Crowd evacuation congestion propagation prediction method and system based on knowledge graph

Abstract

The present application provides a method and system for predicting crowd evacuation congestion propagation based on a knowledge graph, which relates to the technical field of crowd evacuation computer simulation. By constructing a crowd evacuation knowledge graph, the path subgraph at each moment and the observation information of each node are extracted; the path subgraph and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation; and the dynamic time characteristics of congestion propagation are obtained by using an LSTM network; the spatial correlation characteristics and the dynamic time characteristics are fused to obtain the crowd density of the future time slice of the road section; the congestion propagation graph is determined according to the crowd density of the road section and the connectivity of each road section in time and space. In this way, taking into account the current crowd congestion state, it is possible to predict in real time which road sections will be affected in the future, which is conducive to timely formulating corresponding congestion relief strategies and improving the efficiency of crowd evacuation.

Description

Technical Field

[0001] The present application relates to the technical field of crowd evacuation computer simulation, and in particular to a crowd evacuation congestion propagation prediction method and system based on knowledge graph. Background Art

[0002] The statements in this section merely provide background technical information related to the present application and do not necessarily constitute prior art.

[0003] As urban transportation resources become increasingly complete, people travel more conveniently. The flow of people in large public places continues to rise, which often leads to stampede accidents in emergency situations. Crowd congestion spreads rapidly between road sections, causing the congestion to continue to expand. Therefore, based on real-time monitoring of crowd evacuation data, mining the influencing factors of congestion propagation is of great significance for accurately predicting the footprint of crowd congestion propagation in future sections and responding to crowd evacuation in real time in emergency situations.

[0004] Among the existing crowd evacuation methods, it is usually based on group experiments to accurately model crowd dynamics, explore the behavioral characteristics and movement patterns of crowd evacuation in emergency situations, or optimize the destination selection and route selection methods based on instinctive behavior. However, the topological structure of the evacuation scene is complex, the participating roles are diverse, and the resources are diverse. The existing crowd evacuation analysis methods do not fully explore the potential relationship between various factors. Secondly, the real crowd evacuation data depicts the movement patterns of crowds in emergency situations. Traditional crowd evacuation analysis methods cannot model the dynamic spatiotemporal correlation of congestion propagation. In addition, previous crowd evacuation methods only analyze the evacuation process and improve evacuation efficiency from the perspective of complex human behavior, and cannot predict the propagation footprint of congestion on the road section. Summary of the invention

[0005] In order to solve the above problems, the first aspect of the present application provides a crowd evacuation congestion propagation prediction method and system based on knowledge graph, which can predict in real time which road sections will be affected in the future while taking into account the current crowd congestion status, so as to facilitate the timely formulation of corresponding congestion relief strategies and improve the efficiency of crowd evacuation.

[0006] In order to achieve the above objectives, this application mainly includes the following aspects:

[0007] In a first aspect, an embodiment of the present application provides a crowd evacuation congestion propagation prediction method based on a knowledge graph, comprising:

[0008] Obtain crowd evacuation data within a specified evacuation scenario in a preset time period and construct a crowd evacuation knowledge graph;

[0009] At each moment, multiple entities with semantics of road sections are obtained from the crowd evacuation knowledge graph, one of the entities is used as a starting node, a set of nodes having a path connection relationship with the starting node is determined, and a path subgraph between the starting node and other nodes in the set and observation information of each node are extracted;

[0010] The path subgraph extracted at each moment and the observation information of each node are processed by multi-layer graph convolution to obtain spatial features, and the spatial features of the previous moment are fused with the spatial features of the current moment to obtain the spatial correlation features of congestion propagation; and the dynamic time features of congestion propagation are obtained by using the LSTM network; the spatial correlation features and dynamic time features are fused to obtain the crowd density of the road section at the future moment;

[0011] A congestion propagation graph is determined according to the crowd density of the road section and the connectivity of each road section in time and space, and the crowd congestion propagation footprint at a future moment is predicted based on the congestion propagation graph.

[0012] In a possible implementation, the method for acquiring crowd evacuation data includes:

[0013] Obtain the safe evacuation map of the specified evacuation scenario, use intersections and exits as reference points, segment the path of the entire evacuation scenario, obtain multiple sections, and match the static basic data with the corresponding sections;

[0014] Based on the collected video image data of the crowd in each road section, the dynamic road condition information corresponding to each road section is determined, thereby obtaining crowd evacuation data containing a large amount of hidden information; wherein the dynamic road condition information includes crowd density, crowd flow and moving speed.

