates trajectory primitives; The status event unit detects changes in the opening and closing status of facilities based on the opening and closing signals of access control, roller shutters, valves, and shut-off devices in the monitored status data, and generates status primitives.

[0042] The restricted graph generation module includes: The spatial mapping unit maps the primitive set into a dynamic mask layer based on the factory area spatial coordinate system to represent the distribution of road occupancy, enclosure, width restriction, height restriction and obstacles in the spatial and temporal dimensions; The passage core calculation unit is used to obtain the anisotropic passage core corresponding to each side and matched with the direction based on the corresponding action characteristics of personnel passage, equipment transportation and water hose laying. The shape and size of the anisotropic passage core are determined by the net width of the passage, the corner radius, the slope and the minimum passage radius. A morphological calculation unit is used to perform morphological calculations on the dynamic mask layer and the anisotropic passage kernel to obtain a passage window layer. The morphological calculation determines that the passage is not passable in the overlapping area of ​​the anisotropic passage kernel and the dynamic mask layer, and that it is passable in the non-overlapping area. The accessibility labeling unit is used to project the windowing results of the access windowing layer onto the nodes and edges of the baseline accessibility map, and to mark the corresponding nodes and edges as enabled, flow-limited, or disabled in combination with the availability index of the access windowing layer, so as to obtain a restricted accessibility map.

[0043] The restricted graph generation module further includes an update strategy, which is used to dynamically adjust the baseline reachability graph when the primitive set changes. The update strategy also includes mapping logic and decision logic, with the mapping logic configured in the spatial mapping unit and the decision logic configured in the morphological calculation unit.

[0044] The control boundary inference module includes: The target analysis unit analyzes the emergency control target based on the emergency response strategy library stored in the disaster prevention and monitoring platform, and determines the type, priority and time constraints of the target action, wherein the target action includes water outlet coverage, valve shut-off, isolation closure, sprinkler start and stop and evacuation passage. A path filtering unit is used to filter a set of candidate paths that are adapted to the type and priority in the restricted reachability graph based on the enable, rate limiting and disable flags of nodes and edges, wherein the set of candidate paths is used to represent the spatial and sequential combination of executable action sequences. The optimization and solution unit is used to search for the executable action sequence with the minimum comprehensive cost and that meets the time constraints in the candidate path set using a multi-objective optimization algorithm; The boundary calculation unit is used to calculate the parameters of the control boundary and generate a control boundary model for fire scenario construction based on the executable action sequence.

[0045] Next, with reference to the accompanying drawings, the method for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion, as provided in the embodiments of this application, will be introduced. Figure 3 The method shown is applied to a disaster prevention and monitoring platform, wherein the disaster prevention and monitoring platform maintains a baseline reachability map, and the method includes: S1: Acquire multi-source data related to the factory area, and identify and extract a set of primitives to represent temporary states based on the multi-source data; In this embodiment, multi-source data can be acquired in real time using existing information acquisition devices. Natural language semantic recognition and image target detection algorithms are used to extract a set of primitives with spatial and temporal attributes. Each primitive in the primitive set can be understood as a basic unit of expression for a temporary state, used to characterize behaviors such as road occupancy, closure, width restriction, height restriction, displacement, or deactivation in the current environment.

[0046] S2: Project the set of primitives onto the corresponding nodes and edges of the baseline reachability graph to obtain a restricted reachability graph; In this embodiment, the spatial positions of primitives are first transformed and aligned according to the factory's spatial coordinate system, thus forming a dynamic mask layer consistent with the factory layout. Then, combining the directional attributes of different action types, an anisotropic access kernel matching the direction field is generated. Morphological calculations are then performed under the guidance of the direction field to obtain the access window layer. The access window layer reflects the accessibility distribution of the factory area under the influence of temporary states at the current moment. By projecting this result back onto the baseline accessibility graph, the enabled, flow-limited, or disabled flags of nodes and edges can be dynamically updated, forming a restricted accessibility graph.

[0047] S3: Combining the preset emergency control objectives, select executable action sequences from the restricted accessibility map to obtain the control boundary of the corresponding fire scenario; In this embodiment, the emergency control objective can be provided by a contingency plan or strategy library, such as "activating a valve to cut off the fuel supply within 120 seconds" or "completing water coverage of a specific area within 60 seconds". By searching the constrained accessibility graph for paths that satisfy the node states and constraints, executable action sequences that meet the time, personnel, and hydraulic requirements are selected.

[0048] Furthermore, this application employs a multi-objective optimization approach, comprehensively considering the shortest path, minimum water supply flow rate, and operational accessibility, to generate an optimal action sequence that corresponds to an executable emergency response plan under the current operating conditions. Based on this sequence, control boundary parameters are further calculated, including the shortest executable time, minimum water supply flow rate, remaining pressure margin, and the coverage area.

[0049] Before delving into the specific technical details of the steps, the embodiments of this application need to be emphasized again.

[0050] This application does not target a single fire detection, early warning, or control algorithm, but rather addresses the problem of how to achieve real-time understanding and dynamic construction of control boundaries in industrial environments where multiple sources of information coexist for extended periods and states frequently change.

[0051] In a typical industrial plant, disaster prevention and monitoring platforms often receive continuous data streams from video surveillance, access control systems, water supply and smoke exhaust subsystems, personnel positioning systems, and energy control networks. These data are related in physical space but asynchronous in time, and are subject to multiple uncertainties due to factors such as communication delays, obstructions, reflections, or false triggers.

[0052] Traditional solutions often rely on static drawings or contingency plans, simply overlaying or weighting these heterogeneous information to construct a static template for a fire scenario. However, when temporary construction, process maintenance, or partial lockdowns occur within the plant area, this template-based approach fails to accurately reflect actual changes in spatial accessibility, leading to discrepancies between the calculated controllable area and the actual operable area, ultimately resulting in a strategy mismatch.

[0053] The idea behind this application stems from a common observation in engineering practice: The safety status of a plant area is not determined by a single alarm point, but is jointly defined by the currently accessible paths, operable devices, and activatable hydraulic conditions.

