A Networked Operation and Management Method for Urban Rail Transit Based on Graph Neural Networks
By constructing a dynamic operation map and a curvature-density coupled diffusion operator, the problem of identifying and predicting the propagation characteristics of short-term pulse passenger flow in urban rail transit networks was solved, enabling more accurate prediction of operational impacts and collaborative management and control, and improving the adaptability and responsiveness of networked operations.
Patent Information
- Application Number
- CN202611242669.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-17
- Publication Date
- 2026-11-10
AI Technical Summary
Existing urban rail transit operation and management methods are unable to accurately describe the dynamic propagation characteristics of short-term pulse passenger flow across multiple stations and lines, and cannot effectively identify propagation bottlenecks and redundant aggregation areas, resulting in insufficient adaptability and accuracy of operation and management strategies in networked passenger flow scenarios.
A graph neural network-based approach is used to construct a dynamic operation graph. The dynamic operation entropy curvature and propagation bottleneck deviation are determined by a directed two-hop operation propagation path. A curvature-density coupled diffusion operator is constructed by combining the dynamic node propagation density to predict the propagation results of operational impacts and generate collaborative management and control strategies.
It improves the accuracy and identification capability of predicting the propagation of operational impacts, enhances the adaptability and closed-loop nature of networked collaborative management and control of urban rail transit, improves the coordination between train operation adjustments and passenger flow control, and solves the problems of response lag and insufficient strategy adaptability of existing methods under the dispersal of large-scale events.
Smart Images

Figure CN123070570A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent operation and control technology, and in particular to a networked operation and control method for urban rail transit based on graph neural networks. Background Technology
[0002] As the scale of urban rail transit networks continues to expand and the transfer connections between lines become increasingly close, short-term pulse passenger flows generated in scenarios such as the dispersal of large-scale events can easily concentrate and surge into nearby stations in a short period of time, and spread to multiple stations and lines along train operating lines and transfer passages, causing a chain of operational impacts such as platform congestion, train overcrowding, and transfer passage congestion. Existing urban rail transit operation management methods typically set passenger flow control measures or adjust train intervals based on real-time passenger flow at individual stations, the number of people waiting on platforms, or train occupancy rates. These methods do not adequately consider the propagation relationships between different stations, lines, and transfer relationships, making it difficult to accurately describe the directional, time-delayed, and multi-path propagation characteristics of operational impacts within the rail transit network. Furthermore, existing graph-based operational status prediction methods often use fixed adjacency relationships or preset propagation weights for information transmission. They cannot dynamically adjust the information diffusion intensity between nodes based on remaining line capacity, remaining transfer channel capacity, propagation delay, and candidate management strategies. Consequently, they struggle to effectively identify propagation bottlenecks and redundant aggregation areas, and accurately predict the synergistic effect of train interval adjustments and passenger flow control measures. This results in insufficient adaptability and accuracy of the determined operation management strategies in networked passenger flow propagation scenarios. Summary of the Invention
[0003] To address the problem that existing urban rail transit operation management methods struggle to accurately characterize the dynamic propagation of short-term pulse passenger flow across multiple stations and lines, this invention provides a networked operation management method for urban rail transit based on graph neural networks. This method constructs a dynamic operation graph based on short-term pulse passenger flow information and current operational data, generating multiple sets of candidate collaborative management strategies. It determines the dynamic operation entropy curvature, propagation bottleneck deviation, and redundancy aggregation deviation through directed two-hop operation propagation paths, and constructs a curvature-density coupled diffusion operator in conjunction with dynamic node propagation density. The diffusion operator is used to predict the operational impact propagation results of each candidate collaborative management strategy, determining the target collaborative management strategy and generating corresponding management instructions. This invention improves the accuracy and adaptability of operational impact propagation prediction and collaborative management decision-making.
[0004] This invention provides a method for networked operation and management of urban rail transit based on graph neural networks, specifically including the following steps:
[0005] Step S1: Obtain short-term pulse passenger flow information corresponding to the end of large-scale events and the current operation data of the urban rail transit network; determine each station as an operation node, determine the train operation relationship between adjacent stations and the transfer relationship between stations on different lines as directed operation relationships, and construct the basic operation topology; determine the operation node status corresponding to each operation node and the directed operation relationship status corresponding to each directed operation relationship based on the current operation data, and update the corresponding operation node status and associated directed operation relationship status based on the short-term pulse passenger flow information to obtain the current dynamic operation map;
[0006] Step S2: Determine the affected stations and lines based on the current dynamic operation map; determine the train interval adjustment range corresponding to the affected lines and the passenger flow control range corresponding to the affected stations based on preset operation control constraints, and generate multiple sets of candidate collaborative management and control strategies; for each set of candidate collaborative management and control strategies, update the corresponding operation node status and directed operation relationship status to obtain the corresponding control conditional dynamic operation map; wherein, the operation control constraints include the minimum safe train tracking interval constraint, the maximum change in train interval constraint, the station safe capacity constraint, and the control capacity constraint of passenger flow control equipment;
[0007] Step S3: For each control condition-based dynamic operation graph, based on the directed two-hop operation propagation path starting from each initial operation node and its corresponding relationship remaining capacity and relationship propagation delay, determine the dynamic operation entropy curvature corresponding to each operation node; compare the dynamic operation entropy curvature with the preset curvature balance interval to obtain the propagation bottleneck deviation and redundancy aggregation deviation corresponding to each operation node.
