A rail transit emergency personnel dynamic scheduling method based on centrality perception

Abstract

This invention discloses a dynamic scheduling method for emergency personnel in rail transit based on centrality perception, comprising: establishing a time-varying station topology model, assigning time-varying weights to each edge and calculating the shortest travel time between any two nodes; constructing a personnel-task bipartite graph model, establishing bipartite graph edges and calculating edge costs when personnel skills meet task requirements; discretizing the scheduling time window into several time intervals, generating time replicas of personnel nodes and task nodes, and establishing feasible edges when personnel can arrive before the latest task start time; setting the edge cost as the edge cost from the personnel node to the task node, and solving the personnel-task matching scheme using the minimum cost maximum flow algorithm; updating the time-varying weights at fixed time intervals, and recalculating the edge costs and personnel-task matching scheme. This invention reduces arrival delays and improves the reliability of emergency response through centrality-driven personnel selection and dynamic path optimization, and is suitable for complex scenarios such as high passenger flow and large transfer stations.

Description

Technical Field

[0001] This invention relates to the field of rail transit technology, and in particular to a method for dynamic dispatching of emergency personnel in rail transit based on centrality perception. Background Technology

[0002] As passenger density in urban rail transit continues to rise, emergencies within stations (such as passenger fainting, equipment malfunctions, fires, and stampedes) are becoming increasingly diverse, frequent, and sudden. The core objective of emergency response is to dispatch the most suitable personnel to the scene in the shortest possible time and maximize efficiency with limited resources. Currently, emergency dispatching in urban rail transit typically relies on a combination of fixed plans and manual dispatching. Dispatchers, based on the location of the incident and their experience, manually assess the situation and call station staff, security personnel, and maintenance workers to the scene. However, this model has several inherent shortcomings: it relies on human experience and lacks quantitative decision support; dispatching decisions are delayed, making it impossible to track changes in personnel and passenger flow in real time; resource utilization is low, failing to fully consider skill matching and task prioritization; and it lacks modeling for complex network structures, ignoring time-varying traffic conditions within stations. Especially at large transfer stations or during peak passenger flow periods, incidents may require cross-regional, multi-person collaborative handling. In such cases, personnel travel paths, arrival times, and on-site coordination efficiency directly impact the effectiveness of the response, and existing methods struggle to dynamically optimize these key factors. Some advanced integrated monitoring platforms and dispatching systems already possess electronic map visualization dispatching and GPS / indoor positioning functions, capable of displaying personnel locations and supporting task issuance via broadcast, SMS, and intercom. Some systems have introduced static path planning algorithms to calculate arrival times, assisting dispatchers in decision-making. However, these technologies still suffer from problems such as ignoring the time-varying characteristics of station networks, having a single optimization objective, lacking network structure awareness, and insufficient real-time replanning capabilities.

[0003] The "Method and Apparatus for Personnel Allocation in Subway Emergency Events" disclosed in Chinese patent literature, publication number CN116862150A, published on October 10, 2023, includes: determining the location and level of a subway emergency event; searching for the location of each person in a set of subway station staff; calculating the reachability time from each person in the set of subway station staff to the event location based on the personnel location; and allocating emergency response personnel from the set of subway station staff to the subway emergency event based on the reachability time and the event level. This technology, by searching for the location of each person in the set of subway station staff and calculating the reachability time to the event location, can dynamically and flexibly allocate emergency response personnel according to the optimal time plan. However, this technology still belongs to the static path planning method for calculating arrival time, and only uses the shortest time as the objective. It still suffers from problems such as ignoring the time-varying characteristics of the station network, having a single optimization objective, lacking network structure awareness, and insufficient real-time replanning capabilities. Summary of the Invention

[0004] This invention aims to overcome the problems of existing methods for dispatching emergency personnel in rail transit, such as ignoring the time-varying characteristics of station networks, lacking network structure awareness, having a single optimization objective, and insufficient real-time replanning capability. It proposes a dynamic dispatching method for emergency personnel in rail transit based on centrality awareness.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception includes: Establish a time-varying station topology model containing nodes and edges, assign time-varying weights to each edge, and calculate the shortest travel time between any two nodes; Construct a personnel-task bipartite graph model. When personnel skills meet task requirements, establish bipartite graph edges and calculate edge costs. The cost items in the edge cost include at least arrival time, task switching cost, and personnel centrality. Discretize the scheduling time window into several moments, generate time copies of personnel nodes and task nodes, and establish a feasible edge when the personnel can arrive before the latest start time of the task. Introduce source and sink nodes, and set the edge cost to the edge cost from the personnel node to the task node. Solve the personnel-task matching scheme using the minimum cost maximum flow algorithm. Update the time-varying weights at fixed time intervals and recalculate the edge costs and personnel-task matching schemes.

[0006] This invention constructs a time-varying station topology network model, combining real-time passenger flow perception, channel closure status, and equipment operation information to dynamically adjust travel time and achieve path prediction consistent with actual conditions. It introduces network science centrality analysis, quantifying the strategic value of personnel locations in the global network through centrality indicators such as betweenness centrality or proximity centrality, prioritizing the scheduling of personnel with higher global influence to improve overall efficiency during multi-task conflicts. A personnel-task bipartite graph model is established to achieve automatic matching and path synchronization optimization for collaborative tasks involving multiple people and multiple tasks. A time-extended network and rolling optimization strategy are adopted to dynamically adjust the scheduling scheme based on personnel status, task progress, and passenger flow density prediction data to address unforeseen circumstances during execution. Arrival time, task switching cost, and personnel centrality are uniformly incorporated into the minimum cost maximum flow algorithm or assignment optimization framework to achieve globally optimal resource allocation. Through centrality-driven personnel selection, dynamic path optimization, and multi-role personnel collaboration, arrival delays are significantly reduced, improving the reliability and consistency of emergency response, and adapting to complex scenarios such as high passenger flow and large transfer stations.