[0015] In a possible implementation, the crowd evacuation knowledge graph is constructed in a top-down manner. Specifically, the crowd evacuation data is discretized and semanticized into triples based on the knowledge graph pattern layer structure; the divided road sections are used as the organizational units of the evacuation knowledge graph, and all triples are interconnected and stored in the Neo4j graph database to achieve visual storage of the knowledge graph.

[0016] In a possible implementation, the extracted path subgraph and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation, including:

[0017] Determine a dynamic adjacency matrix according to the extracted path subgraph, and determine a node feature matrix according to the observation information of each node;

[0018] The dynamic adjacency matrix and the node feature matrix are used as the input of the graph convolutional neural network. After multi-layer graph convolution processing, the node information is aggregated to obtain the spatial correlation characteristics of congestion propagation.

[0019] In one possible implementation, a resource random walk algorithm is used to obtain the dynamic adjacency matrix; specifically, the head node and tail node of the path subgraph are determined, a predetermined initial amount of resources is placed at the head node, and it is allowed to start a random walk along the path, and the amount of resources flowing to the tail node is used as a measure of the size of the spatial correlation between nodes, thereby obtaining a dynamic adjacency matrix representing the path topology information.

[0020] In a possible implementation, the method for constructing the congestion propagation graph includes:

[0021] According to the crowd congestion index and average crowd density corresponding to each road section, a set of all congested road sections in each time slice is obtained; the crowd congestion index is used to determine the relative value of the congestion degree of the road section in a specific time slice;

[0022] Create a time window of a set size, merge repeated congested sections within the current time window, and record the number of the most recent time slice;

[0023] Combine the time slice number and spatial position to determine whether two congested sections are connected in the current time window, and connect the adjacent congested sections in the time window;

[0024] The time window slides backward by a set number of time slices, and the above steps are repeated until all time slices are traversed.

[0025] In a possible implementation manner, the crowd congestion index is determined in the following manner:

[0026] CCI=(d f -d i ) / d f ;

[0027] Among them, CCI represents crowd congestion index; d f represents the free density of the crowd, that is, the average density of the crowd on section i when the section i remains smooth; d i It represents the average density of people on road section i in a given time slice.

[0028] In a second aspect, an embodiment of the present application provides a crowd evacuation congestion propagation prediction system based on a knowledge graph, comprising:

[0029] An acquisition module is used to acquire crowd evacuation data within a specified evacuation scenario in a preset time period and construct a crowd evacuation knowledge graph;

[0030] An extraction module is used to obtain, at each moment, multiple entities with semantics of road sections from the crowd evacuation knowledge graph, take one of the entities as a starting node, determine a set of nodes that have a path connection relationship with the starting node, and extract a path subgraph between the starting node and other nodes in the set and observation information of each node;

[0031] The fusion module is used to process the path subgraph extracted at each moment and the observation information of each node through multi-layer graph convolution to obtain spatial features, fuse the spatial features of the previous moment with the spatial features of the current moment to obtain the spatial correlation features of congestion propagation; and use the LSTM network to obtain the dynamic time features of congestion propagation; fuse the spatial correlation features and dynamic time features to obtain the crowd density of the road section at the future moment;

[0032] The prediction module is used to determine a congestion propagation map according to the crowd density of the road section and the connectivity of each road section in time and space, and predict the crowd congestion propagation footprint at a future moment based on the congestion propagation map.

[0033] In a third aspect, an embodiment of the present application provides a computer device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the crowd evacuation congestion propagation prediction method as described in the first aspect and any possible implementation scheme of the first aspect are performed.

[0034] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the crowd evacuation congestion propagation prediction method as described in the first aspect and any possible implementation scheme of the first aspect are executed.

