A large-scale group target clustering method integrating dynamic value

By constructing a global value table and improving the K-mean clustering algorithm in large-scale group target coverage scenarios, the problems of limited detector range and reduced target information value are solved, and rapid convergence and high-timed clustering effects are achieved.

CN114970757BActive Publication Date: 2024-11-26THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210703039.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-21
Publication Date
2024-11-26
Estimated Expiration
2042-06-21

AI Technical Summary

Technical Problem

In large-scale group target coverage scenarios, the limited detection range of detectors makes it difficult to obtain the target location of the group in real time, and the dynamic changes in target locations lead to a reduction in the value of information. The existing technology lacks consideration of the importance and value of the target.

Method used

A large-scale group target clustering method that integrates dynamic values ​​is adopted. By constructing a global value table maintained by the detector network, the target value is dynamically updated, and the K-mean clustering algorithm is improved, and weighted clustering is performed based on the target position and value.

Benefits of technology

The clustering effect is improved, the problem of limited detector range and reduced target information value is solved, rapid convergence and high timeliness are achieved, and it is suitable for large-scale group target scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114970757B_ABST
    Figure CN114970757B_ABST
Patent Text Reader

Abstract

The present invention relates to a large-scale group target clustering method integrating dynamic value, and belongs to the field of large-scale target coverage. The method models a group composed of multiple independent target units, establishes a global value table and a value update algorithm for each target in the group, and updates the information in the global value table in real time by a detection set. Based on the information in the global value table, an improved K-means clustering algorithm is used to cluster the large-scale group integrating dynamic value. Based on the improved K-means clustering algorithm, the method proposes a solution that meets practical requirements for the large-scale group target clustering problem, and solves the problem that the current method lacks practicability and lacks consideration of target value for the large-scale group target clustering problem.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of large-scale target coverage, and in particular to a large-scale group target clustering method integrating dynamic values. Background Art

[0002] Large-scale target coverage is a common application in a type of engineering, such as communication base station user coverage, UAV wide-area target search, group target emergency support, etc. Due to the limitations of the limited number of communication base stations, limited UAV search range, and limited emergency equipment, it is necessary to analyze the randomly distributed large-scale group targets, obtain several key locations where relevant resources are most needed to deploy relevant resources, build relevant facilities at these locations, and implement specific tasks.

[0003] Clustering algorithms are usually used to obtain limited key positions of large-scale group targets. Common clustering algorithms include: K-means clustering algorithm, mean shift clustering algorithm, DBSCAN clustering algorithm, hierarchical clustering algorithm, etc. Among them, K-means clustering algorithm is the most commonly used clustering algorithm. Its advantages are fast convergence speed and low algorithm complexity. Its disadvantages are that the number of clusters needs to be determined. Under the premise of using random initial centers, the algorithm results lack consistency; mean shift clustering algorithm is an algorithm that uses sliding windows. Its advantage is that it does not need to select the number of clusters. Its disadvantage is that the radius of the window has a great influence on the classification results of the algorithm, and the window radius needs to be carefully selected; DBSCAN clustering algorithm is a density-based clustering algorithm. Its advantage is that it does not need to select the number of clusters. Its disadvantage is that the density of the target has a great influence on the clustering results. If the density distribution of the target group is greatly different, the clustering effect will be affected; hierarchical clustering algorithms are divided into top-down and bottom-up algorithms. Its advantage is that it does not need to select the number of clusters. Its disadvantage is that the algorithm complexity is high and the convergence speed is slow. In actual application scenarios, especially in clustering algorithms for large-scale target groups, in order to ensure the rapid operation of algorithms deployed on various computing platforms, clustering algorithms with lower algorithm complexity and faster convergence speed are more valuable.

[0004] In traditional large-scale group target coverage scenarios, if drones have both target detection and target coverage functions, they need to detect the real-time position of the target and implement coverage. Most studies assume that the positions of all targets can be obtained in real time for cluster analysis. However, in reality, the acquisition of target positions is achieved by drone-mounted detectors, which have a limited detection range. Targets beyond the detection range cannot provide the latest position information in real time. As the target continues to move, the currently available target data information will lose its value over time, and the value of different targets is often different. Current research lacks consideration of this content. Summary of the invention

[0005] Aiming at the problem that in large-scale group target coverage scenarios, the position of group targets is difficult to obtain in real time due to the limited detection range of detectors, and the value of original target information is continuously reduced due to the continuous movement of targets outside the detection range, the present invention proposes a large-scale group target clustering method integrating dynamic value. The method takes the dynamic value of targets into consideration, can improve the clustering effect, and can be applied to large-scale group target scenarios.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A large-scale group target clustering method integrating dynamic value includes the following steps:

[0008] (1) constructing a group goal, wherein the group goal is composed of multiple independent individual units;

[0009] (2) construct a global value table jointly maintained by the detector network;

[0010] (3) Use the detector network to detect individual units in the group target and dynamically update the global value table;

[0011] (4) Based on the dynamic value information in the global value table, an improved K-means clustering method is used to cluster large-scale groups that integrate dynamic values.

