Cooperative artificial potential field-based unmanned aerial vehicle cluster conflict detection and safety performance evaluation method

Abstract

The invention discloses an unmanned aerial vehicle cluster conflict detection and safety performance evaluation method based on a cooperative artificial potential field. The method comprises the steps of 1, establishing an unmanned aerial vehicle cluster complex network model; 2, unmanned aerial vehicle conflict detection is carried out based on a cooperative artificial potential field method; 3, updating the flight state of the unmanned aerial vehicle; 4, establishing an unmanned aerial vehicle cluster safety evaluation index system; 5, calculating a comprehensive safety evaluation value based on a principal component analysis method; 6, determining a high-risk unmanned aerial vehicle threshold value; and 7, identifying the high-risk unmanned aerial vehicle. According to the method, the problem of large-scale unmanned aerial vehicle cluster conflict detection is solved, and the flight safety performance of the unmanned aerial vehicle cluster is effectively evaluated in real time. Unmanned aerial vehicle cluster conflict detection based on a cooperative artificial potential field method is provided; establishing an unmanned aerial vehicle cluster security evaluation index system according to conflict characteristics and complex network characteristics among unmanned aerial vehicles; and high-risk unmanned aerial vehicles in the cluster are identified by using a principal component analysis method, so that the comprehensive safety quantitative evaluation of the cluster is realized.

Description

Technical Field

[0001] This invention is a method for detecting and assessing the safety performance of drone swarms based on cooperative artificial potential fields, belonging to the field of drone technology assessment. Background Technology

[0002] With the continuous development of navigation, control, and communication technologies, unmanned aerial vehicles (UAVs) have been widely used in both civilian and military fields, including environmental monitoring, agricultural plant protection, and disaster relief. The increasingly complex working environment and the ever-growing scale of UAV swarms pose new challenges to ensuring airspace safety, particularly in how to effectively detect conflicts and assess the safety of UAV swarms.

[0003] In recent years, research methods for UAV conflict detection have been broadly categorized into three types: geometric optimization, intelligent optimization, and Artificial Potential Field (APF). Geometric optimization utilizes the geometric relationship between the aircraft and obstacles to detect conflicts and derive the optimal solution strategy, but it typically involves high computational costs. Heuristic intelligent algorithms, such as genetic algorithms, particle swarm optimization, and pigeon flocking optimization, simulate biological evolution and search for optimal values ​​in the solution space using stochastic methods. Heuristic intelligent algorithms can solve path planning problems in more complex environments, but their high computational complexity makes real-time detection difficult. The APF proposes a virtual force theory, assuming that the UAV moves under the influence of target attraction and obstacle repulsion, ensuring a safe distance between the UAV and obstacles. The APF exhibits good effectiveness and robustness, but as a distributed method, as environmental complexity increases, frequent information exchange between UAVs becomes necessary, making it difficult to guarantee algorithm timeliness.

[0004] The artificial potential field method was first used for robot collision avoidance, establishing a global potential field by calculating obstacles in the system. Once the path's starting and ending points and obstacle positions are determined, a potential field map is constructed around the moving object, simulating potential energy mechanisms found in nature. Moving objects in the environment are considered particles moving within the artificial potential field. This virtual force field consists of attractive forces towards the target point and repulsive forces away from obstacles. The potential field function is typically a gradient function; the negative gradient force of the potential field acts as the guiding force on the moving object. The resultant force of the attractive and repulsive forces constitutes the total guiding force, driving the object to avoid obstacles and move towards the target.

[0005] The traditional artificial potential field method has two main problems that need to be optimized. First, drones are prone to getting stuck in a local equilibrium point between the target's gravity and the obstacle's repulsive force, making it impossible to reach the target point. Second, the repulsive force of the obstacle on the drone only considers the distance factor, but in reality, it should also be adjusted according to the drone's risk level.

[0006] Besides effective drone swarm conflict detection technologies, swarm security assessment methods are also crucial for ensuring the safe flight of drone swarms. Most existing drone security assessment methods assume that drones do not take collision avoidance maneuvers and only consider the conflict characteristics of two-drone systems. For large-scale drone swarms, their security situation is not only affected by local conflicts between any two drones, but also by the globally relevant characteristics of complex networks.

[0007] In summary, UAV swarm conflict detection and security assessment in complex environments is an important research direction. This invention proposes a conflict detection method based on a cooperative artificial potential field, considering the relative motion relationships and priorities among UAVs. Furthermore, based on the conflict characteristics between two UAVs, security assessment indicators are extracted and analyzed. For large-scale UAV swarms, complex network modeling is employed to extract security assessment indicators from the network structure. Finally, principal component analysis is used to identify high-risk UAVs within the swarm, achieving a comprehensive security assessment of the UAV swarm.

