A method and system for improving building fire evacuation efficiency based on pathfinder

Abstract

The application provides a method and system for improving building fire evacuation efficiency based on pathfinder, and belongs to the field of building fire protection. The technical scheme comprises the following steps: S1, analyzing a BIM model to generate an evacuation network model; S2, obtaining time sequence data of an evacuation process based on the evacuation network model through pathfinder; S3, generating dynamic evacuation guiding parameters based on the time sequence data; S4, adjusting the evacuation strategy of pathfinder to obtain an optimized evacuation scheme according to the dynamic evacuation guiding parameters; and S5, writing back and integrating the optimized evacuation scheme into the BIM model to generate an updated BIM model. The application has the beneficial effect that the application realizes a closed loop from building design to fire evacuation optimization by combining BIM model and pathfinder simulation. Through prediction and optimization of dynamic evacuation guiding parameters, the accuracy and real-time performance of the evacuation strategy are improved, and through iterative optimization of the replanning control quantity, the flexibility and effectiveness of the evacuation path and exit selection are improved.

Description

Technical Field

[0001] This invention relates to the field of building fire protection, and in particular to a method and system for improving building fire evacuation efficiency based on pathfinder. Background Technology

[0002] With the increasing number of densely populated buildings such as urban complexes, high-rise buildings, and transportation hubs, fire evacuation assessments are gradually shifting from empirical formulas to computer simulations based on pedestrian dynamics. Software such as Pathfinder, within behavioral modeling frameworks like Steering and SFPE, can characterize pedestrian movement, exit selection, and bottleneck queuing, supporting performance-based evacuation assessments. Meanwhile, BIM provides component-level geometric and attribute data, laying the foundation for evacuation space identification, networked modeling, and scene parameterization input. In recent years, graphical models and temporal learning have been introduced into congestion prediction and strategy optimization, propelling evacuation simulation from "static verification" to "data-driven iterative optimization."

[0003] However, existing Pathfinder-based evacuation analysis largely relies on offline evaluation paradigms such as "setting a scenario—running simulation—outputting evacuation time," which struggles to support refined and verifiable engineering implementation. Firstly, evacuation geometry and access parameters often depend on manual modeling and verification, lacking unified topological and parametric mapping rules from BIM to simulation models, resulting in poor reproducibility and difficulties in component-level traceability. Secondly, most models treat access width, availability, and crowd behavior parameters as static constants, lacking unified time-based resampling and structured extraction of simulation logs, thus failing to generate edge / node time parameters. First, the system lacks sequential characteristics and cannot predict future congestion evolution, leaving path and exit selection only as a passive response. Second, optimization typically employs enumeration or manual parameter tuning, lacking a learning model oriented towards network structures to generate stable dynamic guidance parameters, and also lacking a closed-loop iterative mechanism consisting of replanning triggers, control injection, and convergence criteria. Third, results are often output as reports or static drawings, lacking a mechanism to write back strategies based on source BIM component identifiers and generate updated models, leading to a disconnect between simulation optimization and design delivery, making it difficult to achieve the integrated effect of "prediction-guidance-iterative optimization-BIM solidification" as proposed in this invention. Summary of the Invention

[0004] The purpose of this invention is to provide a closed-loop system for improving building fire evacuation efficiency by combining BIM models with Pathfinder simulation. Through the prediction and optimization of dynamic evacuation guidance parameters, not only is the accuracy and real-time performance of evacuation strategies improved, but also the flexibility and effectiveness of evacuation route and exit selection are enhanced through iterative optimization of replanning control variables.

[0005] This invention is achieved through the following measures: A method for improving building fire evacuation efficiency based on pathfinder, characterized by the following steps: S1. Analyze the BIM model and generate an evacuation network model that includes the topology and attribute parameters of accessible spaces. S2. Based on the evacuation network model, obtain the time-series data of the evacuation process through pathfinder simulation; S3. Based on the time series data, use a graph neural network-based prediction model to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters. S4. Based on the dynamic evacuation guidance parameters, adjust the evacuation strategy of the pathfinder and perform iterative simulation optimization to obtain an optimized evacuation scheme that meets the predetermined conditions. S5. Write back the optimized evacuation plan and integrate it into the BIM model to generate an updated BIM model.

[0006] The invention also has the following specific features: The process of parsing the BIM model and generating an evacuation network model containing the topology and attribute parameters of the accessible space includes: acquiring the components related to evacuation calculations and their geometric and attribute data in the BIM model; performing coordinate system transformation, elevation unification, and geometric error correction on the multi-disciplinary model to ensure the connectivity of the accessible space; and discretizing the accessible space according to preset node and edge mapping rules to construct an evacuation topology map. Based on the geometric and attribute data, parameterized attributes are calculated and assigned to the edges and nodes. The parameterized attributes of the edges include at least spatial length, effective passage width, passage capacity parameter, availability status, passage direction attribute, and elevation difference. The parameterized attributes of the nodes include at least area, floor elevation, and node type code. The parameterized attributes are bound to the corresponding nodes and edges, and associated with the source BIM component identifiers to form the evacuation network model. Among the preset nodes and edges, the nodes include at least room area nodes, corridor segment nodes, stair segment nodes, platform nodes, and exit buffer nodes, and the edges include at least doorway connecting edges, opening connecting edges, and staircase up-down connecting edges.

[0007] Based on the evacuation network model, the time-series data of the evacuation process obtained through pathfinder simulation includes: mapping the structure and parameters of the evacuation network model to the pathfinder simulation environment, configuring the physical width of the corresponding passage connection components and one-way or two-way passage rules based on the effective passage width and passage direction attributes of the edges in the evacuation network model, and configuring the initial crowd distribution and behavior parameters to form a simulation instance. Run the baseline evacuation process of the simulation instance and record the dynamic state logs related to the edges and nodes in the evacuation network model during the simulation process; extract the state indicators of each edge and each node at multiple consecutive time points from the dynamic state logs according to a unified time base, thereby obtaining the edge time series features and node time series features; integrate the edge time series features and node time series features to construct a graph time series input tensor that is aligned with the structure of the evacuation network model and contains the time dimension, as the time series data.

[0008] The status indicators include at least the real-time personnel flow, instantaneous personnel density, average passage speed and / or number of people waiting in queues of the edge, and the real-time number of people staying at the node and / or inflow and outflow.

[0009] Based on the time series data, the dynamic traffic status of the evacuation network is predicted using a graph neural network-based prediction model, and dynamic evacuation guidance parameters are generated, including: based on the graph time series input tensor and the topology of the evacuation network model, the prediction results of the future traffic status of the edges in the evacuation network are output through the graph neural network prediction model. The prediction results are converted into dynamic cost values ​​of edges in the evacuation network, and the dynamic cost values ​​are corrected according to the available state attributes of the edges in the evacuation network model to generate initial dynamic evacuation guidance parameters. Time smoothing and update hysteresis constraints are applied to the initial dynamic evacuation guidance parameters to suppress parameter mutations and frequent updates, and stable dynamic evacuation guidance parameters are generated as output.

[0010] According to the dynamic evacuation guidance parameters, the evacuation strategy of the pathfinder is adjusted and iterative simulation optimization is performed to obtain an optimized evacuation scheme that meets the predetermined conditions. This includes: based on the dynamic evacuation guidance parameters, determining whether the predetermined replanning triggering conditions are met, and generating a replanning control variable for adjusting the evacuation path when the conditions are met. The replanning control variable is injected into the pathfinder simulation environment to drive a new round of evacuation simulation and obtain the performance index of the new round of simulation. The performance index is judged according to the predetermined iteration convergence condition. If the convergence condition is met, the optimal strategy in the historical iteration is output as the optimized evacuation scheme; otherwise, the input is updated with the time series data obtained from the new round of simulation, and the iteration optimization continues.

[0011] Writing back the optimized evacuation plan and integrating it into the BIM model to generate an updated BIM model includes: extracting the parameterized strategy result set from the optimized evacuation plan; writing the strategy data in the parameterized strategy result set into the attribute information of the corresponding component in the BIM model through the source BIM component identifier associated with the evacuation network model; and generating an updated BIM model based on the written attribute information.

[0012] A system employing a pathfinder-based method to improve building fire evacuation efficiency, characterized by comprising: BIM Analysis and Modeling Module: Analyzes the BIM model and generates an evacuation network model that includes the topology and attribute parameters of accessible spaces; Simulation data acquisition module: Based on the evacuation network model, time-series data of the evacuation process are obtained through pathfinder simulation; Prediction and guidance module: Based on the time series data, it uses a graph neural network-based prediction model to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters. Iterative replanning and optimization module: Based on the dynamic evacuation guidance parameters, adjust the evacuation strategy of pathfinder and perform iterative simulation optimization to obtain an optimized evacuation scheme that meets the predetermined conditions. BIM Write-back Integration Module: Writes back the optimized evacuation plan and integrates it into the BIM model to generate an updated BIM model.