[0012] Furthermore, the detector network is composed of a plurality of detectors, and the detectors share information via a wireless network, and each detector obtains target information of individual units within its own detection range.

[0013] Furthermore, the global value table includes target numbers, target values, and target locations of all individual unit targets.

[0014] Furthermore, the specific method of step (3) is as follows: when an individual unit target enters the detection range of a detector of the detector network, its target value is set to a fixed value, and the target position of the individual unit is updated in real time by the detector network; if an individual unit target exceeds the detection range of the detector network, the target position of the individual unit is set to the last detected position, and the target value of the individual unit decreases over time; when the detector network discovers a new target, the new target information is added based on the original global value table.

[0015] Furthermore, the specific method of step (4) is:

[0016] (401) The number of initial cluster centers is set to m, and the positions of the initial cluster centers are selected randomly or according to a preset rule;

[0017] (402) According to the distance relationship between the group target and the initial cluster center, the group target is divided into m sub-groups. If there is no target in a sub-group, the initial cluster center corresponding to the sub-group is reselected, and the sub-group targets belonging to each cluster center are recalculated until the number of targets in all sub-groups is not 0;

[0018] (403) Read the global value table, fuse the target position and the target value and perform K-means clustering by weight, calculate the cluster center of each subgroup, and make the cluster center of the subgroup close to the target with high value in the subgroup;

[0019] (404) Re-dividing subgroups with new cluster centers;

[0020] (405) Repeat steps (403) and (404) until the subgroup divisions no longer change or the maximum number of iterations is reached, and clustering is completed.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. In order to solve the problem of limited detection range of the detector, the present invention designs a global value table, which solves the problem of poor practicality caused by assuming unlimited detection range in most studies by dynamically updating the group target value.

[0023] 2. The present invention proposes an improved K-means clustering algorithm, which can converge quickly and has high timeliness.

[0024] 3. The present invention assigns different initial target position values ​​to different targets, and when the target leaves the detection range, the dynamic value of its position decreases over time, thereby solving the problem of lack of consideration of target importance and target position value in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention is a flowchart of a large-scale group target clustering method integrating dynamic value in an embodiment of the present invention.

[0026] Figure 2 It is the global value table in the embodiment of the present invention.

[0027] Figure 3 Schematic diagram of the target value change rule in the embodiment of the present invention.

[0028] Figure 4 Schematic diagram of the effect of a large-scale group target clustering method integrating dynamic value in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the accompanying drawings to clearly and completely describe the concept, technical solution advantages and technical effects of the present invention to fully understand the purpose, features and effects of the present invention. It should be noted that the specific implementation methods described herein are only used to explain the present invention and do not limit the present invention.

[0030] A large-scale group target clustering method integrating dynamic value is proposed. The method first constructs a group target composed of several independent individual units, establishes a global value table and is jointly maintained and updated by a detector network; then initializes the cluster center to prepare data for the iteration of the improved K-means clustering algorithm; then divides the target into subgroups corresponding to the cluster center based on the distance, and proceeds to the next step when there are targets in all subgroups; based on the dynamic value, the improved K-means clustering algorithm is used to calculate the location of the cluster center; judges whether the loop end condition is reached, that is, the cluster center no longer changes or the maximum number of iterations is reached, and outputs the result if the end condition is reached.

[0031] Specifically, if Figure 1 As shown, the method comprises the following steps:

[0032] Step 1: Construct a group target consisting of several independent individual units

[0033] The target group is expressed as T = {T 1 , …, T n}, each individual unit has a different initial value T init ={T init_1 , …, T init_n}.

[0034] The group target is composed of several individual units, each of which has independent, non-interfering, dynamic properties and can be detected by the detector network.

[0035] Step 2: The detector network jointly maintains a global value table

[0036] The detector network can be expressed as D = {D 1 , …, D m}, and each detector in the detection network has a fixed detection range R = {R 1 , …, R m The global value table stores the ID, location, and value information of each user, which can be expressed as: V = {(T 1_ID , T 1_POS , T 1_VAL ),…,(T n_ID , T n_POS , T n_VAL )},like Figure 2 shown.

