CN-122020217-A - Grouping method and system based on multi-source perception time-varying hypergraph

CN122020217ACN 122020217 ACN122020217 ACN 122020217ACN-122020217-A

Abstract

The invention belongs to the field of intelligent simulation and group behavior modeling, and provides a grouping method and a grouping system based on a multisource perception time-varying hypergraph, which are used for acquiring the position, speed, acceleration, social relationship and psychological state of each individual and preprocessing to obtain a node characteristic matrix; the method comprises the steps of modeling an individual as a hypergraph vertex set, constructing a multi-type hyperedge set according to the relation among nodes to obtain a time-varying hypergraph structure, realizing time-varying optimization of the hyperedge weight in the time-varying hypergraph structure according to a time-adaptive weight updating mechanism to obtain an optimized hyperedge weight, updating the nodes in the time-varying hypergraph structure by adopting a hypergraph attention mechanism to obtain attention adjustment nodes, updating a node characteristic matrix according to comprehensive similarity among the attention adjustment nodes to obtain a comprehensive similarity matrix, and carrying out hierarchical clustering on the comprehensive similarity matrix to complete multi-level crowd grouping. The invention realizes group self-adaptive grouping and collaborative evacuation optimization through a time-varying over-edge weight optimization and attention mechanism.

Inventors

LIU HONG
LI WENHAO
FAN BAOYU
LI XIAOCHUAN
DU PING

Assignees

山东师范大学
浪潮电子信息产业股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. A method of grouping based on multi-source aware time-varying hypergraphs, comprising: acquiring the position, the speed, the acceleration, the social relationship and the psychological state of each individual, and preprocessing to obtain a node characteristic matrix; Modeling an individual as a hypergraph vertex set by taking the individual as a node, and constructing a multi-type hyperedge set according to spatial proximity, time domain accessibility, intention similarity, capacity constraint and association relation among the nodes to obtain a time-varying hypergraph structure; According to a time self-adaptive weight updating mechanism, time-varying optimization of the superside weight in the time-varying supergraph structure is realized by utilizing behavior consistency, risk potential energy and path connectivity indexes, and optimized superside weight is obtained; Updating nodes in the time-varying hypergraph structure by adopting a hypergraph attention mechanism to obtain attention adjustment nodes, and updating a node characteristic matrix according to the comprehensive similarity among the attention adjustment nodes to obtain a comprehensive similarity matrix; And carrying out hierarchical clustering on the comprehensive similarity matrix to complete multi-level crowd grouping.
2. The method for grouping time-varying hypergraphs based on multi-source perception according to claim 1, wherein individual nodes are modeled as hypergraph vertex sets, and a multi-type hyperedge set is constructed according to spatial proximity, time domain accessibility, intention similarity, capacity constraint and association relation among the nodes to obtain a time-varying hypergraph structure, comprising: modeling a hypergraph vertex set by taking an individual as a node; when the Euclidean distance between any two or more nodes is smaller than a distance threshold value, forming an adjacent superside; when a non-shielding visible line exists between the nodes and the angle of view is smaller than the threshold value of the angle, creating a view overrun; if the cosine similarity of the target vectors or strategy embeddings of the two individuals is higher than the intention similarity threshold, forming an intention similarity superside; forming a capacity constraint overrun by nodes which are positioned in a certain area and are of individuals with the current number of people not exceeding the maximum capacity; When the shortest paths of two or more nodes are overlapped on at least one section of area, forming a path association superside; forming a multi-type superside set by using an adjacent superside, a viewing superside, an intended similar superside, a capacity constraint superside and a path association superside; and constructing a time-varying hypergraph structure by using the hypergraph vertex set and the multi-type hyperedge set.
3. The method for grouping time-varying hypergraph based on multi-source perception according to claim 1, wherein the time-varying optimization of the hyperedge weights in the time-varying hypergraph structure is achieved by using behavior consistency, risk potential energy and path connectivity indexes according to a time-adaptive weight update mechanism, so as to obtain optimized hyperedge weights, comprising: Determining an initial superside weight of each superside based on the compactness of each superside critical path and the consistency of the internal behaviors of the group; updating the initial superside weight of each superside according to a time self-adaptive weight updating mechanism to obtain updated superside weights; based on space-time normalization and energy constraint, the updated superside weight is adjusted, and the optimized superside weight is obtained.
4. The method of grouping a time-varying hypergraph based on multi-source perception according to claim 1, wherein updating nodes in the time-varying hypergraph structure by using a hypergraph attention mechanism to obtain attention adjustment nodes, updating a node feature matrix according to comprehensive similarity among the attention adjustment nodes to obtain a comprehensive similarity matrix, comprises: constructing a normalized hypergraph propagation matrix based on the time-varying hypergraph structure; Generating nodes and superside attention weights in a time-varying supergraph structure by using a supergraph attention mechanism, updating the nodes based on the nodes and the superside attention weights, and obtaining attention adjustment nodes; and according to the comprehensive similarity among the attention adjustment nodes, updating the node characteristic matrix to obtain a comprehensive similarity matrix.
5. The method for grouping multiple source-aware time-varying hypergraphs according to claim 1, wherein hierarchical clustering is performed on the integrated similarity matrix to complete multi-level crowd grouping, comprising: determining the distance between the nodes based on the comprehensive similarity between the nodes in the comprehensive similarity matrix; based on the distance between the nodes, calculating the distance between clusters by adopting a weighted average coupling method; and sequentially combining two clusters with minimum inter-cluster distances according to the inter-cluster distance until the number of clusters reaches a set lower limit, and finishing multi-level crowd grouping.
6. The method of claim 1, wherein the node feature matrix comprises position coordinates, velocity vectors, acceleration vectors, social relationship embedding vectors, and mental state feature vectors of the individual.
7. A multi-source aware time-varying hypergraph based packet system comprising: the node characteristic matrix construction module is used for acquiring the position, the speed, the acceleration, the social relationship and the psychological state of each individual and preprocessing the position, the speed, the acceleration, the social relationship and the psychological state to obtain a node characteristic matrix; The time-varying hypergraph structure construction module is used for modeling an individual as a hypergraph vertex set, constructing a multi-type hyperedge set according to spatial proximity, time domain accessibility, intention similarity, capacity constraint and association relation among the nodes, and obtaining a time-varying hypergraph structure; the superside weight dynamic optimization module is used for realizing time-varying optimization of the superside weight in the time-varying supergraph structure by utilizing behavior consistency, risk potential energy and path connectivity indexes according to a time self-adaptive weight updating mechanism to obtain optimized superside weight; The node characteristic optimization module is used for updating nodes in the time-varying hypergraph structure by adopting a hypergraph attention mechanism to obtain attention adjustment nodes, and updating a node characteristic matrix according to the comprehensive similarity among the attention adjustment nodes to obtain a comprehensive similarity matrix; and the hierarchical clustering crowd grouping module is used for performing hierarchical clustering on the comprehensive similarity matrix to complete multi-level crowd grouping.
8. A computer readable storage medium, on which a computer program is stored, which program, when being executed by a processor, implements the steps of a multi-source aware time-variant hypergraph based grouping method according to any of claims 1-6.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of a multi-source aware time-variant hypergraph based grouping method according to any of claims 1-6 when the program is executed.
10. A computer program product, characterized in that the computer program product comprises a computer program which, when executed by a processor, implements the steps in a grouping method based on multisource perception time-varying hypergraph as claimed in any one of claims 1-6.