[0013] The beneficial effects of this invention are as follows: By combining BIM models with Pathfinder simulation, this invention achieves a closed loop from building design to fire evacuation optimization. Through the prediction and optimization of dynamic evacuation guidance parameters, not only is the accuracy and real-time performance of evacuation strategies improved, but also the flexibility and effectiveness of evacuation route and exit selection are enhanced through iterative optimization of replanning control variables. The technical problem addressed by this invention is that existing Pathfinder-based building evacuation simulation and evaluation methods suffer from several issues, including a lack of unified topology and parameterization mapping from BIM to the evacuation model, poor reproducibility and component traceability, difficulty in forming time-series characteristics of simulation logs with a unified time base, inability to predict congestion evolution, lack of dynamic guidance parameter-driven replanning iteration closed loop and result write-back and solidification mechanism, and the problem of how to achieve dynamic evacuation guidance based on time-series prediction and traceably integrate optimization strategies back to the BIM model to form a deliverable updated model. Attached Figure Description

[0014] Figure 1 The overall flowchart of the method for improving building fire evacuation efficiency based on pathfinder provided in the embodiments of the present invention is shown. Detailed Implementation

[0015] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0016] See Example 1 Figure 1 A method for improving building fire evacuation efficiency based on pathfinder includes the following steps: S1. Analyze the BIM model to generate an evacuation network model containing the topology and attribute parameters of accessible spaces, including: Acquire the components and their geometric and attribute data related to evacuation calculations from the BIM model, and perform coordinate system transformation, elevation unification, and geometric error correction on the multi-disciplinary models to ensure the connectivity of passable spaces; discretize the passable spaces according to preset node and edge mapping rules to construct an evacuation topology map; calculate and assign parameterized attributes to edges and nodes based on geometric and attribute data. The parameterized attributes of edges include at least spatial length, effective passage width, passage capacity parameters, availability status, passage direction attributes, and elevation difference. The parameterized attributes of nodes include at least area, floor elevation, and node type code; bind the parameterized attributes to the corresponding nodes and edges, and associate them with the source BIM component identifiers to form an evacuation network model.

[0017] Among the preset nodes and edges, the nodes include at least room area nodes, corridor segment nodes, stair segment nodes, platform nodes, and exit buffer nodes, and the edges include at least doorway connecting edges, opening connecting edges, and staircase up and down connecting edges.

[0018] Step S1 specifically includes: In this embodiment, step S1 is used to parse the building BIM model into an evacuation network model. The evacuation network model uses an evacuation topology graph as its core data structure, consisting of a set of nodes and a set of edges, and assigns parameterized attributes to the nodes and edges respectively. Simultaneously, both nodes and edges are associated with source BIM component identifiers to form traceable and recalculated input objects, providing a unified index basis for the subsequent Pathfinder mapping in step S2 and the BIM write-back in step S5. S1-1 Component Selection and Geometric / Attribute Data Acquisition First, import the BIM model and filter out components and spatial objects related to evacuation calculations. Component objects include at least doors, corridors, stairs, platforms, vestibules, and exits; spatial objects include at least the accessible space boundaries formed by walls, floors, and fire compartments. For each object, read its geometric data (outline, centerline, clear width, key point coordinates, elevation, etc.) and attribute data (component category, door opening direction, normally closed state, access restrictions, floor, zone code, etc.). Simultaneously, write the unique identifier (such as GUID or equivalent) of each object in the BIM into the object index table, which serves as the primary key field for subsequent "node / edge—BIM component" associations, ensuring that any element in the network model can be located back to the original BIM object.

[0019] S1-2. Unification of Multi-Disciplinary Models: Coordinate System Transformation, Elevation Unification, and Geometric Error Correction Since the coordinate datum and elevation datum of multi-disciplinary sub-models may be inconsistent, this embodiment first maps the geometric points of each sub-model to a unified engineering coordinate system. For the coordinate vector of a geometric point in any sub-model, the following formula is used for coordinate transformation:

[0020] in: For point coordinate vectors in a unified coordinate system; This is the coordinate transformation matrix; This is the point coordinate vector in the sub-model coordinate system; It is a translation vector.

[0021] After coordinate transformation, floor elevations are standardized to ensure that spatial objects and component objects on the same floor have a consistent elevation benchmark. Geometric error repair is then performed to guarantee the stable identification of accessible space boundaries, doorways, and openings. Geometric error repair employs an automated "tolerance threshold-driven" criterion: boundary endpoints with distances less than a preset tolerance threshold are snapped together, and areas with gaps between parallel boundaries smaller than the tolerance threshold are closed. Specifically, for any two endpoints to be merged... and The following criteria are used to determine whether adsorption and merging have occurred:

[0022] in, As endpoints Coordinate vectors in a unified coordinate system As endpoints Coordinate vectors in a unified coordinate system For L2 norm operations; This is the geometric adsorption tolerance threshold.

[0023] in, The accuracy of BIM modeling and the allowable deviation of the project can be set to a fixed value (e.g., centimeter or millimeter level) to achieve automated and reproducible error correction. Through the above processing, subsequent connectivity identification based on objects such as doorways, openings, and stairs does not rely on subjective human judgment, avoiding topological disconnection caused by minor misalignments.

[0024] S1-3. Discretize according to the node and edge mapping rules and construct a sparse topology graph. After data unification, this embodiment discretizes the passable space according to preset node and edge mapping rules to construct an evacuation topology map. The rules are derived from BIM object semantics (room, corridor, staircase, etc.), geometric relationships (adjacent, intersecting, connected), and general modeling constraints for engineering evacuation calculations (such as corridor segmentation, staircase segmentation, exit buffer settings, etc.), thereby ensuring the objectivity and operability of the rules. The specific rules are as follows.

[0025] (1) Node generation rules The nodes should include at least room area nodes, corridor segment nodes, staircase segment nodes, platform nodes, and exit buffer nodes: Room area node: For each room / anteroom / lobby and other spatial objects, take the geometric center of its passable area as the node position, and record its floor and space identifier; Corridor Segmentation Nodes: The corridor centerline is segmented, with segmentation rules considering both "length interval thresholds" and "segmentation at turning points." Specifically, when the cumulative length of the corridor centerline exceeds a preset segmentation length threshold... Segment nodes are generated automatically; when the centerline turns and the angle exceeds a preset angle threshold, a segment node is generated at the turning point. The position of the segment node is taken as the center point of the corresponding segment. Staircase segment nodes: Nodes are generated with the starting and ending platforms of the staircase segments as key points, and their floor elevations are recorded to construct vertical connectivity. Platform node: Generate a node by taking the geometric center of the platform region; Export Buffer Node: A buffer area is set up outside the export and a virtual node is generated to represent the consistency between the external reachability status of the export and the statistical caliber. The buffer distance can be set to a fixed threshold and bound to the export object identifier.

[0026] (2) Edge generation rules The edges must include at least the edges connecting the doorway, the edges connecting the opening, and the edges connecting the upper and lower parts of the staircase: Doorway Connecting Edge: When a doorway object geometrically connects two passable space boundaries (such as a room and a corridor), a doorway connecting edge is established between the corresponding nodes on both sides, and the doorway object identifier is recorded as the source BIM component identifier of the edge. Opening connection edge: When there is an opening connection between two corridor segments or between a corridor and a platform / anteroom, an opening connection edge is established between the corresponding nodes, and the component or space boundary identifier where the opening is located is associated. Staircase vertical connection edge: When a staircase object spans different floors and there are stair segments connected, establish a staircase vertical connection edge between the low-elevation staircase segment node and the high-elevation staircase segment node, and associate it with the staircase object identifier.

[0027] Based on this, an evacuation topology map is generated, along with a node table, an edge table, and an adjacency index. The node table must contain at least the node ID, node type code, node coordinates, floor elevation, and source BIM component / space identifier; the edge table must contain at least the edge ID, starting node ID, ending node ID, edge type, and source BIM component identifier; the adjacency index is used to quickly query the set of adjacent edges of any node to support subsequent simulation mapping and graph temporal feature organization.

[0028] S1-4. Calculate and assign parameterized attributes (edge ​​attributes and node attributes). After the evacuation topology is formed, this embodiment calculates and assigns parameterized attributes to each edge and each node, and the obtained attributes are directly written into the edge attribute table and the node attribute table.

[0029] S1-4.1, Spatial length of the edge, effective passage width and passage capacity parameters For any side, first calculate its spatial length. Let the coordinate vectors of the starting and ending nodes of this side in a unified coordinate system be respectively... and The spatial length of the edge is calculated using the following formula:

[0030] in: This is the spatial length of the edge; Let the coordinate vector be the starting node coordinate vector of this edge; Let this be the coordinate vector of the endpoint node of the edge; This refers to the 2-norm operation.