[0008] Step S4: Input the candidate collaborative management and control strategies, propagation bottleneck deviations and redundancy aggregation deviations corresponding to each operation node in the control conditional dynamic operation diagram into the propagation density generation network to obtain the dynamic node propagation density corresponding to each operation node, determine the dynamic diffusion weights corresponding to the directed operation relationship, and construct the curvature-density coupled diffusion operator based on the dynamic diffusion weights.
[0009] Step S5: Use the curvature-density coupled diffusion operator to perform multi-order graph message propagation on the control-conditional dynamic operation graph to predict the operation node state at multiple prediction times; redetermine the dynamic operation entropy curvature based on the current dynamic diffusion weight; and correct the dynamic node propagation density based on the redetermined dynamic operation entropy curvature feedback until the preset feedback update condition is met, and obtain the operation impact propagation prediction results corresponding to each candidate collaborative control strategy.
[0010] Step S6: Determine the target collaborative control strategy based on the propagation prediction results of each operation, generate train interval adjustment instructions and passenger flow control instructions based on the target collaborative control strategy, and update the current dynamic operation map based on the execution results of the corresponding instructions.
[0011] Furthermore, the process of determining the dynamic operational entropy curvature, propagation bottleneck deviation, and redundancy aggregation deviation in step S3 specifically includes the following steps:
[0012] Step C1: For each operation node in the control condition dynamic operation diagram, determine the currently processed operation node as the starting operation node; determine the directed two-hop operation propagation path starting from the starting operation node and passing through two directed operation relationships in sequence according to the train running direction and the transfer passenger flow direction; determine the terminal operation node of each directed two-hop operation propagation path as the corresponding two-hop target operation node, and determine the operation node located between the starting operation node and the two-hop target operation node as the corresponding valid intermediate operation node;
[0013] Step C2: Based on the remaining relationship capacity and relationship propagation delay of the two directed operational relationships in each directed two-hop operational propagation path, determine the effective propagation capacity of the path corresponding to the directed two-hop operational propagation path; for the same two-hop target operational node corresponding to the same starting operational node, add the effective propagation capacity of the path through each effective intermediate operational node to obtain the total two-hop path propagation capacity corresponding to the starting operational node to the two-hop target operational node; divide the effective propagation capacity of the path through each effective intermediate operational node by the total two-hop path propagation capacity to obtain the two-hop propagation share corresponding to each effective intermediate operational node, and form a corresponding two-hop propagation share set from the two-hop propagation shares;
[0014] Step C3: For each two-hop target operation node, determine the corresponding two-hop transmission dispersion based on the dispersion of each two-hop propagation share in the two-hop propagation share set corresponding to the starting operation node and the two-hop target operation node; determine the ratio between the number of effective intermediate operation nodes between the starting operation node and the two-hop target operation node and the number of preset reference intermediate nodes to obtain the relative number of intermediate nodes, and determine the ratio between the sum of the two-hop path propagation capabilities and the preset reference path propagation capabilities to obtain the relative path propagation capability; determine the curvature contribution value corresponding to the starting operation node and the two-hop target operation node based on the two-hop transmission dispersion, the relative number of intermediate nodes, and the relative path propagation capability; after traversing all two-hop target operation nodes corresponding to the starting operation node, combine the curvature contribution values to obtain the dynamic operation entropy curvature corresponding to the currently processed operation node;
[0015] Step C4: The curvature balance interval includes a lower curvature limit and an upper curvature limit. For each operational node, when the corresponding dynamic operational entropy curvature is less than the lower curvature limit, the propagation bottleneck deviation is determined based on the difference between the lower curvature limit and the dynamic operational entropy curvature. When the corresponding dynamic operational entropy curvature is greater than the upper curvature limit, the redundancy aggregation deviation is determined based on the difference between the dynamic operational entropy curvature and the upper curvature limit. When the corresponding dynamic operational entropy curvature is within the curvature balance interval, the corresponding propagation bottleneck deviation and redundancy aggregation deviation are set to zero.
[0016] Furthermore, the process of constructing the curvature-density coupled diffusion operator in step S4 specifically includes the following steps:
[0017] Step D1: For each operational node in each control-conditional dynamic operation diagram, determine the passenger flow control amount and train interval adjustment amount of the candidate collaborative control strategy applied to the operational node and the train operation relationship associated with the operational node; determine the adjacent operational nodes connected to the operational node through directed operational relationships, and determine the minimum, maximum, and average adjacent curvature values based on the dynamic operational entropy curvature corresponding to each adjacent operational node; combine the operational node state, passenger flow control amount, train interval adjustment amount, propagation bottleneck deviation, redundancy aggregation deviation, minimum, maximum, and average adjacent curvature values corresponding to the operational node to obtain the corresponding curvature-conditional operational state;
[0018] Step D2: Input the curvature-conditional operation states into the propagation density generation network to obtain the initial node propagation density corresponding to each operation node; perform non-negative mapping and normalization of the total propagation density on each initial node propagation density to obtain the dynamic node propagation density;
[0019] Step D3: For each directed operational relationship in the control-conditional dynamic operation diagram, determine the basic diffusion weight based on the remaining capacity and propagation delay of the directed operational relationship; determine the density adjustment amount based on the dynamic node propagation density of the operational nodes at both ends of the directed operational relationship, and determine the curvature adjustment amount based on the propagation bottleneck deviation and redundant aggregation deviation of the operational nodes at both ends; adjust the basic diffusion weight using the density adjustment amount and the curvature adjustment amount to obtain the dynamic diffusion weight;
[0020] Step D4: Write each dynamic diffusion weight into the corresponding directed operation relation to obtain the directed weighted adjacency matrix, and construct the curvature-density coupled diffusion operator based on the directed weighted adjacency matrix and the corresponding weighted out-degree matrix.