[0054] Based on this understanding, this embodiment transforms the traditional serial process of detection-alarm-response into a closed-loop structure of state-constraint-boundary, so that the construction of fire scenarios no longer depends on static scene models, but is driven by real-time changes in constraints to generate dynamic and quantifiable scene boundaries.

[0055] Building upon this, the multi-source data fusion method proposed in this embodiment employs a morphological constraint propagation mechanism guided by a directional field. The initial design intention of this mechanism is to enable factors such as obstacles, wind direction, passage, and hydraulics in space to have a unified mathematical expression.

[0056] By constructing a directional field jointly defined by ventilation operation, flue gas flow direction, access control status, and personnel passage patterns, and using this as a constraint input for morphological operations, dynamic evolution patterns of downstream-flow-assisted passage and upstream-flow-limited passage can be adaptively formed in geometric space. This allows for the acquisition of physically consistent spatial accessibility distributions without relying on complex 3D fluid simulations.

[0057] It is important to note that the primitive modeling approach in this embodiment allows complex operational states to be compressed into computable unit representations. For example, temporary states such as scaffolding, pallet stacking, maintenance enclosures, valve tags, or closed access control are independent and difficult to quantify in traditional systems. However, through primitive processing, these states can be uniformly described as entity objects with spatial scope, time windows, and constraint types, and can thus be directly projected onto the nodes and edges of the reachability graph for logical deduction.

[0058] Next, we will further elaborate on the technical content of the baseline accessibility map in this application.

[0059] It is understood that the baseline reachability graph of this application can be understood as a basic graph model structure used to characterize the spatial topological relationships and emergency response path constraints of industrial plant areas.

[0060] In some optional implementations, the construction of the baseline accessibility map is not a one-time static generation, but can be automatically parsed and updated through the plant's building information model, fire protection system drawings, and equipment ledger data. For example, when a partition, ventilation duct, or safety door position is added or adjusted in the plant's building information model, the reconstruction of the corresponding nodes and edges can be automatically triggered; when the fire protection system drawings are updated with water supply circuits or pump station pressure parameters, the hydraulic constraint layer is also reconstructed accordingly.

[0061] It should be noted that the baseline accessibility map in this application is not limited to two-dimensional or three-dimensional representations. Hierarchical or multi-scale graph representations can be adopted depending on the complexity of the plant structure. For example, for multi-story buildings or underground facilities, each floor can be regarded as an independent sub-graph, and cross-floor edges can be established through vertical channels to form a multi-layered nested structure. For comprehensive plant areas with large spatial spans, a regional block approach can be used to construct local graphs, which can then be spliced ​​together through regional boundary nodes to achieve a balance between zonal calculation and global consistency.

[0062] In one example, the method for constructing the baseline reachability graph includes: Based on the building information model, fire protection system drawings, and equipment ledger data of the plant area, fire protection facilities, passages, doors, staircases, valves, shut-off devices, and hazard sources within the plant area are identified as nodes on the baseline accessibility map. According to the preset activity rules, establish the connection relationship between nodes and generate corresponding edges, where the edges are used to represent action paths; According to the preset response rules, constraint parameters are set for each edge. The constraint parameters include geometric constraints, hydraulic constraints, and operation method constraints. The geometric constraints are used to describe the clearance height, width, turning radius, slope, and passage distance of the channel. The hydraulic constraints are used to describe the pressure loss, flow capacity, minimum water supply pressure, and jet range of the fire water supply network. The operation method constraints are used to describe the number of personnel required for operation, operation sequence, and safety conditions. The nodes, edges, and constraint parameters are stored as a baseline reachability graph.

[0063] Next, we will further elaborate on the technical aspects of the method in this application regarding primitive sets.

[0064] In one example, a set of primitives for characterizing temporary states is identified and extracted based on the multi-source data, including: S1.1: Perform semantic analysis on the maintenance work orders, temporary work tickets and lockout records in the text data, identify semantic entities that match road occupation, closure, energy removal, width restriction, height restriction and relocation, extract the corresponding work area, facility identification, start time and end time, and generate text primitives including spatial range and timeliness attributes. Specifically, to accurately map the temporary statuses in maintenance work orders, temporary work permits, and lockout records to both spatial and temporal dimensions without relying on manual data entry, the text data first needs to be preprocessed. This includes character set unification, temporal expression normalization, and proper noun regular expression extraction. Combined with a safety-oriented terminology ontology, semantic tags such as "road occupancy," "closure," "power deactivation," "width restriction," "height restriction," and "displacement" are mapped to text fragments. This mapping is not a simple keyword matching process, but rather an extraction of relationships based on trigger word-constraint word-object word triples, recorded in structured fields for easy compilation and retrieval based on spatial indexes and time windows.

[0065] In this embodiment, semantic parsing employs a combined process of entity recognition and relation extraction: First, dictionary and context-based joint recognition is performed on equipment names, area names, channel names, door numbers, valve numbers, ticket numbers, etc.; then, document-level dependency relations are used to align action words with object entities; next, the effective time and expiration time are obtained through time phrase reduction, and the existence of pre-existing mutual exclusion is determined in conjunction with the status field of the safety permit. For the landing point of the work area, spatial anchoring based on engineering drawings and identification codes is adopted; if ambiguity occurs, disambiguation is performed through the reference relationship of adjacent equipment or the co-occurrence statistics of historical work orders. Finally, text primitives are generated for each text record, and the text primitives include at least: type, affected object identifier, spatial range, time window, intensity level, and confidence level.

[0066] It is understood that the semantic analysis in this application can be achieved through natural language processing techniques, such as using a pre-trained language model based on deep learning combined with named entity recognition and relation extraction models to perform contextual semantic parsing and domain feature recognition on text data. This application will not elaborate on these points here.

