A multi-objective clustering method based on evolutionary clustering

Abstract

The invention discloses a multi-target clustering method based on evolutionary clustering. According to the three-dimensional spatial position coordinates, speed, heading, enemy and friend attributes, military and civilian attributes, and type factors of the targets, a sample set of targets to be clustered is constructed. The three-dimensional spatial position coordinates are used as the measurement subject, and the speed, heading, enemy and friend attributes, military and civilian attributes, and type are used as penalty coefficients. An initial distance measurement function between targets is defined, a multi-factor distance measurement matrix is ​​generated, a time smoothing factor is set, the current calculation result is corrected with the historical calculation result of the distance measurement function, a final distance measurement function is defined, the multi-factor distance measurement matrix is ​​updated, a distance threshold is preset, and the distance between two targets is calculated. If the distance is less than the threshold, the targets are classified into the same group until the clustering of all targets is completed.

Description

Technical Field

[0001] The present invention belongs to the technical field of clustering algorithms, and in particular relates to a group classification technology. Background Art

[0002] Battlefield situation generation is to extract key information from massive battlefield information and analyze and infer it, judge the enemy's intentions and estimate its threat level, thereby reducing the commander's cognitive pressure on the battlefield situation and improving the efficiency of command decision-making.

[0003] Multi-target grouping is an important part of battlefield situation generation. Dividing multiple targets with the same type attributes, similar motion states or consistent combat missions into the same group can effectively simplify the battlefield situation map, which is of great significance for further analyzing enemy intentions and quantifying enemy threats.

[0004] Cluster analysis is the main means of multi-target clustering. Commonly used methods include the nearest neighbor algorithm, K-means algorithm, fuzzy C-means algorithm and density clustering algorithm. The clustering results of the K-means algorithm and the fuzzy C-means algorithm generally depend on the predetermined number of clusters and the selection of the initial cluster center. The target clustering effect is poor in the absence of prior information. The density clustering algorithm has advantages in processing data with uniform density distribution. It is suitable for solving the problem of clustering a large number of dense samples, but is not suitable for the actual battlefield environment of grouped coordinated operations. The nearest neighbor algorithm divides targets with similar spatial positions into the same group by pre-setting thresholds. The principle is simple and easy to implement, and it is widely used in engineering.

[0005] In the context of future joint operations, the battlefield environment is becoming more and more complex. The surge in the number of targets on both sides and the changes in combat styles have posed higher challenges to the battlefield situation generation multi-target grouping technology.

[0006] When the existing technology uses the nearest neighbor algorithm to perform multi-target clustering, it only considers the target position relationship, uses the three-dimensional space distance as the distance measurement function, and ignores the similarity between the target category, attribute and motion state. The existing technology performs nearest neighbor clustering based on the motion state of the target at the current moment, without considering the impact of the clustering results at the historical moment, resulting in the target clustering results not matching the actual battlefield environment.

[0007] For example, after a group relationship is established between multiple targets, if a target performs a combat mission at a certain moment, causing the distance between it and other targets to exceed the threshold in a short period of time, the group relationship will not be maintained. If the movement trajectories of two independent targets intersect, and the position distance between the targets at a certain moment is less than the preset threshold, the two targets will be mistakenly classified into the same group.

[0008] In view of the above problems, a simple and efficient clustering algorithm is urgently needed to realize multi-target clustering technology in complex battlefield environments. This technology is based on the multi-dimensional distance measurement criterion, so that the clustering result at a certain moment is as consistent as possible with the battlefield target situation at the current moment, and the difference with the clustering result at the previous moment is small. After using this technology, the accuracy and stability of battlefield target clustering results can be improved, laying the foundation for group target intention reasoning and threat estimation. Summary of the invention

[0009] In order to solve the problems existing in the prior art, the present invention proposes a multi-objective clustering method based on evolutionary clustering. In order to achieve the above purpose, the present invention adopts the following technical solutions.