[0021] By adopting the above solution, the beneficial effects achieved by the present invention are as follows:
[0022] This method achieves dynamic characterization of the urban rail transit network operation status under the influence of short-term pulse passenger flow. By defining stations as operational nodes and train operation relationships and transfer relationships as directed operational relationships, and combining short-term pulse passenger flow information, operational node status, and directed operational relationship status to construct a current dynamic operation map, it can reflect the propagation process of passenger flow along the train operation direction and transfer direction between different stations and lines. At the same time, the dynamic operation map is updated according to candidate collaborative control strategies to form a control-conditional dynamic operation map, which improves the accuracy of describing the operation status under different train interval adjustment schemes and passenger flow control schemes, and solves the problem that existing methods mainly implement control based on the status of a single station and are difficult to reflect the cross-station and cross-line correlation effects.
[0023] This improves the ability to identify and predict the propagation characteristics of operational impacts. By determining the dynamic operational entropy curvature based on the relationship residual capacity, relationship propagation delay, and propagation share of the directed two-hop operational propagation path, it can identify propagation bottlenecks and redundant aggregation states in the passenger flow propagation process. Furthermore, by combining dynamic node propagation density, propagation bottleneck deviation, and redundant aggregation deviation to adjust the diffusion weights of directed operational relationships, a curvature-density coupled diffusion operator is constructed. This allows the graph message propagation intensity to dynamically change with operational status, network propagation capacity, and candidate collaborative control strategies. This solves the problem that using fixed adjacency relationships or fixed propagation weights is difficult to adapt to changes in the state of the rail transit network, and enhances the ability to predict operational impacts such as platform congestion, train overload, and the scope of congestion.
[0024] This enhances the closed-loop nature and adaptability of networked collaborative management and control of urban rail transit. By utilizing curvature-density coupled diffusion operators to perform multi-order graph message propagation and correcting the dynamic node propagation density based on the redefined dynamic operational entropy curvature feedback, the prediction results of operational impact propagation can be continuously corrected. On this basis, the target collaborative management and control strategy is determined according to the prediction results corresponding to each candidate collaborative management and control strategy, generating train interval adjustment instructions and passenger flow control instructions, and updating the current dynamic operation map based on the instruction execution results. This achieves closed-loop management and control of "state perception - propagation prediction - strategy selection - instruction execution - state update", improving the coordination of train operation adjustment and passenger flow control in short-term pulse passenger flow scenarios such as the dispersal of large events, and solving the problems of response lag and insufficient strategy adaptability of existing management and control methods. Attached Figure Description
[0025] Figure 1 This is a system flowchart of the urban rail transit networked operation and management method based on graph neural networks proposed in this invention;
[0026] Figure 2 This is a schematic diagram of the process for determining dynamic operating entropy curvature, propagation bottleneck deviation, and redundancy aggregation deviation as proposed in Embodiment 4 of the present invention;
[0027] Figure 3 This is a schematic diagram of the process for constructing the curvature-density coupled diffusion operator proposed in Embodiment 5 of the present invention. Detailed Implementation
[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0029] Example 1, as Figure 1 As shown, this invention provides a network-based operation and management method for urban rail transit based on graph neural networks. This method is applied to the network-based operation and management of urban rail transit during the short-term pulse passenger flow propagating across lines via transfer hubs during the dispersal period of large events. Specifically, it includes the following steps:
[0030] Step S1: Obtain short-term pulse passenger flow information corresponding to the end of large-scale events and the current operation data of the urban rail transit network; determine each station as an operation node, determine the train operation relationship between adjacent stations and the transfer relationship between stations on different lines as directed operation relationships, and construct the basic operation topology; determine the operation node status corresponding to each operation node and the directed operation relationship status corresponding to each directed operation relationship based on the current operation data, and update the corresponding operation node status and associated directed operation relationship status based on the short-term pulse passenger flow information to obtain the current dynamic operation map;
[0031] Step S2: Determine the affected stations and lines based on the current dynamic operation map; determine the train interval adjustment range corresponding to the affected lines and the passenger flow control range corresponding to the affected stations based on preset operation control constraints, and generate multiple sets of candidate collaborative management and control strategies; for each set of candidate collaborative management and control strategies, update the corresponding operation node status and directed operation relationship status to obtain the corresponding control conditional dynamic operation map; wherein, the operation control constraints include the minimum safe train tracking interval constraint, the maximum change in train interval constraint, the station safe capacity constraint, and the control capacity constraint of passenger flow control equipment;
[0032] Step S3: For each control condition-based dynamic operation graph, based on the directed two-hop operation propagation path starting from each initial operation node and its corresponding relationship remaining capacity and relationship propagation delay, determine the dynamic operation entropy curvature corresponding to each operation node; compare the dynamic operation entropy curvature with the preset curvature balance interval to obtain the propagation bottleneck deviation and redundancy aggregation deviation corresponding to each operation node.