[0007] Preferably, establishing a time-varying station topology model including nodes and edges includes: The station is divided into a set of nodes, each node corresponds to a key location in the station, and the passable passages between nodes are used as edges to establish a time-varying station topology model. The time-varying weight of each edge represents the time it takes for people to pass through the passable passage corresponding to that edge; The shortest travel time is determined by the minimum sum of time-varying weights among all combinations of edges connecting any two nodes.

[0008] Preferably, the time-varying weight of any edge is: The product of the length of the passable passage corresponding to the edge and the passage obstruction penalty coefficient is divided by the real-time walking speed; The real-time walking speed is the product of the baseline walking speed and the pedestrian density-speed attenuation coefficient; The channel obstruction penalty coefficient reflects the increased cost of passage when passage is obstructed.

[0009] Preferably, the construction of the personnel-task bipartite graph model includes: Set up a personnel set and a task set. Each person in the personnel set should have at least a skill vector, real-time location, and status information. Each task in the task set should have at least a requirement vector, actual location, latest start time, and current completion status. When personnel skills meet task requirements, a bipartite graph edge is established and edge cost is calculated, whereby the edge cost is a weighted sum of several cost items.

[0010] Preferably, the cost item of the edge cost includes: Arrival time is the time it takes for personnel to travel from their location to the task location, calculated from the shortest travel time between any two nodes. The task switching cost is the cost for a person to switch from the current task to the target task, which is calculated based on the correlation value between the current task and the target task and the switching penalty coefficient. Personnel centrality is the centrality value of personnel location in the time-varying station topology model.

[0011] Preferably, the cost item of the edge cost also includes: The skill gap penalty is calculated by normalizing the cosine similarity between the person's skill vector and the task's requirement vector to the positive interval. The higher the matching degree between the skill vector and the task vector, the smaller the skill gap penalty. The urgency of a task is determined by the reciprocal of the difference between the latest start time of the task and the current time. Personnel load balancing is obtained by the ratio of a person's current load to their maximum load, representing the number of tasks performed and fatigue level of each person recently.

[0012] Preferably, the pedestrian density-velocity attenuation coefficient is 1 minus the difference between an exponential function of the flow density deviation with a base of a natural constant and an exponent of the flow density deviation; the flow density deviation is the difference between the reciprocal of the congestion limit density and the reciprocal of the actual pedestrian density, multiplied by the fitting parameter.

[0013] Preferably, the correlation value between the current task and the target task is a weighted sum of skill correlation and spatial correlation; Skill relevance is the cosine similarity between the demand vector of the current task and the demand vector of the target task to be switched to, and normalized to the positive interval. Spatial correlation is the normalized spatial distance between the current task's location and the target task's location in the time-varying station topology model.

[0014] As a preferred option, source nodes and sink nodes are introduced, and the edge cost is set to the edge cost from the personnel node to the task node, including: the edge capacity from the source node to the personnel node is the number of tasks that the personnel can execute simultaneously, and its edge cost is 0; The edge capacity from the task node to the sink node is equal to the number of people required for the task, and its edge cost is 0. An edge from a personnel node to a task node is a feasible edge with a capacity of 1 and a cost equal to the edge cost in the personnel-task bipartite graph.

[0015] As a preferred option, when recalculating edge costs and personnel-task matching schemes, it is permissible to interrupt the current task that is being executed. For the current task after the interruption, the remaining unfinished part will be treated as a new task and personnel will be re-assigned. When replanning, the sum of the time it takes for personnel to arrive at the task and the remaining duration of the task must be less than or equal to the difference between the latest completion time of the task and the current time. The task urgency has been changed to the reciprocal of the difference between the latest completion time of the task and the current time, minus the remaining duration of the task.

[0016] This invention offers the following advantages: It constructs a time-varying station topology network model, dynamically adjusting travel times based on real-time passenger flow perception, channel closure status, and equipment operation information to achieve path prediction consistent with actual conditions; it introduces network science centrality analysis, quantifying the strategic value of personnel locations within the global network through indicators such as betweenness centrality or proximity centrality, prioritizing the scheduling of personnel with higher global influence to improve overall efficiency during multi-task conflicts; it establishes a personnel-task bipartite graph model to achieve automatic matching and path synchronization optimization for collaborative tasks involving multiple personnel and tasks; it employs a time-extended network and rolling optimization strategy, dynamically adjusting scheduling schemes based on personnel status, task progress, and passenger flow prediction data to address unforeseen circumstances during execution; it integrates indicators such as arrival time, skill matching, task urgency, and personnel workload balancing into a minimum cost flow or assignment optimization framework to achieve globally optimal resource allocation; and through centrality-driven personnel selection, dynamic path optimization, and multi-role personnel collaboration, it significantly reduces arrival delays, improves the reliability and consistency of emergency response, and adapts to complex scenarios such as high passenger flow and large transfer stations. Attached Figure Description

[0017] Figure 1 This is a flowchart of the dynamic dispatching method for emergency personnel in rail transit based on centrality perception in this invention.