[0010] Step 1: Construct a sample set of targets to be grouped based on the target's three-dimensional spatial position coordinates, speed, heading, enemy attributes, military and civilian attributes, and type factors;

[0011] Step 2: Take the three-dimensional spatial position coordinates as the measurement subject, and the speed, heading, enemy and friend attributes, military and civilian attributes, and type as the penalty coefficients, define the initial distance measurement function between targets, and generate a multi-factor distance measurement matrix;

[0012] Step 3: Set the time smoothing factor, use the historical calculation results of the distance measurement function to correct the current calculation results, define the final distance measurement function, and update the multi-factor distance measurement matrix;

[0013] Step 4: Preset the distance threshold and calculate the distance between two targets. If the distance is less than the threshold, they are grouped into the same group until all targets are grouped.

[0014] In step 1, set V i k , t k The three-dimensional spatial position coordinates, speed, and heading of the i-th target at the moment, For the enemy attribute, It is a military and civilian property. If type, then among N targets, t k The i-th target at time is

[0015]

[0016] Furthermore, in the enemy and friend attributes, 0 represents enemy, 1 represents friend, 2 represents friend, and 3 represents unknown; in the military and civilian attributes, 0 represents military, 1 represents civilian, and 2 represents unknown; in the type, 0 represents air, 1 represents surface, 2 represents underwater, and 3 represents unknown.

[0017] In step 2, set tk The distance between the i-th target and the j-th target in the three-dimensional rectangular coordinate system at the moment is The speed and heading penalty functions are The enemy and friend attributes, military and civilian attributes, and type penalty functions are Then the initial distance measurement function between targets is

[0018] If the speeds of the two targets are equal, their headings, enemy attributes, military and civilian attributes, and types are the same, then the penalty coefficients are 1 respectively; otherwise, they are greater than 1 and increase as the penalty factor increases.

[0019] In step 3, set t k-1 Two goals at a time The initial distance metric function between k ∈[0,1] is the time smoothing factor, then the final distance metric function is If α k =1, the impact of historical distance measurement on the clustering results is not considered. k =0, then keep the grouping results of the historical moment, if 0<α k <1, the clustering results at historical moments are considered and the influence of the distance measurement at the current moment on the clustering results is balanced.

[0020] The beneficial effects of the present invention are as follows: drawing on the idea of ​​evolutionary clustering algorithm, based on the nearest neighbor criterion, battlefield target clustering is realized, and by defining a new distance measurement function, the accuracy and stability of clustering results are effectively improved, with low implementation complexity and high engineering application value; when calculating the distance measurement between each target, instead of simply calculating the three-dimensional spatial position distance between the targets, the influence of target speed, heading, enemy and friend attributes, military and civilian attributes, and type factors are considered, and a corresponding penalty function is introduced to correct the distance measurement, and the penalty factor of each factor can be adjusted according to the specific application scenario, so that the target clustering result is more accurate; when performing target clustering, historical information is fully utilized, not only the distance measurement between the current moment targets is considered, but also the influence of the distance measurement between the previous moment targets is introduced, and by adjusting the time smoothing factor, the influence of the target state at the current moment and the historical moment on the clustering result is balanced, so that the target clustering result is more in line with the actual battlefield environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the motion trajectory of multiple formations, Figure 2 is the existing technology clustering result of the target trajectory intersection area, Figure 3 This is the result of existing technology grouping where the formation changes and the spacing is unstable. Figure 4 is the clustering result of the present invention in the target track intersection area, Figure 5It is the grouping result of the present invention that the formation changes and the spacing is unstable. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0023] Suppose two enemy helicopter formations are moving towards one of our helicopter formations. When they encounter each other, the enemy formation turns back and our formation pursues.

[0024] The following table gives the basic information of the multi-party formation, and the motion trajectory simulation is as follows: Figure 1 shown.

[0025]

[0026]

[0027] Based on the nearest neighbor algorithm of evolutionary clustering, battlefield targets are grouped. k , construct t k The target sample set at the moment Target number N = 8

[0028]

[0029] Speed ​​penalty factor α in multi-factor distance metric function V =1, heading penalty factor α D =1, the penalty factor α for enemy and friend attributes, military and civilian attributes, and target type T1 =α T2 =α T3 =1, time smoothing factor α k =0.5, distance measurement threshold η=3.