[0033] Step S4: Input the candidate collaborative management and control strategies, propagation bottleneck deviations and redundancy aggregation deviations corresponding to each operation node in the control conditional dynamic operation diagram into the propagation density generation network to obtain the dynamic node propagation density corresponding to each operation node, determine the dynamic diffusion weights corresponding to the directed operation relationship, and construct the curvature-density coupled diffusion operator based on the dynamic diffusion weights.
[0034] Step S5: Use the curvature-density coupled diffusion operator to perform multi-order graph message propagation on the control-conditional dynamic operation graph to predict the operation node state at multiple prediction times; redetermine the dynamic operation entropy curvature based on the current dynamic diffusion weight; and correct the dynamic node propagation density based on the redetermined dynamic operation entropy curvature feedback until the preset feedback update condition is met, and obtain the operation impact propagation prediction results corresponding to each candidate collaborative control strategy.
[0035] Step S6: Determine the target collaborative control strategy based on the propagation prediction results of each operation, generate train interval adjustment instructions and passenger flow control instructions based on the target collaborative control strategy, and update the current dynamic operation map based on the execution results of the corresponding instructions.
[0036] In one specific implementation, the rail transit operation management after a large-scale event at Sports Center A is taken as an example. The event was expected to draw 18,000 people, with passenger flow concentrating at the nearby station A starting at 9:30 PM and peaking at 9:46 PM. A 2-minute control period is defined, and a 20-minute forecast period is defined. At 9:36 PM, the passenger flow entering station A is 950 people per control period, the number of people remaining on the platform is 1,350, the train occupancy rate in the direction of entry is 92%, and the current train interval is 300 seconds.
[0037] Based on the current operational status, three sets of candidate collaborative control strategies are generated: the first set does not adjust train intervals and does not implement passenger flow control; the second set adjusts train intervals from 300 seconds to 240 seconds and restricts entry to station A to 300 people during each control period; the third set adjusts train intervals to 270 seconds and restricts entry to station A to 200 people. The current dynamic operation map is updated according to each of the three candidate collaborative control strategies, resulting in three control-conditional dynamic operation maps.
[0038] Taking the second group of candidate collaborative control strategies as an example, there are two paths, A→B→D and A→C→D, between operation node A and the second-hop target operation node D. The effective propagation capabilities of the two paths are 162.6 and 16.2, respectively, and the second-hop propagation shares are 0.91 and 0.09, respectively. Since the propagation capability is mainly concentrated on the path through operation node B, the dynamic operational entropy curvature of operation node A is calculated to be 0.642. The curvature balance interval is [0.70, 1.00], therefore the propagation bottleneck deviation of operation node A is 0.058, and the redundancy aggregation deviation is zero.
[0039] The entry passenger flow, platform congestion, train occupancy rate, passenger flow control volume, train interval adjustment volume, propagation bottleneck deviation, and dynamic operational entropy curvature of adjacent operating nodes at station A are input into the propagation density generation network to obtain the dynamic node propagation density corresponding to station A. The dynamic node propagation density of other operating nodes is determined using the same method. Combined with the remaining relational capacity of each directed operational relationship, relational propagation delay, and curvature deviation of the two operating nodes, dynamic diffusion weights are obtained, and then curvature-density coupled diffusion operators corresponding to the second set of candidate collaborative control strategies are constructed.
[0040] Operational status was predicted using the curvature-density coupled diffusion operator corresponding to each candidate collaborative control strategy. The prediction results showed that under the first group of candidate collaborative control strategies, the maximum number of people stranded on station A's platform was 2240, exceeding the platform's safe capacity of 1800; under the second group of candidate collaborative control strategies, the maximum number of people stranded was 1680; and under the third group of candidate collaborative control strategies, the maximum number of people stranded was 1770. Considering the number of people stranded on the platform, the duration of train full load, the duration of train delays, and the cost of control adjustments, the second group of candidate collaborative control strategies was selected as the target collaborative control strategy.
[0041] Based on the target-oriented collaborative management and control strategy, train interval adjustment instructions are sent to the train operation control system of the corresponding line, and passenger flow control instructions are sent to the passenger flow control equipment at station A. After one control period, the actual number of people stranded on the platform at station A is 1080, and the train load factor is 89%. Based on the above results, the current dynamic operation map is updated, and the operation and control process for the next control period begins.
[0042] Example 2: This example is based on Example 1 and further explains the process of obtaining the current dynamic operation map in step S1.
[0043] Step A1: Obtain the estimated number of people leaving the large event, the start time of passenger arrival, the peak time of passenger arrival, the duration of passenger flow, and the proportion of passenger flow destinations for the corresponding period of the event's dispersal to obtain short-term pulse passenger flow information; obtain the current operating data of the urban rail transit network;
[0044] Step A2: Identify each station as an operational node, and define the train operation relationships between adjacent stations and the transfer relationships between stations on different lines as directed operational relationships to construct a basic operational topology; determine the station entry passenger flow, platform congestion, train arrival and departure status, and train occupancy rate corresponding to each operational node based on current operational data to obtain the operational node status; determine the remaining transport capacity of the operating section based on the train's rated passenger capacity, train occupancy rate, and train interval, and determine the remaining throughput capacity of the transfer passage based on the transfer passenger flow and transfer passage capacity; define the train travel time, train interval, and remaining transport capacity of the operating section as the directed operational relationship status corresponding to the train operation relationship, and define the transfer time, transfer passenger flow, and remaining throughput capacity of the transfer passage as the directed operational relationship status corresponding to the transfer relationship; load the operational node status and directed operational relationship status into the basic operational topology to obtain the basic dynamic operational diagram;
[0045] Step A3: Determine the time-sharing passenger flow increment corresponding to each operating period based on the short-term pulse passenger flow information, and allocate the time-sharing passenger flow increment to the corresponding train operation relationship and transfer relationship according to the passenger flow destination ratio; update the status of the corresponding operating node in the basic dynamic operation map using the time-sharing passenger flow increment, and update the status of the directed operation relationship associated with the time-sharing passenger flow increment according to the allocation result of the time-sharing passenger flow increment to obtain the current dynamic operation map.