[0031] The effective passage width is then calculated. For the connected sides of the doorway, the clear width of the doorway and the clear width of the connecting area jointly constrain the actual passable width; this embodiment uses the smaller of the two as the basis and deducts a safety margin. Let the clear width of the doorway be... The net width of the connecting area is Safety margin is The effective passage width is determined by the following formula:

[0032] in: This is the effective passage width of this side; The clear geometric shape of a doorway or opening; The geometric clear width of the corridor / connecting area; For safety margin.

[0033] in, In this embodiment, the method is determined based on objective criteria: a fixed value range (e.g., centimeter level) can be selected based on ergonomic data (such as shoulder width and lateral swing margin) or engineering design specifications, and stored in the configuration table as input parameters of the method, so that different engineering projects can complete the parameter implementation under a unified algorithm framework.

[0034] In obtaining Then, the geometric width is mapped to a capacity parameter. Let the capacity coefficient per unit effective width be... The traffic capacity parameter is calculated using the following formula:

[0035] in: This is the passage capacity parameter for that edge; The capacity coefficient per unit effective width; This represents the effective passage width of this side.

[0036] in, The value is determined based on the following criteria in this embodiment: it can be set with reference to the standard recommended values ​​for evacuation projects, authoritative manuals, or existing project statistical data, and written into the configuration table as a configurable parameter (e.g., different values ​​can be set according to building functional zones). This ensures that traffic capacity parameters are both reusable and readily implementable in engineering. The above... , , After being written to the edge attribute table, it will be directly used in step S2 to configure the physical width and flow constraints of the Pathfinder connection component, reducing inconsistencies caused by repeated manual configuration.

[0037] The elevation difference, availability, and traffic direction attributes of side S1-4.2. For any side, calculate the elevation difference between its two endpoints. Let the floor elevation of the starting node be . The floor elevation of the endpoint is The elevation difference is calculated using the following formula:

[0038] in: The difference in elevation between the two nodes on this side; The floor elevation of the starting node; This refers to the floor elevation of the endpoint node.

[0039] In this embodiment, the availability status and passage direction attributes are assigned as discrete fields: Availability status: Primarily determined by BIM attributes (e.g., normally closed, sealed, under maintenance, door type, etc.). When BIM attributes are insufficient to cover fire condition constraints, pre-defined rules are used to supplement them. The pre-defined rules are derived from building fire protection design requirements and engineering evacuation strategy constraints. For example, when a door object is a fire door and its attributes indicate that it should be closed during a fire, the initial availability status of the corresponding door opening connecting edge is marked as "unavailable during a fire." The passage direction attribute is determined by the door opening direction (one-way control), the stair up and down management strategy, or the component attribute, and is written into the edge attribute table so that step S2 can configure one-way / two-way passage rules, and step S3 can perform cost correction or prohibition processing.

[0040] S1-4.3, Node area, floor elevation, and node type code For node attributes, this embodiment calculates at least the area, the floor level, and the node type code: For nodes with closed contours, such as room area nodes and platform nodes, the area is directly calculated from the projected polygon of its passable area and written into the node attribute table. For corridor segment nodes, to ensure consistency between the calculated area ratio and cross-project calculations, this embodiment calculates the area under the engineering simplification assumption that "segments are approximated as rectangles". Let the corridor segment length be... The segmented net width is The estimated area of ​​the corridor segment nodes is calculated using the following formula:

[0041] in: Estimate the area of ​​the segmented nodes of the corridor; The length of each segment of the corridor; The net width of the corridor is divided into segments.

[0042] The aforementioned rectangular approximation is a common simplification method in engineering calculations. Its purpose is to provide stable scale quantities for subsequent node features (such as density, dwell time normalization, etc.) without introducing complex surface region calculations. The floor elevation to which the node belongs is read and written by the BIM floor object; the node type code is assigned and written according to types such as "room area, corridor segment, stair segment, platform, exit buffer", etc., for consistent coding of different node types in subsequent diagram structure input.

[0043] S1-5 parameter binding is associated with the source BIM component identifier to form the evacuation network model output. After completing the parametric attribute calculation for nodes and edges, the parametric attributes are bound to the corresponding nodes and edges in the evacuation topology diagram, and the source BIM component identifier is written for each node and edge (the node corresponds to the spatial object identifier, and the edge corresponds to the component identifier such as door / opening / stairs).

[0044] The final evacuation network model data object includes at least a node table, an edge table, and an adjacency index. This evacuation network model serves as the input for step S2, used to construct the Pathfinder simulation environment. Simultaneously, since both nodes and edges retain the source BIM component identifier, subsequent step S5 can write the optimized evacuation scheme back to the corresponding BIM component via attribute writing, achieving traceable integration from calculation results to the engineering model.

[0045] Parameter configuration table: To ensure the reproducibility of coordinate unification, geometric error repair, corridor segmentation discretization, and capacity parameterization mapping in step S1, this embodiment uses a geometric tolerance threshold. Corridor segment length threshold Safety margin Unit effective width traffic capacity coefficient The parameters are uniformly written into the parameter configuration table, and a default range is given for different building types.

[0046] The values ​​of the above parameters are based on: BIM modeling accuracy and engineering tolerance (…). ), the discretization resolution requirements of the evacuation topology map and the geometric scale of the corridor ( Ergonomic clearance requirements and evacuation route design habits ), empirical data on flow rate per unit width in evacuation dynamics and related standards / manuals ( ).

[0047] In specific project applications, the corresponding default group can be selected according to the building function and population characteristics without changing the algorithm flow, and fine-tuning is allowed within a given range.

[0048]

[0049] Table 1. Default range and value source of key parameters (grouped by building type) in: Criterion for geometric error correction Its function is to eliminate the "pseudo-disconnection" caused by modeling errors, so that the connection relationships such as the doorway connection edge and the opening connection edge can be stably identified.

[0050] Used for generating segmented nodes in corridors, this method balances the "computability" and "detail representation" of corridor topology discretization; it is particularly useful in complex scenarios with dense crowds and long corridors. A larger upper limit is usually chosen to control the node size.

[0051] Used for calculating effective passage width Its value can be appropriately increased in hospital settings to reflect the clearance requirements for the passage of stretchers, wheelchairs, etc.

[0052] Used for calculating traffic capacity parameters The values ​​can be adjusted according to the building's function and the characteristics of people's movement; when the project adopts more conservative evacuation assumptions, the lower limit can be taken from the range in the table.

[0053] S2. Based on the evacuation network model, time-series data of the evacuation process is obtained through Pathfinder simulation, including: mapping the structure and parameters of the evacuation network model to the Pathfinder simulation environment; configuring the physical width of corresponding passage connection components and one-way or two-way passage rules based on the effective passage width and passage direction attributes of the edges in the evacuation network model; configuring initial crowd distribution and behavioral parameters to form a simulation instance; running the baseline evacuation process of the simulation instance and recording dynamic state logs related to the edges and nodes in the evacuation network model during the simulation; extracting state indicators of each edge and node at multiple consecutive moments from the dynamic state logs according to a unified time base to obtain edge time-series features and node time-series features; integrating the edge time-series features and node time-series features to construct a graph time-series input tensor aligned with the structure of the evacuation network model and containing a time dimension, as time-series data. State indicators include at least the real-time pedestrian flow, instantaneous pedestrian density, average passage speed and / or number of people waiting in queues for the edges, and the real-time number of people residing at the nodes and / or inflow and outflow.

[0054] Step S2 specifically includes: In this embodiment, step S2 is used to construct a Pathfinder simulation instance based on the evacuation network model output in step S1, run the baseline evacuation process and record dynamic state logs with consistent indices of nodes and edges in the evacuation network model; then the logs are aligned according to a unified time base, and the state indicators of edges and nodes at continuous time points are extracted to form edge time series features and node time series features; finally, the features are organized into a graph time series input tensor containing the time dimension according to the structure and indexing rules of the evacuation network model, as the time series data output of the evacuation process, so as to directly support the subsequent step S3 for predictive modeling of the dynamic passage status of the evacuation network.

[0055] S2-1 Simulation Example Construction: Mapping and Configuring the Network Model to the Pathfinder Environment (1) Input reading and index fixed The evacuation network model generated in step S1 is read. The evacuation network model includes at least a node table, an edge table, and an adjacency index. To ensure consistency between subsequent log structuring and graph tensor construction, this embodiment uses the node ID sequence... With edge ID sequence The node ID sequence is fixed to a predetermined order in step S1 (e.g., ascending order by ID or generation order) and remains unchanged throughout step S2. With edge ID sequence The threshold for corridor segment length is determined by step S1. After discretization and graph construction are completed under constraints, the following is determined: The corridor segment length threshold is configured in the parameter configuration table for step S1; therefore, The simulation object size and spatial resolution in this embodiment are indirectly constrained by the node set and topological granularity.

[0056] (2) Topological mapping Map the nodes in the node table to room areas or waiting areas in the Pathfinder simulation environment; map the edges in the edge table to access connection components (doorways or connection components) connecting rooms / areas. During mapping, ensure that each mapped object in the simulation environment carries its corresponding node ID or edge ID, so that simulation logs can be written to disk in the format of "object ID - timestamp", avoiding object mismatch during subsequent feature extraction.

