A Multi-Constraint Crowd Management Path Planning Method Based on Congestion Perception

Abstract

This invention discloses a multi-constraint crowd diversion path planning method based on congestion perception, comprising: pre-constructing a diversion road network model of nodes and channels in a target area, and configuring basic attributes for nodes and channels; the method further includes: acquiring congestion perception data of each channel and / or each node in the target area from sensing devices; for each diversion path in a set of multiple candidate diversion paths, mapping it to the operating state parameters of each channel and / or each node based on the congestion perception data and the diversion road network model to obtain dynamic path costs; modeling the constraints involved in diversion path planning to establish a unified multi-constraint model; constructing a multi-objective optimization function, and solving the multiple sets of candidate diversion paths in combination with the unified multi-constraint model and dynamic path costs to determine a set of diversion paths. This method achieves real-time response and global coordination, reduces path conflicts, and has high scenario adaptability.

Description

Technical Field

[0001] This invention belongs to the field of crowd dispersal path planning technology, specifically relating to a multi-constraint crowd dispersal path planning method based on congestion perception. Background Technology

[0002] With the increasing prevalence of high-density crowd scenarios such as large sports stadiums, transportation hubs, and commercial complexes, crowd management and emergency evacuation have become critical issues in the field of public safety. In scenarios such as the end of sporting events, peak holiday traffic, and emergency evacuation, if the crowd control plan cannot respond to changes in crowd flow in real time, it can easily lead to overloaded passageways and path conflicts.

[0003] Pre-defined route planning methods based on static road networks use site plans or topological road network models as a basis, pre-setting fixed parameters such as length, width, direction, and maximum capacity for passages to solve for the shortest path. This method typically only considers static geometric distances and fixed obstacles, ignoring updated congestion perception data such as pedestrian density, which can lead to potential localized congestion along the pre-defined evacuation routes.

[0004] In recent years, research has begun to explore collecting data on pedestrian density or passage speed using devices such as video surveillance, radar, and turnstiles, and using the perceived data to adjust path finding. However, the perceived data is mostly auxiliary or only used to correct local parameters, failing to form a systematic coupling with the path solving process at multiple levels. This results in a significant disconnect between the path planning results and the actual congestion situation on site, easily leading to people from different sources being concentrated and directed to the same key passage, artificially creating new congestion points.

[0005] Furthermore, existing methods lack a unified framework for hierarchical constraint handling during path solving, or typically only address a few types of constraints. Within the set of detoxification paths determined within the solution framework, there may be infeasible paths that violate the constraints. Moreover, under fewer constraints, this can lead to an increase in the number of candidate detoxification paths, causing a rapid expansion of computational complexity, making it difficult to simultaneously guarantee solution efficiency and path quality in large-scale engineering scenarios. Summary of the Invention

[0006] This invention provides a multi-constraint crowd diversion path planning method based on congestion perception. By directly mapping real-time congestion perception data into dynamic path costs and combining it with a unified multi-constraint model, it performs multi-objective collaborative optimization on multiple candidate path sets. This achieves real-time perception-driven dynamic path planning and global collaborative diversion in multi-start and multi-endpoint scenarios, thus solving the problems of poor real-time performance, easy secondary congestion, and low engineering adaptability in existing technologies due to the disconnect between perception and decision-making and incomplete multi-constraint modeling.

[0007] The technical solution adopted in this invention is as follows: A multi-constraint crowd management path planning method based on congestion perception includes: A traffic network model for the nodes and passages in the target area is pre-constructed, and basic attributes are configured for the nodes and passages. The method also includes, Obtain congestion sensing data for each channel and / or node within the target area from sensing devices; For each diversion path in the set of multiple candidate diversion paths, based on the congestion perception data and combined with the diversion road network model, it is mapped to the operating status parameters of each channel and / or each node to obtain the dynamic path cost; the constraints involved in diversion path planning are modeled to establish a unified multi-constraint model. A multi-objective optimization function is constructed, and combined with the multi-constraint unified model and the dynamic path cost, multiple sets of candidate diversion paths are solved to determine a set of diversion paths.

[0008] The multi-constraint crowd management path planning method based on congestion perception adopted in this invention also has the following additional technical features: Multiple sets of candidate diversion paths are as follows: Based on the starting and ending points of the diversion path, and using the diversion road network model, a heuristic shortest path search is employed to generate multiple candidate paths. Based on multiple sets of starting points and ending points, multiple sets of candidate paths are generated. Any one of the candidate paths in each set is combined to obtain a set of multiple candidate diversion paths.

[0009] The dynamic path cost is obtained by mapping the runtime status parameters of each channel and / or each node, specifically: The operational status parameters include at least one of the following: basic travel time, path congestion weight, safety penalty coefficient, and node delay weight. The dynamic path cost is obtained based on the aforementioned operating status parameters; The basic travel time is obtained through the traffic diversion network model, and the path congestion weight, the safety penalty coefficient, and the node stagnation weight are obtained by combining congestion perception data with the traffic diversion network model.

[0010] Construct a multi-objective optimization function, combined with the aforementioned multi-constraint unified model, specifically as follows: The multi-constraint unified model includes hard constraints and / or soft constraints. The hard constraints include at least one of the following: capacity constraints, orientation constraints, and containment constraints. The soft constraint conditions include at least one of efficiency constraints and path overlap constraints. The path overlap constraints are used to constrain the overlap rate of multiple diversion paths in a set of candidate diversion paths. Based on the hard constraints, multiple sets of candidate diversion paths are screened to obtain a set of diversion paths that satisfy the hard constraints, which are then optimized using a multi-objective optimization function. Based on the soft constraints, a penalty term is obtained, which is then added to the multi-objective optimization function to optimize the set of dredging paths.

[0011] Constructing a multi-objective optimization function, combined with the aforementioned multi-constraint unified model, also includes: The multi-objective optimization function includes an efficiency term and a safety term. Adjust the weights of the multi-objective optimization function and the penalty term according to the scenario state, wherein the scenario state includes normal mode and emergency mode; The weight of the efficiency item corresponding to the normal mode is greater than that of the emergency mode, the weight of the penalty item corresponding to the normal mode is less than that of the emergency mode, and the weight of the safety item corresponding to the normal mode is less than that of the emergency mode.