[0067] S1.2: Extract targets from the image data using a target detection algorithm to obtain obstacle targets, calculate the projection range and height information of the obstacle targets in the factory area spatial coordinate system, and convert them into geometric primitives representing temporary occupation and restricted passage status; Specifically, to transform temporary components such as scaffolding, fencing, pallet stacks, and vehicles in the scene into measurable geometric effects, the camera units are first calibrated using extrinsic parameters and basic intrinsic parameters. Identifiable fixed components are used as scale anchors to obtain the mapping relationship between pixels and factory area coordinates. For viewpoints where calibration is insufficient, a weak calibration method based on the proportion of standard components is used to estimate the relative scale. Subsequently, a joint strategy of target detection and instance segmentation is employed to separate temporary components from the background. For metal fencing and transparent materials, cross-frame screening using edge consistency and specular suppression reduces false detections. To obtain height and volume information, monocular geometric priors and multi-frame disparity estimation are combined to estimate the stereoscopic outer contour height based on the shape of the target's projection on the ground and the direction of its vanishing point. When binocular or overlapping fields of view are available, depth is directly inferred from disparity to establish voxelized obstacles.

[0068] In this embodiment, the bounding box, contour, and color texture of the same temporary component in consecutive frames are jointly matched to form the target trajectory. Once it enters a static dwell state and continues for more than the minimum dwell time, the object is upgraded to a geometric primitive and given a timed window. Considering the contour jumps often caused by viewpoint occlusion and backlighting, cross-frame morphological smoothing and topological consistency checks are used to correct occasional gaps, and a hysteresis threshold is introduced into the height field to avoid height misjudgments caused by instantaneous shadows. The final output geometric primitive includes type, planar polygon, estimated height, passability threshold, and confidence level.

[0069] It is understood that the object detection algorithm in this application can be implemented using existing deep learning detection frameworks, such as Faster R-CNN, YOLO, or Mask R-CNN algorithms based on convolutional neural networks, which will not be elaborated upon here.

[0070] S1.3: Based on the vehicle positioning trajectory and personnel positioning trajectory in the trajectory data, the corresponding dense areas are identified through clustering algorithm to generate trajectory primitives; In this embodiment, the generation of trajectory primitives is not limited to density, but comprehensively considers directional consistency and throughput capacity: in narrow passages, opposing flows are separated, and the minimum passage interval for staggered passage is calculated; when the staggered interval exceeds a threshold and the duration exceeds a set window, the corresponding area is marked as flow-limited rather than closed to preserve limited accessibility; for alleyway entrances, ramps, and corners, historical data is superimposed to form a throughput curve through statistics to evaluate the passable time window under the current mixed pedestrian and vehicle flow conditions. If there are vehicles or transport equipment that have been stationary for a long time, a certain safety boundary is expanded according to the geometric outline to upgrade them into approximately static soft obstacles, and trajectory primitives with movable attributes are generated for subsequent composite action deduction of moving obstacles before passage.

[0071] S1.4: Based on the opening and closing signals of access control, roller shutters, valves and shut-off devices in the monitoring status data, detect changes in the opening and closing status of the facilities and generate status primitives; In this embodiment, the spatial location of the state primitive is determined through equipment ledgers and drawing indexes, mapping equipment numbers to specific coordinates or path nodes. For linked equipment, causal references are established to enable the spatial propagation of an event to the accessibility of related areas. For access control, access permissions are combined with attendance records to limit the number of people who can access the system within a specific time period, thereby providing personnel-side boundaries for operational constraints. To improve event credibility, consistency checks are performed on multiple signals from the same device. If a sensing anomaly is detected, the state primitive is assigned a downgraded confidence level and a manual verification flag is triggered.

[0072] Next, we will further elaborate on the technical content of the restricted reachability graph in this application.

[0073] refer to Figure 4 , Figure 4 A flowchart illustrating the steps for generating a restricted reachability graph provided in this application embodiment.

[0074] In one example, projecting the set of primitives onto the corresponding nodes and edges of the baseline reachability graph yields a restricted reachability graph, including: S2.1: Based on the factory area spatial coordinate system, the spatial range of each primitive in the primitive set is mapped on the plane coordinate to generate a dynamic mask layer corresponding to the factory layout. The dynamic mask layer is used to characterize the distribution of temporary states of road occupancy, closure, width restriction, height restriction and obstacles in the spatial and temporal dimensions. Specifically, in order to ensure that temporary road occupancy, closure, width restriction, height restriction and obstacles can be treated uniformly in both spatial and temporal dimensions, the spatial descriptions of various primitives need to be aligned to the same reference coordinates.

[0075] In this embodiment, the measurement points and positioning base station locations of the building information model are first used as control points to complete the unified conversion of multi-source coordinates such as camera plane, access control points, valve coordinates, channel centerline, and stair tread endpoints; then, anchor point parsing is performed on the heterogeneous expressions carried in the primitives to obtain closed polygons or linear buffer zones.

[0076] Furthermore, considering that temporary states usually do not have long duration, they are instantiated on the timeline by time windows consisting of event trigger time, effective delay and cooldown period. This allows the attributes of the same spatial range in different time slices to be indexed frame by frame, so that the subsequent processing logic faces a continuous, superimposed and sliceable scene base map, avoiding the temporary states being scattered in various sub-modules in the form of discrete events and difficult to arrange in a unified manner.

[0077] In one example, the process of mapping the spatial extent of each primitive in the primitive set onto planar coordinates based on the factory spatial coordinate system to generate a dynamic mask layer corresponding to the factory layout includes: Perform coordinate transformation and spatial alignment on the spatial description information, location information and target coordinate information carried by each primitive in the primitive set; Based on the primitive type, corresponding graphic elements are generated, and based on the time-effect attribute, effective time and expiration time are assigned to each graphic element in the time dimension to obtain a spatiotemporal mask sequence. Primitives belonging to the road occupation and closure generate closed polygon regions, primitives belonging to the width restriction and height restriction generate semi-closed polygon regions, and primitives belonging to the obstacle generate volume regions containing height information. The spatiotemporal mask sequences are superimposed and fused according to their spatial positions to obtain a dynamic mask layer.

[0078] In one specific implementation, the formation of the dynamic mask layer can be understood as continuously compiling semantic primitives into spatiotemporal boundaries that can be directly computed by spatial operators: Based on the plant's spatial coordinate system, the reference points, measurement control points, and known facility coordinates in the Building Information Model (BIM) are first used as anchor points. Heterogeneous representations such as area names, equipment identifiers, mileage markers, or relative displacements carried by primitives are mapped to unified coordinates. When different reference systems are involved, such as camera planes, access control points, valves, and passageway centerlines, coordinate transformation and orientation alignment are completed through anchor point correspondence. In cases of ambiguity, disambiguation is achieved using the topology of adjacent facilities and historical co-occurrence relationships. After spatial alignment is completed, computable graphic elements are generated on the plane according to the primitive type. The system transcribes enclosed and closed areas into non-permeable closed polygons, assigns width and height limits to semi-closed polygons with attributes, voxels obstacles in the vertical direction into volume units containing a height field, and adaptively densifies boundaries related to geometric constraints such as channel centerline, corner radius, and slope according to curvature to prevent raster distortion.

