A smart campus multidimensional fire safety monitoring method and system

Abstract

This invention relates to the field of fire safety monitoring technology and discloses a multi-dimensional fire safety monitoring method and system for smart campuses. The method includes: constructing a two-layer graph structure formed by linking a ventilation duct network topology map and a personnel access topology map through location coordinate nodes; calculating the spatiotemporal propagation sequence of the gas pollution front based on the physical parameters of the pipe sections; obtaining the net cumulative flow rate and calculating the cumulative pressure index by differentiating the flow rates at the channel cross sections; inputting the topology deviation vector and the standardized cumulative pressure index time series into a pre-trained LSTM network to output the stampede risk level and remaining time estimate; assigning time-varying availability labels and pressure perception weights to each access edge of the personnel access topology map; and performing a time-dependent shortest safe path search. This invention introduces spatiotemporal dynamic information of gas diffusion into the path availability determination, avoiding guiding personnel to areas already covered by pollution or areas that will be covered before personnel arrive.

Description

Technical Field

[0001] This invention relates to the field of fire safety monitoring technology, and in particular to a multi-dimensional fire safety monitoring method and system for smart campuses. Background Technology

[0002] With the continuous expansion of university laboratory buildings and the increasing diversification of experiments, the types and quantities of hazardous chemicals stored and used in these buildings have significantly increased, making the evacuation needs for personnel in the event of gas leaks more complex. At the same time, the in-depth advancement of smart campus construction has led to the widespread deployment of Building Information Modeling (BIM), indoor positioning systems, and IoT sensor networks in campus buildings, providing the foundation for fire safety monitoring based on multi-source data fusion.

[0003] Regarding gas diffusion prediction, existing solutions typically employ computational fluid dynamics (CFD) simulations or simplified diffusion equations to estimate gas propagation trajectories. While CFD simulations offer high accuracy, they are computationally time-consuming, making real-time updates difficult within the time constraints of emergency response. Simplified diffusion equations often neglect the constraints imposed by the ventilation network topology on gas propagation paths, leading to significant discrepancies between predicted results and actual gas diffusion patterns driven by ventilation. In personnel evacuation route planning, existing solutions largely rely on static graph structures to perform shortest path searches, failing to incorporate the spatiotemporal dynamic evolution of the gas pollution front and the pressure state of personnel in the passageways as joint constraints into path availability determination. This results in planned evacuation routes that may guide personnel into contaminated areas or cause congestion in passageways. Regarding coordinated evacuation of multiple groups, existing solutions lack systematic strategies to mitigate time window conflicts arising from multiple groups converging on shared passageways. The convergence of different evacuation groups on the same passageway may lead to excessive personnel density.

[0004] Current fire safety evacuation schemes do not take into account the constraint effect of ventilation duct topology on gas propagation, the cumulative state of personnel pressure reflected by the flow difference of the channel cross section, and the coupling relationship between the time window conflicts of multiple groups. As a result, the evacuation path generated in the scenario of rapid gas diffusion cannot guarantee the time safety of the entire path, nor can it effectively avoid the risk of channel overcapacity caused by the convergence of multiple groups. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional fire safety monitoring method and system for smart campuses, which solves the technical problems of insufficient accuracy in predicting gas diffusion and propagation paths, lack of perception of congestion pressure in personnel passages, and inability to dynamically resolve conflicts in evacuation paths of multiple groups in existing campus fire evacuation plans.

[0006] This invention provides a multi-dimensional fire safety monitoring method for smart campuses, comprising the following steps: The system acquires the duct layout data and geometric topology data of each passage of the building ventilation system. Based on the duct layout data, it generates a ventilation duct network topology map and a personnel passage topology map. The ventilation duct network topology map and the personnel passage topology map are then linked by the location coordinates of the area space to form a two-layer graph structure. Once the gas leak is confirmed, the fan operating parameters are obtained, and a breadth-first traversal is performed on the ventilation duct network topology map starting from the source node. The transmission delay time is calculated based on the physical parameters of the pipe segments corresponding to each directed edge on the traversal path. The transmission delay time of all pipe segments on the path is accumulated for each downstream node to generate a spatiotemporal progression sequence of the gas pollution front. Real-time personnel location flow data of each evacuation route is obtained, each route is discretized into an ordered cross-section sequence along the axis, the number of people passing through each cross-section per unit time is calculated, and the net cumulative flow rate is obtained by performing differential operation on adjacent cross-sections. The cumulative pressure index of the route section is calculated based on the net cumulative flow rate. Based on the spatiotemporal progression sequence of the gas pollution frontier and the cumulative pressure index, time-varying availability markers and pressure perception weights are assigned to each passage edge in the personnel passage topology map. Time-dependent shortest safe path search is performed on each evacuation group on the personnel passage topology map after the constraint update, and a safe evacuation path is output. Conflict detection is performed on the safe evacuation routes of each group. Based on the time margin between each group and the pollution front, priority is sorted and routes are redistributed to generate a joint evacuation route plan for multiple groups. Wherein, the transmission delay time Calculate using the following formula: ,in For the length of the pipe section, The cross-sectional area of ​​the pipe section. The volumetric air volume of the fan that drives the airflow in this duct section.

[0007] In a preferred embodiment, calculating the cumulative pressure index of the channel segment based on the net cumulative flow rate includes: when there is continuous flow within the channel segment... The net cumulative flow rate between each adjacent section is positive and When this occurs, the channel segment is marked as being in a state of pressure accumulation; the accumulated pressure index According to the formula Calculation, where The starting section number for the continuous positive value region. This represents the number of regions between consecutive positive cross sections. For the first Net cumulative flow rate between cross sections For the first The axial weight corresponding to the region between each cross section is defined as the normalized distance of that region from the channel exit. , This represents the total number of cross sections.

[0008] In a preferred embodiment, a channel flow feature vector is extracted from the channel segment under pressure accumulation, and a baseline flow feature vector is extracted from historical channel flow data of the same period. After Z-score standardization of each corresponding component, a topology deviation vector is calculated. The time series of the topology deviation vector and the accumulated pressure index are input into a pre-trained channel stampede risk time series classification model, which outputs the stampede risk level of each channel segment and an estimated remaining time to reach the dangerous pressure threshold. The topology deviation vector... According to the formula calculate, This is the standardized current channel flow feature vector. This is the standardized baseline flow characteristic vector.

[0009] In a preferred embodiment, the channel flow feature vector This includes the number of people passing through the entrance section of the passageway per unit time, the number of people passing through the exit section per unit time, the ratio of entrance to exit flow rate, the cumulative pressure index, the average population density within the passageway, and the time-varying rate of average passing speed within the passageway. The passageway stampede risk time-series classification model uses an LSTM network, with an input of length [missing information]. The time series is such that the input vector at each time step is the concatenation of the topological deviation vector and the standardized cumulative pressure index at the corresponding time. The output layer contains a classification branch and a regression branch. The classification branch outputs the stampede risk level, and the regression branch outputs the estimated remaining time.