[0018] Figure 2 This is a flowchart of the process of replanning when the current task is interrupted in this invention. Detailed Implementation

[0019] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0020] Example 1, such as Figure 1 As shown, a dynamic dispatching method for emergency personnel in rail transit based on centrality perception includes: Establish a time-varying station topology model containing nodes and edges, assign time-varying weights to each edge, and calculate the shortest travel time between any two nodes; Construct a personnel-task bipartite graph model. When personnel skills meet task requirements, establish bipartite graph edges and calculate edge costs. The cost items in the edge cost include at least arrival time, task switching cost, and personnel centrality. Discretize the scheduling time window into several moments, generate time copies of personnel nodes and task nodes, and establish a feasible edge when the personnel can arrive before the latest start time of the task. Introduce source and sink nodes, and set the edge cost to the edge cost from the personnel node to the task node. Solve the personnel-task matching scheme using the minimum cost maximum flow algorithm. Update the time-varying weights at fixed time intervals and recalculate the edge costs and personnel-task matching schemes.

[0021] This invention constructs a time-varying station topology network model, combining real-time passenger flow perception, channel closure status, and equipment operation information to dynamically adjust travel time and achieve path prediction consistent with actual conditions. It introduces network science centrality analysis, quantifying the strategic value of personnel locations in the global network through centrality indicators such as betweenness centrality or proximity centrality, prioritizing the scheduling of personnel with higher global influence to improve overall efficiency during multi-task conflicts. A personnel-task bipartite graph model is established to achieve automatic matching and path synchronization optimization for collaborative tasks involving multiple people and multiple tasks. A time-extended network and rolling optimization strategy are adopted to dynamically adjust the scheduling scheme based on personnel status, task progress, and passenger flow density prediction data to address unforeseen circumstances during execution. Arrival time, task switching cost, and personnel centrality are uniformly incorporated into the minimum cost maximum flow algorithm or assignment optimization framework to achieve globally optimal resource allocation. Through centrality-driven personnel selection, dynamic path optimization, and multi-role personnel collaboration, arrival delays are significantly reduced, improving the reliability and consistency of emergency response, and adapting to complex scenarios such as high passenger flow and large transfer stations.

[0022] This invention introduces a two-layer network coupling mechanism of "station topology layer - personnel / task bipartite layer" for the first time in emergency personnel dispatching for rail transit. This mechanism allows arrival time to be calculated in real time by the lower-layer path and directly influence the upper-layer assignment decision, avoiding the problem of "separation between path calculation and task allocation" in existing technologies and improving global optimality. The personnel / task bipartite layer is a personnel-task bipartite graph model that incorporates multi-dimensional cost factors such as arrival time, switching costs, and personnel centrality. The station topology layer is a time-varying station topology model that provides real-time time-varying shortest path results from personnel to task locations, achieving multi-layer coupling and endogenous cost calculation.

[0023] By integrating network science centrality analysis, time-varying shortest path algorithm, and minimum cost maximum flow algorithm, optimal matching and rolling time-domain real-time optimization under multidimensional constraints can be achieved, significantly shortening personnel arrival time, improving event handling efficiency, and enhancing the system's adaptability in complex dynamic environments.

[0024] As a specific example, establishing a time-varying station topology model including nodes and edges includes: The station is divided into a set of nodes, each node corresponds to a key location in the station, and the passable passages between nodes are used as edges to establish a time-varying station topology model. The time-varying weight of each edge represents the time it takes for people to pass through the passable passage corresponding to that edge; The shortest travel time is determined by the minimum sum of time-varying weights among all combinations of edges connecting any two nodes.

[0025] The time-varying station topology model of this invention is a time-varying station network driven by passenger flow perception. Based on the station topology, it uses platforms, concourses, entrances and exits, equipment rooms, staircases / elevator entrances, etc. as nodes and passable passages as edges. At the same time, dynamic factors such as passenger flow density prediction, closure status, and equipment failure are introduced to modify the edge weights, forming a time-varying model that supports congestion avoidance and dynamic path planning.

[0026] Specifically, in establishing a time-varying station topology model, the first step is to divide the station into nodes. The entire station can be abstracted into a network composed of several nodes and edges connecting them. Key locations such as platforms, concourses, entrances / exits, equipment rooms, stairwells, and escalator entrances, or locations where multiple passageways intersect, are abstracted as nodes. Then, the passable passageways connecting adjacent nodes are abstracted as edges for model construction. The real-time traversal time for each edge is calculated and used as the corresponding time-varying weight.

[0027] To calculate the shortest travel time between any two nodes, first find all possible edge connections between the two nodes. For each edge connection, calculate the sum of the time-varying weights of all edges in the combination, which is the travel time for that combination. Select the combination with the shortest travel time among all edge connections as the travel path between the two nodes, and its corresponding travel time is the shortest travel time between the two nodes.

[0028] Optionally, the time-varying weight of any edge is: The product of the length of the passable passage corresponding to the edge and the passage obstruction penalty coefficient is divided by the real-time walking speed; Real-time walking speed is the product of the baseline walking speed and the pedestrian density minus the speed attenuation coefficient; The channel obstruction penalty coefficient reflects the increased cost of passage when passage is obstructed.

[0029] In this embodiment, the time-varying weight of an edge is the time required to traverse the passable passage corresponding to that edge, calculated based on real-time predicted pedestrian density. Compared to the fixed passage speed and passage time calculation methods in existing technologies, this embodiment considers the impact of pedestrian density and actual obstacles in the passage on the passage time, making the calculation of passage time more accurate.

