A method and system for evaluating the importance of individual drone swarms based on behavior trajectories

Abstract

The present invention discloses a method and system for evaluating the importance of individuals in a drone cluster based on behavioral trajectories. The method comprises: S1, constructing a time series network under multiple thresholds according to the trajectory behavior data of the drone cluster, and calculating the degree of being surrounded by each individual in the drone cluster; S2, calculating the degree of fluctuation of the individual in the drone cluster during this period of time by the difference of the motion state of the individuals in the drone cluster; S3, constructing a Bayesian network model by combining the degree of being surrounded by each individual in the drone cluster and the degree of fluctuation of the individuals in the drone cluster as features, and using the model to infer whether the individual is a key individual in the drone cluster. The present invention utilizes the motion state of a self-organizing cluster over a period of time, and measures the dynamic correlation between cluster individuals with collaborative characteristics through the motion state characteristics such as surrounding and volatility reflected in the behavioral trajectory, so as to achieve rapid inversion of the internal mechanism of the self-organizing cluster system.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles (UAVs). More specifically, it particularly relates to a method and system for evaluating the importance of individual UAVs in a UAV swarm based on behavioral trajectories. Background Art

[0002] A single UAV has characteristics such as small size, low cost, and strong mobility. However, it also has problems such as poor flexibility and low task execution efficiency. With the continuous improvement of the autonomous intelligence of UAVs, swarm UAV operations are considered an important development trend in the future unmanned combat system. Swarm UAV operations have the advantages of mutual cooperation and complementary advantages, which can significantly improve the success rate of task execution. At the same time, they also pose a huge challenge to traditional air defense systems.

[0003] Existing anti-UAV swarm technologies mainly originate from technologies for single UAVs. However, due to the large number of UAVs in a swarm, traditional methods often cannot attack each UAV within a limited anti-aircraft time. The key to the flexible flight and intelligence of swarm UAVs lies in the information interaction between individuals and the establishment of a communication network. Therefore, a feasible method is to analyze and identify the observation data of the swarm to determine the key individuals in the internal communication structure of the swarm, so as to achieve precise strikes on swarm UAVs. During this strike process, the connectivity of the communication network of the swarm is damaged, and thus the swarm loses its collaborative ability.

[0004] Research on the identification of key individuals in a swarm mainly focuses on natural swarms at present. For example, by long-term tracking of groups such as gray wolves and baboons, analyzing their individual leadership behaviors, and using quantitative methods of statistical physics to infer the leadership structure and key individuals of large-scale swarms such as pigeons and fish schools. Although these works discuss the influence of leader individuals on the entire group in different natural swarms, for unknown swarms where only the motion state information is known, there is still a lack of clear methods to identify key individuals. With the development of complex network science, research on the identification of key nodes in complex systems has emerged one after another, such as methods for ranking the importance of nodes based on neighbors and paths. However, these methods are all based on the known system network topology, and there are not many methods to infer the importance of individuals only from the behavioral data of system individuals. Therefore, it is necessary to develop a method and system for evaluating the importance of individual UAVs in a UAV swarm based on behavioral trajectories. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for evaluating the importance of individual UAVs in a UAV swarm based on behavioral trajectories to overcome the defects of the prior art.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for evaluating the importance of individuals in an unmanned aerial vehicle (UAV) cluster based on behavioral trajectories, comprising the following steps:

[0008] S1. Construct time series networks under multiple thresholds based on the trajectory behavior data of the UAV cluster, and calculate the degree of being surrounded around each individual in the UAV cluster;

[0009] S2. Calculate the degree of fluctuation of an individual within a period of time through the difference in the motion states of individuals in the UAV cluster;

[0010] S3. Combine the degree of being surrounded around each individual in the UAV cluster and the degree of fluctuation of individuals in the UAV cluster as features to construct a Bayesian network model, and use this model to infer whether the individual is a key individual in the UAV cluster.

[0011] Further, the step S1 specifically includes:

[0012] S11. Collect the motion trajectory data of the UAV cluster, preprocess the motion trajectory data to obtain trajectory behavior data with timestamps, and segment the trajectory behavior data into segments of length L where t (t = 1, 2,..., L) represents the index of the time slice;

[0013] S12. Establish a distance matrix based on the motion state according to the observed values of N UAV individuals at time t (t = 1, 2,..., L) where N represents the number of UAV individuals in each time slice, represents the multi-dimensional motion state of UAV individual i observed at time t;

[0014] S13. Take the threshold ∈, and use the Heaviside function and Kronecker symbol to transform the distance matrix D at time t above t into the adjacency matrix G of a specific network t,∈ , where

[0015] S14. Calculate the number of connected individuals around individual i in the time series network at threshold ∈

[0016] S15. Take the maximum and minimum values of the distance matrix D t as the upper and lower bounds of the threshold ∈ respectively, and integrate ∈ and t in the number of connected individuals around individual i respectively to obtain the degree of being surrounded c of individual i i :

[0017]

[0018] Among them, the upper and lower bounds of ∈ and are both normalized;

[0019] S16. Obtain a vector C = {c 1 , c 2 ,..., c N} T that reflects the surroundingness of each individual in the UAV swarm, and record this vector as the degree of being surrounded around each individual. The larger the value of c i , the more important the individual is in the UAV swarm.