[0008] This invention enables real-time and effective inter-drone conflict detection, and establishes a security assessment index system for UAV swarms based on both local conflict characteristics and swarm network structure, identifying high-risk UAVs in the swarm, while also having the advantages of low computational cost. Summary of the Invention

[0009] This invention proposes a method for UAV swarm conflict detection and safety performance evaluation based on cooperative artificial potential field to solve the problem of large-scale UAV swarm conflict and to conduct real-time and effective evaluation of swarm flight safety performance, thereby improving the reliability of UAV swarms and laying the foundation for the development of UAV swarm evaluation technology.

[0010] To address the problem of UAV swarm conflict detection and safety assessment, this invention proposes a method for UAV swarm conflict detection and safety performance assessment. The flowchart of this method is shown below. Figure 1 As shown, the specific implementation steps are as follows:

[0011] Step 1: Establish a complex network model for drone swarms

[0012] A complex network model for unmanned aerial vehicle (UAV) swarms is G = (NOD, E), where node NOD = {nod} i |i=1,2,…,n} represents a drone, and the connected edge E={e i {1, 2, ..., m} represent the data exchange links between drones. When the distance between two drones in the cluster is less than the drone detection threshold R, a connected edge exists between them. The detection threshold is typically chosen as the detection distance of the drone's onboard radar. The adjacency matrix of this network is l = {l...} ij} n×n .

[0013] The complex network structure of UAV swarms is shown in Figures 2(a), 2(b), and 2(c). The arrows indicate the direction of flight speed, the radius of the circles represents the detection distance, and the dashed lines between UAVs represent data mitigation links. In Figure 2(a), the UAVs are evenly distributed with no connection between any two, representing a conflict-free swarm network structure. In Figure 2(b), the UAVs are partially connected, forming two interconnected networks with low aggregation and minimal swarm conflict. In Figure 2(c), the UAVs are fully linked, forming a fully connected network with severe swarm conflict.

[0014] Step 2: Detect UAV conflicts based on cooperative artificial potential fields

[0015] In the traditional artificial potential field method, let the position vectors of the UAV and the target be x and x', respectively. g Classical attraction potential energy function U att for:

[0016]

[0017] Where, k a This is the attraction coefficient.

[0018] By calculating the gradient of the attractive potential function generated by the target on the UAV, the attractive force F can be calculated. att for:

[0019]

[0020] For drones that need to avoid obstacles, a repulsive potential field U is established. rep Its effective range is shown in formula (3). When the distance d between the obstacle and the UAV is greater than a certain value, it can be considered to have no effect.

[0021]

[0022] Where, k r d is the repulsive force coefficient, and d0 is the effective distance of the repulsive potential field. That is, if the distance d between the obstacle and the drone is greater than d0, the obstacle will not generate a repulsive force on the drone.

[0023] Repulsive force F rep For the gradient of the repulsive potential field:

[0024]

[0025] in, Let d be the partial derivative with respect to x, representing the rate of change of d in the x-direction.

[0026] Resultant force F sum From the following formula, we can obtain:

[0027] F sum =F att -F rep (5)

[0028] Based on the traditional artificial potential field method, an approach rate (AR) factor for the UAV is added:

[0029] AR ij =cos(α) ij V i +cos(α ji V j (5)

[0030]

[0031] Where, χ i Let angle be the heading angle of drone i. ij Let α be the azimuth angle of UAV j relative to i. ij Let α be the acute angle formed between the heading of drone i and the line connecting the two drones. Similarly, we can obtain α. ji The calculation formula is:

[0032]

[0033] Where, χ j Let angle be the heading angle of the drone j. ji Let α be the azimuth angle of UAV i relative to j. ji Let be the acute angle formed between the heading of drone j and the line connecting the two drones.

[0034] A high approach rate implies a high risk of collision; when drones are at equal distances, drones with high approach rates should experience greater repulsive forces. Therefore, considering the approach rate factor, the improved formulas for calculating the attractive, repulsive, and resultant forces are as follows:

[0035]

[0036]

[0037] in, Let d be the distance between UAV i and the target point. ij Let i be the distance between drone i and drone j. Let k be the azimuth angle of UAV i relative to the target point. AR l is the proximity coefficient. ij For the elements in the adjacency matrix (when l ij When = 1, the distance between drones i and j is less than the threshold R; when l ijWhen AR = 0, the distance between drones i and j exceeds the threshold R. Therefore, when AR is positive, the repulsive force increases, and vice versa.