[0030] Specifically, the channel obstruction penalty coefficient is a penalty coefficient for obstructions caused by closure, reverse flow, or equipment failure, which increases the travel time. It is used to correct the travel time of restricted channels. When a channel is passable but there are additional obstructions (such as reverse flow of people, local congestion, partial closure, etc.), the penalty coefficient is set to a finite value greater than 1 to reflect the increased travel cost. When a channel is completely impassable (such as full closure, serious equipment failure, etc.), the penalty coefficient can be set to infinity to block the channel in the path calculation.

[0031] The lane obstruction penalty coefficient is not a fixed value, but a configurable strategy parameter, typically set dynamically based on the scenario, rules, or real-time monitoring data. For example: the penalty coefficient for a normal lane is 1; for minor obstruction (slight congestion, partial reverse flow), the penalty coefficient is set to 1.2~1.5; for moderate obstruction (partial closure, narrow lane, partial equipment failure), the penalty coefficient is set to 2~3; and for severe obstruction or complete unavailability (full closure, equipment failure), the penalty coefficient is set to infinity. These values ​​can be pre-configured by operations personnel and traffic management departments based on experience or policy.

[0032] Furthermore, the pedestrian density-velocity attenuation coefficient is 1 minus the difference between an exponential function of the flow density deviation with the natural constant as the base and the exponent as the exponent; the flow density deviation is the difference between the reciprocal of the congestion limit density and the reciprocal of the actual pedestrian density, multiplied by the fitting parameters.

[0033] Actual pedestrian density can be predicted using existing methods. The pedestrian density-speed decay coefficient used in this embodiment is based on an empirical / mechanistic model of speed decreasing with pedestrian density derived from a basic pedestrian map. Alternatively, existing function coefficients can be directly used, so they will not be described in detail. By combining short-term pedestrian density predictions with real-time speed updates for passageways, potential congestion paths can be avoided in advance, increasing the speed at which people reach their destinations and saving time.

[0034] Compared to existing scheduling methods based on static maps or empirical speeds, this embodiment introduces dynamic factors such as crowd density prediction, lockdown status, and equipment failure to construct a time-varying model, which can update the shortest path and arrival time estimates in real time, significantly improving the accuracy and robustness of scheduling.

[0035] As a specific implementation example, constructing a personnel-task bipartite graph model includes: Set up a personnel set and a task set. Each person in the personnel set should have at least a skill vector, real-time location, and status information. Each task in the task set should have at least a requirement vector, actual location, latest start time, and current completion status. When personnel skills meet task requirements, a bipartite graph is constructed and edge costs are calculated. The edge cost is a weighted sum of several cost items.

[0036] Specifically, the skill vector Su of each person u in the personnel set is the specific skill that person possesses in handling practical matters, the real-time position Pu is the specific position of the person in the time-varying station topology model, and the status information can include the person's current idle or working status, current fatigue level, and equipment status, etc.

[0037] The requirement vector Rv for each task v in the task set can include not only skill requirements for personnel, but also location requirements, number of personnel, and equipment requirements, etc. The actual location Qv is the specific location of the task in the time-varying station topology model, the latest start time dv is the absolute time to start processing the task, and the current completion degree is a value between 0 and 1 (used to represent the current progress of the task). In addition, each task also needs a corresponding latest completion time, which can be calculated by adding the latest start time and the standard completion time of the task.

[0038] It should be noted that the task requirement vector includes requirements for personnel location, such as some tasks requiring personnel from specific areas to perform the task, including but not limited to nearby personnel or personnel with area access permissions. Requirements for the number of personnel refer to tasks that require multiple people to work together. Requirements for equipment refer to tasks that require special certifications or equipment to perform, including but not limited to having a high-voltage work permit, obtaining security authorization, or wearing specific protective equipment.

[0039] Furthermore, the edge cost is a weighted sum of several cost items, each with its own weight coefficient. These cost items include arrival time, task switching cost, personnel centrality, skill gap penalty, task urgency, and personnel load balancing. Compared to existing technologies that only schedule personnel tasks based on the shortest path or shortest time, this embodiment calculates personnel task matching from multiple perspectives, achieving globally optimal scheduling and allocation.

[0040] The weighting coefficient for each cost item is mainly determined based on expert experience, or the initial weight is given by operations and scheduling personnel based on their experience, and can be gradually adjusted according to the actual operating conditions.

[0041] Arrival time is the time it takes for personnel to travel from their location to the task location, calculated from the shortest travel time between any two nodes. This cost item reflects whether personnel can reach the task location on time; excessively long arrival times may increase task timeout costs.

[0042] The task switching cost is the cost for a person to switch from the current task to the target task. It is calculated based on the correlation value between the current task and the target task and the switching penalty coefficient, and the task switching cost must be negatively correlated with the correlation value.

[0043] When calculating task switching costs, it's necessary to consider the personnel's status information and task attributes. If the personnel are currently in an idle state, the switching cost can be set to 0, as idle personnel can be directly assigned to the target task. If the personnel are currently performing another task, the task switching cost needs to be calculated based on the correlation value between the current task and the target task. Specifically, it's calculated as 1 minus the difference in correlation value multiplied by the switching penalty coefficient.

[0044] The switching penalty coefficient represents the cost of interrupting the current task and is positively correlated with the task completion rate. Optionally, the switching penalty coefficient can be calculated by multiplying the task completion rate by a scaling constant, which can be set according to actual needs.