[0010] In a preferred embodiment, assigning time-varying availability tags and pressure-aware weights to each passage edge in the personnel access topology graph includes: for any passage edge When the sum of the estimated arrival time of personnel at the starting node and the estimated travel time through the passage edge is not less than the estimated arrival time of contamination at the terminal node, the passage edge is marked as time-restricted and unavailable. For passage edges that are in a pressure accumulation state and are not marked as time-restricted and unavailable, their path search weights are determined according to the formula... Upward adjustment, of which Based on the weights, This represents the cumulative pressure index for the corresponding channel segment. The pressure perception coefficient is used; when the trampling risk level reaches high risk, the corresponding passage edge is marked as high risk and restricted from passage.

[0011] In a preferred embodiment, the time-dependent shortest safe path search includes: expanding from the starting node of the evacuation group using a time-dependent Dijkstra algorithm, performing time safety verification and stress safety verification for each candidate node expanded; the time safety verification verifies whether the estimated time to reach the candidate node along the current path is less than the estimated time of contamination arrival at that node, and if not; the stress safety verification checks whether the cumulative stress index of each channel traversed by the current path is lower than a preset safety threshold, and prunes any channel segments that exceed the preset safety threshold.

[0012] In a preferred embodiment, the prioritization and path reallocation based on the time margin between each group and the pollution front includes: groups time margin According to the formula Calculation, where For groups The set of all nodes on the safe evacuation route. For nodes The pollution is expected to arrive at the time of [time]. For groups Reaching the node The estimated time is determined; the group with the smallest time margin is given the highest priority and its original path remains unchanged, while the group with a larger time margin re-executes the time-dependent shortest safe path search. During the search, conflicting channel segments are marked as unavailable within the conflict time window. The process is iterated until all conflicts are resolved.

[0013] In a preferred embodiment, during the evacuation process, the latest pedestrian flow data and gas concentration sensor data are acquired every preset update cycle; the actual arrival time of nodes where gas concentration exceeds the standard is compared with the predicted arrival time of pollution in the spatiotemporal advancement sequence of the gas pollution front; when the gas diffusion prediction deviation exceeds the preset tolerance, or the trampling risk level of a certain passage section is upgraded to high risk, the complete calculation process from generating the spatiotemporal advancement sequence of the gas pollution front to generating the multi-group joint evacuation path scheme is re-executed with the current time as the new starting time, and the updated multi-group joint evacuation path scheme is pushed to the evacuation guidance equipment in each area.

[0014] In a preferred embodiment, the ventilation duct network topology is a directed graph, where the direction of each directed edge is consistent with the set flow direction of the airflow within the corresponding duct segment, and each directed edge is associated with three physical attributes: duct segment length, duct segment cross-sectional area, and fan volume and airflow. The personnel passage topology is an undirected graph, where each passage edge is associated with channel length, effective channel width, and rated passage capacity. The two-layer graph structure achieves the association between the two topology layers through a one-to-one mapping relationship of shared nodes, so that the gas pollution status at the ventilation duct network level is transmitted to the passage edge associated with the corresponding node at the personnel passage level.

[0015] A multi-dimensional fire safety monitoring system for smart campuses includes a two-layer graph construction module, a gas diffusion prediction module, a channel pressure calculation module, a graph constraint update module, a path search module, a conflict resolution module, and a dynamic update and push module. The two-layer graph construction module is configured to acquire duct layout data and geometric topology data of each channel in the building ventilation system, generate a ventilation duct network topology map and a personnel passage topology map, and connect the two through node association using the location coordinates of the area space to form a two-layer graph structure. The gas diffusion prediction module is configured to acquire fan operating parameters after gas leakage is confirmed, perform breadth-first traversal on the ventilation duct network topology map, calculate the transmission delay time of each duct segment, and generate a spatiotemporal progression sequence of the gas pollution front. The channel pressure calculation module is configured to acquire real-time personnel location flow data, discretize the channel into an ordered cross-sectional sequence, calculate the cross-sectional flow difference to obtain the net flow accumulation. The system calculates the cumulative pressure index of the passage segment; the graph constraint update module is configured to assign time-varying availability markers and pressure perception weights to each passage edge in the personnel passage topology map based on the spatiotemporal advancement sequence of the gas pollution front and the cumulative pressure index; the path search module is configured to perform time-dependent shortest safe path search on the personnel passage topology map after constraint update for each evacuation group and output a safe evacuation path; the conflict resolution module is configured to perform conflict detection on the safe evacuation paths of each group, prioritize and reallocate paths based on time margin, and generate a multi-group joint evacuation path scheme; the dynamic update and push module is configured to periodically acquire the latest sensor data and perform deviation detection, and when the trigger condition is met, call the gas diffusion prediction module to the conflict resolution module to recalculate, and push the updated multi-group joint evacuation path scheme to the evacuation guidance equipment in each area.

[0016] The beneficial effects of this invention are: 1. This invention forms a two-layer graph structure by associating the ventilation duct network topology map and the personnel passage topology map with nodes through regional spatial location coordinates. This allows the spatiotemporal advancement sequence of gas pollution front calculated based on the physical parameters of the pipeline segment to be directly mapped to the passage edge constraints at the personnel passage level. This introduces the spatiotemporal dynamic information of gas diffusion into the path availability determination, avoiding guiding personnel to areas that are already covered by pollution or will be covered before personnel arrive.

[0017] 2. This invention obtains the net cumulative flow rate by performing differential calculation on the flow rate of the channel cross section, and calculates the cumulative pressure index with exit distance weighting when the net cumulative flow rate of the flow rate is positive across multiple consecutive cross sections. This allows the impact of people staying in areas far from the channel exit on the overall evacuation smoothness to be quantitatively characterized, providing numerical basis for subsequent path weight adjustment and stampede risk assessment.

[0018] 3. This invention feeds the topological deviation vector and the standardized cumulative pressure index time series as joint inputs into a pre-trained LSTM network, enabling the capture of the time-series pattern of the channel evolving from a normal flow state to a dangerous congestion state. The output stampede risk level and remaining time estimate provide time-dimensional constraint parameters for path search, so that high-risk channel segments are marked as impassable in the path search.

[0019] 4. This invention prioritizes multiple groups with time window conflicts based on time margin. The group with the smallest time margin keeps its original path unchanged, while the group with a larger time margin obtains an alternative detour path by re-executing the path search with the unavailable marker of the conflicting channel segment. This allows the combined pedestrian flow density of multiple groups on the shared channel segment to be controlled within the safe density corresponding to the rated traffic capacity.