[0035] The beneficial effects of this application are:

[0036] 1. This application constructs a crowd evacuation knowledge graph to extract the path subgraph at each moment and the observation information of each node; the path subgraph and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation; and the dynamic time characteristics of congestion propagation are obtained by using the LSTM network; the spatial correlation characteristics and the dynamic time characteristics are fused to obtain the crowd density of the future time slice of the road section; the congestion propagation graph is determined according to the crowd density of the road section and the connectivity of each road section in time and space. In this way, by fusing the spatial correlation characteristics and the dynamic time characteristics, the potential correlation between the road sections can be fully excavated to improve the accuracy of crowd congestion propagation prediction. Moreover, taking into account the current crowd congestion status, it is possible to predict in real time which sections will be affected in the future, which is conducive to the timely formulation of corresponding congestion relief strategies and improve the efficiency of crowd evacuation.

[0037] 2. By using the idea of ​​random walk algorithm to extract dynamic evacuation airspace map, the correlation between road sections in the knowledge graph can be automatically mined to obtain more diversified spatiotemporal information. In addition, sub-graph sampling based on the evacuation knowledge graph solves the problem of explosion of the number of adjacent points in the knowledge graph, which affects the query efficiency.

[0038] 3. Construct congestion propagation diagrams on multiple historical and future time slices, characterize the dynamic propagation process of crowd congestion on multiple consecutive time slices, and be able to predict how these congestions will spread. This can provide valuable information for crowd evacuation in emergency situations to help alleviate emerging congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings in the specification, which constitute a part of this application, are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0040] Figure 1 It is a flow chart of a crowd evacuation congestion propagation prediction method provided by an embodiment of the present invention;

[0041] Figure 2 It is a structural diagram of the model layer of the crowd evacuation knowledge graph provided in the embodiment of the present application;

[0042] Figure 3 is a schematic diagram of a crowd evacuation knowledge graph provided in an embodiment of the present application;

[0043] Figure 4 It is a schematic diagram of extracting a dynamic evacuation airspace map provided in an embodiment of the present application;

[0044] Figure 5 It is a schematic diagram of the GCN mining spatial correlation feature provided by the embodiment of the present application;

[0045] Figure 6 It is a schematic diagram of LSTM mining dynamic time features provided in an embodiment of the present application;

[0046] Figure 7 This is a diagram of the architecture of the congestion propagation spatiotemporal neural network model CPSTNN provided in an embodiment of the present application;

[0047] Figure 8 It is a schematic diagram of constructing a congestion propagation graph to predict crowd congestion propagation provided by an embodiment of the present application;

[0048] Fig. 9 It is a structural schematic diagram of a crowd evacuation congestion propagation prediction system provided in an embodiment of the present application;

[0049] Fig.10 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The present application is further described below in conjunction with the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0053] Embodiment 1

[0054] See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a crowd evacuation congestion propagation prediction method provided by an embodiment of the present invention. Figure 1 As shown in , the crowd evacuation congestion propagation prediction method includes the following steps:

[0055] S101: Obtain crowd evacuation data within a specified evacuation scenario in a preset time period and construct a crowd evacuation knowledge graph.

[0056] In a specific implementation, the crowd evacuation data includes static infrastructure, dynamic road condition information and path topology. The static infrastructure includes but is not limited to one or more of safety signs, fire-fighting equipment, and emergency lighting. The dynamic road condition information includes crowd density, flow of people and moving speed.

[0057] As an optional implementation, the method for acquiring crowd evacuation data includes:

[0058] Obtain a safe evacuation map for a specified evacuation scenario, use intersections and exits as reference points, segment the path of the entire evacuation scenario, obtain multiple sections, and match the static basic data with the corresponding sections; determine the dynamic road condition information corresponding to each section based on the collected video image data of the crowd in each section, thereby obtaining crowd evacuation data containing a large amount of hidden information; wherein the dynamic road condition information includes crowd density, crowd flow and moving speed.

[0059] Among them, you can download the safe evacuation instructions map of the specified evacuation scene, collect a large amount of video image data through the cameras installed on the road section, and calculate the dynamic road condition information such as crowd density, pedestrian flow, moving speed, etc. corresponding to the road section based on the video image data.

[0060] As an optional implementation, the crowd evacuation knowledge graph is constructed in a top-down manner. Specifically, the crowd evacuation data is discretized and semantically converted into triples based on the knowledge graph pattern layer structure. The divided road sections are used as the organizational units of the evacuation knowledge graph, and all triples are interconnected and stored in a Neo4j graph database to achieve visual storage of the knowledge graph.