[0038] Furthermore, when two conflicting drones share the same target point, they become trapped in a state of mutual attraction and repulsion, creating a local minimum point from which they can never reach their target. This phenomenon is known as the "conflict dilemma." To solve this problem, the two conflicting drones can cooperate. Before resolving the conflict, the right-of-way of one drone is prioritized. The drone with priority does not need to take collision avoidance maneuvers; only the other drones need to adjust their speed and heading. This effectively reduces computational load and drone maneuverability.

[0039] Priority is given to the drones that are closer to the intersection of the original flight paths, and the α of each group of conflicting drones is calculated according to formulas (6) and (7). ij With α ji Thus, the priority matrix M is obtained:

[0040]

[0041] The formula for calculating repulsive force has been improved to:

[0042]

[0043] Step 3: Update the drone's flight status

[0044] The position of UAV i at time (k+1) is calculated by the following kinematic equations:

[0045] X i (k+1)=X i (k)+V i (k)Δt (13)

[0046] Where t is the step time, X i With V i Let be the position and velocity vectors of UAV i, respectively.

[0047] The calculation of the heading angle of UAV i at time (k+1) can be transformed into the heading deflection steer. i calculate:

[0048]

[0049] χ i (k+1)=χ i (k)+steer i (16)

[0050] Where, ω maxThis represents the maximum turning rate. Due to the inherent performance limitations of the UAV, the angle deflection should not exceed the product of the maximum turning rate and the step size. i ∈(-ω max Δt,ω max Δt).

[0051] The velocity of drone i at time (k+1) is calculated by the following formula:

[0052]

[0053] in, χ represents the weight of drone i. i (k+1) is the heading angle of UAV i at time (k+1), a max V is the maximum acceleration of the drone. max This is the maximum speed of the drone.

[0054] Step 4: Establish a security assessment index system for drone swarms

[0055] The aforementioned drone swarm security assessment index system includes the approach rate index (ARI), drone potential energy index (EU), maximum potential energy index (MEU), repulsive force index (RF), maximum repulsive force index (MRF), node degree index (DG), average strength index (AS), and average clustering coefficient index (ACC).

[0056] (1) Approach Rate Index (ARI): The higher the approach rate between two drones, the higher the risk of conflict. The approach rate assessment index for drone i is the maximum approach rate between drone i and surrounding drones:

[0057] ARI i =max j (AR ij (18)

[0058] AR ij =cos(α) ij V i +cos(α ji V j (19)

[0059]

[0060] Where, χ i Let angle be the heading angle of drone i. ij Let χ be the azimuth angle of UAV j relative to UAV i. j Let angle be the heading angle of the drone j. ji Let be the azimuth angle of UAV i relative to UAV j.

[0061] (2) Potential Energy Index (EU) of UAV: ​​The potential energy generated by all UAVs within the detection range of UAV i is the potential energy of UAV i. i That is, the sum of the potential energy generated by the surrounding drones on it:

[0062]

[0063] Among them, E ij EU represents the potential energy generated by drone j on drone i. i This represents the energy of drone i, and num represents the total number of drones within drone i's detection range.

[0064] (3) Maximum potential energy index (MEU): When EU i =EU j To differentiate conflict risks, a maximum potential energy value index is introduced. When the maximum potential energy value of UAV i is greater than that of UAV j, i.e., when the maximum potential energy value of MEU is greater... i MEU j At that time, the risk of conflict for drone i is greater than that for drone j.

[0065] MEU i =max j (E ij ) (twenty four)

[0066] (4) Repulsion Force Index (RF): The greater the repulsion force experienced by the UAV, the higher its risk of conflict. The formula for calculating the repulsion force index RF of UAV i is as follows:

[0067]

[0068] (5) Maximum Repulsive Force Index (MRF): Similar to the maximum potential energy index MEU, the maximum repulsive force index is defined as the maximum repulsive force when RF... i =RF j At that time, the maximum repulsive force is greater, and the risk of conflict is higher:

[0069] MRF i =max j (rf ij (27)

[0070] (6) Degree (DG): The degree of drone i is the number of connected edges connected to it. The weighted degree of drone i is the sum of the weights of the connected edges connected to it. Generally, for each pair of drones, the closer they are, the higher the risk of collision. Therefore, the reciprocal of the distance between two drones is used as the weight of the connected edges. Drones with higher degrees interact more with other drones, resulting in a greater safety risk.

[0071] k i =∑ j l ij(28)

[0072]

[0073] Where, ω ij Let ω be the weight of the edge connecting drone i to drone j. ij =1 / d ij k i and These represent the degree and weighted degree of drone i, respectively.