[0020] 5. This invention periodically compares the arrival time of nodes where the actual gas concentration exceeds the standard with the predicted value. When the relative deviation exceeds the preset tolerance or the stampede risk level is unexpectedly upgraded, the entire process is recalculated and the updated path plan is pushed to the evacuation guidance equipment in each area. This allows the path planning results to continuously reflect the real-time status changes of gas diffusion and channel pressure. Attached Figure Description

[0021] Figure 1 This is a flowchart of the smart campus multi-dimensional fire safety monitoring method provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the two-layer graph structure and data processing flow provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the spatiotemporal propagation sequence of the gas pollution front (the expected arrival time of pollution at each node) provided in an embodiment of the present invention; Figure 4This is a schematic diagram of the flow distribution and net cumulative flow rate of the main evacuation channel cross section provided in an embodiment of the present invention; Figure 5 This is a schematic diagram showing the comparison between the current value and the baseline value of the main evacuation channel flow characteristic vector provided in this embodiment of the invention. Figure 6 This is a schematic diagram of the time-dependent shortest safe path search and conflict resolution process provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the time window for the occupancy of the three-layer evacuation group passage provided in an embodiment of the present invention; Figure 8 This is a schematic diagram comparing the estimated arrival times of contamination at each node before and after dynamic updates, provided in an embodiment of the present invention. Figure 9 This is a schematic diagram showing the comparison of the weights after updating the constraints of various typical access edges provided in the embodiments of the present invention. Detailed Implementation

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

[0023] Example 1 Reference Figures 1-9 This invention discloses a multi-dimensional fire safety monitoring method for smart campuses, comprising the following steps S100~S800; Step S100: Construct a two-layer graph structure; It should be noted that in step S100, the duct layout data and geometric topology data of each passage of the experimental building ventilation system are obtained. A ventilation duct network topology map is generated based on the duct layout data, and a personnel passage topology map is generated based on the passage geometric topology data. The ventilation duct network topology map and the personnel passage topology map are linked by nodes through the location coordinates of the area space to form a two-layer graph structure. For example, a ventilation duct network topology diagram is defined as a directed graph. ,in Each node in the diagram represents a spatial area (such as a laboratory, corridor, or stairwell). Each directed edge in the diagram represents a pipe segment connecting two spatial regions, and the direction of the directed edge is consistent with the set flow direction of the airflow within that pipe segment. Associated with three physical properties: pipe segment length (Unit: m) Pipe section cross-sectional area (Unit: m) 2The volumetric air volume provided by the fan driving the airflow in that duct section. (Unit: m) 3 / s).

[0024] For example, a people access topology graph is defined as an undirected graph. ,in Each node in the diagram represents a space accessible to people (such as room entrances / exits, passageway intersections, and stair landings). Each edge in the diagram represents a passable path connecting two nodes. Each passable edge... The associated attributes include channel length. Effective width of the channel and rated traffic capacity .

[0025] In one embodiment of the present invention, the two-layer graph structure associates the ventilation duct network topology graph and the personnel access topology graph through a shared set of nodes. Specifically, the nodes in the ventilation duct network topology graph... Nodes in the personnel access topology diagram When the corresponding spatial areas of the two locations have the same three-dimensional coordinates in the building information model, a one-to-one mapping relationship is established. Through this mapping relationship, the gas pollution status calculated at the ventilation duct network level can be directly transmitted to the passage edge associated with the corresponding node at the personnel passage level, thereby realizing the linkage between the two layers of constraints.

[0026] For example, in one instance, a university chemistry lab building has five floors. The system retrieves the building information model data of the building and generates a ventilation duct topology diagram containing 120 spatial nodes and 250 directed edges of pipes. And a topology diagram of pedestrian access, including 120 access nodes and 85 access routes. By linking the 120 nodes of the ventilation duct network topology map with the personnel passage topology map through floor numbers and room coordinates, a two-layer map structure is formed.

[0027] Step S200: Calculate the spatiotemporal propagation sequence of the gas pollution front; It should be noted that in step S200, after the gas leak in the source area is confirmed, the current speed and corresponding air volume parameters of each fan are obtained. A breadth-first traversal is performed on the ventilation duct network topology map starting from the source node. Based on the physical parameters of the pipe segment corresponding to each directed edge on the traversal path, the transmission delay time of the gas through each pipe segment is calculated. The transmission delay time of all pipe segments on the path is accumulated for each downstream node to generate a spatiotemporal advancement sequence of the gas pollution front.

[0028] For example, let the leak source node be... The leak was confirmed at the time of the leak. For ventilation duct network topology diagram From the node Starting from any downstream node reachable via breadth-first search. Let from arrive The traversal path passes through the sequence of directed edges as follows: ,in To start from the source node The first directed edge from which to start, It is the second directed edge on the path. To reach the node The last directed edge, then the gas passes through a single directed edge. Transmission delay time Calculate using the following formula:

[0029] in, Pipeline section Length (unit: m) Pipeline section Cross-sectional area (unit: m) 2 ), For driving pipeline section Internal airflow fan volumetric air volume (unit: m³) 3 / s). Molecules The dimension of m 3 , and the denominator Dimension m 3 After dividing by / s The dimension of is seconds (s), and its physical meaning is the time required for the gas to fill the pipe section, which is the gas volume in the pipe section divided by the volumetric flow rate through the pipe section.

[0030] node Expected arrival time of gas pollution for:

[0031] in, To start from the source node To the node A breadth-first traversal of the set of all directed edges on the path is performed. Leakage confirmation time (unit: seconds). The transmission delay time (in seconds) for each pipeline segment. The dimension of is seconds (s).

[0032] In one embodiment of the present invention, when there are multiple paths from a source node to a downstream node in the ventilation duct network topology, the path selected by breadth-first traversal is the shortest hierarchical path, and the accumulated transmission delay time corresponding to this shortest hierarchical path is the earliest time when the gas arrives at that node. The spatiotemporal progression sequence of the gas pollution front is defined as a set. ,in In the ventilation duct network topology diagram, it can be started from the source node. The node reached, For nodes The expected arrival time of pollution, the spatiotemporal progression sequence of the gaseous pollution front is as follows: Arranged in ascending order, it represents the complete spatiotemporal distribution of the gaseous pollution front in the built space as it progresses over time.

[0033] For example, in one instance, the leak source was located in the 3rd floor chemical synthesis chamber (node). Leakage confirmation time Seconds. The system obtains the current fresh air fan speed as 1200 rpm, corresponding to an air volume of... From node To the second floor stairwell (node) The traversal path of ) passes through the main pipeline segment ( , The main pipeline section has a transmission delay of 1.5 times that of the main pipeline and two branch pipeline sections. Seconds, summing the delays of each segment to get The second refers to the predicted gas reaching the second-floor stairwell 120 seconds after the leak is confirmed.

[0034] Step S300: Calculate the channel cumulative pressure index; It should be noted that in step S300, real-time personnel location flow data of each evacuation channel is obtained, and each channel is discretized into an ordered cross-section sequence along the axis based on the channel geometric topology data. The number of people passing through each cross-section per unit time within the current time window is calculated, and the net cumulative flow rate between each cross-section is obtained by differential operation on the number of people passing through adjacent cross-sections per unit time. Based on the net cumulative flow rate, the pressure accumulation state of the channel segment is determined and the cumulative pressure index is calculated.