[0079] It is understood that the graphical elements generated in this application are geometrical carriers used to express the occupancy and constraint relationships of temporary states on the passable geometry in two-dimensional space. The core purpose is to enable abstract semantic events to participate in subsequent morphological calculations and topological constraints as continuous and computable spatial entities.

[0080] Specifically, in planar representation, closed polygons can form closed connected regions, allowing pixels within the mask to be identified as blocked spaces during computation, ensuring that operators such as morphological erosion and dilation have clear internal and external definitions at the boundaries. For example, when a region is completely closed due to construction, its extent should possess the geometric property of a closed boundary, so that it can be determined whether the kernel has crossed the obstacle when sliding through the kernel.

[0081] Furthermore, the primitives for width and height limits generate semi-closed polygonal regions with attributes to represent partially accessible but restricted physical situations. Such regions typically have passable windows, but their geometric constraints only apply in one direction or dimension. The semi-closed boundary design allows the mask to allow passage in one direction while retaining constraints in the restricted direction.

[0082] Furthermore, generating volumetric units containing height information for obstacle primitives aims to reflect local obstruction effects in three-dimensional space. Unlike closures, obstacles often only block passage within a local area, and their accessibility is closely related to height. For example, objects such as stacked pallets or hoisted equipment only affect passage at a certain height. By voxelizing in the vertical direction, obstacles are divided into several volumetric units with height labels, allowing the mask to not only have a two-dimensional planar projection but also reflect the vertical passage cross-section.

[0083] The aforementioned graphical elements are instantiated into a spatiotemporal sequence based on the effective and ineffective times of the primitives. Different thresholds for hysteresis are set for appearance and decay to suppress short-term jitter. Under insufficient confidence or fuzzy descriptions of neighboring / surrounding classes, a direction-related buffer band is generated based on positioning accuracy and calibration error. Finally, the elements are superimposed and fused according to spatial location and priority. During superposition, the coverage rule of closure is higher than flow limiting and isolation is higher than soft barrier is followed. The covered records are retained at the overlapping areas for traceability. For cross-layer structures, the vertical correspondence is maintained through hierarchical indexing. Local gaps are filled with holes by connectivity and boundary consistency constraints. The output is a continuous mask surface that can be directly manipulated by the morphological kernel at any time slice.

[0084] S2.2: Based on the corresponding action characteristics of personnel passage, equipment transportation and water hose laying, anisotropic passage cores matching the direction are obtained for each side, wherein the shape and size of the anisotropic passage cores are determined by the net width of the passage, the corner radius, the slope and the minimum passage radius; Specifically, the spatial occupancy and tolerance for personnel passage, equipment transportation, and hose laying differ. Direct morphological calculations can easily lead to overly conservative or overly optimistic judgments at corners, ramps, and narrow passages. Therefore, an anisotropic passage core matching the direction is generated on each side for each type of action. The major axis of the core is aligned along the side direction, while the minor axis is dynamically adjusted based on the net width threshold, carrying width, hose diameter, minimum corner radius, and slope limit. Where steps or flanges exist, the core shape introduces gaps at corresponding locations to express the constraint that the load can be carried but not dragged.

[0085] S2.3: Perform morphological calculations on the dynamic mask layer and the anisotropic passage kernel to obtain a passage window layer, wherein the morphological calculations determine that the area where the anisotropic passage kernel and the dynamic mask layer overlap is impassable, and the area where they do not overlap is passable. Specifically, simply determining passability based on the overlap between the core and the mask is insufficient to reflect the difference between downstream-flow-assisted passage and upstream-flow-limited passage, and it is also unsuitable for progressive evaluation near the critical boundary. Therefore, anisotropic passage cores slide along streamlines on the dynamic mask layer, recording the overlap ratio between the core and the mask, the relative orientation of the boundary, and the local directional intensity. When the core slides downstream and the overlap ratio is below a threshold, progressive erosion is performed, treating minor surrounding obstructions as resolvable. When the core slides upstream and the overlap ratio approaches the threshold, segmented expansion is performed, generating a buffer zone in front of the obstacle boundary to represent the expansion of reverse obstruction. The two types of results are superimposed at the same location with different weights and iterated within a small range until the change falls below the convergence threshold, forming a direction-dependent passage probability distribution.

[0086] It is understandable that the morphological computation in this application is not a simple dilation or erosion operation on the binary region of an image in the traditional sense, but rather a direction-sensitive morphological modeling method for the dynamic spatial constraints of industrial plant areas. The design aims to solve the three major technical limitations of conventional morphological methods in complex spatial scenes: orientation distortion, excessive erosion, and discontinuous response.

[0087] Specifically, traditional morphological operations are based on isotropic structural elements, with the fundamental assumption that the effects of spatial barriers are equivalent in all directions. However, in typical emergency scenarios in industrial plants, this assumption clearly does not hold: Fires or deflagrations can cause significant directional differences in airflow, smoke plumes, and the direction of personnel evacuation. Geometric constraints of structures such as passageways, ramps, and stairs result in varying accessibility in different directions; The same obstacle can impede a path of movement to varying degrees depending on whether it is downstream or upstream.

[0088] Based on this, if the traditional morphological kernel (i.e., symmetric kernel, fixed scale, static morphology) is still used, the expansion or erosion will act equally in all directions, resulting in the incorrect erosion of narrow channels in the downstream direction and the excessive neglect of local obstacles in the upstream direction, leading to misjudgment of spatial connectivity. When the obstacle boundary forms a certain angle with the flow direction, the traditional erosion operation will cut off the actually passable area at the boundary, causing the passage path to be misjudged as broken. The binarization operation only outputs two states: passable / unpassable, which cannot express the gradual passability under near-critical conditions, nor can it quantify important engineering states such as partially passable or passable after buffering.