[0020] Furthermore, the specific steps of step S2 include:

[0021] S21. Represent the trajectory behavior data as where are L observations of the i-th individual in the swarm changing with time, and represents the multi-dimensional motion state of the UAV individual i observed at time t;

[0022] S22. Calculate the change in the motion state at adjacent times through the difference method

[0023]

[0024] S23. Combine the change in the motion state at each time to obtain the volatility value R of the motion state of individual i i :

[0025]

[0026] S24. Positive normalize the volatility vector R = {R 1 , R 2 ,..., R N} to obtain the volatility index F i :

[0027]

[0028] In the formula, max{R} and min{R} are the maximum and minimum values among the elements of the volatility vector respectively.

[0029] The present invention also provides a system according to the above-mentioned method for evaluating the importance of individuals in a UAV swarm based on behavioral trajectories, including:

[0030] A surroundingness calculation module, configured to construct a time-series network under multiple thresholds based on the trajectory behavior data of the UAV swarm and calculate the degree of being surrounded around each individual in the UAV swarm;

[0031] A fluctuation degree calculation module, which is used to calculate the fluctuation degree of an individual within a certain period of time by taking the difference of the motion states of individuals in the UAV cluster;

[0032] A Bayesian network model, which is used to construct a Bayesian network model by combining the degree of being surrounded around each individual in the UAV cluster and the fluctuation degree of individuals in the UAV cluster as features, and use this model to infer whether the individual is a key individual in the UAV cluster.

[0033] Compared with the prior art, the advantages of the present invention are as follows: Based on the surroundability and volatility of the cluster motion state, the present invention infers the key individuals in the self-organizing cluster. This process only requires the observation data of the cluster, does not depend on a specific cluster type, and does not require any prior information, so it has a wide range of applications. As the amount of cluster data increases, the surroundability and volatility indexes calculated by using the motion state are regarded as behavioral trajectory features respectively, and two different motion state features are combined for comprehensive inference by constructing a Bayesian inference network. And a distance matrix is constructed on the time slice of the cluster time series, and this distance matrix is used to represent the motion state distance between individuals. Then, by thresholding the distance matrix, the specific network structure at this moment can be obtained. The operation of constructing a network on the time slice can not only make full use of the collected information, but also reduce the influence of outliers in the time series on the inference result. Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0035] Figure 1 It is a flowchart of the method for evaluating the importance of UAV cluster individuals based on behavioral trajectories of the present invention.

[0036] Figure 2 It is a schematic diagram of the inference result of key individuals based on the surroundability index of different cluster models of the present invention.

[0037] Figure 3 It is a schematic diagram of the inference result of key individuals based on the volatility index of different cluster models of the present invention.

[0038] Figure 4 It is a schematic diagram of the inference result of key individuals based on different indexes of different cluster models of the present invention.

[0039] Figure 5 It is a framework diagram of the system for evaluating the importance of UAV cluster individuals based on behavioral trajectories of the present invention. Detailed Embodiments

[0040] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0041] The technical solution of the present invention is as follows: First, use devices such as radar to collect the motion trajectories of the UAV swarm, and preprocess the data to obtain trajectory data with timestamps, that is, a set of time series. The time series is segmented into segments of length L and expressed as where t (t = 1, 2,..., L) represents the index of the time slice, is the observation value of N UAV individuals at time t, N represents the number of UAV individuals in each time slice, represents the multi-dimensional motion state (such as position, speed or heading angle) of UAV individual i observed at time t. In this process, the stage where the motion state tends to be stable is removed. Secondly, based on the collected UAV swarm trajectory data, a multi-threshold time series network is constructed, and the surrounding index (Surrounding) of different individuals in the swarm is calculated through the change degree of the number of network edges under different thresholds. The difference calculation is performed on the trajectory of the motion state of the individuals in the swarm, that is, the change amount at adjacent times is calculated. The change amount is integrated to obtain an index reflecting the fluctuation (Fluctuation) of the motion state. Finally, the surrounding and fluctuation indexes obtained from the motion state are regarded as the characteristics of the individual respectively, and it is inferred whether the individual is a key individual in the swarm system through the Bayesian network inference framework.