[0035] For example, for a people access topology map Any channel edge in Let the length of this channel segment be... Along the axial direction of the channel at intervals Channel edge Discretize into There are ordered cross sections, among which , For the channel edge The total number of cross sections obtained by discretization along the axial direction, with the cross section numbers sequentially from the inlet side to the outlet side: ,in The cross-section is the one closest to the entrance of the passage. This is the cross-section closest to the channel exit side. Within the current time window. Inside, cross section The number of people passing through per unit of time is recorded as (Unit: people / s) The value is obtained by statistically analyzing the cross-section within this time window. Number of people divided by Obtain. Adjacent sections and Net cumulative rate of flow between Calculate using the following formula:

[0036] in, cross section The number of people passing through per unit of time (unit: people / s). cross section The number of people passing through per unit of time (unit: people / s). The total number of cross sections, The dimension of is human units per second. When, it indicates the cross section. and The number of people in the area is increasing, indicating that there is a backlog of people and accumulated stress in the area.

[0037] In one embodiment of the present invention, when there are continuous channels within a channel segment Between adjacent sections ( When the net cumulative flow rate of a channel segment is positive, the channel segment is marked as being in a state of pressure accumulation. The cumulative pressure index of this channel segment is... Calculate using the following formula:

[0038] in, The starting section number for the continuous positive value region. The number of regions between consecutive positive cross sections ( ), For the first Net cumulative flow rate between cross sections (unit: person / s). For the first The axial weight corresponding to the region between each cross section is defined as the normalized distance of that region from the channel exit, i.e. , The total number of cross sections, It is a dimensionless scalar, and its range of values ​​is 1. Its physical meaning is that the farther away from the exit, the greater the negative impact on the overall smoothness of evacuation when people are stranded. Cumulative Pressure Index The dimension of is human / s, reflecting the weighted intensity of personnel congestion within the passage segment.

[0039] For example, in one example, the main evacuation corridor from the 1st to the 2nd floor is 30m long. The data is discretized into 15 ordered sections. Within a 5-second time window, the sections... to The number of people passing through per unit time are respectively Persons per second. Difference calculation yields... to All values ​​are positive, marking the cross-section. to The region is in a state of pressure accumulation; the weighted summation yields the cumulative pressure index for this channel segment. Person / s.

[0040] Step S400: Assess the risk of stampedes in the passageway; It should be noted that in step S400, channel flow feature vectors are extracted for each channel segment, baseline flow feature vectors are extracted from historical channel flow data of the same period, and topology deviation vectors are calculated. The time series of topology deviation vectors and cumulative pressure index are input into the pre-trained channel stampede risk time series classification model, and the stampede risk level of each channel segment and the estimated remaining time to reach the dangerous pressure threshold are output.

[0041] For example, for any channel segment in a state of pressure accumulation Channel flow characteristic vector Defined as:

[0042] in, This represents the number of people passing through the entrance section of the passage per unit time (unit: people / s). The number of people passing through the exit section per unit time (unit: people / s). The ratio of inlet to outlet flow (i.e. (dimensionless) The cumulative pressure index (unit: person / s) calculated in step S300. Average population density within the passageway section (unit: people / m²) 2 ), The rate of change of the average passing speed within the channel segment over time (unit: m / s) 2 ), Indicates transpose. Channel flow characteristic vector. The dimension is 6.

[0043] Reference flow eigenvector The baseline flow feature vector is extracted from channel flow data from the same historical period (e.g., the same break time during weekdays). The six components and the channel flow characteristic vector The six components correspond one-to-one, namely: , , , , , The corresponding statistical mean in the historical data also has a dimension of 6.

[0044] Before participating in the calculation of the topology deviation vector, the channel flow characteristic vector is... and the reference flow eigenvector Each corresponding component is standardized using Z-score, with the standardization parameters (mean and standard deviation) derived from historical data statistics. This ensures that the current channel flow characteristic vector and the baseline flow characteristic vector are in the same numerical space, eliminating the influence of differences in the dimensions of various physical quantities and inconsistencies in numerical scales on subsequent vector subtraction operations and model input. Topology deviation vector Calculate using the following formula:

[0045] in, This is the standardized current channel flow feature vector. The two are standardized baseline flow feature vectors, both with a dimension of 6. This is a 6-dimensional topological deviation vector. Each component represents the degree of deviation of the current channel flow state from the historical normal state in each feature dimension.

[0046] In one embodiment of the present invention, the time-series classification model for trampling risk in passageways is implemented using an LSTM network. The input of the LSTM network is a sequence of lengths... The time series, where The length of the time series (i.e., the number of time steps) is given, and the input vector for each time step in the series is the topological deviation vector. Standardized cumulative stress index at the corresponding time point The input dimension is 7 (6-dimensional topological deviation vector plus 1-dimensional standardized cumulative pressure exponent scalar). The hidden states of the LSTM network are mapped to the output space through a fully connected linear mapping layer. The output layer contains two branches: a classification branch outputs the stampede risk level (low risk, medium risk, and high risk), and a regression branch outputs the estimated remaining time to reach the dangerous pressure threshold. (Unit: seconds). The LSTM network uses historical evacuation event data and simulation data as training sets. The training labels include manually labeled risk levels and remaining time values ​​determined based on simulation. The classification branch uses the cross-entropy loss function, and the regression branch uses the mean squared error loss function. The weighted sum of the two is used as the joint loss function, and the parameters are updated using the Adam optimization algorithm.

[0047] It should be noted that the stampede risk level output by the classification branch is a discrete category label (low risk, medium risk, high risk), while the stampede risk level output by the regression branch is... Both are continuous values ​​measured in seconds, and can be directly used in the passage edge marking logic of step S500 and the priority sorting of step S700 without additional decoding transformation. The channel stampede risk time-series classification model, by simultaneously receiving a topological deviation vector reflecting the degree of flow deviation and a standardized cumulative pressure index reflecting the degree of personnel congestion as joint inputs, can capture the time-series pattern of channel evolution from a normal flow state to a dangerous congestion state, and outputs... This provides time-dimensional constraint parameters for subsequent path planning.

[0048] Step S500: Update the personnel access topology constraints; It should be noted that in step S500, based on the spatiotemporal progression sequence of the gas pollution front and the stampede risk assessment results, time-varying availability markers and pressure perception weights are assigned to each passage edge in the personnel passage topology map.

[0049] For example, for a people access topology map Any passable edge in Let the terminal node corresponding to this passage edge be... The pollution is expected to arrive at the following time: (Obtained from step S200). When a person departs from the current location and arrives at the node. The estimated time is via the passage The estimated travel time is If the following conditions are met, then the passable edge will be... Marked as time-limited and unavailable:

[0050] in, For personnel to reach the starting node of the passage edge The estimated time (in seconds). To pass through the border The estimated travel time (in seconds, calculated based on the channel length and the current average travel speed). For terminal nodes The estimated arrival time of pollution (in seconds). This inequality holds when personnel are crossing the passageway. Reaching the terminal node At that time, terminal node The area is now covered by gas pollution.