[0045] The relevance between the current task and the target task is a weighted sum of skill relevance and spatial relevance, with the sum of the weights of skill relevance and spatial relevance being 1.

[0046] Skill relevance is the cosine similarity between the demand vector of the current task and the demand vector of the target task to be switched to, and normalized to the positive interval. One possible method to normalize the cosine similarity to the positive interval is to add 1 to the calculated cosine similarity value and then divide by 2 to obtain the normalized result.

[0047] Spatial correlation is the normalized spatial distance between the current task's location and the target task's location in the time-varying station topology model. One possible method for normalizing the spatial distance is to calculate an exponential function with the natural constant e as the base and the negative value of the distance coefficient as the exponent. The distance coefficient is the product of the topological distance between the current task's location and the target task's location and a scaling attenuation coefficient, which is determined according to actual needs.

[0048] Personnel centrality is the centrality value of a person's location in a time-varying station topology model. When calculating edge costs, 1 minus the centrality value is used as the actual cost item, making the personnel centrality cost item negatively correlated with the actual centrality value. A higher centrality value indicates a higher strategic importance of the corresponding personnel location, and consequently, a lower cost. In this embodiment, the centrality value of a personnel location can be calculated using various commonly used centrality indicators; for example, betweenness centrality can be calculated as the percentage of shortest paths passing through a personnel location.

[0049] The skill gap penalty is calculated by normalizing the cosine similarity between a person's skill vector and the task's requirement vector to a positive interval. The higher the matching degree between the skill vector and the task vector, the smaller the skill gap penalty. Its practical significance lies in the fact that insufficient skills reduce the task success rate, thus requiring increased costs to prioritize scheduling personnel with suitable skills for execution. In this embodiment, the premise for calculating edge costs is that the personnel's skills meet the task requirements, but this does not conflict with the calculation of the skill gap penalty, because meeting the task requirements is only the minimum compliance standard. The level of skill proficiency, the number of skills, etc., all affect the task success rate, resulting in different cost differences. The skill gap penalty can be set by selecting the most suitable personnel from multiple personnel who simultaneously meet the task requirements to execute the corresponding task. One optional method for normalizing the cosine similarity to a positive interval is to add 1 to the calculated cosine similarity value, then divide by 2 to obtain the normalized result.

[0050] Task urgency is determined by the reciprocal of the difference between the task's latest start time and the current time. A smaller difference indicates higher urgency, requiring immediate action. However, a very small difference can lead to an extremely large reciprocal, severely impacting the overall consideration of other cost items. Therefore, an urgency cap is needed to ensure calculation stability. One possible approach is to take the reciprocal of the maximum value between the difference between the task's latest start time and the current time and a preset parameter, ensuring the urgency cap is the reciprocal of the preset parameter.

[0051] Personnel load balancing, calculated as the ratio of a person's current load to their maximum load, represents the number of tasks each person has recently executed and their fatigue level. Setting personnel load balancing parameters can help prevent excessive fatigue. A person's current load can be defined as the number of tasks they have executed up to the current time within the scheduling window, while the maximum load can be defined as the maximum number of tasks a person can execute within the scheduling window. These definitions are optional; the specific ratio of current load to maximum load can be defined and selected according to actual needs, but its meaning in expressing personnel fatigue levels remains unchanged.

[0052] As a specific implementation, the scheduling time window is discretized into several moments, and time copies of personnel nodes and task nodes are generated. When a personnel can arrive before the latest start time of the task, a feasible edge is established.

[0053] In this embodiment, the scheduling time window H is discretized into multiple moments t1, t2, etc. through a time extension network. The scheduling time window H is the overall time range considered for personnel scheduling. For example, when scheduling personnel within the next 2 hours, the scheduling time window H is set to [0, 120] minutes, and this window is discretized into several moments.

[0054] After discretization, a corresponding copy of the personnel node and task node is generated at each time step, and waiting edges and execution edges are connected on the time axis. The waiting edge represents "person / task waiting in place". For example, if a person does not accept a task at time t1, the personnel node at time t1 and time t2 can be connected, keeping the state unchanged. The execution edge represents "person executing a task at time t". Once a task is assigned, the personnel node at that time will be connected to the task node.

[0055] In a time-extended network, a feasible edge is established between a person node and a task node at the current time only if the shortest travel time from person u to task v is less than or equal to the difference between the latest start time of the task and the departure time of person u. This is a feasible scheduling constraint; a feasible edge can only be connected if the person can reach the task location before the deadline. Otherwise, it cannot be connected, and other person nodes need to be checked again.

[0056] Furthermore, after establishing feasible edges for personnel nodes and task nodes, source nodes and sink nodes are introduced, and the edge cost is set to the edge cost from the personnel node to the task node: The edge capacity from the source node to the personnel node is the number of tasks that the personnel can execute simultaneously (usually 1), and its edge cost is 0. The edge capacity from the task node to the sink node is equal to the number of people required for the task, and its edge cost is 0. An edge from a personnel node to a task node is a feasible edge with a capacity of 1. The edge cost is the edge cost in the personnel-task bipartite graph, which is the weighted sum of various cost items, including arrival time, task switching cost, personnel centrality, skill gap penalty, task urgency, and personnel load balancing.