[0061] In the specific implementation, we first design the pattern layer structure of the crowd evacuation knowledge graph based on the professional knowledge in the crowd evacuation field and the collected data, and then construct the crowd evacuation knowledge graph according to the designed pattern layer structure. Figure 2 As shown in the figure, the spatial data part is divided into three entity types: road sections, intersections, and exits, which are the most basic elements of the crowd evacuation knowledge graph; static infrastructure such as safety signs, fire-fighting equipment, and accident lighting in the evacuation scene are stored as point of interest entities in the crowd evacuation knowledge graph; dynamic road condition data such as crowd density, pedestrian flow, and moving speed are stored by timestamp as attributes of the road section entity; the topological relationship between evacuation sections includes "upstream and downstream" and "intersection", and the relationship between sections and other static spatial data includes "nearby", "connected", and "existent".

[0062] The collected data is discretized and semantically transformed into triples, and the divided road sections are used as the organizational units of the evacuation knowledge graph. All triples are connected and stored in the Neo4j graph database, such as Figure 3 As shown in , a crowd evacuation knowledge graph containing a large amount of hidden knowledge is constructed.

[0063] S102: At each moment, multiple entities with semantics of road sections are obtained from the crowd evacuation knowledge graph, one of the entities is taken as the starting node, a set of nodes having a path connection relationship with the starting node is determined, and a path subgraph between the starting node and other nodes in the set and observation information of each node are extracted.

[0064] In the specific implementation, subgraph sampling is performed on the static spatial part along the path of the evacuation knowledge graph to extract the dynamic evacuation airspace graph connected by multiple spatial relationships, such as Figure 4 As shown in , the method for constructing a dynamic evacuation airspace graph includes: finding all entities with semantics of road sections in the crowd evacuation knowledge graph; randomly selecting a node h 1 As the starting node, find all nodes with h 1 The triplet of the head node (h 1 , R, E), where R and E represent the relationship set and node set in the first-hop query result respectively; if there is a node t in E with the semantics of a road segment 1 , then record h 1 and t 1 The path subgraph is taken as the head node for the second hop query. The entire subgraph sampling process ends after the fifth hop, thus obtaining a dynamic evacuation airspace graph with a more streamlined scale and higher association density.

[0065] S103: The path subgraph extracted at each moment and the observation information of each node are processed by multi-layer graph convolution to obtain spatial features, and the spatial features of the previous moment are fused with the spatial features of the current moment to obtain the spatial correlation features of congestion propagation; and the dynamic time features of congestion propagation are obtained by using the LSTM network; the spatial correlation features and the dynamic time features are fused to obtain the crowd density of the road section at the future moment.

[0066] In the specific implementation, a congestion propagation spatiotemporal neural network model CPSTNN is established, specifically: GCN is used to mine the spatial correlation characteristics of congestion propagation, LSTM is used to mine the dynamic time characteristics of congestion propagation, and the obtained spatial correlation characteristics and dynamic time characteristics are fused to obtain the crowd density of the road section in the future K time slices.

[0067] In the embodiment of the present application, as an optional implementation, the extracted path subgraph and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation, including:

[0068] Determine a dynamic adjacency matrix based on the extracted path subgraph, and determine a node feature matrix based on the observed information of each node; use the dynamic adjacency matrix and node feature matrix as inputs to a graph convolutional neural network, and after multi-layer graph convolution processing, aggregate node information to obtain spatial correlation characteristics of congestion propagation. Optionally, use a resource random walk algorithm to obtain the dynamic adjacency matrix; specifically, determine the head node and tail node of the path subgraph, place a predetermined initial amount of resources at the head node, let it start a random walk along the path, and use the amount of resources flowing to the tail node as a measure of the size of the spatial correlation between nodes to obtain a dynamic adjacency matrix representing the path topology information.

[0069] In the specific implementation, the adjacency matrix is ​​used to represent the relationship between nodes. The random walk of resources is performed based on the path of the dynamic evacuation airspace graph to obtain the spatial correlation weight between any two nodes, which is the dynamic adjacency matrix A at time slice t. t The three observed values ​​of crowd density, crowd flow, and moving speed generated by the road section at time slice t are standardized to [0,1] as the feature vector of each road section.