[0046] Example 3, based on Example 1, further explains the process of generating the control-conditional dynamic operation diagram in step S2. The current dynamic operation diagram is generated using the method described in Example 2.
[0047] Step B1: Based on the number of passengers stranded on the platform and the train occupancy rate of each operating node in the current dynamic operation map, determine the degree of passenger flow accumulation for each operating node; based on the remaining transport capacity of the operating section or the remaining throughput capacity of the transfer channel corresponding to the directed operating relationship connected to each operating node, determine the corresponding passenger flow dispersal capacity; identify stations where the degree of passenger flow accumulation reaches the preset accumulation judgment value or the passenger flow dispersal capacity is lower than the preset dispersal capacity judgment value as affected stations, and identify lines with train operation relationships or transfer relationships with affected stations as affected lines;
[0048] Step B2: Based on the minimum safe following interval constraint and the maximum change in train interval constraint, determine the train interval adjustment range corresponding to each affected line; based on the station safety capacity constraint and the control capacity constraint of passenger flow control equipment, determine the passenger flow control range corresponding to each affected station within a unit control period; whereby, the passenger flow control quantity represents the number of passengers restricted from entering the affected station or restricted from passing through the corresponding transfer relationship within a unit control period.
[0049] Step B3: According to the preset control step size, determine multiple train interval adjustment amounts from each train interval adjustment range and multiple passenger flow control amounts from each passenger flow control range; combine the train interval adjustment amounts and passenger flow control amounts, and eliminate combinations that do not meet the operational control constraints to obtain multiple sets of candidate collaborative management and control strategies.
[0050] Step B4: For each group of candidate collaborative control strategies, update the train arrival and departure status, train load factor and remaining transport capacity of the affected line according to the train interval adjustment amount, and update the inbound passenger flow, platform lingering number and transfer passenger flow of the affected station according to the passenger flow control amount, to obtain the corresponding control conditional dynamic operation map.
[0051] Example 4, as Figure 2 As shown, this embodiment, based on Embodiment 1, further explains the process of determining the dynamic operating entropy curvature, propagation bottleneck deviation, and redundancy aggregation deviation in step S3. The control-conditional dynamic operating diagram is generated using the method described in Embodiment 3.
[0052] Step C1: For each operation node in the control condition dynamic operation diagram, determine the currently processed operation node as the starting operation node; determine the directed two-hop operation propagation path starting from the starting operation node and passing through two directed operation relationships in sequence according to the train running direction and the transfer passenger flow direction; determine the terminal operation node of each directed two-hop operation propagation path as the corresponding two-hop target operation node, and determine the operation node located between the starting operation node and the two-hop target operation node as the corresponding valid intermediate operation node;
[0053] Step C2: Based on the remaining relationship capacity and relationship propagation delay of the two directed operational relationships in each directed two-hop operational propagation path, determine the effective propagation capacity of each directed two-hop operational propagation path; for the same two-hop target operational node corresponding to the same starting operational node, add the effective propagation capacity of the path through each effective intermediate operational node to obtain the total two-hop path propagation capacity corresponding to the starting operational node to the two-hop target operational node; divide the effective propagation capacity of the path through each effective intermediate operational node by the total two-hop path propagation capacity to obtain the two-hop propagation share corresponding to each effective intermediate operational node, and form a corresponding two-hop propagation share set from each two-hop propagation share;
[0054] The process for determining the effective propagation capability of the path is as follows: For each directed two-hop operation propagation path, the remaining transport capacity of the operating section corresponding to the train operation relationship or the remaining throughput capacity of the transfer channel corresponding to the transfer relationship is determined as the remaining capacity of the corresponding directed operation relationship, and the train operation time corresponding to the train operation relationship or the transfer time corresponding to the transfer relationship is determined as the propagation delay of the corresponding directed operation relationship; the smaller value of the remaining capacity of the two directed operation relationships is determined as the basic propagation capability of the path, and the basic propagation capability of the path is attenuated according to the sum of the propagation delays of the two directed operation relationships to obtain the corresponding effective propagation capability of the path, so that when the sum of the propagation delays increases, the effective propagation capability of the path decreases.