[0051] In one embodiment of the invention, for a passageway that is in a state of pressure accumulation but is not marked as time-limited unavailable... Passage Path search weights Adjust upwards using the following formula:

[0052] in, For passage edge The basic weights (determined based on channel length and rated capacity). For passage edge The cumulative pressure index of the corresponding channel segment (unit: person / s). The stress perception coefficient (a preset normal value, in seconds per person, used to...) This is converted into a dimensionless amplification factor, thereby adjusting the amplification of the pressure index on the weights. This formula ensures that channel segments with more severe pressure accumulation are assigned a larger weight cost in the path search, thus guiding the path search algorithm to tend to select channels with lower pressure accumulation.

[0053] Meanwhile, when the trampling risk level output by step S400 reaches high risk, the corresponding passable edge is marked as high risk restricted and is considered impassable in the path search.

[0054] Step S600: Perform a time-dependent shortest safe path search; It should be noted that in step S600, for each group of people in the evacuation starting area, a time-dependent shortest safe path search is performed on the personnel passage topology map with time-varying availability constraints and pressure perception weights, and the safe evacuation path of each group of people and the estimated passage time window of each passage are output.

[0055] For example, the search algorithm employs Dijkstra's time-dependent algorithm. For each evacuation group... Set up evacuation groups The starting node is The target node set is the building safety exit node set. The time-dependent Dijkstra algorithm from Start expanding, expanding by one candidate node at a time. Perform the following two verifications: Time safety verification: Calculate the number of nodes reached along the current path. Expected time ,verify Whether it is valid, among which For nodes The estimated arrival time of pollution is determined (obtained from the spatiotemporal propagation sequence of the gas pollution front in step S200). If this is not the case, the expansion is pruned and not included in the search queue.

[0056] Pressure safety verification: Check whether the cumulative pressure index corresponding to each segment of the current path is lower than the preset safety threshold. (Unit: people / s). If the cumulative pressure index of any channel segment exceeds... If so, the path is pruned.

[0057] In one embodiment of the present invention, the arrival time of each node It is obtained by summing the estimated travel times of all accessible edges along the path segment by segment. Estimated travel time Based on channel length Compared with the current real-time average traffic speed The ratio is determined, that is ,in The dimension of is m. The dimension is m / s. The dimension of is s. The time-dependent Dijkstra's algorithm outputs the path with the minimum cumulative weight reaching the safe exit node as the evacuation group. The system records the safe evacuation routes and the estimated entry and exit times for each section of the route, forming evacuation groups. The channel occupancy time window sequence.

[0058] For example, in one instance, when performing a search on a 3-story evacuation group, starting from the initial node of the 3-story chemical synthesis room, the west stairwell passageway is... The path branch was marked as time-limited and unavailable after a few seconds, and the time-dependent Dijkstra algorithm automatically pruned the branch; the cumulative pressure index of the backup stairwell on the east side was 0.8 people / s (below the safety threshold). The number of people per second (in seconds) and the arrival time meet the time safety constraints are output as a safe evacuation path for a 3-layer group.

[0059] Step S700: Multi-group conflict detection and resolution; It should be noted that in step S700, conflict detection is performed on the safe evacuation routes of each group. When a conflict exists, priority is sorted and routes are reallocated based on the time margin between each group and the pollution front, and a multi-group joint evacuation route scheme after conflict resolution is generated.

[0060] For example, the criterion for collision detection is: for any two groups and If the group With Group The safe evacuation routes are in the same passageway section The time windows for occupancy of the passageways overlap, and the combined expected pedestrian density of the two groups within the overlapping time windows... Exceeding the rated capacity of this passage segment The corresponding maximum safe density At that time, it is marked that multiple groups of conflicts exist within that channel segment within that time window. Based on rated traffic capacity Divide by the effective passage area of ​​the passage segment (i.e. (Identified, unit: person / m) 2 .

[0061] In one embodiment of the invention, conflict resolution is performed based on a priority ordering of time margins. (Group) time margin Defined as a group The estimated arrival time of contaminants at each node along the safe evacuation route and the group The minimum difference between the estimated times of arrival at the corresponding node:

[0062] in, For groups The set of all nodes on the safe evacuation route. For nodes The estimated arrival time of the pollution (in seconds). For groups Reaching the node The estimated time (in seconds). The dimension of is s. The smaller the time margin, the more urgent the threat to the gas pollution front for that group.

[0063] For a set of conflicting groups, the group with the smallest time margin is given the highest priority and its original path remains unchanged; groups with larger time margins are required to re-execute the time-dependent shortest safe path search in step S600. During the search, conflicting channel segments are marked as unavailable within the conflict time window, thereby obtaining alternative detour paths. This process is iterated until all conflicts are resolved, generating a joint evacuation route scheme for multiple groups.

[0064] For example, in one instance, the paths of the 3rd and 4th floor groups overlapped in the time window of the 1st floor lobby exit passage segment, resulting in a merging density exceeding the safety capacity. Calculate the time margin of the 3-level group. Seconds, 4-layer group time margin Seconds. The 3rd layer group has high priority and maintains its original path; the 4th layer group re-executes the time-dependent shortest safe path search and is then guided to the outdoor corridor detour path.

[0065] Step S800: Dynamic update and command push.

[0066] It should be noted that in step S800, during the evacuation process, the latest channel flow data and gas concentration sensor data are obtained every preset update cycle. Based on the deviation detection results, it is determined whether to trigger the recalculation of the route plan, and the route change instruction is pushed to the evacuation guidance equipment in each area through the security terminal.

[0067] For example, the system is set to update cycle as follows: (e.g., 10 seconds). At the end of each update cycle, the system acquires the real-time concentration readings of the gas concentration sensors distributed at each pipeline node, and compares the actual arrival time of the node where the gas concentration actually exceeds the standard with the predicted arrival time of the pollution in step S200. When the relative deviation between the actual diffusion rate and the predicted value exceeds the preset tolerance... If the relative deviation exceeds 10%, the system determines that the current spatiotemporal progression sequence of the gas pollution front can no longer accurately reflect the actual diffusion state.

[0068] In one embodiment of the present invention, the conditions for triggering recalculation include two categories: the first category is that the gas diffusion prediction deviation exceeds the tolerance. The second category involves an unexpected accumulation of pressure in a certain passageway, causing the stampede risk level to escalate from low or medium to high. When any of the above conditions are met, the system uses the current moment as the new starting moment, with the latest fan operating parameters and passageway pedestrian flow data as input, and re-executes the complete calculation process from steps S200 to S700, outputting an updated multi-group joint evacuation route plan. The updated multi-group joint evacuation route plan is pushed to the smart evacuation signs, emergency broadcast terminals, and mobile communication terminals on each floor via wired or wireless networks. The evacuation guidance equipment in each area updates the displayed evacuation direction and voice guidance content in real time according to the received route change instructions.

[0069] For example, in one instance, 10 seconds after evacuation is initiated, the system acquires data from the gas concentration sensor and finds that the actual diffusion rate is 15% faster than the prediction in step S200 (exceeding the preset tolerance of 10%). Steps S200 to S700 are immediately re-executed to update the estimated arrival times of each node in the spatiotemporal progression sequence of the gas pollution front, recalculate the time-varying availability markers and path search weights for each access edge, and generate a new multi-group joint evacuation route plan. The updated multi-group joint evacuation route plan is pushed to the smart evacuation signs on the 4th floor area via the campus security network. The smart evacuation signs change from displaying "Evacuate to the west staircase" to "Evacuate towards the connecting corridor," and the emergency broadcast simultaneously announces the route change information.