[0012] A multi-objective optimization function is constructed, and the set of dredging paths is determined by combining the multi-constraint unified model and the dynamic path cost, specifically as follows: When the dynamic path cost of any diversion path exceeds the first threshold, based on the optimized set of multiple diversion paths, the priority of the corresponding diversion path is adjusted according to the dynamic path cost to determine a set of diversion paths. The priority adjustment of the diversion path is negatively correlated with the cost of the dynamic path.

[0013] The method further includes: When the rate of change of the dynamic path cost of any of the determined set of diversion paths exceeds a second threshold, and / or, Any one of the given set of diversion paths does not satisfy the hard constraints, and / or When the scenario state changes, the diversion path is replanned.

[0014] The replanning of the diversion routes will be implemented, specifically as follows: When the rate of change of the dynamic path cost of any of the determined set of diversion paths exceeds a second threshold, and / or, If any one of the given diversion paths does not meet the hard constraints, the other diversion paths in the set are retained, and local replanning is performed based on the corresponding diversion path. When the scenario state changes, all diversion paths are globally replanned.

[0015] A second aspect of the present invention provides a multi-constraint crowd management path planning system based on congestion perception, for executing the method, comprising: The road network modeling module is used to construct a traffic network model of nodes and passages in the target area, and to configure basic attributes for the nodes and passages. The sensing access module is used to acquire congestion sensing data of each channel and / or each node in the target area from the sensing device; The state mapping module is used to map each diversion path in the set of multiple candidate diversion paths to the operating state parameters of each channel and / or each node based on the congestion perception data and the diversion road network model, so as to obtain the dynamic path cost. The multi-constraint modeling module is used to model the constraints involved in the planning of each diversion path in the set of multiple candidate diversion paths, based on the congestion perception data and combined with the diversion road network model, and establish a unified multi-constraint model. The multi-objective optimization module is used to construct multi-objective optimization functions; The path planning module is used to solve multiple sets of candidate diversion paths based on the multi-objective optimization function, combined with the multi-constraint unified model and the dynamic path cost, and to determine a set of diversion paths. The results output module is used to output a set of determined diversion paths for guidance.

[0016] A third aspect of the present invention provides an electronic device, including a processor and a memory. The memory contains computer programs. The processor implements the method when executing the computer program.

[0017] Due to the adoption of the above technical solution, the beneficial effects achieved by this invention are as follows: 1. In this invention, congestion perception data of each channel and / or each node within a target area is acquired from sensing devices. Based on the congestion perception data and the traffic management network model, it is mapped to the operating status parameters of each channel and / or each node to obtain dynamic path costs. By directly converting the real-time congestion perception data collected by the sensing devices into dynamic cost parameters in path planning, the path costs can accurately reflect the current traffic status, and the planning results can be quickly updated according to changes in pedestrian flow, significantly improving the timeliness of traffic management planning.

[0018] Furthermore, the constraints involved in the diversion route planning are modeled, a multi-constraint unified model is established, and a multi-objective optimization function is constructed. Combining the multi-constraint unified model and the dynamic path cost, multiple sets of candidate diversion routes are solved to determine a set of diversion routes. By constructing a collaborative solution framework, constraints such as path overlap are incorporated into the unified model, making the planning results more balanced at the global level. This effectively reduces the possibility of congestion exacerbated by path overlap among multiple pedestrian flows, and reduces the risk of secondary congestion and path conflicts.

[0019] Furthermore, by pre-constructing a traffic network model of nodes and passages in the target area and configuring basic attributes for these nodes and passages, and then driving the path planning process based on real-time congestion perception data, the methodology is highly modular and adaptable to various scenarios. It supports the access of multiple sensing devices and the construction of different types of site models, demonstrating strong deployment capabilities in various high-density public places. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the multi-constraint crowd diversion path planning method based on congestion perception according to one embodiment of the present invention. Figure 2 This is a flowchart illustrating the multi-constraint crowd diversion path planning method based on congestion perception, according to one embodiment of the present invention. Figure 3 This is a schematic diagram of the traffic diversion network model according to one embodiment of the present invention, wherein blue dots represent nodes and green lines represent traffic diversion paths. Detailed Implementation

[0021] To more clearly illustrate the overall concept of the present invention, a detailed description will be provided below with reference to the accompanying drawings and examples.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0023] like Figures 1 to 3 As shown, a multi-constraint crowd management path planning method based on congestion perception includes: S1: Pre-construct a traffic network model for nodes and channels in the target area, and configure basic attributes for the nodes and channels; The main purpose of this step is to establish a standardized data structure foundation for the target area, providing a unified digital basis for subsequent path search and multi-constraint solution. This model must simultaneously meet two requirements: first, it must be able to fully represent the spatial topology of the site (i.e., which nodes are connected by which channels); second, it must be able to store various static physical constraint parameters (such as channel width, rated capacity, safety level, etc.), providing a data foundation for subsequent dynamic cost calculation and constraint checking.

[0024] Based on the spatial structure map of the target area (CAD drawings, GIS data, or manually marked floor plan), the following locations are automatically or semi-automatically identified as nodes, such as... Figure 3 As shown: All entrances and exits (including regular exits and emergency exits); Intersection / fork in the road; Staircase entrances and exits, both ends of escalators and elevators, and elevator door locations; Functional area boundary points (such as grandstand exits, platform edges, and shop entrances / exits); Evacuation assembly point / safe zone entrance.

[0025] For any two reachable nodes, if there exists a physical path unit (including corridors, stairs, escalators, ramps, doorways, etc.) that people can pass through, the system establishes a directed or bidirectional passage. The basic attribute data structure of each passage is defined as follows: A unique identifier used for indexing and retrieval; The starting node ID and ending node ID of the channel; The passage length (in meters) is directly measured from the spatial structure data; The width of the passage (in meters) determines the physical throughput capacity of the passage. Rated throughput capacity (persons / second·meter or person / minute) can be configured according to the width of the passageway, the type of passageway (stairs / escalator / straight passageway) and industry standards; The traffic direction attribute can take values ​​including two-way traffic, one-way traffic (forward), and one-way traffic (reverse). One-way traffic only allows passage in the preset direction and prohibits reverse traffic. Safety levels can be set from 1 to 5 (or higher granularity), with higher numbers indicating greater safety. Safety levels can be pre-assessed based on static factors such as the availability of fire-fighting equipment, structural stability, lighting conditions, and historical accident records. The basic passage time (in seconds) is calculated using the following formula: =length / ,in This is a normal walking speed (usually 1.2-1.5 meters per second). A Boolean flag indicating whether special groups are allowed to pass through. A channel set to TRUE indicates that its width meets the requirements for wheelchair access, has a gentle slope, and has an accessible threshold, making it suitable for use by special groups such as the elderly, children, the injured, and people with disabilities. The channel's open status can be either open, restricted, or blocked, corresponding to three scenarios: normal passage, partially restricted passage, and completely prohibited passage, respectively.