[0089] In this embodiment, morphological operations no longer use isotropic structural elements with fixed shapes. Instead, a directional field is constructed by fusing multi-source data, and based on this, a convection structural element is generated that is adaptively stretched along the main direction and automatically corrected in shape at corners. This makes the range of expansion and erosion operations dynamically related to the angle between the flow field, the flow trend, and the obstacle normal.

[0090] As the convection structure element slides along the directional streamlines, the passage gradient is calculated based on the overlap ratio between the convection structure element and the mask, the local directional angle, and the flow intensity. When flowing downstream, a progressive erosion approach is used; when flowing upstream, a segmented expansion approach is used. A continuous access probability field, rather than a binary result, is formed through iterative fusion. This achieves the directionality of spatial constraints, the gradual change in access judgment, and the continuity of boundary responses. It extends accessibility assessment from static geometric determination to dynamic, direction-sensitive modeling, improving the accuracy of access judgment in narrow passages and complex corner areas, and significantly enhancing the real-time performance and physical plausibility of restricted access analysis in fire or smoke extraction scenarios.

[0091] In one example, morphological calculations are performed on the dynamic mask layer and the anisotropic passage kernel to obtain a passage window layer, including: S2.3.1: Based on the spatial coordinate system of the factory area, a direction field is constructed to guide morphological operations based on the anisotropic access kernel. The direction field is obtained by fusing the opening and closing status of access control and roller shutters, the operating status of ventilation and smoke exhaust equipment, wind speed and differential pressure monitoring data, real-time trajectories of vehicles and personnel, and smoke plume flow clues. Specifically, the orientation field is used to characterize the dominant trends of passage / airflow / plume in space and to provide guiding vectors for anisotropic morphology.

[0092] To avoid propagation bias caused by inconsistencies in coordinates and time between data from different sources, the alignment of multi-source events and continuous quantities is first completed in the plant area spatial coordinate system: The opening and closing events of access control and roller shutters are mapped to the opening factor of the channel nodes. The operating conditions of ventilation and smoke exhaust equipment are mapped to the supply / exhaust air indicators of the corresponding air outlets. The wind speed and differential pressure monitoring points use the Thiessen polygon weighted diffusion within the measuring point radius as local flow direction prompts. The trajectories of vehicles and personnel are adsorbed according to the center line of the channel and the instantaneous speed and dominant direction are extracted. The smoke plume flow clues are estimated as local drift vectors by the smoke plume texture displacement or thermal image gradient displacement in the video.

[0093] The aforementioned discrete and continuous information are uniformly projected onto a hybrid carrier of regular grids and channel skeletons: A skeleton vector is formed on the channel skeleton, and a surface vector is formed in the open area outside the skeleton. The two are combined into a single candidate vector in the overlapping area by prioritizing the skeleton and compensating for the surface.

[0094] In this embodiment, the weight sources of the orientation field are fused according to a reliable sequence: If the access control is unlocked and a stable flow of people is observed on the channel screen, the weight of the flow trajectory is increased; if the smoke exhaust fan is running at high speed and the differential pressure monitoring shows a unidirectional differential pressure, the wind direction weight is prioritized; if the two conflict, the plume drift clue is used as arbitration, retaining a small amount of random disturbance to prevent inertial errors caused by long-term locking. For areas with missing measurement points, the propagation direction is determined by the shortest path in the channel topology, with attenuation applied to the propagation length; outliers are reduced in weight or removed after joint discrimination based on neighborhood consistency and device status. The direction field outputs a triplet of dominant direction, direction strength, and confidence for each grid and skeleton segment.

[0095] S2.3.2: Based on the direction field and the action type of each side, generate the corresponding convection structure element. The convection structure element is stretched along the main direction of the direction field and its shape is corrected at the corner. Specifically, convection structural elements are used to embed the differences in how actions occupy space into the kernel shape, so that the kernel adapts to the action type, geometric constraints and orientation field of the edges.

[0096] For three common actions—personnel passage, equipment handling, and hose laying—a set of master kernel parameters is defined respectively: The personnel core uses shoulder width and safe side distance as the short axis reference, and stride and turning radius as the long axis and end fillet reference; the equipment core uses the equipment outline and turning radius as the short axis and end reference, and adds restricted flanks to the side of the core to represent the additional clearance to the wall side during handling; the hose core uses hose diameter, joint turning space and minimum straight section length as geometric reference, and inserts the joint compartment shape at the corner of the core to reserve the turning area.

[0097] The core's major axis is aligned with the geometric direction of the side, while the minor axis is dynamically scaled based on the channel's net width and the width-limiting mask. During ramp and platform stages, the core inserts segmented advancement markers to indicate composite actions that require stopping and then advancing.

[0098] The aforementioned nuclear prototype was then modulated by the directional field: when the intensity was high in the downstream direction, the long axis of the nuclear extended to reflect the improved propulsion efficiency; when the intensity was high in the upstream direction, the long axis of the nuclear shortened and the short axis slightly increased to reflect the avoidance requirements.

[0099] In this embodiment, the convection structure element is spatially segmented based on edges: Each edge is divided into a geometrically consistent segment and a turning segment. The geometrically consistent segment uses a fixed end fillet core, while the turning segment automatically generates a fan-shaped transition core based on the radius of curvature. When the edge crosses a height-restricted area, a height label is assigned to the core to participate in the three-dimensional passage determination.

[0100] S2.3.3: Based on the convection structure element, perform erosion and expansion operations on the dynamic mask layer along the streamlines of the directional field to obtain a passage window layer. When the barrier boundary of the dynamic mask layer is consistent with the main direction of the directional field, the erosion operation is performed. When the barrier boundary of the dynamic mask layer is inconsistent with the main direction of the directional field, the expansion operation is performed. Specifically, to transform orientation sensitivity and kernel shape differences into a continuous spatial accessibility representation, a morphological process combining streamline sliding and iterative fusion is employed. Streamline clusters are generated on the orientation field starting from the edges, with the streamline spacing adaptively set based on directional intensity and channel width. For each streamline, a structural element is slid once in the forward and backward directions, recording the overlap ratio, the angle between the kernel principal axis and the boundary normal, the directional intensity, and the local rotation margin at each sliding position, and calculating the position-level passage gradient accordingly. When the streamline aligns with the orientation field and the overlap ratio is below the downstream threshold, progressive erosion is performed, allowing the kernel to make limited intrusions into the mask edge to simulate downstream passage. When the streamline is opposite to the orientation field or the angle exceeds the limit and the overlap ratio approaches the counter-current threshold, segmented expansion is performed, generating a buffer zone outside the obstacle boundary, indicating that counter-current propulsion requires more space.