[0057] Furthermore, the evacuation topology map has been prepared in step S1 based on... Geometric error repair and endpoint adsorption closure were completed, among which The geometric adsorption tolerance threshold is set in the parameter configuration table for step S1. Therefore, the connectivity between the doorway connecting edge, the opening connecting edge, and the staircase connecting edge is stable when mapped to Pathfinder, which can reduce the risk of simulation topology distortion caused by "pseudo-disconnection" due to BIM geometric misalignment.

[0058] (3) Configuration of the physical width and passage direction of the passage connection components For any side, directly read the effective passage width from the side attribute table in step S1. and will This serves as the physical width configuration value for the corresponding Pathfinder connection component; simultaneously, it reads the passage direction attribute field of the edge and maps it to a one-way or two-way passage rule.

[0059] in, For step S1 Calculate and write the effective passage width field in the edge attribute table. This is the safety margin specified in the parameter configuration table for step S1. Therefore, the configuration of the physical width of the connecting components in S2 not only inherits the BIM geometric constraints ( It also explicitly inherits the safety margin. The ergonomic clearance corrections incorporated in the simulation make the simulated accessibility more closely resemble actual engineering conditions. in, For effective passage width, The geometric clear width of the doorway or opening. The geometric clear width of the corridor or connecting area. For safety margin.

[0060] (4) Initial population distribution configuration and behavioral parameter configuration An initial crowd distribution is generated for the room / area corresponding to each node. The node area field is taken from the node attribute table in step S1, denoted as... ; Among them: room area nodes, platform nodes, etc. The projected area of ​​the passable area; the corridor segment nodes. The formula for calculating the approximate area of ​​the rectangle in step S1 Obtain and write to the node attribute table. For the node... The initial number of people is determined using the following formula:

[0061] in, For nodes Initial number of people in the corresponding area For nodes Personnel load parameters per unit area of ​​the functional zone For the node in the node attribute table of step S1 The area field; This is the rounding operator.

[0062] in, Based on the load settings for engineering personnel or relevant specifications, these parameters can be written into the simulation instance configuration table to ensure the objectivity and verifiability of personnel distribution settings under different projects or scenarios. Subsequently, behavioral parameters such as movement speed distribution, reaction time delay, and individual scale parameters are configured for different population categories and fixed as the initialization configuration of the simulation instance. To ensure the comparability of subsequent iteration simulation results, this embodiment records and fixes the random seed value during simulation instance initialization, so that the individual attributes of randomly sampled populations remain reproducible across different rounds of simulation.

[0063] S2-2 Run the baseline evacuation simulation and record the dynamic status log. After completing the simulation instance construction, run the Pathfinder baseline evacuation process. The simulation process follows the spatial connectivity, connection component width, and traffic direction rules described by the evacuation network model. During the simulation run, record dynamic state logs corresponding one-to-one with the edges and nodes in the evacuation network model. The dynamic state logs must contain at least the following fields: Timestamp field (corresponding to simulation time); Edge ID field and node ID field (used for...) , Alignment); Edge-related status fields: real-time personnel flow related count information, instantaneous personnel density related number of people on the edge information, average passage speed information, and queue waiting number information; Node-related status fields: real-time number of residents, and count information related to inflow and outflow.

[0064] The above log field design ensures that any log entry can be located to a unique edge or node in the evacuation network model through the edge ID or node ID, thus providing an object-level consistency basis for subsequent extraction of status indicators based on a unified time benchmark.

[0065] S2-3, Log field and index consistency verification (including capacity consistency verification, explicitly introduced) ) To avoid errors in time-series feature construction due to missing logs, index drift, or inconsistent object mapping, this embodiment performs a consistency check between log fields and the index before extracting status metrics. The check includes at least the following: (1) ID completeness check: Check whether there are missing edge ID or node ID fields in the dynamic status log. If missing, the record is determined to be unusable for structured extraction and marked as invalid. (2) ID validity check: Check whether the set of edge IDs in the log is valid. Are the subsets and node ID sets of ? If an ID not in the set is found in a subset of the set, a mapping error is determined and an error list is recorded. (3) Coverage check: For each sampling time window, check whether there is a complete missing case of "available edges / nodes"; if an edge or node has no log record in the continuous sampling window, it is recorded as a missing object and missing value processing is triggered; (4) Consistency check: Check whether there are multiple conflicting records under the same timestamp and the same side ID (or node ID); if there is a conflict, merge them according to the preset priority rules (e.g., using a higher confidence source or using average / maximum value aggregation); (5) Missing value handling method: When a certain edge or node is missing a certain state field at the sampling time, the value of the field is obtained by nearest neighbor preservation (forward filling) or time interpolation, and the missing value handling flag is recorded so that the field can be selectively used or masked during subsequent model training; (6) Capacity consistency verification (based on the capacity parameters in step S1): To verify the consistency between the counting aperture and the mapping of the opposite edge, this embodiment is based on the unit effective width capacity coefficient in the parameter configuration table of step S1. The effective passage width in the edge attribute table of step S1 Calculate the edges Traffic capacity parameters And further calculate the edges In the interval Saturation ratio within The following formula is used: ; ; in: For the edge Traffic capacity parameters; The capacity coefficient per unit effective width (from the parameter configuration table in step S1); For the edge Effective passage width (from the edge attribute table in step S1); For the edge In the interval Saturation ratio within; For the edge In the interval Real-time pedestrian flow within the area.

[0066] when When multiple consecutive time windows are significantly greater than 1, it is determined that the log counting caliber, time window division, or edge ID mapping of that edge may be abnormal, triggering a review and aggregation correction of the edge ID mapping relationship, counting window, and conflict records. Through the above verification and processing, the edge / node feature sequences constructed subsequently are consistent in terms of object index, field meaning, and time alignment, avoiding the mistaken input of log anomalies into the graph time series tensor.

[0067] S2-4. Unified Time Base Alignment and Status Indicator Extraction Pathfinder dynamic status logs may be event-triggered outputs with uneven timestamp intervals. To transform the logs into a time-series input with equal-step size that can be used for graph neural network learning, this embodiment sets a uniform sampling time interval. Construct a sampling time sequence.

[0068] No. Each sampling time is determined according to the following formula:

[0069] in: For the first Each sampling time; This is the start time of sampling; The sampling sequence number; To standardize the sampling time interval.

[0070] At each sampling time Align the log fields of edges and nodes: when the original log timestamp does not fall exactly on At that time, the interpolation alignment method of adjacent timestamps is used to obtain... The corresponding state values ​​are used to form a sequence of edge / node states with equal step sizes. This process ensures that each feature dimension has a consistent time step size and a strict time index during subsequent time window modeling.

[0071] Then, the state indicators of edges and nodes are extracted: 1) Real-time pedestrian flow on the edge opposite side In the interval Internal statistics are achieved through counting. The real-time pedestrian flow is calculated using the following formula:

[0072] in: For the edge In the interval Real-time pedestrian flow within the area; For the edge In the interval The number of people inside is counted; To standardize the sampling time interval.

[0073] 2) Instantaneous population density at the edge To ensure that the density index is completely consistent with the geometric and passage parameters of step S1, this embodiment specifies the sampling time. Read edges The number of people on the side at an instant and use the edge length from the edge attribute table in step S1 With effective passage width Calculate instantaneous population density:

[0074] in: For the edge At sampling time Instantaneous population density; For the edge At sampling time The number of people on the side at any given moment; For the edge Spatial length; For the edge Effective passage width.

[0075] 3) Average passage speed and number of people waiting in line At sampling time Read edges from the dynamic state log average traffic speed and the number of people waiting in line The above indicators and , Together they form an edge At any moment The temporal characteristics of the edges.

[0076] 4) Real-time number of users, inflow and outflow of nodes. For nodes At the sampling time Read the number of users staying in the game in real time and in the interval Internal statistics inflow count With outflow count These serve as indicators of node inflow and outflow. All node and edge indicators use the same sampling time base to ensure consistency in subsequent graph structure alignment.

[0077] S2-5 Feature Organization and Construction of Graph Temporal Input Tensors To generate time-series data that can be directly used as input for the subsequent step S3, this embodiment performs sampling at each sampling time. Construct node feature matrices and edge feature matrices, and use a fixed index sequence. , Ensure the row order remains unchanged across time steps: Node feature matrix The row correspondence The Middle One node; Edge feature matrix The row correspondence The Middle Edge.

[0078] The node feature matrix and edge feature matrix are defined as follows: ; ; in: Sampling time The node feature matrix, For nodes At any moment The real-time number of resident users; For nodes In the interval Inflow count within; For nodes In the interval Outflow count within; A fixed sequence of node IDs; Sampling time The edge feature matrix, For the edge In the interval Real-time pedestrian flow within the area For the edge At any moment Instantaneous population density For the edge At any moment Average traffic speed, For the edge At any moment Number of people waiting in line It is a fixed sequence of edge IDs (derived from the edge table order in step S1).