[0042] Combined with Figure 1 shown in, this embodiment provides a method for evaluating the importance of UAV swarm individuals based on behavioral trajectories, including the following steps:

[0043] Step S1, construct a time series network under multiple thresholds according to the trajectory behavior data of the UAV swarm, and calculate the degree of being surrounded by each individual in the UAV swarm, specifically including the following steps:

[0044] Step S11, collect the motion trajectory data of the UAV swarm, preprocess the motion trajectory data to obtain trajectory behavior data with timestamps, and segment the trajectory behavior data into segments of length L In the formula, t (t = 1, 2,..., L) represents the index of the time slice.

[0045] Step S12, establish a distance matrix based on the motion state according to the observation values of N UAV individuals at time t In the formula, N represents the number of UAV individuals in each time slice, Denote the multi - dimensional motion state of the UAV individual \(i\) observed at time \(t\).

[0046] In this embodiment, when calculating the surrounding index, the surrounding degree between individuals is measured by constructing a distance matrix of the motion state. In this process, the most basic Euclidean distance is used. Of course, other distance metrics can also be selected, such as Manhattan distance, Chebyshev distance, or Minkowski distance.

[0047] Step S13: Take a threshold \(\epsilon\), and use the Heaviside function and Kronecker symbol to transform the distance matrix \(D\) at time \(t\) above t into the adjacency matrix \(G\) of a specific network t,∈ , where

[0048] Step S14: Calculate the number of connected individuals around the individual \(i\) in the temporal network at time \(t\) under the threshold \(\epsilon\)

[0049] Step S15: In order to more comprehensively measure the surrounded degree of an individual, take the maximum and minimum values of the distance matrix \(D\) t as the upper and lower bounds of the threshold \(\epsilon\) respectively, and integrate \(\epsilon\) and \(t\) in the number of connected individuals around the individual \(i\) respectively to obtain the surrounded degree \(c\) of the individual \(i\) i :

[0050]

[0051] where the upper and lower bounds of \(\epsilon\) and are both normalized;

[0052] Step S16: Finally, obtain the vector \(C=\{c\) 1 , \(c\) 2 , \(\cdots\), \(c\) N \}\) T reflecting the surrounding degree of each individual in the UAV cluster, and denote this vector as the surrounded degree around each individual. The larger the value of \(c\) i , the more important the individual is in the UAV cluster. The experimental results of identifying key individuals in different cluster models are as Figure 1 shown.

[0053] Step S2: Calculate the fluctuation degree of the individual within this period of time by taking the difference of the motion states of individuals in the UAV cluster, which specifically includes the following steps:

[0054] Step S21. Different from calculating from the cross-sectional perspective of the time series in Step S1, the differential index focuses on the motion state of a single individual. Therefore, it is necessary to reorganize the cluster data. Represent the trajectory behavior data as where are L observations of the i-th individual in the cluster changing over time, represents the multi-dimensional motion state (such as position, velocity or heading angle) of the UAV individual i observed at time t.

[0055] Step S22. Calculate the change amount D of the motion state at adjacent times by the difference method i t :

[0056]

[0057] The larger the value, the greater the change in the motion state of this individual at this time.

[0058] Step S23. Combine the change amounts of the motion state at each time to obtain the volatility value R of the motion state of individual i i :

[0059]

[0060] Step S24. Different from the surrounding index (Surrounding) being a maximization index, while the above volatility value is a minimization index. For the convenience of display and calculation, the volatility vector R = {R 1 , R 2 ,..., R N} is normalized to obtain the volatility index F i :

[0061]

[0062] In the formula, max{R} and min{R} are respectively the maximum and minimum values in the elements of the volatility vector. Based on the identification experiment results of key individuals in different cluster models as Figure 2 shown.

[0063] Step S3. As the amount of collected data increases, combine the degree of being surrounded by each individual in the UAV cluster and the degree of fluctuation of the individuals in the UAV cluster as features to construct a Bayesian network model, and use this model to infer whether this individual is a key individual in the UAV cluster. As Figure 1 shown, make a more comprehensive speculation. The comparison results of separately using the surrounding (Surrounding), fluctuation (Fluctuation) indexes and comprehensive inference using the Bayesian network are as Figure 4 shown.

[0064] In this embodiment, when using a Bayesian network for comprehensive inference, the metrics of surroundness and volatility are adopted as features. However, other motion characteristics reflecting the differences between leaders and followers can also be mined from the behavioral data of the cluster. For example, followers show lag in motion state relative to leaders. All motion characteristic metrics calculated based on the behavioral data can be used as features of the Bayesian inference network to comprehensively infer the roles of individuals in the cluster.