[0057] After setting the edge costs and capacities between each node, the minimum cost maximum flow algorithm (MCMF) is used to solve the personnel-task matching problem. Assume task T1 requires two people, and the candidates are P1, P2, P3, and P4. When solving the personnel-task matching problem, all possible combinations of two people can be listed, the total cost of each combination can be calculated, and finally, the combination with the smallest total cost corresponds to the optimal scheduling for task T1. The MCMF algorithm performs a one-time global search using a "flow network" approach, automatically finding the "minimum total cost to satisfy the task's personnel requirements" among all combinations of personnel-task edges, thus obtaining the globally optimal matching result.

[0058] Finally, after calculating the personnel-task matching scheme, the corresponding scheduling method is output, and the time-varying weights in the time-varying station topology model are updated at fixed time intervals. At the same time, the personnel-task bipartite graph and the corresponding edge costs are updated, and the personnel-task matching scheme is recalculated.

[0059] The fixed time interval in this embodiment can be set according to actual needs. That is, each time interval is based on the predicted passenger flow density, and the time-varying weight of each edge in the time-varying station topology model (the time of passing through the corresponding passable channel) is updated. Based on the updated personnel and task information, the personnel-task bipartite graph is updated, the edge cost of the bipartite graph edges is recalculated, and then the personnel-task matching scheme is replanned. The fixed time interval rolling optimization strategy can replan in real time in the event of sudden task, personnel status changes, or path obstruction. The introduction of task switching costs and personnel load balancing balances personnel load, prevents individual personnel from working under high pressure for a long time, and improves the sustainability of the emergency system. At the same time, the short-term prediction of passenger flow density is included in the path cost calculation, which can avoid future high-density channels and potential congestion areas in advance, and realize preventive task allocation rather than simply passive response.

[0060] The centrality-aware dynamic scheduling method for emergency personnel in rail transit in this embodiment allows for the reassignment of tasks that have been assigned but not yet executed. Specifically, for example, during the previous personnel-task matching calculation, personnel and tasks were matched and assigned. Personnel A moves towards task a according to the planned path to prepare for the corresponding task. After a fixed time interval, if personnel A has not yet reached task a, the time-varying weights, edge costs, and task states are updated again, and replanning is performed. Personnel A can still be assigned to other tasks, and task a, still in an unexecuted state, can still undergo personnel-task matching, considering both personnel A and other personnel. Whether to reassign is primarily determined by the task switching cost in the edge cost, and minimizing the edge cost during minimum-cost maximum flow calculation. Even if the task switching cost of switching personnel to task a or switching personnel A from task a to other tasks is higher when the final edge cost is minimized, task reassignment is still necessary from a globally optimal perspective.

[0061] The following are the core problems existing in the current handling of urban rail transit emergencies: Untimely dispatch response: After an emergency occurs, the existing system relies on manual judgment and experience to allocate personnel, lacking comprehensive analysis of multiple factors such as station topology, real-time pedestrian flow, and traffic status, which results in the best rescue or disposal personnel being unable to arrive at the scene in the shortest possible time.

[0062] Low efficiency in personnel-task matching: Current scheduling strategies mostly use "shortest distance" or "shortest time" as the target for matching, which cannot fully consider the multi-dimensional matching of personnel skills, location, task requirements and urgency. There is a lack of scientific optimal allocation algorithms, which easily leads to inefficiencies such as "skill mismatch" and "not prioritizing the dispatch of nearby personnel".

[0063] Rigid scheduling under dynamic changes: Existing solutions mostly use static travel time and do not consider the impact of dynamic factors such as passenger congestion, channel closures, and equipment failures on path time, resulting in calculation results that do not match reality. At the same time, during the incident handling process, station passenger density, channel closures, and personnel task status will change continuously. Existing solutions lack real-time replanning capabilities and cannot adjust scheduling strategies according to the latest situation during execution.

[0064] Lack of a critical node priority strategy: The existing scheduling model does not incorporate centrality indicators from network science to identify the "critical locations" and "critical personnel" that play a decisive role in rescue / response. It cannot prioritize the allocation of personnel with the greatest global influence when multiple tasks conflict, resulting in resource allocation that cannot guarantee optimal overall network efficiency.

[0065] Compared with the prior art, the technical solution of the present invention has the following advantages: Dynamic nature: By utilizing real-time path calculation based on time-varying weights, dynamic factors such as changes in pedestrian flow, channel closures, and equipment status can be reflected in real time, and the scheduling results are highly consistent with the actual situation.

[0066] To achieve global optimality, a two-layer network model (time-varying station topology model and personnel-task bipartite graph model) and a minimum cost maximum flow algorithm are adopted, so that path planning and task allocation are completed within the same optimization framework, avoiding local optima caused by step-by-step optimization in alternative solutions.

[0067] Its collaborative capabilities enable it to handle collaborative tasks involving multiple roles and multiple participants, supporting multiple constraints such as skill matching, task switching, and personnel centrality, which is significantly superior to the single matching mode of existing technologies.

[0068] Rapid response and sustainability are achieved through rolling time-domain optimization and fairness control. Real-time replanning is possible in the event of sudden tasks or environmental changes, while ensuring balanced personnel load. It is suitable for emergency scenarios that require long-term operation.

[0069] Predictive-driven approaches, which incorporate crowd density prediction and centrality analysis into scheduling decisions, can proactively avoid potential congestion and achieve preventative rather than reactive responses.

[0070] When facing multi-task concurrency, highly dynamic environments, and rail transit emergency scenarios with extremely high requirements for response speed and accuracy, the rail emergency personnel dynamic scheduling method provided in this embodiment, which integrates network scientific centrality analysis, time-varying shortest path algorithm, and minimum cost maximum flow solution, can achieve optimal matching and rolling time-domain real-time optimization under multi-dimensional constraints, significantly shorten personnel arrival time, improve event handling efficiency, and enhance the system's adaptability in complex dynamic environments.