[0079] Finally, they will be arranged in chronological order. Encapsulated together with the topological information (adjacency index or edge start and end node index) output in step S1, we obtain the graph temporal input tensor. . As the evacuation process time series data output by step S2, the topology remains unchanged, the state indices of nodes and edges evolve over time, and the feature matrix at any time has a clear and verifiable index correspondence with the nodes and edges in the evacuation network model, thus completing the deterministic transformation from Pathfinder simulation logs to learnable graph time series data.

[0080] S3. Based on time-series data, a graph neural network-based prediction model is used to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters. This includes: based on the graph time-series input tensor and the topology of the evacuation network model, the graph neural network prediction model outputs the prediction results of the future traffic status of edges in the evacuation network; the prediction results are converted into the dynamic cost value of edges in the evacuation network, and the dynamic cost value is corrected according to the available state attributes of edges in the evacuation network model to generate initial dynamic evacuation guidance parameters; time smoothing and update hysteresis constraints are applied to the initial dynamic evacuation guidance parameters to suppress parameter mutations and frequent updates, generating stable dynamic evacuation guidance parameters as output.

[0081] Step S3 specifically includes: Based on the evacuation network model formed in step S1, a prediction model based on graph neural network is constructed using the evacuation process time series data obtained in step S2. The prediction results of the future passage status of each edge in the evacuation network model are output, and the prediction results are converted into the dynamic cost value of the edge. At the same time, time smoothing and update lag constraints are applied to form dynamic evacuation guidance parameters that can be used to adjust the Pathfinder evacuation strategy in the subsequent step S4.

[0082] S3-0 Input Objects and Reference Basics (1) Structural basis from step S1: Step S1 has discretized the passable space to form an evacuation topology graph consisting of a set of nodes and a set of edges, and formed an evacuation network model; wherein the nodes include at least room area nodes, corridor segment nodes, stair segment nodes, platform nodes and exit buffer nodes, and the edges include at least doorway connecting edges, opening connecting edges and staircase connecting edges; and for each edge, step S1 calculates and binds its parameterized attributes, including at least the spatial length. Effective passage width Available status With the passage direction attribute; where the effective passage width Safety margin parameters have been introduced in the net clearance correction according to step S1. The capacity mapping coefficient is .

[0083] (2) Data basis from step S2: Step S2 has recorded and parsed the dynamic status log during the Pathfinder simulation, and extracted and aligned it according to a unified time base and a unified sampling time interval to obtain the graph time-series input tensor consistent with the evacuation topology graph index as the time-series data. To avoid conflict with the parameter table in step S1... The (corridor segment length threshold) is ambiguous; in this embodiment, the unified sampling time interval in step S2 is denoted as... and in discrete time steps Indicates the first Each sampling time corresponds to an actual time of [number]. .

[0084] (3) Basis for topological matrix representation: Adjacency matrix is ​​derived from the sparse topological graph in step S1. Where any node and another node have a direct connectivity relationship represented by an edge set in the evacuation topology graph. ,otherwise This matrix is ​​merely an equivalent representation of the sparse topology graph and does not change the definition or type of nodes / edges.

[0085] S3-1 Constructing the node feature matrix and hidden states from the graph temporal input tensor (improved expression) The graph time-series input tensor obtained in step S2 contains state indicators of each node at multiple consecutive time steps (e.g., real-time number of users, inflow, outflow, etc.); this step involves each discrete time step... Organize node features into a matrix Its row index is consistent with the node index of the evacuation topology graph in step S1, and the node hidden state matrix is ​​initialized. (For example, set to zero or obtained by linear mapping of features from the first frame). Wherein... and All indexes are kept consistent with the constraints of "log field and index consistency verification" in step S2 to ensure that subsequent prediction outputs can be backmapped to the specific nodes and edges in step S1.

[0086] S3-2 Time-varying attention adjacency with topological constraints (improved expression) To enable the model to adaptively characterize the "strength of congestion impact" without altering the connectivity of the evacuation topology, this embodiment modifies the adjacency matrix. The time-varying attention weights are calculated for adjacent node pairs to obtain the time-varying adjacency weights. : ; ; Formula S3-1 in: For node indexing; For nodes Index of directly connected nodes in the evacuation topology graph; It is a discrete time step and corresponds to the actual time. ; These are the elements of the adjacency matrix derived from the sparse topology graph; For matrix Middle node At time step The feature vectors are derived from the temporal features of node S2 in step S2; For matrix Middle node At time step The hidden state vector; The unnormalized attention score; To normalize attention weights; The adjacency weights are subject to topological constraints and are time-varying. and It is a trainable mapping matrix; These are trainable projection vectors; The hidden state dimension scaling factor; It is the hyperbolic tangent function; It is an exponential function.

[0087] Explanation of the range: And for fixed satisfy , Furthermore, it is non-zero only on adjacencies allowed in the evacuation topology graph. This ensures that the attention propagation strength is learnable without violating engineering topology constraints.

[0088] Spatiotemporal Coupling Update of Gated Loop Unit in Figure S3-3 exist Under the constraints, we first perform neighborhood aggregation on the hidden state of nodes, and then perform gated update to obtain the hidden state of nodes that includes the influence of historical evolution and neighborhood propagation:

[0089] in, For time steps The node feature matrix is ​​composed of node state indicators extracted in step S2; For time steps The hidden state matrix of the nodes; The time-varying adjacency weight matrix obtained from (Equation S3-1); It is a neighborhood aggregation matrix; To reset the gate matrix; To update the gate matrix; The candidate hidden state matrix; The updated hidden state matrix for the nodes; The input is the mapping weight matrix; The neighborhood state mapping weight matrix; This is the bias matrix; It is a matrix of all 1s; For the Sigmoid function; It is the hyperbolic tangent function; This is an element-wise product.

[0090] Explanation of the range: and element-wise Internally, selective memory updates are implemented. The normalized hidden representation is used to carry dynamic traffic situation information of the node layer.

[0091] S3-4 Predicting the future travel state of edges generated from the hidden states of nodes. Step S1 has clarified the edge set and its corresponding endpoint node relationships, so for any edge in the edge set... (its endpoint node is) and In this embodiment, the predicted future traffic status of an edge is constructed using the hidden state vectors of the two endpoints. The predicted value is defined as the normalized congestion intensity or delay risk.

[0092] To enable subsequent "time smoothing" to adaptively adjust based on predictive reliability, this embodiment employs... The predicted mean and variance are obtained through sub-random perturbation inference: ; Formula S3-5; in, The edge index is a value that belongs to the edge set in step S1; and For the edge The endpoint node indexes and the connectivity relationships derived from the evacuation topology graph; For matrix Middle node At time step The hidden state vector; For matrix Middle node At time step The hidden state vector; For the edge The predicted traffic status at the next time step; It is a trainable bilinear mapping matrix; and These are trainable projection vectors; It is a biased scalar; The number of random perturbation samplings; For the first Predicted values ​​obtained through random perturbation inference; To predict the mean; To predict variance; For the Sigmoid function; For summation operators.

[0093] Explanation of the range: , This indicates the normalized congestion intensity or delay risk; the closer the value is to 1, the more congested the future will be. It indicates the degree of uncertainty.

[0094] S3-5 Conversion of Prediction Results to Dynamic Cost Values Step S1 has provided the effective passage width of the edge. (Its calculation already includes a safety margin) ) and spatial length The parameter configuration table should specify the capacity coefficient per unit effective width. The values ​​are based on the default range. Therefore, this step first calculates the theoretical travel capacity parameters of the edges according to the mapping method in step S1:

[0095] Then predict the mean Dynamic cost of edges via nonlinear fractional mapping And based on the available state of the edges. Apply an unavailability penalty to ensure consistency with the "Availability Status" attribute in step S1:

[0096] in, The edge obtained in step S1 Effective passage width and safety margin already reflected Net airspace deduction; The unit effective width passage capacity coefficient is the parameter configuration table for step S1; For the edge Theoretical traffic capacity parameters; The edge obtained in step S1 Spatial length; The mean is predicted using equation (S3-5); For the edge The dynamic value of the next time step; and The nonlinear mapping shape parameter is used as the cost. Edges bound for step S1 Available status flags; This is an unavailable penalty constant; It is the natural logarithm function; It is an exponential function.

[0097] Interpretation of the range: When hour Furthermore, it reflects the engineering constraint of "increased length—increased cost, enhanced traffic capacity—decreased cost"; fractional terms make The cost increases non-linearly to enhance sensitivity to congestion; when Time through To create strong penalties to achieve unusable edge avoidance.

[0098] S3-6 Time Smoothing and Update Hysteresis Constraints To meet the requirements of "time smoothing and update hysteresis constraints", this embodiment uses prediction variance. Generate smoothing coefficient The greater the uncertainty, the more conservatively historical values ​​are adopted.