[0026] The above node set N and channel set E are constructed into a graph structure G=(N,E), which is stored in memory or a database as an adjacency list or adjacency matrix.

[0027] It should be noted that the road network model is not limited to a node-channel topology. It can also be implemented using two-dimensional planar grid diagrams, three-dimensional spatial topology diagrams, regional unit diagrams, or digital twin scene diagrams. There are no restrictions on this, as long as it can express the passability relationship, capacity attributes, and directional attributes of personnel.

[0028] Existing static road network models often only record two types of parameters: distance and connectivity. They lack the ability to pre-store parameters such as passage safety levels, accessibility for special groups, and real-time closure status. This results in many management rules that must be followed in real-world scenarios being unable to be recognized and adhered to by the algorithm during subsequent route planning, leading to the generation of invalid paths that violate constraints.

[0029] This step pre-configures multi-dimensional basic attributes for channels, which are then integrated with subsequent constraint construction and multi-objective optimization. On one hand, the path search algorithm can directly call these attributes for constraint verification during execution, without needing to recalculate or rely on external rule tables each time. On the other hand, when the channel state changes, the system only needs to update the attribute fields of the corresponding channel, without needing to reconstruct the entire road network model, thereby improving the system's dynamic response capability.

[0030] Those skilled in the art will understand that the specific types and quantities of the aforementioned basic attributes can be added, deleted, or adjusted according to the actual management needs of the target area, and there are no restrictions on this.

[0031] The method also includes, S2: Obtain congestion sensing data for each channel and / or node within the target area from the sensing device.

[0032] The main purpose of this step is to collect pedestrian flow data at various locations within the target area, providing real, accurate, and real-time input variables for subsequent dynamic path cost mapping and multi-constraint solution.

[0033] Connect one or more of the following sensing devices to continuously collect pedestrian flow information at various nodes, channels, and local spaces within the target area: The video surveillance equipment uses the Single Frame Detector (SSD) algorithm or the YOLO series algorithms for real-time pedestrian detection, and combines the Deep-SORT algorithm for multi-target tracking to extract the real-time number, instantaneous speed and direction of movement of pedestrians from the building surveillance video stream; Millimeter-wave radar / liDAR, using FMCW radar or LiDAR equipment, can detect the number of people and their movement speed in an area in all weather conditions and without contact. It does not depend on lighting conditions and is suitable for emergency evacuation scenarios at night or in severe weather. Turnstile counting equipment is deployed at main entrances or passage nodes to count the number of people entering and exiting in one direction and to accumulate the number of people remaining in the passage in real time. Infrared / thermal imaging equipment is suitable for nighttime or low-visibility environments, detecting the presence and degree of gathering of people through differences in thermal radiation; Bluetooth / WiFi probes and UWB positioning devices can estimate the number of people in an area by detecting the MAC address or beacon signal of mobile terminal devices, and can also obtain detailed information such as the dwell time and movement trajectory of people; Electronic fence and mobile phone signaling data can be used to define electronic fence areas and count the number of devices entering and leaving the area, thus helping to estimate the number of people and the flow rate in the area.

[0034] It should be noted that the aforementioned sensing devices can be combined or integrated in any way. The system can perform fusion processing on multi-source sensing data based on Kalman filtering or weighted averaging algorithms to improve the accuracy and robustness of the data.

[0035] Each sensing record obtained from the sensing device contains at least the following fields: Timestamps (with millisecond precision) are used to identify the timeliness of data; Location type, which can take the values ​​node (node) or edge (channel), represents the entity type corresponding to the data; The corresponding node ID or channel ID is used to associate the sensed data with specific entities in the road network model; The real-time crowd density (people / square meter) of the passageway is calculated by dividing the number of people in the area collected by the sensing device by the area of ​​the area. The real-time actual number of people can be directly obtained or estimated based on the counting results of the sensing devices, which is particularly suitable for dynamic filling calculations with capacity constraints. Channel occupancy rate (a real number between 0 and 1) is the ratio of the actual number of people in the current channel to the maximum capacity of the channel. Inflow rate and outflow rate (people / second) reflect the intensity of people inflow / outflow per unit time in the current channel or node; Average passage speed (m / s) reflects the actual movement speed of people in the passage and can be calculated through video tracking or trajectory estimation. Congestion levels (e.g., 0-5) can be given directly by the sensing system or determined by a combination of real-time crowd density and average traffic speed in the passageway; level 0 indicates unobstructed traffic, and level 5 indicates severe congestion. An exception event flag (0 or 1) indicates whether an exception event has occurred at this location; Anomaly types (such as fire, stampede warning, equipment failure, area lockdown, etc.) are used to trigger the corresponding emergency response logic.

[0036] For multi-source sensing data at the same location and time, the system employs a weighted average method or Kalman filtering for data fusion to ensure the accuracy and stability of the input data. If a channel or node is not directly covered by sensing devices, the system estimates the number of people at that location by using the inflow and outflow rates of adjacent nodes. Specifically, for channels without direct sensing devices, the system can estimate the approximate number of people in the channel at the current time by subtracting the accumulated outflow from the accumulated inflow at the upstream nodes and combining this with the passage time delay parameter.

[0037] In existing technologies, data acquisition in sensing systems is often independent of the path planning module. Sensing data is mostly used for offline statistics or post-event analysis and fails to participate in real-time path decision-making. This step, through a standardized data access and fusion mechanism, enables the real-time data collected by sensing devices to directly participate as decision variables in subsequent dynamic path cost mapping and multi-objective solution processes. This achieves coupling between the sensing system and the planning system, solving the bottleneck of the disconnect between sensing and decision-making.

[0038] Taking the exit scenario of a large sports stadium as an example, as the event nears its end, the crowd density in the passageways around the stands may rapidly increase from a normal 0.5 people / square meter to 3 people / square meter. This step can be updated every second or every few seconds, allowing the system to immediately perceive this trend and provide timely and accurate input data for subsequent dynamic path adjustments.

[0039] S3: For each diversion path in the set of multiple candidate diversion paths, based on the congestion perception data and combined with the diversion road network model, map it to the operating status parameters of each channel and / or each node to obtain the dynamic path cost; model the constraints involved in diversion path planning and establish a unified multi-constraint model.