[0065] Combined with Figure 5 As shown, the present invention also provides a system for evaluating the importance of individuals in a UAV cluster based on the above-mentioned behavior trajectory. The system includes: a surroundness calculation module 1, configured to construct a time-series network under multiple thresholds according to the trajectory behavior data of the UAV cluster and calculate the degree of surroundness around each individual in the UAV cluster; a volatility calculation module 2, configured to calculate the volatility of an individual during this period of time through the difference in the motion states of individuals in the UAV cluster; a Bayesian network model 3, configured to construct a Bayesian network model by combining the degree of surroundness around each individual in the UAV cluster and the volatility of the individuals in the UAV cluster as features, and use the model to infer whether the individual is a key individual in the UAV cluster.

[0066] Based on the surroundness and volatility of the cluster motion state, the present invention infers the key individuals in the self-organizing cluster. This process only requires the observed data of the cluster, does not depend on a specific cluster type, and does not require any prior information, with a wide range of applications. As the amount of cluster data increases, the metrics of surroundness and volatility calculated using the motion state are regarded as behavioral trajectory features respectively. By constructing a Bayesian inference network, the two different motion state features are combined for comprehensive inference, and a distance matrix is constructed on the time slice of the cluster time series. The distance matrix is used to represent the motion state distance between individuals. Then, by thresholding the distance matrix, the specific network structure at this moment can be obtained. The operation of constructing a network on the time slice can not only make full use of the collected information, but also reduce the influence of outliers in the time series on the inference result.

[0067] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, the patent owner can make various deformations or modifications within the scope of the appended claims. As long as it does not exceed the protection scope described in the claims of the present invention, it should be within the protection scope of the present invention.

Claims

1. A method for evaluating the importance of individual drone clusters based on behavior trajectories, characterized in that: The following steps are involved: S1. Construct a time series network under multiple thresholds based on the trajectory behavior data of the drone cluster, and calculate the degree of surrounding around each individual in the drone cluster; S2, by differentiating the motion states of individuals in the drone cluster, calculate the fluctuation degree of the individual during this period; S3. Combining the degree of surrounding of each individual in the drone cluster and the degree of fluctuation of individuals in the drone cluster as features, a Bayesian network model is constructed, and the model is used to infer whether the individual is a key individual in the drone cluster; The step S1 specifically includes: S11, collect the motion trajectory data of the drone cluster, pre-process the motion trajectory data, obtain the trajectory behavior data with timestamps, and divide the trajectory behavior data into segments of length L Where t represents the index of the time slice, t = 1, 2, ..., L; S12, based on the observation values ​​of N drone individuals at time t Establish a distance matrix based on motion status In the formula, N represents the number of drone individuals in each time slice, represents the multi-dimensional motion state of drone individual i observed at time t, t = 1, 2, ..., L; S13, take the threshold ∈, use the Heaviside function and Kronecker symbol to transform the distance matrix D at time t t Transformed into a specific network adjacency matrix G t，∈ ,in, S14. Calculate the number of connected individuals around individual i in the temporal network at time t under the threshold ∈ S15. Get the distance matrix D t The maximum and minimum values ​​of are used as the upper and lower bounds of the threshold ∈, respectively, and the number of connected individuals around individual i is integrated, and the degree of surrounding of individual i is obtained. i : Among them, the upper and lower bounds of ∈ and All were normalized; S16, obtain the vector C that reflects the individual surrounding properties of the drone cluster = {c1, c2, ..., c N } T , and record the vector as the degree of surrounding each individual, c i Larger values ​​indicate that the individual is more important in the drone swarm; The step S2 specifically includes: S21, the trajectory behavior data is expressed as in, are the L observations of the i-th individual in the cluster over time, represents the multi-dimensional motion state of drone individual i observed at time t; S22. Calculate the change of motion state at adjacent moments by difference method S23, combine the changes in the motion state at each moment to obtain the volatility value R of the motion state of individual i i : S24, the volatility vector R={R1,R2,...,R N } Positive transformation to obtain volatility index F i : Where max{R} and min{R} are the maximum and minimum values ​​of the volatility vector elements, respectively.

2. The system of the method for evaluating the importance of individual drone clusters based on behavior trajectories according to claim 1 is characterized in that: include: The surrounding degree calculation module is used to construct a time series network under multiple thresholds based on the trajectory behavior data of the drone cluster, and calculate the degree of surrounding around each individual in the drone cluster; The fluctuation degree calculation module is used to calculate the fluctuation degree of an individual during this period of time by taking the difference of the motion state of an individual in the drone cluster; The Bayesian network model is used to construct a Bayesian network model by combining the degree of surrounding of each individual in the drone cluster and the degree of fluctuation of individuals in the drone cluster as features, and use the model to infer whether the individual is a key individual in the drone cluster.

Images