[0099]

[0100] And set the minimum effective interval for update lag. (Based on distance from walking distance, actual corresponding) ):like Then let Otherwise update and make .

[0101] in, For the edge At time step The smoothing coefficient is located in [0,1); It is a smoothness sensitivity parameter and is a positive number; Equation (S3-5) predicts the variance; For the edge At time step The effective smoothing value; For the edge At time step The smoothing cost of the output; The original dynamic cost of equation (S3-7); For the edge Last effective update time step index; The number of steps corresponding to the minimum effective interval; It is an exponential function.

[0102] Explanation of the range: Follow The increase in the value enhances the reliance on historical effective cost values ​​and suppresses frequent jumps under high uncertainty conditions; the hysteresis rule further ensures that the guiding parameters are stable and effective within a short time window, avoiding policy oscillations after subsequent S4 injection into Pathfinder.

[0103] Encapsulation and Output of S3-7 Dynamic Evacuation Guidance Parameters The final output of the dynamic evacuation guidance parameters in step S3 includes at least the effective dynamic cost of each edge in the edge set. and its availability status And maintain a one-to-one correspondence with the source BIM component identifiers bound in step S1. Through this correspondence, the subsequent step S4 can... The parameters are mapped to the passage cost or path preference parameters of the corresponding connecting components in the Pathfinder simulation environment, thereby enabling evacuation strategy adjustment and iterative simulation optimization based on the predicted dynamic passage status.

[0104] To ensure the reproducibility of graph neural network prediction, dynamic cost mapping, and time smoothing and update hysteresis constraints in step S3, this embodiment maps the cost nonlinearity to shape parameters. Unusable penalty constant Uncertainty-driven smoothing sensitivity Random perturbation sampling number Minimum number of steps to take effect They are all written as configuration parameters into the parameter configuration table.

[0105] The values ​​of the above parameters are based on the following: congestion prediction values. To dynamic value The "steepness" requirement of the mapping curve The need for forced avoidance of unusable edges Regarding the uncertainty of forecasting The resulting demand for value updates and stability Trade-off between the accuracy and computational cost of predicting the mean and variance And the engineering stability requirements for avoiding policy oscillations and frequent switching after Pathfinder policy injection. .

[0106] In specific project applications, the corresponding default group can be selected according to building function, population characteristics and acceptance indicators without changing the algorithm flow, and fine-tuning is allowed within a given range.

[0107] Table 2 Default range and value source of key parameters in step S3

[0108] Note: In the table The "relative magnitude range" is given; the actual values ​​are recommended to be based on the project specifications. The typical magnitude is determined after normalization, such that when The time penalty term dominates when comparing path costs, thereby enabling the forced avoidance of unusable edges.

[0109] S4. Based on the dynamic evacuation guidance parameters, adjust the evacuation strategy of the pathfinder and perform iterative simulation optimization to obtain an optimized evacuation scheme that meets predetermined conditions. This includes: based on the dynamic evacuation guidance parameters, determining whether predetermined replanning trigger conditions are met, and generating replanning control variables for adjusting the evacuation path when met; injecting the replanning control variables into the pathfinder simulation environment to drive a new round of evacuation simulation and obtaining the performance indicators of the new round of simulation; judging based on the performance indicators and predetermined iterative convergence conditions, if the convergence conditions are met, outputting the optimal strategy in the historical iterations as the optimized evacuation scheme; otherwise, updating the input with the time-series data obtained from the new round of simulation and continuing iterative optimization.

[0110] Step S4 specifically includes: Based on the evacuation network model formed in step S1 and the evacuation process time series data obtained in step S2, step S3 outputs dynamic evacuation guidance parameters that are updated with discrete time steps, which at least include the time step of each edge in the edge set. The dynamic value of the effective date It maintains a one-to-one correspondence with the edge indexes of the evacuation topology map and the identifiers of the source BIM components, while also monitoring the availability status of the edges. As one of the edge parameterization attributes of step S1, it is inherited and used in step S3 for the unavailability penalty constraint.

[0111] This step is used to: determine whether the predetermined replanning trigger condition is met, generate a replanning control variable when triggered, and inject the control variable into the simulation environment through the Pathfinder application programming interface, thereby running a new round of evacuation simulation and evaluating performance indicators. Finally, based on the iterative convergence condition, the optimal strategy in the historical iteration is output as the optimized evacuation scheme.

[0112] Determining the S4-1 Replanning Trigger Condition at discrete time step (corresponding to actual time) , (From step S2, the unified sampling time interval), this embodiment first is based on the output of step S3. A "current guidance situation assessment" is performed on the evacuation topology. To ensure that the triggering determination is feasible and does not depend on additional external variables, this embodiment uses two complementary criteria: one reflects whether the overall cost level has significantly deteriorated, and the other reflects whether the relative advantages between exits have undergone structural changes. Specifically, for any exit buffer node in the evacuation topology, the effective dynamic cost is determined based on the edge. Find the minimum path cost from any regional node to the exit; when the change in the minimum path cost of the optimal exit compared to the previous time exceeds a threshold, or the difference in the minimum path cost of different exits exceeds a threshold, it is determined that replanning needs to be triggered. By applying dynamic cost to edges on the evacuation topology graph in step S1. The shortest path is calculated using the edge weights, thus giving the triggering criterion a clear and computable basis.

[0113] To formalize the above triggering criteria, let Indicates at time step The "optimal exit minimum path cost" is given by: Let "dispersion of minimum path cost between exits" be represented. Then, the replanning trigger decision can be expressed as: ; ; when or When the replanning trigger condition is met, it is determined that the replanning trigger condition is met. For the preset threshold, To avoid extremely small positive numbers with a denominator of zero.

[0114] in: It is a discrete time step and corresponds to the actual time. , To standardize the sampling time interval for step S2; For edge-based dynamic value The minimum path cost of the optimal exit is calculated on the evacuation topology map; This corresponds to the value from the previous time step; The optimal cost relative rate of change; A discrete measure of the minimum path cost for each exit; The discreteness of the previous time step; This represents the increment of dispersion. To prevent zero from being the smallest positive number; The threshold is triggered by the deterioration at the cost of the outcome; The threshold for triggering export dispersion; It is an absolute value operator.

[0115] Explanation of the range: Furthermore, the larger the value, the more significant the deterioration of the path cost at the optimal exit. The larger the value, the more pronounced the widening of the relative advantage differences among exports, implying that dispersal preferences should be reallocated to balance export utilization.

[0116] S4-2 Generation of Replanning Control Quantities When the replanning triggering conditions are met, this embodiment generates a replanning control quantity for adjusting evacuation routes.

[0117] The design of replanning control quantities follows two constraints: First, the control quantity must be directly received by the Pathfinder simulation environment and applied to the connection component or output selection logic; Second, the control quantity must remain consistent with the edge index of the evacuation topology graph in step S1 so that the edge-level value can be mapped to the corresponding door or connecting component in the Pathfinder.

[0118] Therefore, this embodiment will dynamically calculate the edge activation value. Normalized to edge-travel "preference weights" This results in a higher cost value and a lower weight, while also addressing unusable edges. Apply zero weight directly, thus aligning with the semantics of the unavailability penalty in step S3:

[0119] in This is a scaling parameter used to scale different projects. The dimensions and magnitudes are converted into stable weight ranges. The resulting set is one of the core contents of the replanning control quantity, which can be used to "modify the access cost attribute of connecting components" or "modify personnel's selection preferences for connecting components".

[0120] In engineering implementation, control variables can be further grouped into: suppression of critical bottleneck edges (reduction) Enhancement of alternative channel edges (improvement) And synchronous adjustment of export attractiveness parameters (the path weights to the export are normalized and then written into the export selection logic).

[0121] The key design point is that the control quantity is not specified based on experience, but is obtained by strictly mapping the predicted driving cost output of step S3. Therefore, it can reflect "future congestion" in the current path preference in advance, making the Pathfinder's strategy adjustment interpretable and verifiable.

[0122] in: The edge index in the evacuation topology graph in step S1; For discrete time steps; The edge availability status marker is bound to step S1 and used for unavailability penalty in step S3; The dynamic cost of the edge activation output in step S3; The edge passage preference weight in the replanning control quantity; For weighting scale parameters; It is an exponential function.

[0123] Interpretation of the range: When hour And follow Increase monotonically decrease; when hour This indicates that the edge is forcibly avoided during replanning.

[0124] S4-3 Control Variable Injection into the Pathfinder Simulation Environment and Execution of a New Round of Simulation After generating the replanning control variable, this embodiment performs the injection operation through the application programming interface provided by Pathfinder.

[0125] The injected object maintains the same mapping relationship as in step S2, "mapping the evacuation network model to the Pathfinder simulation environment": in step S2, "edges are mapped to doorways or connecting components," therefore, this step maps each edge... corresponding Write the passage cost or preference field for the doorway or connecting component; for the exit area corresponding to the exit buffer node, the exit attractiveness can be calculated based on the comprehensive weight of the path to the exit and written into the exit parameter field.