[0070] The congestion propagation path in crowd evacuation is essentially a graph structure, and GCN is designed to learn graph features, such as Figure 5 As shown in , a self-loop is added to each node in the dynamic adjacency matrix to integrate its own road features. Where I is the N×N identity matrix, A t is the dynamic adjacency matrix, Represents the dynamic adjacency matrix after adding the self-loop;

[0071] Initialize the 0th layer embedding as the original node feature matrix of the evacuation airspace graph at time slice t where F t is the node feature matrix at time slice t, Represents the embedding of the 0th layer at time slice t; superimpose L layers of convolutional layers, and perform convolution aggregation on the features of itself and its neighbors layer by layer. The multi-layer graph convolution operation is as follows:

[0072]

[0073] in, represents the output of the lth convolutional layer, using ReLU as the activation function σ, represents the dynamic adjacency matrix after adding the self-loop, For the matrix The degree matrix of represents the hidden nodes of the l-1th layer, is the weight parameter learned in the l-1th layer.

[0074] After L layers of neighborhood information aggregation, the final embedding representation is obtained in is the output of the Lth convolutional layer at time slice t, H t Represents the spatial feature vector at time slice t;

[0075] The spatial feature vector H of all historical time slices 1 ,H 2 ,…,H T ∈R T×d Merge as the dynamic spatial feature F of crowd congestion propagation s =H 1 ⊙H 2 ⊙…⊙H T ;

[0076] The evacuation data generated by adjacent road sections at consecutive time stamps are dynamically related, so crowd congestion propagation prediction is a typical time series data prediction problem. Figure 6 As shown in the figure, LSTM is used to mine the dynamic time characteristics of congestion propagation, including:

[0077] The observed values ​​of road crowd density are input into the LSTM network model. The forget gate first decides how to discard the information of the previous hidden state. The forget gate can be expressed as:

[0078] f t =σ(w f ·x t +u f ·h t-1 +b f );

[0079] Among them, Sigmoid is used as the activation function σ, w f 、u f and b f is a learnable parameter of the forgetting unit, which is used to control the degree of forgetting. t is the observed value of the crowd density at time slice t, h t-1 represents the hidden state of the previous layer, f t is the output of the forget gate; the input gate determines how to update the information of the current input data to the new unit state. The input gate can be expressed as:

[0080] i t =σ(w i ·x i +u i ·h t-1 +b i );

[0081] Among them, w i 、ui and b i is the learnable parameter of the input unit; the input data x is processed using the tanh function t , generate candidate cell states The memory of the internal cell is obtained by combining the old cell state C t-1 and candidate cell status Generate a new cell state C t , the process can be expressed as:

[0082]

[0083]

[0084] Among them, w c 、u c and b c are the learnable parameters of the unit, represents the Hadamard product of corresponding elements, i t is the output of the input gate; the output gate will decide how to retain the information of the new unit state and generate the hidden state h t As the output of this layer and the input of the next layer, the output gate and the generation method of the new hidden state can be expressed as:

[0085] o t =σ(w o ·x t +u o ·h t-1 +b o );

[0086]

[0087] Among them, w o 、u o and b o is the learnable parameter of the output unit, o t Output gate output; Finally, the LSTM model converts the hidden state h of the last layer T , as the dynamic time characteristic of crowd congestion propagation F T .

[0088] Figure 7 CPSTNN architecture diagram of the congestion propagation spatiotemporal neural network model provided by the embodiment of the present application. As shown in the figure, the embodiment of the present application first uses GCN to capture the spatial correlation characteristics of crowd congestion propagation, then uses LSTM to mine the dynamic time characteristics of congestion propagation, and finally uses the graph convolution layer to fuse the spatiotemporal features to predict the crowd density D of the road section in the next K time slices. KSpecifically, the dynamic adjacency matrix and node feature matrix obtained from the dynamic evacuated spatial domain graph are used as the input of GCN, and the neighbor node information is aggregated through multi-layer graph convolution to capture the spatial correlation feature F of the entire graph. S ; Input the congestion time series data of the road section in the historical T time slices into LSTM to mine the dynamic time characteristics of congestion propagation F T ; Use a layer of graph convolution to fuse the spatial correlation features and dynamic time features as the complete features of the road section. The graph convolution process is as follows:

[0089]

[0090] Among them, F ST represents the spatiotemporal characteristics of crowd congestion propagation, A is the fixed adjacency matrix obtained from the evacuation knowledge graph, Add the degree matrix after self-loop to the A matrix, and use the ReLU function as the activation function σ.