[0101] In one example, based on the convection structural element, erosion and expansion operations are performed on the dynamic mask layer along the streamlines of the directional field to obtain a passage window layer, including: Guided by the streamlines of the directional field, a cluster of streamlines adjacent to each side is selected in the dynamic mask layer; The streamline cluster is slid by the convection structure element, and the overlap ratio between the convection structure element and the barrier boundary is calculated at each sliding position to obtain the local impedance value. The local impedance value is then used to determine whether the streamline cluster is downstream. If the streamline cluster is in the direction of flow, the sliding positions are processed by erosion operation and then connected to obtain the smooth flow channel; If the streamline cluster is countercurrent, the sliding position is processed by expansion calculation and buffering operation is performed to obtain the stagnation range; By traversing all streamline clusters, the smooth passageway and the obstruction range are merged to obtain the passage window layer.

[0102] S2.4: Project the opening result of the access window layer onto the nodes and edges of the baseline reachability graph, and mark the corresponding nodes and edges as enabled, flow-limited, or disabled in combination with the availability index of the access window layer to obtain a restricted reachability graph; Specifically, to transform spatial-level accessibility assessment into executable constraints on a graph structure, the availability index of the accessibility window layer needs to be mapped to nodes and edges. The specific approach is as follows: Within the path buffer zone of each edge, sample the passage probability and calculate the minimum connectivity. Edges below the activation threshold are marked as disabled, and those between the activation and disable thresholds are marked as flow-limited. Record the flow-limiting level. At the node, aggregate the indicators of adjacent edges and weight the access control status, valve position operability, and personnel accessibility to generate the node availability status. For edges containing clearable channel markers, derive supplementary composite action edges to represent alternative paths for performing obstacle removal and re-passage, and attach estimated time, number of cooperating personnel, and failure fallback conditions to the edge.

[0103] In one example, by combining preset emergency control objectives, executable action sequences are filtered from the restricted accessibility map to obtain the control boundaries of the corresponding fire scenario, including: Based on the emergency response strategy library stored in the disaster prevention and monitoring platform, the emergency control target is analyzed to determine the type, priority and time constraints of the target action, wherein the target action includes water outlet coverage, valve shut-off, isolation closure, sprinkler start and stop and evacuation passage. In the restricted reachability graph, a set of candidate paths adapted to the type and priority is filtered based on the enabled, rate-limited, and disabled flags of nodes and edges, wherein the set of candidate paths is used to represent the spatial and sequential combination of executable action sequences; The multi-objective optimization algorithm searches the candidate path set for an executable action sequence that has the minimum overall cost and meets the time constraints. Based on the executable action sequence, the parameters of the control boundary are calculated and a control boundary model for fire scenario construction is generated, wherein the parameters include the shortest executable time, minimum water supply flow rate, remaining pressure margin, covered space range, and corresponding inaccessible areas.

[0104] In one specific implementation, the generation of control boundaries can be understood as mapping the logic of goal-constraint-action sequentially to a restricted reachability graph and searching and measuring it based on executability: First, preset objectives are parsed from the emergency response strategy library. Statement-based objectives such as water coverage within a limited time, valve shut-off within a limited time, and isolation or evacuation within a limited time are reduced to a set of action types, their priorities, time windows, and necessary preconditions. Each type is then bound to the start and end nodes, allowed action edge sets, and operation method constraints on the accessibility graph using an action dictionary. Subsequently, the enabled, flow-limited, and disabled flags of nodes and edges, as well as availability indicators such as flow-limited level, remaining headroom, and turning margin, are read from the restricted accessibility graph. Parameters such as minimum water supply pressure, expected remaining pressure, and minimum range from the hydraulic side are overlaid. Candidate path sets are constructed according to the objective type. Composite action edges that satisfy preconditions such as permission, obstacle removal, joint arrangement, and docking segmentation are automatically inserted into the paths. Finally, the three-dimensional comprehensive cost of time, hydraulics, and operation method, as well as the penalty for breach of contract, are calculated for each candidate path. Penalties are imposed. Based on this, a multi-objective optimization algorithm is used to search the candidate path set, prioritizing the action sequence that is reachable within the target time window and has the lowest overall cost. Equivalent solutions are scored and ranked according to priority and robustness. Hydraulic closed-loop verification and resource availability verification are performed on the selected action sequences. If violations are found, backtracking to the suboptimal sequence or triggering a compound action alternative branch is performed. Finally, based on the optimal action sequence that passes the verification, control boundary parameters are calculated and solidified, including the shortest executable time from the starting node to the target node, the minimum water supply flow required to meet the coverage target and the corresponding remaining pressure margin, the coverable space range under the current constraints, and the unreachable area caused by disabling or disconnection. The control boundary model is stored with the version number of the edge set and node set, the trigger primitive identifier, and the effective time window, so that the control boundary model can be called and refreshed online by downstream scenarios.

[0105] For example, the following numerical example is given to illustrate how to calculate the passage coefficient of the passage window layer. This example is only used to explain the relationship between the operation link and the dimensions. The selected parameters and values ​​are illustrative values ​​and do not represent the actual calibration results or engineering recommended values.