[0074] (7) Average Strength Index (AS): Average Strength Indicates the proximity of drone i to other drones:

[0075]

[0076] (8) Average Clustering Coefficient (ACC): Complex networks exhibit obvious clustering characteristics, and the clustering coefficient c ω (i) The degree of interconnectivity between adjacent drones was quantified.

[0077]

[0078] Among them, c ω (i) represents the clustering coefficient of drone i. The average clustering coefficient of the network. Quantify the density of all drones.

[0079]

[0080] in, N is the number of drones in the network, when The drones in the network are isolated from each other, so there is no risk of conflict. There is at least one connecting edge between the two drones in the network, which poses a high risk of conflict.

[0081] Step 5: Calculate the comprehensive safety assessment value based on principal component analysis.

[0082] To achieve rapid processing of evaluation data and analyze the structural relationships among various evaluation indicators, principal component analysis is used. Through linear transformation, the original high-dimensional evaluation data is combined into a new set of uncorrelated comprehensive evaluation values. The original evaluation data matrix can be represented as X = (x ab ) n×p ,a=1,2,…,n; b=1,2,…,p,x ab Let represent the value of the b-th evaluation indicator for the a-th sample, where n is the total number of samples and p is the total number of evaluation indicators. The original evaluation data is then standardized.

[0083]

[0084] The correlation matrix R of the original evaluation data matrix is \(r_{bk}\) bk ) p×p , where \(b = 1, 2, \ldots, p\); \(k = 1, 2, \ldots, p\), and \(r_{bk}\) bk is the correlation coefficient between evaluation index \(b\) and evaluation index \(k\):

[0085]

[0086] where \(r_{aa}\) bb = 1, \(r_{bk}\) bk = \(r_{kb}\), \(a = 1, 2, \ldots, n\); \(b = 1, 2, \ldots, p\); \(k = 1, 2, \ldots, p\).

[0087] From the characteristic equation \(|\lambda\) p - R| = 0, \(p\) characteristic roots \(\lambda_g\) g (\(g = 1, 2, \ldots, p\)) are obtained. Sort \(\lambda\) from largest to smallest as \(\lambda_1\geq\lambda_2\geq\cdots\geq\lambda\) p \(\geq0\), which is the variance of the principal component. The magnitude of its value indicates the size of the role played by each principal component in describing the evaluated object. Each characteristic root corresponds to a characteristic vector \(L_g\) g = (\(l_{g1}\) g1 , \(l_{g2}\) g2 , \(\ldots\), \(l_{gp}\) gp ), \(g = 1, 2, \ldots, p\).

[0088] Convert the standardized evaluation index variables into principal components:

[0089] SAF_g g = \(l_{g1}z_1 + l_{g2}z_2+\cdots + l_{gp}z_p\) g1 + \(l_{g2}z_2+\cdots + l_{gp}z_p\) g2 + \(\cdots + l_{gp}z_p\) gp z_p p , \(g = 1, 2, \ldots, p\) (37)

[0090] where \(SAF_g\) p becomes the \(p\)th principal component. The principal component analysis method selects as few as \(k\) principal components (\(k < p\)) for comprehensive evaluation. The value of \(k\) is determined by the variance contribution rate and is generally taken as 0.85, 0.9 or 0.95.

[0091] The comprehensive security evaluation value is:

[0092]

[0093] Step 6: Determine the threshold of high-risk UAVs

[0094] ​​The residual spatial statistic Squared Prediction Error (SPE) and Hotelling's statistic (T) 2 This can be used to determine whether data is abnormal. The SPE and T are calculated using the principal component eigenvectors. 2 :

[0095] SPE = ||xx r || 2 (39)

[0096] T 2 =x T SAFΛ -1 SAF T x (40)

[0097] Where SAF is the principal component matrix, Λ is the diagonal matrix of eigenvalues ​​of the principal components, and x is the original evaluation value. t This is the evaluation value reconstructed using principal component analysis.

[0098] SPE and T 2 The threshold is defined as:

[0099]

[0100]

[0101] in, c represents the eigenvalues ​​of the covariance matrix. μ To satisfy the confidence limit of the standard normal distribution, 1-μ is the confidence level, F μ (δ,τ-δ) is the F-distribution with δ as the first degree of freedom and τ-δ as the second degree of freedom.

[0102] Step 7: Identify high-risk drones

[0103] When a certain drone i satisfies or SPE i SPE μ If a drone is detected at a certain time, it can be identified as a high-risk drone. The number of high-risk drones in a drone swarm reflects the airspace safety situation.