[0091] ⊙ represents the dynamic time feature F T and dynamic spatial features F S Merge, θ is a learnable parameter;

[0092] Finally, the fully connected layer is used to adjust the output dimension and predict the crowd density D of the road section in the next K time slices. K :

[0093] D K =FC(F ST );

[0094] Where D K is the predicted value of crowd density, F ST is the spatiotemporal characteristics of crowd congestion propagation, and FC is the fully connected layer function.

[0095] S104: Determine a congestion propagation graph according to the crowd density of the road section and the connectivity of each road section in time and space, and predict crowd congestion propagation footprints at future moments based on the congestion propagation graph.

[0096] In the present application, as an optional implementation, Figure 8 As shown in , the method for constructing the congestion propagation graph includes the following steps:

[0097] Step b11: according to the crowd congestion index and average crowd density corresponding to each road section, obtain a set of all congested road sections in each time slice; the crowd congestion index is used to determine the relative value of the congestion degree of the road section in a specific time slice; wherein the crowd congestion index can be determined by the following method:

[0098] CCI=(d f -di ) / d f ;

[0099] Among them, CCI represents crowd congestion index; d f represents the free density of the crowd, that is, the average density of the crowd on section i when the section i remains smooth; d i It represents the average density of people on road section i in a given time slice.

[0100] Step b12: Create a time window of a set size, merge repeated congested sections in the current time window, and record the number of the most recent time slice.

[0101] In the specific implementation, the set of congested sections of T historical time slices and K future time slices is obtained, and a time window is created with a window size of μ+1. The μ time slices in the current window are merged into a new time slice. The congestion status of the section is usually maintained over multiple time slices. The repeated congested sections in the current window are merged, and the sequence number of the most recent time slice of these congested sections is recorded.

[0102] Step b13: judging whether two congested sections are connected in the current time window by combining the time slice number and the spatial position, and connecting the adjacent congested sections in the time window;

[0103] Step b14: The time window slides backward by a set number of time slices, and the above steps are repeated until all time slices are traversed. Using this method, the dynamic propagation process of crowd congestion is characterized in multiple historical and future time slices, and the congestion propagation footprint of future road sections is predicted.

[0104] The crowd evacuation congestion propagation prediction method provided by the present application constructs a crowd evacuation knowledge graph to extract the path subgraph at each moment and the observation information of each node; the path subgraph and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation; and the dynamic time characteristics of congestion propagation are obtained by using the LSTM network; the spatial correlation characteristics and the dynamic time characteristics are fused to obtain the crowd density of the future time slice of the road section; the congestion propagation graph is determined according to the crowd density of the road section and the connectivity of each road section in time and space. In this way, by fusing the spatial correlation characteristics and the dynamic time characteristics, the potential correlation between the road sections can be fully excavated, and the accuracy of the crowd congestion propagation prediction can be improved. Moreover, considering the current crowd congestion state, it is possible to predict in real time which road sections will be affected in the future, which is conducive to timely formulating corresponding congestion relief strategies and improving the efficiency of crowd evacuation.

[0105] Embodiment 2

[0106] See also Fig. 9, Fig. 9 The crowd evacuation congestion propagation prediction system provided by the embodiment of the present application is as follows: Fig. 9 As shown in , the crowd evacuation congestion propagation prediction system 900 provided in the embodiment of the present application includes:

[0107] An acquisition module 910 is used to acquire crowd evacuation data within a specified evacuation scenario in a preset time period and construct a crowd evacuation knowledge graph;

[0108] The extraction module 920 is used to obtain, at each moment, multiple entities with semantics of road sections from the crowd evacuation knowledge graph, take one of the entities as a starting node, determine a set of nodes that have a path connection relationship with the starting node, and extract a path subgraph between the starting node and other nodes in the set and observation information of each node;

[0109] Fusion module 930 is used to process the path subgraph extracted at each moment and the observation information of each node through multi-layer graph convolution to obtain spatial features, fuse the spatial features of the previous moment with the spatial features of the current moment to obtain spatial correlation features of congestion propagation; and use LSTM network to obtain dynamic time features of congestion propagation; fuse the spatial correlation features and dynamic time features to obtain the crowd density of the road section at the future moment;

[0110] The prediction module 940 is used to determine a congestion propagation graph according to the crowd density of the road section and the connectivity of each road section in time and space, and predict the crowd congestion propagation footprint at a future moment based on the congestion propagation graph.