[0106] Assume a floor in the factory area is discretized into a 1m x 1m grid, and the accessibility baseline map shows a main passage running from west to east. After multi-source data fusion, the main direction of the directional field for this area is "east-northeast". For ease of calculation, the angle of the main direction is taken to be approximately 30 degrees, and a direction-sensitive convection structure element is constructed accordingly: covering 5 grid lengths along the main direction and 1 grid width laterally. Simultaneously, a dynamic mask layer is generated jointly by the access control status and smoke concentration: the access control shows a door is half-open, and the fusion of video and smoke detection infers that the smoke plume near the doorway has significant scouring along the main direction, thus forming a semi-blocked dynamic mask strip at the doorway. To convert the semi-blockage into a computable quantity, the mask value is taken as an occupancy intensity from 0 to 1, where 0 represents complete unobstructed and 1 represents complete obstruction; the mask value is 0.8 at the center of the doorway, 0.5 and 0.3 for the transition grids on both sides of the doorway respectively, and 0.1 for the remaining passage grids (indicating slight smoke but not constituting a physical blockage).

[0107] Furthermore, when evaluating the sliding motion of the convection structure elements along the main direction in the channel, each slide falls within a 5-grid sampling window. Taking the window with the doorway center as the sliding center as an example, the mask values ​​of the 5 grids within the window are 0.1, 0.3, 0.8, 0.5, and 0.1, respectively. To obtain the overlap ratio, we use the most intuitive area representation: considering the mask value of each grid as the proportion of the area that grid is occluded, the average occlusion ratio of the window is equal to the average of these 5 values, which is 0.1 + 0.3 + 0.8 + 0.5 + 0.1 divided by 5, resulting in 0.36. Next, we convert the occlusion ratio into local impedance, taking a linear mapping that is easy to interpret: impedance equals 1 plus 4 times the occlusion ratio, so the window impedance equals 1 plus 4 times 0.36, resulting in 2.44. To convert the impedance into a continuous passability factor (the smaller the value, the more difficult it is to pass through), we take the passability factor as 1 divided by 1 plus the impedance. Therefore, the passability factor for this window is approximately 1 divided by 3.44, which is about 0.29. In contrast, for a clean window far from the doorway, assuming all five grid mask values ​​are 0.1, the average occlusion percentage is 0.1, the impedance is 1.4, and the passability factor is 1 divided by 2.4, approximately 0.42. This yields a clear result: when the doorway is half-open and the smoke plume washes over the window, the passability factor evaluated along the main direction of the sliding window will show a significant dip near the doorway.

[0108] Furthermore, during progressive erosion in the downstream direction (along a 30-degree direction), instead of directly deleting the binarized mesh, the passability coefficient is treated as a retainable quantity. For each mesh, the lowest passability coefficient within the length range of the preceding structural element is taken as the update value to simulate the propagation and amplification of unfavorable passability factors in the downstream direction. Taking the area near the center of the doorway as an example, the original passability coefficient of the mesh at the center of the doorway is 0.29. If the four meshes in front of it are 0.35, 0.38, 0.40, and 0.41 respectively, then the update value of the center of the doorway after downstream erosion is taken as the minimum of these five values, which is still 0.29. The first mesh after the doorway, which might originally have a value of 0.35, will have its window coverage reveal a value of 0.29 at the center of the doorway, so it will be lowered to 0.29 after the update, thus forming a continuous low-passability zone in the downstream direction. Conversely, when performing segmented expansion in the counter-current direction, to avoid local anomalies like doorways contaminating the entire passageway, a segmented threshold is introduced: when the average occupancy percentage of a sliding window is below 0.2, backfilling is allowed based on the highest neighboring throughput factor; otherwise, the original value is maintained. For example, if the average occupancy percentage of a window further in front of the doorway is 0.15, corresponding to an impedance of 1.6 and a throughput factor of approximately 0.38, and meets the threshold of below 0.2, then counter-current expansion can raise the local throughput factor back to a higher level in the neighborhood, ensuring that the throughput window layer converges but does not diverge near the doorway, thus maintaining the true bottleneck at the doorway without infinitely expanding the bottleneck to the point where the entire passageway becomes unusable.

[0109] Furthermore, when ultimately merging the results of downstream erosion and downstream expansion, a "more conservative" strategy can be adopted, where the final accessibility coefficient of each grid is the smaller of the two values, thus obtaining the accessibility window layer. Sampling along the channel, the final accessibility coefficient is mostly around 0.40 far from the doorway, while the doorway and a small downstream section are lowered to approximately 0.29. If this accessibility window layer is then projected back onto the baseline accessibility map, an easily interpretable activation threshold can be set, for example, grids with an accessibility coefficient not lower than 0.33 are considered "accessible." The doorway and its low downstream accessibility zone will then form a narrow "semi-discontinuity" on the restricted accessibility map, allowing subsequent fire scenario construction and evacuation route deduction to automatically bypass or reduce the priority of this channel segment. Since every quantity in the entire link comes from continuous quantities of the same dimension such as "mask ratio, impedance, and accessibility coefficient," the example clearly demonstrates the numerical conversion relationship from multi-source states to spatial access constraints.