[0040] The main purpose of this step is twofold: first, to transform real-time congestion perception data into dynamic cost parameters that can participate in path search and optimization through mathematical mapping, so that the planning results can truly reflect the current traffic status; second, to systematically model and hierarchically manage various constraints that must be followed and can be flexibly adjusted during the path planning process, so that complex and diverse actual management constraints can be handled uniformly and efficiently during candidate path selection and multi-objective optimization.

[0041] The system maps the real-time number of people and flow status collected by the perception layer into dynamic cost parameters in path planning, specifically including the calculation of the following parameters: The channel congestion weight ω(e,t) reflects the severity of congestion in channel e at time t. The channel congestion weight is determined by combining one or more of the following sensing data: real-time crowd density ρ(e,t), channel occupancy rate occ(e,t), and congestion level. Average traffic speed (e,t) and normal speed The ratio, etc.

[0042] In one embodiment, the channel congestion weight can be calculated according to the following formula, but the present invention is not limited thereto: ω(e,t)=α·(ρ(e,t) / ρ max )+β·occ(e,t)+γ·(1-v curr (e,t) / v normal ), Where, ρ max The maximum safe density threshold allowed for the channel is defined by α, β, and γ, which are preset weighting coefficients. The sum of these three coefficients can be 1. The higher the quality of the congestion perception data, the larger the coefficients can be. If the perception system directly provides the congestion level, the system can also use a preset mapping table to map the congestion level to the corresponding congestion weight (e.g., level 0 → 0, level 1 → 0.2, level 2 → 0.4, level 3 → 0.6, level 4 → 0.8, level 5 → 1.0). This method is simpler and faster, and is suitable for edge computing scenarios with limited computing resources.

[0043] The safety penalty coefficient ψ(e,t) is used to reflect the safety correction factor of channel e at the current moment. This coefficient includes both static attribute components (safety level of channel infrastructure) and dynamic components (whether there are abnormal events).

[0044] If there are anomalies in the area covered by the passage flag =1), the system can set ψ(e,t) to a positive value much greater than 1 (such as ψ risk =10.0), which causes a sharp increase in the dynamic cost of the channel; if there are no abnormal events, the calculation formula can be: ψ(e,t)=ψ static ×ψ dynamic , Where ψ static For the safety correction base (e.g., 1 / safety level), ψ dynamic Further adjustments can be made based on whether any abnormal information is generated.

[0045] By combining the above parameters, namely channel congestion weight, safety penalty coefficient, and basic travel time, the final edge weight used for path search, i.e., dynamic path cost, is calculated.

[0046] Furthermore, this invention consolidates the management constraints of traditionally decentralized processing into a unified modeling framework. Specifically, it includes: Capacity constraints limit the number of people passing through a channel or node per unit of time to its rated capacity. Specifically, for any channel e, its currently occupied capacity (calculated based on real-time sensing data) plus the planned allocated flow must not exceed the capacity constraint. Capacity constraints are essentially dynamic constraints; their currently occupied capacity is dynamically determined by the actual number of people and flow status provided in real-time by the sensing equipment.

[0047] Directional constraints mean that path units with directional restrictions, such as one-way channels, escalator directions, and dedicated diversion lanes, cannot be used for reverse traffic planning. During path search, if the current search direction does not match the preset direction of the channel, the path is not feasible.

[0048] The lockdown constraint means that for passages marked by the sensing system as temporarily closed, dangerous area accidents, or management restrictions, they are directly set to an impassable state when the path is deployed.

[0049] In addition, efficiency constraints are included, which set optimization targets for indicators such as overall traffic diversion time, average travel time, and average congestion risk. Instead of requiring the indicators to be lower than the absolute threshold, the goal is to minimize the value of the objective function.

[0050] Path overlap constraints are used to limit the degree of overlap among multiple diversion paths in a set of candidate diversion paths. Path overlap can be represented as how many planned paths share the same passage. The higher the path overlap rate, the more diversion paths pass through the same passage or node, and the higher the risk that passage or node will become a congestion bottleneck. Path overlap constraints are optimized by incorporating the overlap rate as a penalty term into a multi-objective optimization function, rather than rigidly limiting it to a fixed threshold, thus achieving flexible adjustment.

[0051] In addition, it should be noted that it may also include: Balanced traffic distribution constraints set optimization targets for the traffic distribution balance of channels to avoid situations where some channels bear too much traffic while adjacent channels are idle.

[0052] Special group constraints are implemented for special groups such as the elderly, children, the injured, and people with mobility impairments. When planning pathways, priority is given to those with sufficient width and better accessibility. This constraint can be designed as a bonus or penalty factor and incorporated into the objective function value of each pathway. For example, a negative reward (i.e., reduced cost) can be given to pathways that support special groups as an incentive to prioritize them.

[0053] Understandably, depending on the specific management needs of the target area, the multi-constraint unified model can incorporate one or more of the above constraints into a unified framework for processing.

[0054] Traditional route planning methods often rely on manually preset constraints or offline statistical data, which fail to reflect the actual number of people, density, and traffic load at a given location at any given moment. This step maps the multi-dimensional data collected by sensing devices in real time into dynamic parameters such as congestion weight (ω), stagnation weight (η), and safety penalty coefficient (ψ). This allows the route cost to adjust according to changes in the actual number of people, occupancy rate, and traffic speed, preventing deviations between routes planned using static historical values ​​and the actual congestion situation.

[0055] S4: Construct a multi-objective optimization function, combine the multi-constraint unified model and the dynamic path cost, solve multiple sets of candidate diversion paths, and determine a set of diversion paths.

[0056] The main purpose of this step is to select the globally optimal set of path solutions from all candidate path combinations in complex concurrent traffic management scenarios with multiple starting points and multiple ending points. This step must consider both the individual path efficiency of each starting point and ensure that no new congestion points are caused by multiple paths occupying the same channel at the global level, thereby effectively solving the serious problem of artificially created additional congestion when solving multiple paths independently in existing technologies.

[0057] Before formally executing multi-objective optimization, the system first generates multiple sets of candidate diversion paths. A multi-objective optimization function is then constructed. Taking into account the following objectives: Time cost objective item This reflects the total travel time of the entire group of solutions. Specifically, for each path within the group, the dynamic path cost Cost(e,t) of each channel it contains is accumulated. Congestion risk target item This reflects the degree of congestion concentration on the entire set of routes. It can be designed as a weighted sum of the congestion weights of routes used by multiple paths simultaneously; the more routes use a route, the greater its contribution. Security Targets It reflects the security level of the entire group of schemes and can be calculated as the security weighted sum of all paths in the group. The security level of each path is determined by the security level of each channel it contains. Path overlap penalty target item Penalize the degree of path overlap on the channel, such as =The sum of the squares of the number of paths that pass through the entire set of solutions.