[0071] Example 2: Based on the technical content of Example 1, such as... Figure 2 As shown, the centrality-based dynamic scheduling method for emergency personnel in rail transit in this embodiment also allows for interruption and switching of the current task in progress after personnel-task matching is completed.

[0072] Establish a time-varying station topology model containing nodes and edges, assign time-varying weights to each edge, and calculate the shortest travel time between any two nodes; Construct a personnel-task bipartite graph model. When personnel skills meet task requirements, establish bipartite graph edges and calculate edge costs. The cost items in the edge cost include at least arrival time, task switching cost, and personnel centrality. Discretize the scheduling time window into several moments, generate time copies of personnel nodes and task nodes, and establish a feasible edge when the personnel can arrive before the latest start time of the task. Introduce source and sink nodes, and set the edge cost to the edge cost from the personnel node to the task node. Solve the personnel-task matching scheme using the minimum cost maximum flow algorithm. Update the time-varying weights at fixed time intervals and recalculate the edge costs and personnel-task matching schemes.

[0073] In this embodiment, it is necessary to determine whether the currently executing task is interrupted after replanning. If there is no interruption, the task is scheduled according to the replanned personnel-task matching scheme. If there is an interruption of the current task, the remaining unfinished part of the current task after the interruption is added back to the personnel-task bipartite graph model as a new task requirement for replanning. This process continues until all tasks are matched and there are no more interruptions, at which point the corresponding personnel-task matching scheme is output.

[0074] Specifically, taking the case where personnel A has already been assigned to execute task a and has completed a portion of the task, after a fixed-time replanning process, the planning result shows that for overall optimization, personnel A needs to interrupt task a and switch to other tasks added within the fixed time interval. In this case, task a needs to be changed to an incomplete task, and its remaining portion is reassigned to new personnel as a new task. At this point, the replanned result is not directly scheduled for execution; instead, the remaining new tasks need to be updated in the current replanned personnel-task bipartite graph, and the current replanning is performed again; new personnel are then assigned to execute the remaining tasks of task a. In practice, this interruption is usually considered a high-cost situation and only occurs in extremely urgent situations.

[0075] Specifically, the cost can be determined by the switching penalty coefficient in the task switching cost. The switching penalty coefficient represents the cost of interrupting the current task and is positively correlated with the task completion rate. The switching penalty coefficient is calculated by multiplying the task completion rate by a scaling constant. Setting the scaling constant to a larger value or setting the weight corresponding to the task switching cost to a larger value can increase the cost of interrupting the task and switching.

[0076] When replanning, the sum of the time it takes for personnel to arrive at the task and the remaining duration of the task must be less than or equal to the difference between the latest completion time of the task and the current time.

[0077] In this embodiment, the latest start time is a hard constraint for tasks that have not yet started execution; no feasible edge connections are made for personnel nodes and task nodes that exceed the latest start time. For remaining tasks that have started but were interrupted, the latest start time has already been consumed, and the latest completion time should be considered next. The latest completion time can be obtained by adding the latest start time to the task's standard working hours.

[0078] If task a is interrupted and person A is replanned to go to another target task, the remaining part of task a needs to be added as a new task for replanning. At this time, for any candidate person, the sum of the shortest travel time to the task location of task a, the current time, and the remaining duration of the task must be less than or equal to the latest completion time of the task. Only when this constraint is met can a feasible edge between person and task be established. If no other person besides person A can meet this constraint, a feasible edge cannot be established, and the task will be marked as "unable to complete / timeout". In this case, there are two options: one is to schedule personnel according to the current replanning result, issue an alarm, and manually dispatch suitable personnel to complete the remaining task; the other is to mark the match between task a and person A as uninterruptible and replan again to obtain the final scheduling result. The specific choice can be made according to actual needs.

[0079] In the event of a current task interruption, it is also necessary to adjust the cost items in the edge cost. The weight coefficient of each cost item is mainly determined based on expert experience, or the initial weight is given by the operation and scheduling personnel based on their experience, and can be gradually adjusted according to the actual operation situation.

[0080] Arrival time is the time it takes for personnel to travel from their location to the task location, calculated from the shortest travel time between any two nodes.

[0081] The task switching cost is the cost for a person to switch from the current task to the target task, which is calculated based on the correlation value between the current task and the target task and the switching penalty coefficient.

[0082] When calculating the cost of task switching, it is necessary to calculate the cost based on the correlation value between the current task and the target task to be switched to. Specifically, the cost is calculated by subtracting the difference in correlation value from 1 and multiplying it by the switching penalty coefficient.

[0083] The switching penalty coefficient represents the cost of interrupting the current task and is positively correlated with the task completion rate. Optionally, the switching penalty coefficient can be calculated by multiplying the task completion rate by a scaling constant.

[0084] Personnel centrality is the centrality value of a person's location in the time-varying station topology model.

[0085] The skill gap penalty is calculated by normalizing the cosine similarity between the person's skill vector and the task's requirement vector to the positive interval. The higher the matching degree between the skill vector and the task vector, the smaller the skill gap penalty.