[0111] The crowd evacuation congestion propagation prediction system provided by the embodiment of the present application includes an acquisition module, an extraction module, a fusion module and a prediction module. By constructing a crowd evacuation knowledge graph, the path subgraph at each moment and the observation information of each node are extracted; the path subgraph and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation; and the dynamic time characteristics of congestion propagation are obtained by using an LSTM network; the spatial correlation characteristics and the dynamic time characteristics are fused to obtain the crowd density of the future time slice of the road section; the congestion propagation graph is determined according to the crowd density of the road section and the connectivity of each road section in time and space. In this way, by fusing the spatial correlation characteristics and the dynamic time characteristics, the potential correlation between the road sections can be fully excavated, and the accuracy of the crowd congestion propagation prediction can be improved. Moreover, considering the current crowd congestion state, it is possible to predict in real time which sections will be affected in the future, which is conducive to timely formulating corresponding congestion relief strategies and improving the efficiency of crowd evacuation.

[0112] Embodiment 3

[0113] See also Fig.10 , Fig.10 Schematic diagram of a computer device according to an embodiment of the present application. Fig.10 As shown in , the computer device 1000 includes a processor 1010 , a memory 1020 and a bus 1030 .

[0114] The memory 1020 stores machine-readable instructions executable by the processor 1010. When the computer device 1000 is running, the processor 1010 communicates with the memory 1020 via the bus 1030. When the machine-readable instructions are executed by the processor 1010, the above-mentioned Figures 1 to 8 The specific implementation of the steps of the crowd evacuation congestion propagation prediction method in the method embodiment shown can be found in the method embodiment, which will not be repeated here.

[0115] Embodiment 4

[0116] Based on the same application concept, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the crowd evacuation congestion propagation prediction method described in the above method embodiment are executed.

[0117] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0118] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A crowd evacuation congestion propagation prediction method based on knowledge graph, characterized in that: include: Obtain crowd evacuation data within a specified evacuation scenario in a preset time period and construct a crowd evacuation knowledge graph; The spatial data part is divided into three entity types: road section, intersection, and exit, which serve as the most basic elements of the crowd evacuation knowledge graph; Static infrastructure in evacuation scenarios is stored as point of interest entities in the crowd evacuation knowledge graph; dynamic road condition data is stored by timestamp as the attribute of the road segment entity; the topological relationship between evacuation road segments includes "upstream and downstream" and "intersection", and the relationship between road segments and other static spatial data includes "nearby", "connected" and "existent"; At each moment, multiple entities with semantics of road sections are obtained from the crowd evacuation knowledge graph, one of the entities is used as a starting node, a set of nodes having a path connection relationship with the starting node is determined, and a path subgraph between the starting node and other nodes in the set and observation information of each node are extracted; The path subgraph extracted at each moment and the observation information of each node are processed by multi-layer graph convolution to obtain spatial features. The spatial features of the previous moment are fused with the spatial features of the current moment to obtain the spatial correlation features of congestion propagation. And using the LSTM network to obtain the dynamic time characteristics of congestion propagation; fusing the spatial correlation characteristics and the dynamic time characteristics to obtain the crowd density of the road section at a future time; Determine a congestion propagation graph according to the crowd density of the road section and the connectivity of each road section in time and space, and predict the crowd congestion propagation footprint at a future moment based on the congestion propagation graph; The crowd congestion index is determined by: ； Among them, CCI represents crowd congestion index; d f represents the free density of the crowd, that is, the average density of the crowd on road section i when road section i remains smooth; d i It represents the average density of people on road section i in a given time slice.

2. The crowd evacuation congestion propagation prediction method according to claim 1, characterized in that: The method for acquiring crowd evacuation data comprises: Obtain the safe evacuation map of the specified evacuation scenario, use intersections and exits as reference points, segment the path of the entire evacuation scenario, obtain multiple sections, and match the static basic data with the corresponding sections; Based on the collected video image data of the crowd in each road section, the dynamic road condition information corresponding to each road section is determined, thereby obtaining crowd evacuation data containing a large amount of hidden information; wherein the dynamic road condition information includes crowd density, crowd flow and moving speed.