[0126] After the injection is completed, start or continue running a new round of evacuation simulation, and record dynamic status logs with the same caliber as in step S2, so as to ensure that the output of the new round of simulation can directly reuse the "unified time base extraction - edge / node temporal features - graph temporal input tensor" process of step S2, without causing inconsistencies between fields and indexes.

[0127] S4-4 Performance Indicator Acquisition and Iterative Convergence Determination After the new round of simulation is completed, this embodiment extracts performance indicators for iterative evaluation. To maintain consistency with the abstraction level in step S4, this embodiment does not strictly limit the unique set of performance indicators, but at least includes global quantities that can support the judgment of "whether it has improved" and "whether it has converged", such as total evacuation completion time, maximum congestion level in key areas, and exit utilization balance.

[0128] For a formal convergence criterion, let... Indicates the first The overall performance score of the iterative simulation (including the baseline cycle) is calculated, with a lower score indicating a better solution; let... This represents the relative improvement amount. When the improvement amount is lower than the threshold or the maximum number of iterations is reached, convergence is determined, and the strategy corresponding to the historical best round is output as the optimized evacuation scheme.

[0129] It can be expressed as follows:

[0130] If convergence is not achieved, the dynamic state log generated in this round of simulation will be used to generate a new graph time series input tensor according to step S2, and returned to step S3 as the update input to update the prediction and dynamic cost, thus entering the next closed-loop iteration of "prediction-trigger-injection-simulation-evaluation".

[0131] in: For iteration round index; For the first The overall performance score of the wheel simulation; This refers to the previous round of scoring; To improve the quantity; This is the convergence threshold; This represents the maximum number of iterations. Interpretation of the range: When This indicates an improvement over the previous round, with larger values ​​indicating more significant improvements; when This indicates that the improvement is insufficient and the marginal benefit of continuing the iteration is low, thus satisfying the engineering convergence criterion.

[0132] The key difference between step S4 and the conventional approach that relies solely on Pathfinder's internal static parameters or a single simulation is that the replanning trigger and control variable generation are directly derived from the dynamic prediction cost of step S3. The driver and control quantity mapping explicitly consider the edge availability state. Therefore, when both "edge unavailability constraint" and "future congestion evolution" exist, an injectable path preference correction amount can be automatically generated; and by unifying the consistency constraint between the time base and the log index in step S2, the simulation output of each iteration can be directly fed back to step S3 for updating, thus forming a verifiable and reproducible experimental link.

[0133] S5. Write back the optimized evacuation plan and integrate it into the BIM model to generate an updated BIM model, including: extracting the parameterized strategy result set from the optimized evacuation plan; writing the strategy data in the parameterized strategy result set into the attribute information of the corresponding component in the BIM model through the source BIM component identifier associated with the evacuation network model; and generating an updated BIM model based on the written attribute information.

[0134] Step S5 specifically includes: Based on step S1, which has generated an evacuation network model from the BIM model and established a binding relationship between "nodes / edges and source BIM component identifiers", step S2 completes the mapping from the evacuation network model to the Pathfinder simulation environment and generates traceable dynamic status logs and time-series data. Step S3 generates the dynamic value of edge activation under a unified time base. Step S4 involves using dynamic evacuation guidance parameters to trigger replanning and generate replanning control variables. The final output is the evacuation strategy with the best performance in historical iterations, designated as the "optimized evacuation plan." Step S5 aims to write this optimized evacuation plan into the attribute information of BIM components in a parameterized and traceable manner without altering existing BIM geometric entities. This ensures that the updated BIM model carries executable evacuation decision information, thereby realizing the transformation from "simulation optimization results" to "engineering design information."

[0135] Extraction and Structured Encapsulation of S5-1 Parameterized Strategy Result Set This embodiment first extracts the parameterized policy result set from the optimal policy determined in step S4. Since the replanning control quantity obtained in step S4 is indexed by edges... and discrete time step For the basic index unit (e.g., by (This represents the adjustment of the access preference for connecting components). Therefore, this embodiment organizes the policy result set into a structured set of entries that can be directly written back. Each policy entry includes at least: policy object type (edge ​​or node), object index (such as edge index). The policy includes the node index, policy content (e.g., recommended travel preferences or exit selection suggestions), effective time window (corresponding to the discrete time step interval), and source policy round identifier (corresponding to the iteration round determined as the historical best in step S4). The object index is a unique sequence number assigned to the node set and edge set when constructing the evacuation topology graph in step S1, thereby ensuring that the policy entries correspond one-to-one with the objects in the evacuation network model and are verifiable.

[0136] In engineering implementation, the strategy result set can be further divided into two categories: 1) Edge-level strategy: used to describe the recommended passage preferences or passage cost adjustments of connecting components such as doorway connecting edges, opening connecting edges, and staircase connecting edges under different time windows; 2) Node-level strategy: Used to describe recommended exits or diversion suggestions for areas such as room area nodes, corridor segment nodes, stair segment nodes, platform nodes, and exit buffer nodes under different time windows.

[0137] The technical significance of the above encapsulation method is that it maintains consistency with the evacuation topology map index system in step S1, and enables strategy entries to be automatically located to specific components through "object index → ​​source BIM component identifier", thereby achieving automation and verifiability of the write-back process, without the need for manual interpretation of which door, corridor, or staircase corresponds to each strategy.

[0138] S5-2 Write-back positioning and attribute field mapping based on source BIM component identifiers In step S1, the nodes and edges of the evacuation network model are associated with source BIM component identifiers (e.g., unique identifiers of components such as doors, corridors, staircases, platform areas, exit areas, or spatial objects enclosed by components).

[0139] Therefore, this embodiment does not directly use geometric coordinates for matching during write-back, but uses the source BIM component identifier as the primary key for positioning: for any strategy entry, the corresponding node or edge in the evacuation network model is first located according to its object index, and then the source BIM component identifier bound to the node or edge is read, thereby determining a unique target component object in the BIM model.

[0140] This method avoids the risk of mismatch caused by minor geometric differences due to geometric error repair and coordinate system transformation in step S1, and can maintain the stability of component traceability relationship after unifying multi-disciplinary models.

[0141] At the attribute writing level, this embodiment uses a "field mapping table" to map the semantics of strategy entries to attribute field values ​​of BIM components. To ensure compatibility across different BIM platforms or project templates, the field mapping table does not require a fixed field name, but rather maps using a preset set of fields. For example, for edge-level strategies, fields such as "EvacCostAdj", "EvacPref", and "EvacTimeWin" can be written; for node-level strategies, fields such as "RecExit", "EvacGuide", and "EvacTimeWin" can be written.

[0142] If the target BIM template does not have a corresponding field, a new attribute field with the same name can be added to the BIM attribute set and a value can be assigned. In this way, even if different projects use different BIM standards, the strategy information can still be written down and read by subsequent tools through the field mapping table.

[0143] Normalized encoding and write consistency verification of S5-3 policy values Since the strategy obtained in step S4 usually has a time dimension (as... (Change), while BIM attribute fields generally carry static text or values. Therefore, this embodiment standardizes and encodes the strategy values ​​before writing them to ensure that they can express time-varying characteristics while maintaining the readability and parsability of the fields.

[0144] Specifically, the time window is expressed as a discrete time step interval, and the sampling time reference is unified with that of step S2. This allows for the conversion between "field value and simulation time"; for example, writing the effective window of a certain strategy entry as "[t_a,t_b]" allows for the conversion between the actual time interval. Unique and definitive. For edge-level preference / cost strategies, normalized preference values ​​or cost levels can be written to facilitate understanding by engineers and manual review when necessary.

[0145] Meanwhile, to ensure the correctness of the "strategy entry → component attribute" writing process, this embodiment sets up a write-back consistency check: At least three items must be verified: First, whether the source BIM component identifier is uniquely matched in the BIM model; second, whether the policy entry object type is consistent with the target component type (e.g., edge-level policies cannot be written to non-connected components); and third, whether the policy entry's effective time window aligns with the unified sampling time benchmark. Whether it is consistent (to prevent the writing content from being inconsistent with the S2 time base); Only after the consistency check passes will the attribute write to disk be executed, thereby avoiding the problem of "inconsistency between strategy information and simulation basis" in the updated BIM model caused by incorrect strategy writing.

[0146] To facilitate the provision of verifiable write-back rules in the specification, this embodiment further provides a formalized expression for writing strategy entries to component attributes. Let the source BIM component identifier of a certain strategy entry be... Its corresponding strategy data encoding is The write operation can then be abstracted as:

[0147] in, This means that without deleting the original attribute set, the strategy code will be merged and written as a new field or an updated field, thus maintaining the integrity of the original BIM attribute information.

[0148] in: The identifier of the source BIM component obtained from the association of the evacuation network model; For writing pre-component The set of attributes; For post-write components The set of attributes; For components The generated parameterized strategy data encoding; Write the attribute merging operator.