[0058] The combined form of a multi-objective optimization function can be expressed as: , Where λ1, λ2, λ3, and λ4 are the weight coefficients of each objective item.

[0059] The system, having completed multiple sets of candidate diversion paths for the constrained model, normalizes the objective values ​​corresponding to each combination scheme and then inputs them into the multi-objective optimization function Obj to calculate the overall objective function value. The set of paths with the smallest (or largest, depending on the defined direction of the objective terms) objective function value is selected as the final output set of diversion paths.

[0060] This step transforms multi-dimensional decision-making factors such as time cost, congestion risk, safety level, and path overlap rate into a single measurable optimization objective through a multi-objective optimization function. This enables the path planning process to comprehensively and balancedly consider factors such as efficiency and safety, flexibly respond to the actual needs of different scenarios, and significantly enhance the adaptability of the solution to different times and events.

[0061] In a preferred embodiment of the present invention, the set of multiple candidate diversion paths is specifically as follows: Based on the starting and ending points of the diversion path, and using the diversion road network model, a heuristic shortest path search is employed to generate multiple candidate paths. Based on multiple sets of starting points and ending points, multiple sets of candidate paths are generated. Any one of the candidate paths in each set is combined to obtain a set of multiple candidate diversion paths.

[0062] In complex traffic management scenarios with multiple start and end points, there are a massive number of possible path combinations between each pair of start and end points. If all possible combinations are evaluated one by one, the computational scale will explode exponentially with the number of start and end points, making it difficult to meet real-time requirements. For example, for a large venue with 8 start points and 12 end points, if approximately 10 candidate paths are generated for each pair of start and end points, the total number of candidate paths would be approximately 10^(8×12).

[0063] Therefore, this implementation uses a pre-defined heuristic strategy to select and construct a candidate solution set of manageable size from a massive number of possibilities, serving as the basis for subsequent multi-objective optimization solutions. This candidate set should simultaneously meet two requirements: first, it should cover a sufficient number of high-quality alternatives to ensure that potential optimal solutions are not filtered out; second, it should be of moderate size to ensure the computational efficiency of subsequent hard constraint screening and multi-objective optimization.

[0064] For each single origin-endpoint path pair, heuristic shortest path search is used to independently generate K candidate paths that meet basic feasibility requirements. One path is selected from the candidate path set corresponding to each origin and combined to form a set of multiple candidate diversion schemes.

[0065] In different embodiments of the present invention, the core search algorithm of the candidate path generation layer may be any one or more of the Yen algorithm (K-shortest path algorithm), the improved A algorithm, the Dijkstra algorithm, and its variants, without limitation.

[0066] In the candidate path set P(S) for each origin-destination pair i , T j After generation, multiple sets of candidate diversion paths are obtained by combining them. Let there be m starting points S1, S2, ..., S... m (Each starting point corresponds to an evacuation source area), with n endpoints T1, T2, ..., T n (Each endpoint corresponds to an exit or safe zone). For each starting point S i The system in the corresponding candidate path set P(S) i Choose a path from P(S, T). i There are n×K candidate paths for each of the m starting points. Selecting from each candidate path yields a path combination scheme.

[0067] Each scheme can be expressed in the following form: {p i1 , p i2 , …, p im}, where p ij Let represent a candidate path starting from the i-th starting point. The set of all possible assignment schemes constitutes the solution space.

[0068] It should be noted that, in order to control the overall computational load without overlooking high-quality candidate solutions, the system can control the number of candidate paths K for each origin-endpoint pair. The value of K is usually between 3 and 8, limiting the number of candidate paths for each origin-endpoint pair. If the value of K is too large, the number of combinations will inflate; if the value of K is too small, key alternative paths may be missed.

[0069] In a preferred embodiment of the present invention, the dynamic path cost is obtained by mapping the operating status parameters of each channel and / or each node, specifically as follows: The operational status parameters include at least one of the following: basic travel time, path congestion weight, safety penalty coefficient, and node delay weight. The dynamic path cost is obtained based on the aforementioned operating status parameters; The basic travel time is obtained through the traffic diversion network model, and the path congestion weight, the safety penalty coefficient, and the node stagnation weight are obtained by combining congestion perception data with the traffic diversion network model.

[0070] The purpose of this implementation is to map the accessed raw congestion perception data into dynamic path cost parameters that can be directly used by the path search algorithm.

[0071] The dynamic path cost mapping mechanism explicitly incorporates both congestion and safety dimensions into the evaluation system: These three parameters correspond to the basic traffic efficiency dimension ( ), congestion level dimension ( ) and security dimensions ( ).

[0072] To further improve the accuracy of global path planning, in a preferred embodiment, node stagnation weights can be added as an additional penalty term to the dynamic path cost calculation. Specifically: , in For nodes At any moment The real-time number of people gathered. For nodes The preset maximum safe capacity. The value of is between 0 and 1. If the node If the congestion is severe, approaching or exceeding the capacity limit, then It is close to or even equal to 1.

[0073] The node dwell weight is incorporated into the dynamic path cost, specifically as follows: , in It is a passage The endpoint This is a preset node penalty coefficient, typically ranging from 0.1 to 0.5, representing the proportion of the impact of node congestion risk on the total cost relative to the channel congestion risk. By adding a node stagnation weight penalty item, the system can effectively avoid severely congested nodes near exits or intersections, guiding crowds to entrances with stronger capacity in surrounding nodes, thus achieving a more balanced global distribution of pedestrian flow load.

[0074] The system operates at fixed time intervals. (e.g., every 1 or 5 seconds) periodically retrieve the latest congestion sensing data, recalculate the dynamic path cost of all channels and nodes, and ensure that the computational load is stable and controllable.