[0055] Step C3: For each two-hop target operation node, determine the corresponding two-hop transmission dispersion based on the dispersion of each two-hop propagation share in the two-hop propagation share set corresponding to the starting operation node and the two-hop target operation node; determine the ratio between the number of effective intermediate operation nodes between the starting operation node and the two-hop target operation node and the number of preset reference intermediate nodes to obtain the relative number of intermediate nodes, and determine the ratio between the sum of the two-hop path propagation capabilities and the preset reference path propagation capabilities to obtain the relative path propagation capability; determine the curvature contribution value corresponding to the starting operation node and the two-hop target operation node based on the two-hop transmission dispersion, the relative number of intermediate nodes, and the relative path propagation capability; after traversing all two-hop target operation nodes corresponding to the starting operation node, combine the curvature contribution values to obtain the dynamic operation entropy curvature corresponding to the currently processed operation node; wherein, under the same conditions, the more concentrated the two-hop propagation share is in one or a few effective intermediate operation nodes, the smaller the corresponding curvature contribution value, and the dynamic operation entropy curvature corresponding to the currently processed operation node is reduced accordingly;
[0056] Dynamic operating entropy curvature formula:
[0057] ;
[0058] ;
[0059] ;
[0060] in, This indicates the number of the candidate collaborative management and control strategy and its corresponding control condition dynamic operation diagram; Indicates the starting operational node number currently being used for calculation; Indicates the starting operating node The corresponding second-hop target operation node number; Indicates that it is located at the starting operation node With the second jump target operation node The valid intermediate operational node number between them; Indicates the traversal number used when summing the valid intermediate operational nodes; Indicates the starting operation node The traversal number used when summing the corresponding two-hop target operation nodes; Indicates the first In a dynamic operation diagram with control conditions, the starting operation node To the second hop target operation node The set of effective intermediate operation nodes between them; Indicates the number of effective intermediate operating nodes; Indicates the first In a dynamic operation diagram with control conditions, the starting operation node The set of two-hop target operation nodes that can be reached through the directed two-hop operation propagation path; Indicates the first Under the collaborative management and control strategy for group candidates, the effective propagation capability of the directed two-hop operation propagation path i→m→j; Indicates the first Under the collaborative management and control strategy for candidate groups, the effective propagation capability of the directed two-hop operation propagation path i→n→j; This represents the effective propagation capability of the path i→n→l; Indicates a valid intermediate operating node The corresponding second-hop propagation share; Indicates the target operation node of the second hop. The corresponding curvature contribution weight; Indicates the starting operating node The corresponding dynamic operational entropy curvature; Indicates the preset number of intermediate reference nodes; Indicates the propagation capability of the preset reference path; , , The weights representing the two-hop transmission dispersion, the relative number of intermediate nodes, and the relative path propagation capability are non-negative, and the sum of the three is 1.
[0061] Step C4: The curvature balance interval includes a lower curvature limit and an upper curvature limit. For each operational node, when the corresponding dynamic operational entropy curvature is less than the lower curvature limit, the propagation bottleneck deviation is determined based on the difference between the lower curvature limit and the dynamic operational entropy curvature. When the corresponding dynamic operational entropy curvature is greater than the upper curvature limit, the redundancy aggregation deviation is determined based on the difference between the dynamic operational entropy curvature and the upper curvature limit. When the corresponding dynamic operational entropy curvature is within the curvature balance interval, the corresponding propagation bottleneck deviation and redundancy aggregation deviation are set to zero.
[0062] Example 5, such as Figure 3 As shown, this embodiment, based on Embodiment 1, further explains the process of constructing the curvature-density coupled diffusion operator in step S4. Specifically, the propagation bottleneck deviation and redundancy aggregation deviation are generated using the method described in Embodiment 4.
[0063] Step D1: For each operational node in each control-conditional dynamic operation diagram, determine the passenger flow control amount and train interval adjustment amount of the candidate collaborative control strategy applied to the operational node and the train operation relationship associated with the operational node; determine the adjacent operational nodes connected to the operational node through directed operational relationships, and determine the minimum, maximum, and average adjacent curvature values based on the dynamic operational entropy curvature corresponding to each adjacent operational node; combine the operational node state, passenger flow control amount, train interval adjustment amount, propagation bottleneck deviation, redundancy aggregation deviation, minimum, maximum, and average adjacent curvature values corresponding to the operational node to obtain the corresponding curvature-conditional operational state;
[0064] Step D2: Input each curvature-conditional operation state into the propagation density generation network to obtain the initial node propagation density corresponding to each operation node; perform non-negative mapping and normalization of the total propagation density of each initial node propagation density to obtain the dynamic node propagation density; the propagation density generation network is a small neural network defined in this scheme, which is used to map the curvature-conditional operation state into a feedforward neural network for node information propagation density, including a state mapping layer and a density output layer connected in sequence.
[0065] Step D3: For each directed operational relationship in the control-conditional dynamic operation diagram, determine the basic diffusion weight based on the remaining capacity and propagation delay of the directed operational relationship; determine the density adjustment amount based on the dynamic node propagation density of the operational nodes at both ends of the directed operational relationship, and determine the curvature adjustment amount based on the propagation bottleneck deviation and redundant aggregation deviation of the operational nodes at both ends; adjust the basic diffusion weight using the density adjustment amount and curvature adjustment amount to obtain the dynamic diffusion weight, so that the dynamic diffusion weight of the effective propagation direction increases when the propagation bottleneck deviation increases, and the dynamic diffusion weight decreases when the redundant aggregation deviation increases;
[0066] Step D4: Write each dynamic diffusion weight into the corresponding directed operation relation to obtain the directed weighted adjacency matrix, and construct the curvature-density coupled diffusion operator based on the directed weighted adjacency matrix and the corresponding weighted out-degree matrix.