[0149] Explanation of the range: and For a collection of properties; when When empty Indicates no strategy update; when Non-empty time It includes newly added or updated evacuation strategy fields, thereby achieving strategy integration of the BIM model.

[0150] S5-4 Generate updated BIM model and version fixation After writing back all strategy entries, this embodiment generates an updated BIM model. The generation method can be: A derived version with attribute updates is created based on the original BIM model, and a version identifier is recorded for this derived version. The version identifier includes at least the "strategy generation time", "corresponding optimal iteration round identifier", "evacuation network model version identifier", and "unified sampling time benchmark". Configuration identifiers, etc., are used to ensure that the updated model can be traced back to the generation basis of steps S1-S4. The updated BIM model version identifier maintains a traceable association with the model version number associated with the generation of the evacuation network model in step S1, so that when the BIM model is iteratively updated or multiple versions are released in parallel, the evacuation network model, parameter configuration and strategy source on which each version is based can still be clearly distinguished; The updated BIM model is then subjected to an overall consistency check, including: whether the attribute fields are completely written according to the field mapping table, whether the strategy entries cover the expected key edges and key nodes, and whether there are any abnormal entries that do not match the component identifier. After the check passes, the updated BIM model is output as step S5 for subsequent design delivery or review and verification.

[0151] The technical problem addressed in step S5 is not simply "generating a report" in an offline output format. Instead, it solidifies the dynamic prediction and iterative optimization results from steps S3 / S4 into executable attribute data at the BIM component level. This enables the engineering design model to possess an evacuation strategy carrier that is "readable, auditable, and reusable." Unlike conventional practices that only output evacuation times, route maps, or static suggestions, this embodiment achieves one-to-one binding between strategies and components through source BIM component identifiers and encodes and writes time-varying strategies using a unified time base. Therefore, it can maintain the traceability and recalculation of strategies during subsequent review, modification, or model version changes, avoiding the problem of "disconnect between simulation results and engineering models," and thus forming a closed loop from "simulation optimization" to "engineering implementation."

[0152] Example 2 A system employing a pathfinder-based method to improve building fire evacuation efficiency includes: BIM Analysis and Modeling Module: Analyzes the BIM model and generates an evacuation network model that includes the topology and attribute parameters of accessible spaces; Simulation data acquisition module: Based on the evacuation network model, it obtains the time series data of the evacuation process through pathfinder simulation; Predictive guidance module: Based on time series data, it uses a graph neural network-based prediction model to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters. Iterative replanning and optimization module: Based on the dynamic evacuation guidance parameters, adjust the evacuation strategy of the pathfinder and perform iterative simulation optimization to obtain an optimized evacuation scheme that meets the predetermined conditions. BIM Write-back Integration Module: Writes back the optimized evacuation plan and integrates it into the BIM model to generate an updated BIM model.

[0153] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A method for improving building fire evacuation efficiency based on pathfinder, characterized in that, Includes the following steps: S1. Analyze the BIM model and generate an evacuation network model that includes the topology and attribute parameters of accessible spaces. S2. Based on the evacuation network model, obtain the time-series data of the evacuation process through pathfinder simulation; S3. Based on the time series data, use a graph neural network-based prediction model to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters. S4. Based on the dynamic evacuation guidance parameters, adjust the evacuation strategy of the pathfinder and perform iterative simulation optimization to obtain an optimized evacuation scheme that meets the predetermined conditions. S5. Write back the optimized evacuation plan and integrate it into the BIM model to generate an updated BIM model.

2. The method for improving building fire evacuation efficiency based on pathfinder according to claim 1, characterized in that, Parsing the BIM model to generate an evacuation network model containing the topology and attribute parameters of accessible spaces includes: Obtain the components and their geometric and attribute data related to evacuation calculations in the BIM model, and perform coordinate system transformation, elevation unification and geometric error correction on the multi-disciplinary models to ensure the connectivity of passable spaces; The passable space is discretized according to the preset node and edge mapping rules to construct an evacuation topology graph; Based on the geometric and attribute data, parameterized attributes are calculated and assigned to the edges and nodes. The parameterized attributes of the edges include at least spatial length, effective passage width, passage capacity parameter, availability status, passage direction attribute, and elevation difference. The parameterized attributes of the nodes include at least area, floor elevation, and node type code. The parameterized attributes are bound to the corresponding nodes and edges, and associated with the source BIM component identifiers to form the evacuation network model.

3. The method for improving building fire evacuation efficiency based on pathfinder according to claim 2, characterized in that, Among the preset nodes and edges, the nodes include at least room area nodes, corridor segment nodes, stair segment nodes, platform nodes and exit buffer nodes, and the edges include at least doorway connecting edges, opening connecting edges and staircase up and down connecting edges.

4. The method for improving building fire evacuation efficiency based on pathfinder according to claim 3, characterized in that, Based on the aforementioned evacuation network model, the time-series data of the evacuation process obtained through pathfinder simulation includes: The structure and parameters of the evacuation network model are mapped to the pathfinder simulation environment. Based on the effective passage width and passage direction attributes of the edges in the evacuation network model, the physical width of the corresponding passage connection components and the one-way or two-way passage rules are configured. At the same time, the initial crowd distribution and behavior parameters are configured to form a simulation instance. Run the baseline evacuation process of the simulation instance and record the dynamic state logs related to the edges and nodes in the evacuation network model during the simulation process; From the dynamic state log, state indicators of each edge and each node at multiple consecutive moments are extracted according to a unified time base, thereby obtaining edge temporal characteristics and node temporal characteristics. By integrating the temporal features of the edges and nodes, a graph temporal input tensor that is aligned with the structure of the evacuation network model and includes a time dimension is constructed as the temporal data.

5. The method for improving building fire evacuation efficiency based on pathfinder according to claim 4, characterized in that, The status indicators include at least the real-time personnel flow, instantaneous personnel density, average passage speed and / or number of people waiting in queues of the edge, and the real-time number of people staying at the node and / or inflow and outflow.

6. The method for improving building fire evacuation efficiency based on pathfinder according to claim 5, characterized in that, Based on the aforementioned time-series data, a graph neural network-based prediction model is used to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters, including: Based on the graph temporal input tensor and the topology of the evacuation network model, the graph neural network prediction model outputs the prediction results of the future travel status of the edges in the evacuation network. The prediction results are converted into dynamic cost values ​​of edges in the evacuation network, and the dynamic cost values ​​are corrected according to the available state attributes of the edges in the evacuation network model to generate initial dynamic evacuation guidance parameters. Time smoothing and update hysteresis constraints are applied to the initial dynamic evacuation guidance parameters to suppress parameter mutations and frequent updates, thereby generating stable dynamic evacuation guidance parameters as output.

7. The method for improving building fire evacuation efficiency based on pathfinder according to claim 6, characterized in that, Based on the dynamic evacuation guidance parameters, the evacuation strategy of the pathfinder is adjusted and iterative simulation optimization is performed to obtain an optimized evacuation scheme that meets predetermined conditions, including: Based on the dynamic evacuation guidance parameters, it is determined whether the predetermined replanning triggering conditions are met, and if they are met, a replanning control quantity for adjusting the evacuation path is generated. The replanning control variable is injected into the pathfinder simulation environment to drive a new round of evacuation simulation and obtain the performance indicators of the new round of simulation. The performance indicators and predetermined iteration convergence conditions are used to determine whether the optimal strategy in the previous iteration is met. If the convergence conditions are met, the optimal strategy in the previous iteration is output as the optimized evacuation scheme. Otherwise, the input is updated with the time series data obtained from the new round of simulation, and the iteration optimization continues.

8. The method for improving building fire evacuation efficiency based on pathfinder according to claim 7, characterized in that, Writing back the optimized evacuation plan and integrating it into the BIM model to generate an updated BIM model includes: extracting the parameterized strategy result set from the optimized evacuation plan; writing the strategy data in the parameterized strategy result set into the attribute information of the corresponding component in the BIM model through the source BIM component identifier associated with the evacuation network model; and generating an updated BIM model based on the written attribute information.

9. A system employing the pathfinder-based method for improving building fire evacuation efficiency as described in any one of claims 1-8, characterized in that, include: BIM Analysis and Modeling Module: Analyzes the BIM model and generates an evacuation network model that includes the topology and attribute parameters of accessible spaces; Simulation data acquisition module: Based on the evacuation network model, time-series data of the evacuation process are obtained through pathfinder simulation; Prediction and guidance module: Based on the time series data, it uses a graph neural network-based prediction model to predict the dynamic traffic status of the evacuation network and generate dynamic evacuation guidance parameters. Iterative replanning and optimization module: Based on the dynamic evacuation guidance parameters, adjust the evacuation strategy of pathfinder and perform iterative simulation optimization to obtain an optimized evacuation scheme that meets the predetermined conditions. BIM Write-back Integration Module: Writes back the optimized evacuation plan and integrates it into the BIM model to generate an updated BIM model.

Images