[0037] Step 3: Dynamically update group target information based on the detection information of the detector network

[0038] For all targets, when target T i Located at any detector D i Within the detection range R i When the global value table V contains information about the target, T i_POS Update to the current position of the target, T i_VAL Update to T init_i When the target T i Not in any detector D i Within the detection range R i When the value is T i_VAL According to f(T init_i , T t_i ) results are updated. Figure 3 As shown, the value change function f(T inii_i , T t_i ), can be defined according to actual needs. t_i The target is not in any detector D i Within the detection range R i The location information is still the original information in the global value table T i_POS When a new target is found, after the original global value table,

[0039] Added new target information, and updated the global value table to:

[0040] V={(T 1_ID , T 1_POS , T 1_VAL ),…,(T n_ID , T n_POS , T n_VAL ), (T n+1_ID , T n+1_POS , T n+1_VAL )}.

[0041] The detector network consists of multiple detectors, which can share information through wireless networks. Each detector can obtain target information within a certain detection range.

[0042] Step 4: Based on the dynamic value information in the global value table, the improved K-means clustering algorithm is used to cluster large-scale groups with dynamic values.

[0043] For the updated global value table V, the number of initial cluster centers is m. The positions of the initial cluster centers are randomly selected or selected according to the rules, and the result is POS = {POS 1 ,…,POS m}, calculate the distance from each target in the target set T to each point of the cluster center POS, select the nearest cluster center as the classification result, and form a sub-cluster set G = {G 1 , …, G m}, where G i The target location and value information belonging to sub-cluster i is stored in G. i If there is no target in the POS i Randomly select or select a new location according to the rules, and re-execute the above process until there is a target in each subcluster. For the problem of updating the location of the cluster center of each subcluster, use the following formula to update:

[0044]

[0045] T i ∈G i

[0046] For two-dimensional data or high-dimensional data, the data dimension can be increased or decreased according to the above method. Then update the cluster center position set POS and repeat step 4 until the classification result does not change (such as Figure 4 ) or the maximum number of iterations is reached.

[0047] In summary, the method of the present invention models a group composed of multiple independent target units, establishes a global value table and a value update algorithm for each target in the group, and uses the detection set to update the information in the global value table in real time. Based on the information in the global value table, an improved K-means clustering algorithm is used to cluster large-scale groups that integrate dynamic values. Based on the improved K-means clustering algorithm, the present invention proposes a solution that meets practical requirements for the problem of large-scale group target clustering, which solves the problem that the current method lacks practicality and consideration of target value for the problem of large-scale group target clustering.

Claims

1. A large-scale group target clustering method integrating dynamic value, characterized in that: The following steps are involved: (1) constructing a group goal, wherein the group goal is composed of multiple independent individual units; (2) construct a global value table jointly maintained by the detector network; The detector network is composed of multiple detectors, and the detectors share information through a wireless network. Each detector obtains target information of individual units within its own detection range; the global value table contains the target number, target value, and target position of all individual unit targets; (3) Use the detector network to detect individual units in the group target and dynamically update the global value table; the specific method is: when an individual unit target enters the detection range of a detector in the detector network, its target value is set to a fixed value, and the detector network updates the target position of the individual unit in real time; if an individual unit target exceeds the detection range of the detector network, the target position of the individual unit is set to the position where it was last detected, and the target value of the individual unit decreases over time; when the detector network discovers a new target, the new target information is added to the original global value table; (4) Based on the dynamic value information in the global value table, the improved K-means clustering method is used to cluster large-scale groups that integrate dynamic values; the specific method is as follows: (401) The number of initial cluster centers is set to m, and the positions of the initial cluster centers are selected randomly or according to a preset rule; (402) According to the distance relationship between the group target and the initial cluster center, the group target is divided into m sub-groups. If there is no target in a sub-group, the initial cluster center corresponding to the sub-group is reselected, and the sub-group targets belonging to each cluster center are recalculated until the number of targets in all sub-groups is not 0; (403) Read the global value table, fuse the target position and the target value and perform K-means clustering by weight, calculate the cluster center of each subgroup, and make the cluster center of the subgroup close to the target with high value in the subgroup; (404) Re-dividing subgroups with new cluster centers; (405) Repeat steps (403) and (404) until the subgroup divisions no longer change or the maximum number of iterations is reached, and clustering is completed.