[0075] As a preferred embodiment of the present invention, a multi-objective optimization function is constructed, which, in conjunction with the aforementioned multi-constraint unified model, specifically includes: The multi-constraint unified model includes hard constraints and / or soft constraints. The hard constraints include at least one of the following: capacity constraints, orientation constraints, and containment constraints. The soft constraint conditions include at least one of efficiency constraints and path overlap constraints. The path overlap constraints are used to constrain the overlap rate of multiple diversion paths in a set of candidate diversion paths. Based on the hard constraints, multiple sets of candidate diversion paths are screened to obtain a set of diversion paths that satisfy the hard constraints, which are then optimized using a multi-objective optimization function. Based on the soft constraints, a penalty term is obtained, which is then added to the multi-objective optimization function to optimize the set of dredging paths.

[0076] The core objective of this implementation method is to incorporate various practical constraints involved in the planning of traffic diversion routes into a unified modeling framework. By adopting a hierarchical processing strategy of prior screening of hard constraints and optimization of soft constraints, the feasibility of candidate traffic diversion route sets is screened and the objectives are optimized. Finally, a set of traffic diversion routes that meets both the hard bottom line requirements of safety management and is optimal in terms of soft indicators such as efficiency and coordination is selected.

[0077] This invention unifies the constraints involved in dredging path planning into two levels: hard constraints and soft constraints. Hard constraints are conditions that a path solution must absolutely satisfy; any path that violates a hard constraint is considered infeasible and should be directly excluded from the candidate set. Soft constraints, on the other hand, represent ideal but not mandatory requirements, reflected in the multi-objective optimization function in the form of penalty terms, to guide the solution towards a better direction rather than directly excluding solutions.

[0078] This hierarchical approach organically unifies constraints and optimization objectives. Hard constraints screen candidate solutions for validity, ensuring the safety of the final solution. Soft constraints drive the solution to move further towards optimality within the valid space, achieving a balance between efficiency, safety, and collaboration. This two-tiered mechanism considers both the safety and feasibility of the path solution and its engineering advantages.

[0079] Hard constraints include, but are not limited to, the following three types: capacity constraints, direction constraints, and containment constraints. Iterate through each set of multiple candidate diversion paths; for each path p, sequentially check whether each of its constituent channels e violates any of the above three types of hard constraints. If any path violates any hard constraint, that set is directly eliminated from the candidate set; all sets that pass the hard constraint screening continue to the next stage of soft constraint penalty and multi-objective optimization.

[0080] Candidate solutions selected through hard constraints are essentially feasible, but they still require further optimization through soft constraint penalty terms. Soft constraints include at least efficiency constraints and path overlap constraints, or collaborative constraints.

[0081] After hard constraint screening and soft constraint penalty term construction, a multi-objective optimization function is constructed. The construction of the multi-objective optimization function, combined with the aforementioned multi-constraint unified model, also includes: The multi-objective optimization function includes an efficiency term and a safety term. Adjust the weights of the multi-objective optimization function and the penalty term according to the scenario state, wherein the scenario state includes normal mode and emergency mode; The weight of the efficiency item corresponding to the normal mode is greater than that of the emergency mode, the weight of the penalty item corresponding to the normal mode is less than that of the emergency mode, and the weight of the safety item corresponding to the normal mode is less than that of the emergency mode.

[0082] The purpose of this embodiment is to construct a dual-mode switching mechanism for normal and emergency scenarios, and to dynamically adjust the weight coefficients and penalty term priorities of the multi-objective optimization function according to the scenario status, so as to flexibly adapt to the changing needs of different operational scenarios and meet the requirements of normal and emergency path planning.

[0083] The normalized crowd control mode is applicable to general scenarios such as venue dispersal, holiday crowd management, post-event crowd diversion, and daily high-volume operations in commercial complexes. Under the normalized mode, the overall threat within the target area is controllable. While ensuring basic safety, it maximizes traffic efficiency and achieves balanced traffic flow through designated channels to maintain order on site.

[0084] For multi-objective optimization functions, the default weighting strategy in normal mode is: efficiency term weighting. A higher value makes the algorithm prioritize channels with shorter travel times and lower dynamic path costs. Congestion risk item weight Choose appropriate values ​​and moderately avoid congested areas to maintain traffic efficiency and balanced flow distribution; Security item weight The value is relatively low, and the security level of the channel exists as a reference indicator, but its impact on path selection is relatively limited; Penalty weight The value should be set from moderate to high, and increased appropriately according to management needs to avoid too many paths occupying the same channel.

[0085] A typical implementation example is the collaborative crowd control experiment conducted in a large sports stadium dispersal scenario, involving 8 starting areas in the stands and 12 ending areas in the exits. With the total weight of 1, efficiency-oriented goals dominate under this weight configuration, while the penalty term also has a significant inhibitory effect on the main channel.

[0086] Emergency evacuation mode triggers an alarm when the sensing system detects an anomaly. flag =1) or automatically triggered when the population density rapidly exceeds the safety threshold in a short period of time. In emergency mode, personnel in the danger zone are transferred to a safe area in the shortest possible time while ensuring their safety.

[0087] Corresponding to the multi-objective optimization function, the weight adjustment strategy for emergency mode is as follows: Efficiency Term Weight While efficiency goals remain important, they have taken a backseat in terms of overall priority. Congestion risk item weight Maintaining or moderately improving accessibility to avoid secondary risks of stampedes caused by congestion is still being considered, but its importance is slightly lower than that of safe accessibility. Security item weight Significantly improved, becoming the dominant objective, the priority of high-security passages was significantly amplified, and routes passing through dangerous areas were severely penalized; Penalty weight The limit can be appropriately increased or maintained, and may be further increased depending on the management's preferences if further mandatory diversion is needed to prevent all people from being directed to the same exit.

[0088] If the system detects an alarm from a smoke sensor in the eastern circular passage area, the weighting combination will be adjusted to... ( It dropped to 0.15. (Rising to 0.40). At this point, the path cost to the eastern exits increases sharply due to passing through alarm areas, and the multi-objective optimization solver will prioritize alternative paths leading to the western and southern safe exits. The weight adjustment mechanism in this embodiment reduces the number of people stranded in high-risk areas, although the global evacuation time partially increases, trading key safety benefits for an acceptable efficiency cost.

[0089] Furthermore, a multi-objective optimization function is constructed, and the set of dredging paths is determined by combining the multi-constraint unified model and the dynamic path cost, specifically as follows: When the dynamic path cost of any diversion path exceeds the first threshold, based on the optimized set of multiple diversion paths, the priority of the corresponding diversion path is adjusted according to the dynamic path cost to determine a set of diversion paths. The priority adjustment of the diversion path is negatively correlated with the cost of the dynamic path.