[0110] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion, applied to a disaster prevention and monitoring platform, characterized in that: The disaster prevention monitoring platform is used to refresh the control boundaries of fire scenarios and to construct and control typical fire scenarios in real time. The disaster prevention monitoring platform maintains a baseline reachability graph. The system includes: The multi-source data acquisition module is used to acquire multi-source data related to the factory area, including text data, image data, trajectory data, and monitoring status data. A temporary state identification module is used to identify and extract a set of primitives for characterizing temporary states based on the multi-source data. A restricted graph generation module is used to project the primitive set onto the corresponding nodes and edges of the baseline reachability graph to generate a restricted reachability graph; The control boundary reasoning module is used to combine preset emergency control objectives and filter executable action sequences from the restricted accessibility map to obtain the control boundary of the corresponding fire scenario. The temporary state identification module includes: The text parsing unit is used to perform semantic analysis on the maintenance work orders, temporary work permits and lock-up records in the text data, identify semantic entities that match the terms "road occupation", "closure", "energy removal", "width restriction", "height restriction" and "displacement", extract the corresponding work area, facility identification, start time and end time, and generate text primitives that include spatial range and timeliness attributes. The image recognition unit is used to extract targets from the image data using a target detection algorithm, obtain obstacle targets, calculate the projection range and height information of the obstacle targets in the factory area spatial coordinate system, and convert them into geometric primitives representing temporary occupation and restricted passage status. The trajectory clustering unit, based on the vehicle positioning trajectory and personnel positioning trajectory in the trajectory data, identifies the corresponding dense areas through a clustering algorithm and generates trajectory primitives; The status event unit detects changes in the opening and closing status of facilities based on the opening and closing signals of access control, roller shutters, valves, and shut-off devices in the monitoring status data, and generates status primitives. The restricted graph generation module includes: The spatial mapping unit maps the primitive set into a dynamic mask layer based on the factory area spatial coordinate system to represent the distribution of road occupancy, enclosure, width restriction, height restriction and obstacles in the spatial and temporal dimensions; The passage core calculation unit is used to obtain the anisotropic passage core corresponding to each side and matched with the direction based on the corresponding action characteristics of personnel passage, equipment transportation and water hose laying. The shape and size of the anisotropic passage core are determined by the net width of the passage, the corner radius, the slope and the minimum passage radius. A morphological calculation unit is used to perform morphological calculations on the dynamic mask layer and the anisotropic passage kernel to obtain a passage window layer. The morphological calculation determines that the passage is not passable in the overlapping area of ​​the anisotropic passage kernel and the dynamic mask layer, and that it is passable in the non-overlapping area. The accessibility labeling unit is used to project the windowing results of the access windowing layer onto the nodes and edges of the baseline accessibility map, and to mark the corresponding nodes and edges as enabled, flow-limited, or disabled in combination with the availability index of the access windowing layer to obtain a restricted accessibility map. The morphological calculation unit is equipped with decision logic, which is configured as follows: Based on the spatial coordinate system of the factory area, a direction field is constructed to guide morphological operations based on the anisotropic access kernel. The direction field is obtained by fusing the opening and closing status of access control and roller shutters, the operating status of ventilation and smoke exhaust equipment, wind speed and differential pressure monitoring data, real-time trajectories of vehicles and personnel, and smoke plume flow clues. Based on the direction field and the action type of each side, a corresponding convection structure element is generated. The convection structure element is stretched along the main direction of the direction field and its shape is corrected at the corner. Based on the convection structure element, erosion and expansion operations are performed on the dynamic mask layer along the streamlines of the directional field to obtain a passage window layer. When the barrier boundary of the dynamic mask layer is consistent with the main direction of the directional field, erosion operation is performed; when the barrier boundary of the dynamic mask layer is inconsistent with the main direction of the directional field, expansion operation is performed.

2. The system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion as described in claim 1, characterized in that, The method for constructing the baseline reachability graph includes: Based on the building information model, fire protection system drawings, and equipment ledger data of the plant area, fire protection facilities, passages, doors, staircases, valves, shut-off devices, and hazard sources within the plant area are identified as nodes on the baseline accessibility map. According to the preset activity rules, establish the connection relationship between nodes and generate corresponding edges, where the edges are used to represent action paths; According to the preset response rules, constraint parameters are set for each edge. The constraint parameters include geometric constraints, hydraulic constraints, and operation method constraints. The geometric constraints are used to describe the clearance height, width, turning radius, slope, and passage distance of the channel. The hydraulic constraints are used to describe the pressure loss, flow capacity, minimum water supply pressure, and jet range of the fire water supply network. The operation method constraints are used to describe the number of personnel required for operation, operation sequence, and safety conditions. The nodes, edges, and constraint parameters are stored as a baseline reachability graph.

3. The system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion as described in claim 1, characterized in that, The restricted graph generation module further includes an update strategy, which is used to dynamically adjust the baseline reachability graph when the primitive set changes. The update strategy also includes mapping logic and decision logic, and the mapping logic is configured within the spatial mapping unit.

4. The system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion according to claim 3, characterized in that, The mapping logic is configured as follows: Perform coordinate transformation and spatial alignment on the spatial description information, location information and target coordinate information carried by each primitive in the primitive set; Based on the primitive type, corresponding graphic elements are generated, and based on the time-effect attribute, effective time and expiration time are assigned to each graphic element in the time dimension to obtain a spatiotemporal mask sequence. Primitives belonging to the road occupation and closure generate closed polygon regions, primitives belonging to the width restriction and height restriction generate semi-closed polygon regions, and primitives belonging to the obstacle generate volume regions containing height information. The spatiotemporal mask sequences are superimposed and fused according to their spatial positions to obtain a dynamic mask layer.

5. The system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion as described in claim 1, characterized in that, Based on the convection structure element, erosion and expansion calculations are performed on the dynamic mask layer along the streamlines of the directional field to obtain a passage window layer, including: Guided by the streamlines of the directional field, a cluster of streamlines adjacent to each side is selected in the dynamic mask layer; The streamline cluster is slid by the convection structure element, and the overlap ratio between the convection structure element and the barrier boundary is calculated at each sliding position to obtain the local impedance value. The local impedance value is then used to determine whether the streamline cluster is downstream. If the streamline cluster is in the direction of flow, the sliding positions are processed by erosion operation and then connected to obtain the smooth flow channel; If the streamline cluster is countercurrent, the sliding position is processed by expansion calculation and buffering operation is performed to obtain the stagnation range; By traversing all streamline clusters, the smooth passageway and the obstruction range are merged to obtain the passage window layer.

6. The system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion according to claim 1, characterized in that, The control boundary inference module includes: The target analysis unit analyzes the emergency control target based on the emergency response strategy library stored in the disaster prevention and monitoring platform, and determines the type, priority and time constraints of the target action, wherein the target action includes water outlet coverage, valve shut-off, isolation closure, sprinkler start and stop and evacuation passage. A path filtering unit is used to filter a set of candidate paths that are adapted to the type and priority in the restricted reachability graph based on the enable, rate limiting and disable flags of nodes and edges, wherein the set of candidate paths is used to represent the spatial and sequential combination of executable action sequences. The optimization and solution unit is used to search for the executable action sequence with the minimum comprehensive cost and that meets the time constraints in the candidate path set using a multi-objective optimization algorithm; The boundary calculation unit is used to calculate the parameters of the control boundary and generate a control boundary model for fire scenario construction based on the executable action sequence.

7. The system for constructing typical fire scenarios in industrial enterprises based on multi-source data fusion according to claim 6, characterized in that, The parameters of the control boundary include the shortest executable time, minimum water supply flow rate, remaining pressure margin, covered spatial range, and corresponding inaccessible areas.

Images