[0086] Task urgency is the reciprocal of the difference between the latest completion time and the current time, minus the remaining duration of the task. The remaining duration can be obtained by subtracting the executed time from the standard task hours. Furthermore, if the difference between the latest completion time and the current time minus the task duration is small, the reciprocal calculated using only this difference will be extremely large, severely impacting the overall consideration of other cost items. Therefore, an upper limit for urgency needs to be set to ensure the stability of the calculation. One possible setting is to take the reciprocal of the maximum value between the difference between the latest start time and the current time minus the task duration and a preset parameter, so that the upper limit of urgency is the reciprocal of the preset parameter.

[0087] Personnel load balancing is obtained by the ratio of a person's current load to their maximum load, representing the number of tasks performed and fatigue level of each person recently.

[0088] The above embodiments are further elaborations and descriptions of the present invention to facilitate understanding, and are not intended to limit the present invention in any way. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception, characterized in that, include: Establish a time-varying station topology model and assign time-varying weights to each edge; Construct a personnel-task bipartite graph model. When personnel skills meet task requirements, establish bipartite graph edges and calculate edge costs. The cost items include at least arrival time, task switching cost, and personnel centrality. Discretize the scheduling time window, generate time copies of personnel nodes and task nodes, and establish feasible edges when personnel can arrive before the latest start time of the task. Introduce source and sink nodes, and set the edge cost to the edge cost of feasible edges. Solve the personnel-task matching scheme using the minimum cost maximum flow algorithm. Update the time-varying weights at fixed time intervals and recalculate the edge cost and personnel-task matching scheme.

2. The method for dynamic dispatching of emergency personnel in rail transit based on centrality perception according to claim 1, characterized in that, The establishment of the time-varying station topology model includes: The station is divided into a set of nodes, each node corresponds to a key location in the station, and the passable passages between nodes are used as edges to establish a time-varying station topology model. The time-varying weight of each edge represents the time it takes for people to pass through the passable passage corresponding to that edge; The shortest travel time is determined by the minimum sum of time-varying weights among all combinations of edges connecting any two nodes.

3. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception, as described in claim 1 or 2, characterized in that, The time-varying weight of any edge is: The product of the length of the passable passage corresponding to the edge and the passage obstruction penalty coefficient is divided by the real-time walking speed; The real-time walking speed is the product of the baseline walking speed and the pedestrian density-speed attenuation coefficient; The channel obstruction penalty coefficient reflects the increased cost of passage when passage is obstructed.

4. The method for dynamic dispatching of emergency personnel in rail transit based on centrality perception according to claim 1, characterized in that, The construction of the personnel-task bipartite graph model includes: Set up a personnel set and a task set. Each person in the personnel set should have at least a skill vector, real-time location, and status information. Each task in the task set should have at least a requirement vector, actual location, latest start time, and current completion status. When personnel skills meet task requirements, a bipartite graph edge is established and edge cost is calculated, whereby the edge cost is a weighted sum of several cost items.

5. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception, as described in claim 1 or 4, characterized in that, In the cost item of the aforementioned edge cost: Arrival time is the time it takes for personnel to travel from their location to the task location, calculated from the shortest travel time between any two nodes. The task switching cost is the cost for a person to switch from the current task to the target task, which is calculated based on the correlation value between the current task and the target task and the switching penalty coefficient. Personnel centrality is the centrality value of personnel location in the time-varying station topology model.

6. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception, as described in claim 1 or 4, characterized in that, The cost item of the edge cost also includes: The skill gap penalty is calculated by normalizing the cosine similarity between the person's skill vector and the task's requirement vector to the positive interval. The higher the matching degree between the skill vector and the task vector, the smaller the skill gap penalty. The urgency of a task is determined by the reciprocal of the difference between the latest start time of the task and the current time. Personnel load balancing is obtained by comparing the current load of personnel with the maximum load, representing the number of tasks recently performed and the level of fatigue for each person.

7. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception according to claim 3, characterized in that, The pedestrian density-velocity attenuation coefficient is the difference between 1 and an exponential function of the flow density deviation with a base of the natural constant and an exponent of the flow density deviation; the flow density deviation is the difference between the reciprocal of the congestion limit density and the reciprocal of the actual pedestrian density, multiplied by the fitting parameters.

8. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception according to claim 5, characterized in that, The correlation value between the current task and the target task is a weighted sum of skill correlation and spatial correlation; Skill relevance is the cosine similarity between the demand vector of the current task and the demand vector of the target task to be switched to, and normalized to the positive interval. Spatial correlation is the normalized spatial distance between the current task's location and the target task's location in the time-varying station topology model.

9. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception according to claim 1, 2, or 4, characterized in that, Introduce source nodes and sink nodes, and set the edge cost to the edge cost of feasible edges, including: the edge capacity from the source node to the personnel node is the number of tasks that the personnel can execute simultaneously, and its edge cost is 0; The edge capacity from the task node to the sink node is equal to the number of people required for the task, and its edge cost is 0. An edge from a personnel node to a task node is a feasible edge with a capacity of 1 and a cost equal to the edge cost in the personnel-task bipartite graph.

10. A method for dynamic dispatching of emergency personnel in rail transit based on centrality perception according to claim 9, characterized in that, When recalculating edge costs and personnel-task matching schemes, it is permissible to interrupt the current task that is being executed. For the interrupted current task, the remaining unfinished part will be treated as a new task and personnel will be re-assigned. When replanning, the sum of the time it takes for personnel to arrive at the task and the remaining duration of the task must be less than or equal to the difference between the latest completion time of the task and the current time. The task urgency has been changed to the reciprocal of the difference between the latest completion time of the task and the current time, minus the remaining duration of the task.

Images