[0090] The core objective of this embodiment is to establish a dynamic path cost threshold monitoring mechanism to adaptively adjust path priorities while maintaining the overall scheme framework formed by multi-objective optimization, thereby avoiding artificial secondary congestion during the implementation of the diversion scheme.

[0091] First threshold This represents the maximum dynamic path cost that the system can tolerate. Paths exceeding this threshold indicate that their current actual transit cost has exceeded the system's tolerance range.

[0092] about The specific value is determined by setting a fixed upper limit for the safety cost for each type of passage based on management experience. For example, when the main passage length L is approximately 100 meters and the basic passage time... In a normal operating scenario of seconds, when When the growth exceeds 2.0 times the normal state of this path, it is considered... The base value of .

[0093] Of course, real-time congestion sensing data can also be used to dynamically adjust during operation. The system can calculate the average congestion level of all channels by setting a sliding time window, based on the fluctuations in the current global congestion level. and standard deviation Then set .

[0094] The system operates at a fixed time step. Periodically scan the dynamic costs of all candidate diversion paths to determine if they exceed a threshold. For each diversion path... Assigned a priority scalar value The higher the value, the higher the priority of the path service. That is, the earlier the path is in the planning order of the current plan, and the more likely it is to be assigned to the evacuation population that has just entered the area.

[0095] Priority and dynamic path cost are set to have a negative correlation. In other words, priority is determined based on the dynamic path cost of all candidate paths. The paths are sorted from lowest to highest priority, with the path having the lowest dynamic path cost being assigned the highest priority. And so on.

[0096] The entire priority list is continuously updated during system operation. Whenever a new batch of congestion sensing data arrives, the priority list for all paths is recalculated. Sort list.

[0097] In a preferred embodiment of the present invention, the method further includes: When the rate of change of the dynamic path cost of any of the determined set of diversion paths exceeds a second threshold, and / or, Any one of the given set of diversion paths does not satisfy the hard constraints, and / or When the scenario state changes, the diversion path is replanned.

[0098] The core objective of this implementation method is to establish a set of replanning trigger judgment matrices containing multiple complementary rules, enabling the system to intelligently identify the failure level of the current diversion plan and automatically execute the corresponding replanning operation.

[0099] Example 1: Dynamic path cost change rate triggers replanning, which measures the rate of change in the cost of the diversion path based on the relative cost change per unit time. Second threshold. This directly controls the upper limit of the acceptable relative range of dynamic path costs per unit of time. It should be noted that a rapid increase in dynamic path costs indicates a deterioration in the current diversion path, necessitating replanning and adjustment.

[0100] Example 2: In the hard constraint screening of the aforementioned multi-constraint unified model, the system has eliminated schemes that violate the hard constraints from the candidate diversion path set. However, during the implementation of the diversion scheme, the state of hard constraint channels that previously met the conditions may change in real time.

[0101] Violation of hard constraints during execution includes capacity constraint failure, meaning that a channel that was originally balanced in terms of inflow and outflow and had sufficient capacity may experience a real-time occupancy rate exceeding the rated capacity limit after the issuance of the diversion order due to overload caused by multiple groups of people. Sensor monitoring and judgment. This is a hard or immediate trigger for the judgment.

[0102] Directional constraints fail when management temporarily changes a two-way passage to one-way traffic based on site conditions. Since the sensing system receives this rule change, the original intended direction of travel no longer matches the new management rule, constituting a hard constraint violation.

[0103] The lockdown measures failed, meaning that due to the sudden spread of dense smoke or other reasons, fire management personnel were forced to temporarily physically block a certain passage included in the original plan, and the path immediately became invalid.

[0104] The system monitors the satisfaction status of hard constraints in real time in two ways: on the one hand, it periodically receives the updated values ​​of the channel status field transmitted back by the sensing device; on the other hand, when an emergency occurs, the system directly processes it with the highest priority and immediately starts the replanning process.

[0105] Example 3: Supports a switching mechanism between normal evacuation mode and emergency evacuation mode.

[0106] The switching of scene status can be divided into two main types: switching from normal mode to emergency mode. This is triggered by conditions such as abnormal event alarm information or the continuous exceedment of the safety warning threshold by the monitored crowd density. At this time, the system cancels the normal rules and switches to emergency mode.

[0107] When the abnormal event has been resolved and the population density has fallen below the preset safety level, the system switches back to normal mode.

[0108] After the mode switch, the weight matrix of the multi-objective optimization function is reloaded and the constraint priority is reconstructed to perform replanning.

[0109] It should be noted here that the replanning of the diversion routes specifically involves: When the rate of change of the dynamic path cost of any of the determined set of diversion paths exceeds a second threshold, and / or, If any one of the given diversion paths does not meet the hard constraints, the other diversion paths in the set are retained, and local replanning is performed based on the corresponding diversion path. When the scenario state changes, all diversion paths are globally replanned.

[0110] When a path is eliminated due to the dynamic path cost exceeding a second threshold or hard constraint failure, the system can adopt a local replanning strategy. The remaining unaffected paths in the candidate diversion path set are retained, and path recalculation is performed only on one or more affected starting areas. This generates new alternative paths to form a new candidate diversion path set, which is then resubmitted into the multi-objective optimization screening framework to obtain the updated overall diversion path set. This approach offers faster computation speed, minimizes disturbance to already operating paths, adapts to scenarios requiring localized acceleration of pedestrian flow changes, and significantly optimizes computational overhead and real-time response.

[0111] When the scenario state changes, multiple paths fail simultaneously during path planning, hard constraints are violated on a large scale, or local replanning fails to provide a feasible solution after a limited number of attempts, the current set of diversion paths is discarded. Based on the current dynamic road network weights and constraints, the complete candidate path generation and multi-objective solution process is re-executed for multiple starting and ending point regions.

[0112] A second aspect of the present invention provides a multi-constraint crowd management path planning system based on congestion perception, for executing the method, comprising: The road network modeling module is used to construct a traffic network model of nodes and passages in the target area, and to configure basic attributes for the nodes and passages. The sensing access module is used to acquire congestion sensing data of each channel and / or each node in the target area from the sensing device; The state mapping module is used to map each diversion path in the set of multiple candidate diversion paths to the operating state parameters of each channel and / or each node based on the congestion perception data and the diversion road network model, so as to obtain the dynamic path cost. The multi-constraint modeling module is used to model the constraints involved in the planning of each diversion path in the set of multiple candidate diversion paths, based on the congestion perception data and combined with the diversion road network model, and establish a unified multi-constraint model. The multi-objective optimization module is used to construct multi-objective optimization functions; The path planning module is used to solve multiple sets of candidate diversion paths based on the multi-objective optimization function, combined with the multi-constraint unified model and the dynamic path cost, and to determine a set of diversion paths. The results output module is used to output a set of determined diversion paths for guidance.