3. The crowd evacuation congestion propagation prediction method according to claim 1, characterized in that: The crowd evacuation knowledge graph is constructed in a top-down manner. Specifically, the crowd evacuation data is discretized and semanticized into triples based on the knowledge graph pattern layer structure; the divided road sections are used as the organizational units of the evacuation knowledge graph, and all triples are connected and stored in the Neo4j graph database to achieve visual storage of the knowledge graph.

4. The crowd evacuation congestion propagation prediction method according to claim 1, characterized in that: The path subgraph extracted at each time and the observation information of each node are processed by multi-layer graph convolution to obtain the spatial correlation characteristics of congestion propagation, including: Determine a dynamic adjacency matrix according to the extracted path subgraph, and determine a node feature matrix according to the observation information of each node; The dynamic adjacency matrix and the node feature matrix are used as the input of the graph convolutional neural network. After multi-layer graph convolution processing, the node information is aggregated to obtain the spatial correlation characteristics of congestion propagation.

5. The crowd evacuation congestion propagation prediction method according to claim 4, characterized in that: The dynamic adjacency matrix is ​​obtained by adopting a resource random walk algorithm; specifically, the head node and the tail node of the path subgraph are determined, a predetermined initial amount of resources is placed at the head node, and it is allowed to start a random walk along the path, and the amount of resources flowing to the tail node is used as a measure of the size of the spatial correlation between nodes, so as to obtain a dynamic adjacency matrix representing the path topology information.

6. The crowd evacuation congestion propagation prediction method according to claim 1, characterized in that: The method for constructing the congestion propagation graph includes: According to the crowd congestion index and average crowd density corresponding to each road section, a set of all congested road sections in each time slice is obtained; the crowd congestion index is used to determine the relative value of the congestion degree of the road section in a specific time slice; Create a time window of a set size, merge repeated congested sections within the current time window, and record the number of the most recent time slice; Combine the time slice number and spatial position to determine whether two congested sections are connected in the current time window, and connect the adjacent congested sections in the time window; The time window slides backward by a set number of time slices, and the above steps are repeated until all time slices are traversed.

7. A crowd evacuation congestion propagation prediction system based on knowledge graph, characterized in that: include: An acquisition module is used to acquire crowd evacuation data within a specified evacuation scenario in a preset time period and construct a crowd evacuation knowledge graph; The spatial data part is divided into three entity types: road section, intersection, and exit, which serve as the most basic elements of the crowd evacuation knowledge graph; Static infrastructure in evacuation scenarios is stored as point of interest entities in the crowd evacuation knowledge graph; dynamic road condition data is stored by timestamp as the attribute of the road segment entity; the topological relationship between evacuation road segments includes "upstream and downstream" and "intersection", and the relationship between road segments and other static spatial data includes "nearby", "connected" and "existent"; An extraction module is used to obtain, at each moment, multiple entities with semantics of road sections from the crowd evacuation knowledge graph, take one of the entities as a starting node, determine a set of nodes that have a path connection relationship with the starting node, and extract a path subgraph between the starting node and other nodes in the set and observation information of each node; The fusion module is used to process the path subgraph extracted at each moment and the observation information of each node through multi-layer graph convolution to obtain spatial features, and fuse the spatial features of the previous moment with the spatial features of the current moment to obtain the spatial correlation features of congestion propagation; And using the LSTM network to obtain the dynamic time characteristics of congestion propagation; fusing the spatial correlation characteristics and the dynamic time characteristics to obtain the crowd density of the road section at a future time; A prediction module, used to determine a congestion propagation graph according to the crowd density of the road section and the connectivity of each road section in time and space, and predict the crowd congestion propagation footprint at a future moment based on the congestion propagation graph; The crowd congestion index is determined by: ； Among them, CCI represents crowd congestion index; d f represents the free density of the crowd, that is, the average density of the crowd on road section i when road section i remains smooth; d i It represents the average density of people on road section i in a given time slice.

8. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the crowd evacuation congestion propagation prediction method as described in any one of claims 1 to 6 are performed.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the crowd evacuation congestion propagation prediction method according to any one of claims 1 to 6 are executed.

Images