[0067] Example 6, based on Example 1, further explains the process of feedback correction in step S5 to obtain the operational impact propagation prediction result. The curvature-density coupled diffusion operator is generated using the method described in Example 5.
[0068] Step E1: Based on the curvature-density coupled diffusion operator, perform graph message diffusion of different propagation orders on the state of the operation nodes in each control conditional dynamic operation graph, and combine the operation node diffusion states corresponding to different propagation orders to obtain the predicted state of the operation nodes at each prediction time.
[0069] Step E2: Use the current dynamic diffusion weight as the basis for determining the effective propagation capability of the directed two-hop operation propagation path. Use the same determination method as in step S3 to re-determine the dynamic operation entropy curvature corresponding to each operation node. Based on the difference between the re-determined dynamic operation entropy curvature and the preset curvature balance interval, obtain the updated propagation bottleneck deviation and the updated redundancy aggregation deviation.
[0070] Step E3: When the update propagation bottleneck deviation or update redundancy aggregation deviation is greater than the preset curvature deviation acceptance value, and the current feedback update count is less than the preset maximum feedback update count, increase the dynamic node propagation density of the corresponding operation node and its effective propagation direction according to the update propagation bottleneck deviation, and decrease the dynamic node propagation density of the corresponding operation node according to the update redundancy aggregation deviation. After reconstructing the curvature-density coupled diffusion operator, return to step E1.
[0071] Step E4: When the deviations of each update propagation bottleneck and the aggregation deviations of each update redundancy are not greater than the curvature deviation acceptance value, or when the current number of feedback updates reaches the maximum number of feedback updates, determine the changes in the number of people stranded on the platform, the duration of train full load, the changes in train delays, and the range of stations affected by congestion based on the predicted status of the operational nodes corresponding to each prediction time, and obtain the operational impact propagation prediction results of the corresponding candidate collaborative control strategies; wherein, the preset feedback update conditions include that the deviations of each update propagation bottleneck and the aggregation deviations of each update redundancy are not greater than the curvature deviation acceptance value, or when the current number of feedback updates reaches the maximum number of feedback updates.
[0072] Example 7, based on Example 1, further explains the process of determining the target collaborative control strategy and generating corresponding instructions in step S6. The operational impact propagation prediction result is generated using the method described in Example 6.
[0073] Step F1: For each candidate collaborative control strategy, determine the maximum number of people stranded on the platform, the duration of full train load, the cumulative train delay time, and the number of stations affected by congestion within the prediction time domain based on the corresponding operational impact propagation prediction results;
[0074] Step F2: Determine whether each candidate collaborative management strategy meets the constraints of station capacity, minimum train interval, transport capacity of operating section, and passenger flow control volume. The candidate collaborative management strategy that meets all constraints is determined as an executable collaborative management strategy.
[0075] Step F3: Based on the maximum number of people stranded on the platform, the duration of full train load, the cumulative duration of train delays, the number of stations affected by congestion, and the control adjustment costs corresponding to the adjustment of train intervals and passenger flow control, determine the comprehensive control cost of each executable collaborative control strategy, and determine the executable collaborative control strategy with the minimum comprehensive control cost as the target collaborative control strategy.
[0076] Step F4: Generate train interval adjustment instructions based on the train interval adjustment amount in the target collaborative management and control strategy, generate passenger flow control instructions based on the passenger flow control amount in the target collaborative management and control strategy, send the train interval adjustment instructions to the train operation control system of the corresponding line, send the passenger flow control instructions to the passenger flow control equipment of the corresponding station, and update the current dynamic operation schedule based on the execution results returned by the train operation control system and the passenger flow control equipment.
[0077] The present invention and its embodiments have been described above. This description is not restrictive. The accompanying drawings are only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the present invention, such design should fall within the protection scope of the present invention.
Claims
1. A method for networked operation and management of urban rail transit based on graph neural networks, characterized in that, The method specifically includes the following steps: Step S1: Obtain short-term pulse passenger flow information and current operation data, determine operation nodes and directed operation relationships, and obtain the current dynamic operation map; Step S2: Determine the affected stations and lines based on the current dynamic operation map; determine the train interval adjustment range corresponding to the affected lines and the passenger flow control range corresponding to the affected stations based on the preset operation control constraints, and obtain the control-conditional dynamic operation map; Step S3: Determine the dynamic operating entropy curvature corresponding to each operating node in the control conditional dynamic operation diagram; compare the dynamic operating entropy curvature with the preset curvature balance interval to obtain the propagation bottleneck deviation and redundancy aggregation deviation corresponding to each operating node; Step S4: Input the candidate collaborative management and control strategies, propagation bottleneck deviations, and redundancy aggregation deviations corresponding to the operation nodes in the control-conditional dynamic operation diagram into the propagation density generation network to obtain the dynamic node propagation density and construct the curvature-density coupled diffusion operator. Step S5: Use the curvature-density coupled diffusion operator to perform multi-order graph message propagation on the control-conditional dynamic operation graph, redetermine the dynamic operation entropy curvature, and obtain the operation impact propagation prediction results; Step S6: Determine the target collaborative management strategy based on the propagation prediction results of each operation impact, generate train interval adjustment instructions and passenger flow control instructions, and update the current dynamic operation map based on the execution results of the corresponding instructions.