[0070] Example 2 A smart campus multi-dimensional fire safety monitoring system includes a two-layer graph construction module, a gas diffusion prediction module, a channel pressure calculation module, a risk assessment module, a graph constraint update module, a path search module, a conflict resolution module, and a dynamic update and push module. The two-layer graph construction module is configured to acquire the duct layout data and geometric topology data of each passage of the laboratory building's ventilation system, generate a ventilation duct network topology map based on the duct layout data, generate a personnel passage topology map based on the passage geometric topology data, and associate the ventilation duct network topology map and the personnel passage topology map through the location coordinates of the area space to form a two-layer graph structure. The gas diffusion prediction module is configured to acquire fan operating parameters after gas leaks are confirmed, perform breadth-first traversal on the ventilation duct network topology map and calculate the transmission delay time of each duct segment, and generate a spatiotemporal propagation sequence of the gas pollution front. The channel pressure calculation module is configured to acquire real-time personnel location flow data, discretize the channel into an ordered cross-section sequence, calculate the cross-sectional flow difference to obtain the net flow accumulation rate, determine the pressure accumulation state of the channel segment, and calculate the cumulative pressure index. The risk assessment module is configured to extract the channel flow feature vector and the baseline flow feature vector, perform Z-score standardization on each corresponding component to calculate the topology deviation vector, and input the topology deviation vector and the standardized cumulative pressure index time series into a pre-trained LSTM network to output the stampede risk level and the estimated remaining time. The graph constraint update module is configured to assign time-varying availability markers and pressure perception weights to each passage edge in the personnel passage topology graph based on the spatiotemporal advancement sequence of the gas pollution frontier and the stampede risk assessment results. The path search module is configured to perform time-dependent shortest safe path search for each evacuation group on the personnel access topology map after the constraint update, and output the safe evacuation path and the time window sequence of passage occupancy. The conflict resolution module is configured to perform conflict detection on the safe evacuation paths of each group, perform path reallocation based on priority sorting with time margin, and generate a joint evacuation path scheme for multiple groups. The dynamic update and push module is configured to periodically acquire the latest sensor data and perform deviation detection. When the triggering conditions are met, it calls the gas diffusion prediction module to the conflict resolution module to recalculate and pushes the updated multi-group joint evacuation path plan to the evacuation guidance equipment in each area.

[0071] The following provides a further explanation of this embodiment using specific parameter scenarios: This embodiment is applied to a chemistry laboratory building (hereinafter referred to as "Building A") in a university. It has 5 floors, with 4 chemistry laboratories, 1 main corridor and 2 stairwells on each floor. The system extracts the ventilation duct layout data and geometric topology data of each passage from the building information model.

[0072] Step S100: Construct a two-layer graph structure; The ventilation duct layout data includes the length, cross-sectional area, and volumetric air volume of each duct section; the geometric topology data of each channel includes the channel length, effective channel width, and rated passage capacity.

[0073] Based on ventilation duct layout data, the system generates a directed graph. Its node set It contains 120 regional spatial nodes and a directed edge set. It contains 250 directed edges for pipes, each associated with a pipe segment length. Cross-sectional area and volumetric air volume Three physical properties. Based on the geometric topology data of each channel, the system generates an undirected graph. Its node set It contains 120 access nodes and an edge set. It contains 85 channel edges, each associated with a channel length. Effective width and rated traffic capacity By comparing the coordinates of the 3D building information model corresponding to the nodes in the two layers of the graph, and A one-to-one mapping relationship is established between nodes with the same coordinates, forming a two-layer graph structure.

[0074] Table 1 below shows some typical nodes in Building A and their key topological attributes to reflect the node relationships between the two layers of the topological graph in the two-layer graph structure.

[0075] Table 1. Attributes of the two-layer diagram of typical nodes in Building A.

[0076] Step S200: Calculate the spatiotemporal propagation sequence of the gas pollution front; At 10:22 AM on a weekday morning during the spring semester of a certain year, in the chemical synthesis laboratory on the 3rd floor of Building A (node ​​11, corresponding to...) Nodes in = VV-311) Gas sensor alarm confirmed ammonia leak; the leak confirmation time is recorded as . Seconds. The system obtains the current operating parameters of the fresh air fan: the main fan speed is 1200 rpm, corresponding to a volumetric air volume. .

[0077] The system is based on nodes Starting point Perform a breadth-first traversal, and for each directed edge on the traversal path, use the formula... Calculate the transmission delay time, then sum it along the path for each downstream node to obtain the estimated arrival time of the contamination at each node. .

[0078] From node Taking (VV-311) to node VV-214 (the second-floor stairwell platform, the path passes through the main pipe section and two branch pipe sections) as an example, the parameters of the three pipe sections are shown in Table 2.

[0079] Table 2. Path pipe segment parameters and transmission delay from the 3rd floor chemical synthesis laboratory to the 2nd floor stairwell.

[0080] The estimated arrival time of contamination at node VV-214 is:

[0081] After performing the same calculation on all reachable nodes and according to After ascending order, a spatiotemporal progression sequence of the gas pollution front is generated, and some results are shown in Table 3.

[0082] Table 3. Spatiotemporal progression sequence of the gas pollution front (partial nodes)

[0083] Step S300: Calculate the channel cumulative pressure index; After the leak alarm was triggered, students and staff began moving towards the various evacuation exits. The system obtained real-time pedestrian flow data from personnel location terminals in each area. This data was used to track the main evacuation route (passage path) from floors 1 to 2 of Building A. Channel length Taking the cross-sectional spacing as an example, the system uses... Discretize the channel along the axial direction, the total number of cross sections Section numbering starts from the entrance side. To the exit side Within the time window Within this section, the number of people passing through per unit time for each cross-section is statistically analyzed. (Unit: people / s), perform differential calculation on adjacent sections. To obtain the net cumulative flow rate, some cross-sectional data are shown in Table 4.

[0084] Table 4. Flow rate and net cumulative flow rate of main evacuation routes

[0085] section to The net cumulative flow rate between all 5 adjacent sections is positive. The system determines that the channel segment is in a state of pressure accumulation, and... , , Calculate axial weight Then, by weighted summation, the cumulative stress index is obtained:

[0086] Step S400: Assess the risk of stampedes in the passageway; The system targets the main evacuation route (passage side) in a state of pressure accumulation. Extract the channel flow feature vector at the current moment. Simultaneously, historical channel flow data for the same weekday and break time (10:00 AM to 11:00 AM) are retrieved from the historical database to extract the baseline flow feature vector. The values ​​of each component are shown in Table 5.