[0030] Step 1: For the data in the sample set, calculate the time-smoothed multi-factor distance between the targets to form an N×N dimensional distance measurement matrix Dis. Since the distance matrix is ​​symmetric, only the upper triangular matrix needs to be calculated in practical applications.

[0031]

[0032] Step 2: Compare the distance values ​​of the two targets in the distance measurement matrix with the distance threshold in turn. If the distance is less than the threshold, they are divided into the same group and the group number is set.

[0033] Step 2.1: Initialize the distance metric matrix Dis, the target group number i_Id = 0 (1≤i≤8), the number of groups Num = 0;

[0034] Step 2.2: Select the initial target i=1; if i_Id=0, that is, the target is not classified into any group, then set it as an independent group target, Num=Num+1, i_Id=Num;

[0035] Step 2.3: Select target j=i+1. If Dis(i,j)<η, divide it into the same group as the i-th target, j_Id=i_Id, j=j+1 until all (Ni) targets are traversed.

[0036] Step 2.4: Let i=i+1, repeat steps 2.2-2.4 until i=N, and all target grouping is completed.

[0037] Step 3: Count the target grouping results based on the number of groups and the sequence number of each target group and output them.

[0038] When the distance between the formations is large and stable, the existing technology can achieve a good grouping effect, but in areas where the target trajectories intersect, such as Figure 2 As shown, or when the formation changes within the formation cause the distance between targets to be unstable, such as Figure 3 As shown, errors occurred in the prior art.

[0039] The present invention takes into account multiple factors such as target type, attribute, speed, heading, and the influence of the grouping result at the previous moment, effectively improving the accuracy and stability of the grouping result. Figure 4 and 5 shown.

[0040] The above are embodiments of the present invention and do not limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims

1. A multi-objective clustering method based on evolutionary clustering, characterized in that: include: Step 1: Construct a sample set of targets to be grouped based on the target's three-dimensional spatial position coordinates, speed, heading, enemy attributes, military and civilian attributes, and type factors, including: V i k , t k The three-dimensional spatial position coordinates, speed, and heading of the i-th target at the moment, For the enemy attribute, It is a military and civilian property. If type, then among N targets, t k The i-th target at time is Step 2: Take the three-dimensional spatial position coordinates as the measurement subject, and the speed, heading, enemy and friend attributes, military and civilian attributes, and type as the penalty coefficients, define the initial distance measurement function between targets, and generate a multi-factor distance measurement matrix, including: t k At the moment, the distance between the i-th target and the j-th target in the three-dimensional rectangular coordinate system is The speed and heading penalty functions are The enemy and friend attributes, military and civilian attributes, and type penalty functions are Then the initial distance measurement function between targets is If the speeds of the two targets are equal, their headings, enemy attributes, military and civilian attributes, and types are the same, then the penalty coefficients are 1 respectively. Otherwise, they are greater than 1 and increase as the penalty factor increases. Step 3: Set the time smoothing factor, use the historical calculation results of the distance measurement function to correct the current calculation results, define the final distance measurement function, and update the multi-factor distance measurement matrix, including: t k-1 Two goals at a time The initial distance metric function between k ∈[0,1] is the time smoothing factor, then the final distance metric function is If α k =1, the impact of historical distance measurement on the clustering results is not considered. k =0, then keep the grouping results of the historical moment, if 0<α k <1, the clustering results at historical moments are considered and the influence of the distance measurement at the current moment on the clustering results is balanced; Step 4: Preset the distance threshold and calculate the distance between two targets. If the distance is less than the threshold, they are grouped into the same group until all targets are grouped.

2. The multi-objective clustering method based on evolutionary clustering according to claim 1, characterized in that: Said enemy and friend attributes, 0 represents enemy, 1 represents us, 2 represents friend, and 3 represents unknown; the military and civilian attributes are 0 for military, 1 for civilian, and 2 for unknown; the type is 0 for air, 1 for surface, 2 for underwater, and 3 for unknown.

Images