2. The urban rail transit network operation and management method based on graph neural networks according to claim 1, characterized in that: Each station is designated as an operational node, and the train operation relationship between adjacent stations and the transfer relationship between stations on different lines are defined as directed operational relationships.
3. The urban rail transit network operation and management method based on graph neural networks according to claim 2, characterized in that, Step S3, which determines the dynamic operational entropy curvature, propagation bottleneck deviation, and redundancy aggregation deviation, specifically includes the following steps: Step C1: For each operation node in the control condition dynamic operation diagram, determine the currently processed operation node as the starting operation node; determine the directed two-hop operation propagation path according to the train running direction and the transfer passenger flow direction; determine the end operation node of the directed two-hop operation propagation path as the two-hop target operation node, and determine the operation node located between the starting operation node and the two-hop target operation node as the effective intermediate operation node; Step C2: Construct a set of two-hop propagation shares based on the remaining capacity and propagation delay of the two directed operational relationships in each directed two-hop operational propagation path; Step C3: Determine the two-hop transmission dispersion based on the dispersion of each two-hop transmission share in the two-hop transmission share set; obtain the relative number of intermediate nodes and the relative path propagation capability, and combine them with the two-hop transmission dispersion to obtain the dynamic operational entropy curvature; Step C4: The curvature balance interval includes the lower curvature limit and the upper curvature limit; when the dynamic operating entropy curvature is less than the lower curvature limit, the propagation bottleneck deviation is determined; when the dynamic operating entropy curvature is greater than the upper curvature limit, the redundancy aggregation deviation is determined.
4. The urban rail transit network operation and management method based on graph neural networks according to claim 3, characterized in that, Step C2 specifically involves: determining the effective propagation capacity of the path based on the remaining capacity and propagation delay of the two directed operational relationships in each directed two-hop operational propagation path; for the same two-hop target operational node corresponding to the same starting operational node, adding the effective propagation capacity of the path through each effective intermediate operational node to obtain the total two-hop path propagation capacity; dividing the effective propagation capacity of the path through each effective intermediate operational node by the total two-hop path propagation capacity to obtain the two-hop propagation share, and forming a two-hop propagation share set from the various two-hop propagation shares.
5. The urban rail transit network operation and management method based on graph neural networks according to claim 3, characterized in that, Step C3 specifically involves: determining the two-hop transmission dispersion based on the dispersion of each two-hop propagation share in the two-hop propagation share set; determining the ratio between the number of effective intermediate operating nodes between the starting operating node and the target operating node of the two-hop path and the number of preset reference intermediate nodes to obtain the relative number of intermediate nodes; and determining the ratio between the total two-hop path propagation capability and the preset reference path propagation capability to obtain the relative path propagation capability. Based on the two-hop transmission dispersion, the relative number of intermediate nodes, and the relative path propagation capability, the curvature contribution value is determined and combined to obtain the dynamic operational entropy curvature.
6. The urban rail transit network operation and management method based on graph neural networks according to claim 1, characterized in that, The process of constructing the curvature-density coupled diffusion operator in step S4 specifically includes the following steps: Step D1: For each operating node in the control-conditional dynamic operation diagram, determine the passenger flow control quantity and train interval adjustment quantity for that operating node; obtain the adjacent operating nodes connected to that operating node through directed operation relationships, and determine the minimum, maximum, and average adjacent curvature values based on the dynamic operation entropy curvature corresponding to each adjacent operating node; combine the operating node state, passenger flow control quantity, train interval adjustment quantity, propagation bottleneck deviation, redundancy aggregation deviation, minimum, maximum, and average adjacent curvature values corresponding to that operating node to obtain the curvature-conditional operation state; Step D2: Input the curvature-conditional operation states into the propagation density generation network to obtain the initial node propagation density; perform non-negative mapping and normalization of the total propagation density on each initial node propagation density to obtain the dynamic node propagation density; Step D3: Based on the remaining relational capacity and relational propagation delay of each directed operational relation in the dynamic operation diagram of each control condition, obtain the dynamic diffusion weight; Step D4: Write each dynamic diffusion weight into the corresponding directed operation relation to obtain the directed weighted adjacency matrix, and then use the curvature-density coupled diffusion operator based on the directed weighted adjacency matrix and the corresponding weighted out-degree matrix.
7. The urban rail transit network operation and control method based on graph neural networks according to claim 6, step D3 specifically comprises: determining the basic diffusion weight based on the remaining relation capacity and relation propagation delay of each directed operation relation in the dynamic operation graph of each control condition; determining the density adjustment amount based on the dynamic node propagation density of the operation nodes at both ends of the directed operation relation, and determining the curvature adjustment amount based on the propagation bottleneck deviation and redundancy aggregation deviation of the operation nodes at both ends; adjusting the basic diffusion weight using the density adjustment amount and the curvature adjustment amount to obtain the dynamic diffusion weight.
8. The urban rail transit network operation and management method based on graph neural networks according to claim 1, characterized in that, Step S5 specifically involves: using the curvature-density coupled diffusion operator to perform multi-order graph message propagation on the control-conditional dynamic operation graph, predicting the operation node state at multiple prediction times for each operation node; redetermining the dynamic operation entropy curvature based on the dynamic diffusion weight; and correcting the dynamic node propagation density based on the redetermined dynamic operation entropy curvature feedback until the preset feedback update conditions are met, thereby obtaining the operation impact propagation prediction results corresponding to each candidate collaborative control strategy.