[0087] Table 5 Comparison of Flow Characteristic Vectors in Main Evacuation Channels and Reference Flow Characteristic Vectors

[0088] right and Each corresponding component is Z-score standardized using the mean and standard deviation from historical data statistics, so that the two are in the same numerical space, and then calculated according to the formula. Calculate the topological deviation vector, whose six components are as follows: (Each component is a dimensionless standardized deviation value). [The last part appears to be incomplete and possibly refers to a past event or a separate topic.] The topological deviation vector at each time step is concatenated with the corresponding standardized cumulative pressure index to form a 7-dimensional input sequence, which is then fed into a pre-trained LSTM network. The classification branch of the LSTM network outputs that the stampede risk level for this passage segment is medium risk, while the regression branch outputs an estimate of the remaining time before the dangerous pressure threshold is reached. .

[0089] Step S500: Update the personnel access topology constraints; Based on the output results of steps S200 and S400, the system... The time-varying availability flags and pressure-aware weights of each pass edge are updated.

[0090] The west staircase passage on the 3rd floor (passage side) Taking the connection nodes VP-313 and VP-214 as an example: the estimated arrival time of personnel departing from the 3rd floor chemical synthesis laboratory (VP-311). The estimated travel time through this passage. The predicted arrival time of contamination for node VP-214 obtained from step S200 is... The following conditions are met:

[0091] Therefore, the passage edge Marked as time-limited and unavailable. For the main evacuation route (passage side) (The area is under pressure and the risk level of a stampede is medium risk; it has not been marked as unavailable), so the base weights are used. Pressure perception coefficient Cumulative stress index Update the path search weights according to the formula:

[0092] The constraint update results for each typical through edge are summarized in Table 6.

[0093] Table 6. Update Results of Passage Edge Constraints in the Personnel Access Topology Graph

[0094] Step S600: Perform a time-dependent shortest safe path search; The system applies a time-dependent Dijkstra algorithm to each evacuation group in Building A. Taking the evacuation group on the third floor (starting node VP-311, totaling 38 people) as an example, the algorithm expands the candidate nodes progressively from VP-311, performing a time-dependent safety verification on each candidate node. ) and pressure safety verification (the cumulative pressure index of all channel segments along the path is lower than ) ).

[0095] The west staircase path passes through node VP-214 ( The candidate expansion path corresponds to The time safety verification failed, and the path branch was pruned. The backup staircase path on the east side passes through node VP-212 (east side 2nd floor staircase platform). ), under this path The time safety verification was passed; the cumulative pressure index of the passageway corresponding to the east-side backup staircase was [value missing]. The pressure safety verification was passed. The algorithm outputs the safe evacuation route for the 3-story evacuation group as "VP-311 → VP-312 → VP-212 (East Side Backup Staircase) → VP-115 (1st Floor Lobby Exit)", and records the estimated entry and exit times for each passage segment, forming a sequence of passage occupancy time windows, as shown in Table 7. The estimated passage time for each segment is also shown in Table 7. The channel average speed is taken as the current real-time average value.

[0096] Table 7. Time window sequence of evacuation group passage occupancy on the 3rd floor.

[0097] Step S700: Detection and resolution of conflicts among multiple groups; The system performs conflict detection on the safe evacuation routes of all evacuation groups. The safe evacuation route of the 4th-floor evacuation group (45 people, starting at VP-411) also passes through the 1st-floor lobby exit passage (VP-212 → VP-115). This passage is occupied by the 4th-floor group from second 35 to second 68, overlapping with the 3rd-floor group's occupancy window (second 32.50 to second 59.77). Within this overlapping window, the combined expected crowd density of the two groups is [not specified]. Exceeding the rated maximum safety density of this channel segment (Based on rated capacity) Divide by effective passage area (Confirmed), the system marks the channel segment as having multiple group conflicts within the aforementioned time window.

[0098] According to the formula Calculate the time margin for each group: the minimum residual margin of each node on the path for a 3-layer group. (Here, we take the node with the minimum margin on the path), the minimum residual margin of the 4-level group at each node on the path. The 3-level group has a smaller time margin and higher priority, so its original path remains unchanged. The 4-level group re-executes the time-dependent shortest safe path search, marking the overlapping time window of the VP-212 → VP-115 passage segment as unavailable during the search, and outputs an alternative path to the outdoor connecting corridor. The multi-group joint evacuation route scheme after conflict resolution no longer has simultaneous high-density conflicts in passage segments.

[0099] Step S800: Dynamic update and command push.

[0100] Ten seconds after the evacuation was initiated, the system entered its first update cycle, acquiring real-time readings from gas concentration sensors distributed across various pipe nodes in Building A. Sensor data showed that node VV-312 (middle section of the main corridor on the 3rd floor) had exceeded the ammonia concentration limit at 6.10 seconds, while step S200 predicted the contamination arrival time at this node to be... The actual diffusion rate was faster than predicted, with a relative deviation of:

[0101] The deviation exceeds the preset tolerance. This triggers a recalculation. The system uses the current moment as the new starting moment and the latest fan operating parameters (the main fan speed is reduced to 1050 rpm due to overload protection, corresponding to the volumetric air volume). Using the latest pedestrian flow data as input, the complete calculation process from steps S200 to S700 is re-executed, updating the estimated arrival time of contamination for each downstream node and re-evaluating the time-varying availability markers for each passage edge. After the update, the estimated arrival time of contamination for node VP-115 (1st floor lobby exit) is changed from the original... shortened to approximately The time-varying availability constraints of the corresponding access edges are updated accordingly. The updated multi-group joint evacuation route plan is pushed to the smart evacuation sign on the 4th floor through the campus security network. The sign changes from displaying "Evacuate to the west staircase" to "Evacuate to the east corridor". The emergency broadcast terminal broadcasts the route change information simultaneously, completing the dynamic response of this update cycle.

[0102] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A multi-dimensional fire safety monitoring method for smart campuses, characterized in that, Includes the following steps: The system acquires the duct layout data and geometric topology data of each passage of the building ventilation system. Based on the duct layout data, it generates a ventilation duct network topology map and a personnel passage topology map. The ventilation duct network topology map and the personnel passage topology map are then linked by the location coordinates of the area space to form a two-layer graph structure. Once the gas leak is confirmed, the fan operating parameters are obtained, and a breadth-first traversal is performed on the ventilation duct network topology map starting from the source node. The transmission delay time is calculated based on the physical parameters of the pipe segments corresponding to each directed edge on the traversal path. The transmission delay time of all pipe segments on the path is accumulated for each downstream node to generate a spatiotemporal progression sequence of the gas pollution front. Real-time personnel location flow data of each evacuation route is obtained, each route is discretized into an ordered cross-section sequence along the axis, the number of people passing through each cross-section per unit time is calculated, and the net cumulative flow rate is obtained by performing differential operation on adjacent cross-sections. The cumulative pressure index of the route section is calculated based on the net cumulative flow rate. Based on the spatiotemporal progression sequence of the gas pollution frontier and the cumulative pressure index, time-varying availability markers and pressure perception weights are assigned to each passage edge in the personnel passage topology map. Time-dependent shortest safe path search is performed on each evacuation group on the personnel passage topology map after the constraint update, and a safe evacuation path is output. Conflict detection is performed on the safe evacuation routes of each group. Based on the time margin between each group and the pollution front, priority is ranked and routes are redistributed to generate a joint evacuation route plan for multiple groups.

2. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 1, characterized in that, The methods for calculating the cumulative pressure index of a channel segment based on the net cumulative flow rate include: When there are consecutive channels in the segment The net cumulative flow rate between each adjacent section is positive and When this occurs, the channel segment is marked as being in a state of pressure accumulation; The cumulative pressure index Calculate using the following formula: ,in The starting section number for the continuous positive value region. This represents the number of regions between consecutive positive cross sections. For the first Net cumulative flow rate between cross sections For the first The axial weight corresponding to the region between each cross section is defined as the normalized distance of that region from the channel exit. , This represents the total number of cross sections.

3. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 1, characterized in that, It also includes the following steps: For channel segments under pressure accumulation, extract channel flow feature vectors, extract benchmark flow feature vectors from historical channel flow data of the same period, and calculate topology deviation vectors after performing Z-score standardization on each corresponding component. The time series of the topological deviation vector and the cumulative pressure index are input into the pre-trained channel stampede risk time series classification model, which outputs the stampede risk level of each channel segment and the estimated remaining time to reach the dangerous pressure threshold. Wherein, the topological deviation vector Calculate using the following formula: ,in This is the standardized current channel flow feature vector. This is the standardized baseline flow characteristic vector.

4. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 3, characterized in that, The channel flow characteristic vector This includes the number of people passing through the entrance section of the passageway per unit time, the number of people passing through the exit section per unit time, the ratio of inlet to outlet flow, the cumulative pressure index, the average personnel density within the passageway, and the time-varying rate of average passing speed within the passageway. The time-series classification model for the risk of stampedes in the passageway uses an LSTM network, with an input of length [length missing]. The time series is such that the input vector at each time step is the concatenation of the topological deviation vector and the standardized cumulative pressure index at the corresponding time. The output layer contains a classification branch and a regression branch. The classification branch outputs the stampede risk level, and the regression branch outputs the estimated remaining time.

5. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 3, characterized in that, The method of assigning time-varying availability markers and pressure perception weights to each passage edge in the personnel passage topology map based on the spatiotemporal progression sequence of the gas pollution front and the cumulative pressure index includes: For any passing edge When the sum of the estimated time of personnel arriving at the starting node and the estimated travel time through the passage edge is not less than the estimated arrival time of pollution at the terminal node, the passage edge is marked as time-restricted and unavailable. For accessible edges that are under pressure accumulation and are not marked as time-limited and unavailable, their path search weights are increased according to the following formula: ,in Based on the weights, This represents the cumulative pressure index for the corresponding channel segment. This refers to the pressure perception coefficient. When the risk level of trampling reaches high risk, the corresponding passage edge is marked as high risk and restricted from passage.

6. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 1, characterized in that, The execution time-dependent shortest safe path search methods include: The time-dependent Dijkstra algorithm is used to expand from the starting node of the evacuation group, and time safety verification and stress safety verification are performed when expanding each candidate node; The time safety verification is to verify whether the estimated time to reach the candidate node along the current path is less than the estimated time of contamination of that node; if not. The pressure safety verification involves checking whether the cumulative pressure index of each channel traversed by the current path is lower than a preset safety threshold. If there is a channel segment that exceeds the preset safety threshold, it is pruned.

7. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 1, characterized in that, Prioritization and path reallocation methods based on the time margin between each group and the pollution front include: Group time margin Calculate using the following formula: ,in For groups The set of all nodes on the safe evacuation route. For nodes The pollution is expected to arrive at the time of [time]. For groups Reaching the node The expected time; The group with the smallest time margin is given the highest priority and its original path remains unchanged. The group with a larger time margin re-executes the shortest safe path search based on time dependencies. During the search, conflicting channel segments are marked as unavailable within the conflict time window. The process is repeated until all conflicts are resolved.

8. The method for multi-dimensional fire safety monitoring in a smart campus according to claim 1, characterized in that, Also includes: During the evacuation process, the latest channel flow data and gas concentration sensor data are acquired at preset update cycles. The actual arrival time of the nodes where the gas concentration actually exceeds the standard is compared with the predicted arrival time of the pollution in the spatiotemporal advancement sequence of the gas pollution front. When the gas diffusion prediction deviation exceeds the preset tolerance, or the trampling risk level of a certain passage section is upgraded to high risk, the complete calculation process from generating the spatiotemporal advancement sequence of the gas pollution front to generating the multi-group joint evacuation path scheme is re-executed with the current time as the new starting time, and the updated multi-group joint evacuation path scheme is pushed to the evacuation guidance equipment in each area.

9. A multi-dimensional fire safety monitoring method for smart campuses according to claim 1, characterized in that, The ventilation duct network topology is a directed graph, where the direction of each directed edge is consistent with the set flow direction of the airflow within the corresponding duct segment. Each directed edge is associated with three physical attributes: duct segment length, duct segment cross-sectional area, and fan volume and airflow. The personnel passage topology is an undirected graph, where each passage edge is associated with the channel length, effective channel width, and rated passage capacity. The two-layer graph structure achieves the association between the two topology graphs through a one-to-one mapping relationship of shared nodes, so that the gas pollution status at the ventilation duct network level is transmitted to the passage edge associated with the corresponding node at the personnel passage level.

10. A smart campus multi-dimensional fire safety monitoring system, used to execute the smart campus multi-dimensional fire safety monitoring method according to any one of claims 1 to 9, characterized in that, Includes the following modules: The two-layer graph construction module acquires the duct layout data and geometric topology data of each passage of the building ventilation system, generates a ventilation duct network topology map and a personnel passage topology map, and connects the two nodes through the location coordinates of the area space to form a two-layer graph structure. The gas diffusion prediction module obtains the fan operating parameters after the gas leak is confirmed, performs a breadth-first traversal on the ventilation duct network topology map and calculates the transmission delay time of each duct segment, and generates a spatiotemporal propagation sequence of the gas pollution front. The channel pressure calculation module acquires real-time personnel location flow data, discretizes the channel into an ordered cross-section sequence, calculates the cross-sectional flow difference to obtain the net cumulative flow rate, and calculates the cumulative pressure index of the channel segment. The graph constraint update module assigns time-varying availability markers and pressure perception weights to each passage edge in the personnel passage topology graph based on the spatiotemporal advancement sequence of the gas pollution frontier and the cumulative pressure index. The path search module performs a time-dependent shortest safe path search on the personnel access topology map after the constraint update for each evacuation group, and outputs a safe evacuation path. The conflict resolution module performs conflict detection on the safe evacuation routes of each group, prioritizes and redistributes routes based on time margin, and generates a joint evacuation route plan for multiple groups. The dynamic update and push module periodically acquires the latest sensor data and performs deviation detection. When the triggering conditions are met, it calls the gas diffusion prediction module to the conflict resolution module to recalculate and pushes the updated multi-group joint evacuation path plan to the evacuation guidance equipment in each area.

Images