Description

Grouping method and system based on multi-source perception time-varying hypergraph Technical Field The invention belongs to the technical field of intelligent simulation and group behavior modeling, and particularly relates to a grouping method and system based on a multisource perception time-varying hypergraph. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. In complex building environments or emergency situations, the safety and efficiency of crowd evacuation are highly dependent on collaborative behavior and path selection strategies between individuals. The traditional crowd evacuation modeling method mostly adopts a structure based on graph theory or cellular automaton, the model generally abstracts individuals into nodes, and the interaction among the individuals is simplified into edges, so that only binary corresponding relations can be described. However, in the actual evacuation process, the group behaviors often show high-order interaction characteristics such as multi-main aggregation, dynamic evolution, multi-level cooperation and the like, for example, the influence of social relationship networks, psychological coupling effects, spatial proximity, common target guidance and other factor interweaving effects form complex group dynamics. The traditional graph model can only express the relation between every two, and is difficult to effectively describe the high-dimensional group interaction mode related to simultaneous interaction of a plurality of individuals, so that the traditional graph model has remarkable limitation in the aspects of dynamic group division and path collaborative optimization. In addition, existing crowd grouping strategies mostly rely on clustering algorithms based on static features (such as local density or euclidean distance), e.g. K-Means, DBSCAN, etc. Although the method can realize group identification to a certain extent, the method can not reflect the dynamic characteristics of the group structure evolving along with time in the evacuation process, and is difficult to fuse the mutual coupling relation between multi-source heterogeneous information, including building space structures, individual psychological states, social relation strength, real-time visual perception, environmental risks and the like. Because the space-time evolution and multi-factor synergistic effect are not fully considered, the precision of the existing model in the aspects of dynamic group identification, behavior prediction and collaborative evacuation modeling is often insufficient, so that the reliability and safety of the overall evacuation path planning are affected. Disclosure of Invention In order to solve the problems, the invention provides a grouping method and a grouping system based on multi-source perception time-varying hypergraph, which can dynamically model the high-order interaction and space-time evolution rule of crowd behaviors in a multi-dimensional feature space, and realize the self-adaptive identification and dynamic update of a crowd structure by introducing a time-varying hyperedge weight optimizing mechanism and an attention mechanism, thereby supporting the optimization of a collaborative evacuation path and finally improving the efficiency and the safety of the whole evacuation process. According to some embodiments, the first scheme of the present invention provides a grouping method based on multi-source perception time-varying hypergraph, which adopts the following technical scheme: A method of grouping based on multi-source aware time-varying hypergraphs, comprising: acquiring the position, the speed, the acceleration, the social relationship and the psychological state of each individual, and preprocessing to obtain a node characteristic matrix; Modeling an individual as a hypergraph vertex set by taking the individual as a node, and constructing a multi-type hyperedge set according to spatial proximity, time domain accessibility, intention similarity, capacity constraint and association relation among the nodes to obtain a time-varying hypergraph structure; According to a time self-adaptive weight updating mechanism, time-varying optimization of the superside weight in the time-varying supergraph structure is realized by utilizing behavior consistency, risk potential energy and path connectivity indexes, and optimized superside weight is obtained; Updating nodes in the time-varying hypergraph structure by adopting a hypergraph attention mechanism to obtain attention adjustment nodes, and updating a node characteristic matrix according to the comprehensive similarity among the attention adjustment nodes to obtain a comprehensive similarity matrix; And carrying out hierarchical clustering on the comprehensive similarity matrix to complete multi-level crowd grouping. Further, individual is used as node modeling to form hypergraph vertex set, and