[0113] Understandably, the output module outputs the following results: the diversion routes corresponding to each starting point, the estimated travel time, the current congestion risk level, diversion suggestions, multi-terminal guidance instructions, and auxiliary decision-making information.

[0114] Furthermore, the relevant results can be further linked with digital twin platforms, electronic screens, broadcasting systems, mobile terminals, and emergency command systems to form a closed-loop management and control system guided by multiple terminals.

[0115] It can achieve any effect in the multi-constraint crowd diversion path planning method based on congestion perception, which will not be elaborated here.

[0116] A third aspect of the present invention provides an electronic device, including a processor and a memory. The memory contains computer programs. The processor implements the method when executing the computer program.

[0117] Therefore, it can achieve any effect in the multi-constraint crowd diversion path planning method based on congestion perception, which will not be elaborated here.

[0118] For any parts not mentioned in this invention, existing technologies can be used or referenced.

[0119] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0120] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A congestion-aware multi-constraint crowd evacuation path planning method, characterized in that, include: A traffic network model for the nodes and passages in the target area is pre-constructed, and basic attributes are configured for the nodes and passages. The method also includes, Obtain congestion sensing data for each channel and / or node within the target area from sensing devices; For each diversion path in the set of multiple candidate diversion paths, based on the congestion perception data and combined with the diversion road network model, it is mapped to the operating status parameters of each channel and / or each node to obtain the dynamic path cost; the constraints involved in diversion path planning are modeled to establish a unified multi-constraint model. A multi-objective optimization function is constructed, and combined with the multi-constraint unified model and the dynamic path cost, the solution is obtained for multiple sets of candidate diversion paths to determine a set of diversion paths.

2. The method of claim 1, wherein, Multiple sets of candidate diversion paths are as follows: Based on the starting and ending points of the diversion path, and using the diversion road network model, a heuristic shortest path search is employed to generate multiple candidate paths. Based on multiple sets of starting points and ending points, multiple sets of candidate paths are generated. Any one of the candidate paths in each set is combined to obtain a set of multiple candidate diversion paths.

3. The method of claim 1, wherein, The dynamic path cost is obtained by mapping the runtime status parameters of each channel and / or each node, specifically: The operational status parameters include at least one of the following: basic travel time, path congestion weight, safety penalty coefficient, and node delay weight. The dynamic path cost is obtained based on the aforementioned operating status parameters; The basic travel time is obtained through the traffic diversion network model, and the path congestion weight, the safety penalty coefficient, and the node stagnation weight are obtained by combining congestion perception data with the traffic diversion network model.

4. The method of claim 1, wherein, Construct a multi-objective optimization function, combined with the aforementioned multi-constraint unified model, specifically as follows: The multi-constraint unified model includes hard constraints and / or soft constraints. The hard constraints include at least one of the following: capacity constraints, orientation constraints, and containment constraints. The soft constraint conditions include at least one of efficiency constraints and path overlap constraints. The path overlap constraints are used to constrain the overlap rate of multiple diversion paths in a set of candidate diversion paths. Based on the hard constraints, multiple sets of candidate diversion paths are screened to obtain a set of diversion paths that satisfy the hard constraints, which are then optimized using a multi-objective optimization function. Based on the soft constraints, a penalty term is obtained, which is then added to a multi-objective optimization function to optimize the set of dredging paths.

5. The method of claim 4, wherein, Constructing a multi-objective optimization function, combined with the aforementioned multi-constraint unified model, also includes: The multi-objective optimization function includes an efficiency term and a safety term. Adjust the weights of the multi-objective optimization function and the penalty term according to the scenario state, wherein the scenario state includes normal mode and emergency mode; The weight of the efficiency item corresponding to the normal mode is greater than that of the emergency mode, the weight of the penalty item corresponding to the normal mode is less than that of the emergency mode, and the weight of the safety item corresponding to the normal mode is less than that of the emergency mode.

6. The method of claim 4, wherein, A multi-objective optimization function is constructed, and the set of dredging paths is determined by combining the multi-constraint unified model and the dynamic path cost, specifically as follows: When the dynamic path cost of any diversion path exceeds the first threshold, based on the optimized set of multiple diversion paths, the priority of the corresponding diversion path is adjusted according to the dynamic path cost to determine a set of diversion paths. The priority adjustment of the diversion path is negatively correlated with the cost of the dynamic path.

7. The method of claim 5, wherein, Also includes: When the rate of change of the dynamic path cost of any of the determined set of diversion paths exceeds a second threshold, and / or, Any one of the given set of diversion paths does not satisfy the hard constraints, and / or When the scenario state changes, the diversion path is replanned.

8. The method of claim 7, wherein, The replanning of the diversion routes will be implemented, specifically as follows: When the rate of change of the dynamic path cost of any of the determined set of diversion paths exceeds a second threshold, and / or, If any one of the given diversion paths does not meet the hard constraints, the other diversion paths in the set are retained, and local replanning is performed based on the corresponding diversion path. When the scenario state changes, all diversion paths are globally replanned.

9. A multi-constraint crowd management path planning system based on congestion perception, characterized in that, For performing the method according to any one of claims 1 to 8, comprising: The road network modeling module is used to construct a traffic network model of nodes and passages in the target area, and to configure basic attributes for the nodes and passages. The sensing access module is used to acquire congestion sensing data of each channel and / or each node in the target area from the sensing device; The state mapping module is used to map each diversion path in the set of multiple candidate diversion paths to the running state parameters of each channel and / or each node based on the congestion perception data and the diversion road network model, so as to obtain the dynamic path cost. The multi-constraint modeling module is used to model the constraints involved in the planning of each diversion path in multiple sets of candidate diversion paths, based on the congestion perception data and combined with the diversion road network model, and establish a unified multi-constraint model. The multi-objective optimization module is used to construct multi-objective optimization functions; The path planning module is used to solve multiple sets of candidate diversion paths based on the multi-objective optimization function, combined with the multi-constraint unified model and the dynamic path cost, and to determine a set of diversion paths. The results output module is used to output a set of determined diversion paths for guidance.

10. An electronic device, characterized in that, Including processor and memory, The memory contains computer programs. When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

Images