[0104] This invention presents a method for UAV swarm conflict detection and safety performance assessment based on cooperative artificial potential fields. Its advantages and effectiveness lie in: improving the repulsive force calculation of the traditional artificial potential field method by considering the relative motion relationships between UAVs, and effectively solving the local minimum problem of the traditional artificial potential field method by setting conflict priorities, thus reducing the number of UAVs in the swarm that need to perform maneuvers. Furthermore, by analyzing the micro-conflict characteristics of UAVs and the complex network characteristics of macro-UAVs, a safety assessment index is proposed. Principal component analysis is used to comprehensively evaluate all assessment indexes to identify high-risk UAVs in the swarm, and the number of high-risk UAVs is used to measure the swarm's flight safety. Attached Figure Description

[0105] Figure 1 A flowchart for drone swarm conflict detection and safety performance assessment.

[0106] Figures 2(a), 2(b), and 2(c) are schematic diagrams of complex network structures of unmanned aerial vehicle (UAV) swarms.

[0107] Figure 3 This is a complex network diagram of a drone swarm.

[0108] Figure 4 The graph shows the numerical curve of the squared prediction error.

[0109] Figure 5 This is a graph of the numerical values ​​of the Hotelling statistic.

[0110] The labels and symbols in the diagram are explained as follows:

[0111] x — the horizontal axis of the coordinate system;

[0112] y—the vertical axis of the coordinate system. Detailed Implementation

[0113] The effectiveness of the proposed method is verified below through a specific example of drone swarm security assessment. The experimental computer was configured with an Intel Core i7-8750H processor, 2.20GHz clock speed, 16GB of memory, and MATLAB 2020a software.

[0114] The specific steps of this method are as follows:

[0115] Step 1: Establish a complex network model for unmanned aerial vehicles (UAVs)

[0116] Within a 2000×2000 meter area, 80 drones are randomly and uniformly generated. The drone velocities are randomly generated and uniformly distributed across the range [0, 20 m / s²]. The drone accelerations are within the range [0, 10 m / s²]. 2 Randomly generated within the range, the drone's turning speed is [0, π / 18] rad / s, and the exploration threshold R is 200 meters.

[0117] Complex network diagram of drone swarms as follows Figure 3 As shown, each drone is abstracted as a node. When the distance between drones is less than the detection distance R, there is a data exchange link, which is abstracted as a connected edge.

[0118] Step 2: Detect UAV conflicts based on cooperative artificial potential fields

[0119] (1) Determine the priority among conflicting drones

[0120] For every two conflicting drones, to ensure the appearance of the "conflict dilemma" in the line graph, α is calculated for each group of conflicting drones according to formulas (6) and (7). ij With α ji Priority is given to drones that are closer to the intersection of the original flight path, and priority matrix M is obtained according to formula (11).

[0121] (2) Calculate the forces acting on the UAV in the cooperative artificial potential field.

[0122] The approach speed of each UAV is calculated according to formulas (5)-(7). Considering the approach speed factor, the attractive and repulsive forces experienced by the UAV in the artificial potential field method are calculated according to formulas (8) and (12), respectively, where the attractive force coefficient k a =5, repulsive force coefficient k r =15, close to the rate coefficient k AR =10, the effective distance of the repulsive potential field is d0 = 200 meters, and the potential field force on the UAV is calculated according to formula (10).

[0123] Step 3: Update the drone's flight status

[0124] The position of UAV i at time (k+1) is calculated by formula (13). The step size is selected as Δt = 0.5 seconds. The heading angle calculation can be converted into heading deflection according to formulas (14)-(16). i The velocity of UAV i at time (k+1) is calculated using formula (17).

[0125] Step 4: Establish a security assessment index system for drone swarms

[0126] Based on the UAV flight data, calculate the values ​​of various safety assessment indicators. Calculate the UAV approach rate index (ARI) using formulas (18)-(21). The higher the approach rate between two UAVs, the higher the risk of conflict. Calculate the UAV potential energy index (EU) using formulas (22) and (23). When the potential energies of two UAVs are equal, calculate the UAV maximum potential energy index (MEU) using formula (24) to differentiate the risk of conflict. Calculate the UAV repulsion force index (RF) using formulas (25) and (26). The greater the repulsion force experienced by a UAV, the higher its risk of conflict. When the repulsion force assessment index values ​​of two UAVs are equal, calculate the repulsion force index to differentiate the risk of conflict. To assess conflict risk, the maximum repulsion force index (MRF) is calculated according to formula (27). Drones with a greater maximum repulsion force have a higher conflict risk. The node degree index (DG) is calculated according to formulas (28) and (29) with the reciprocal of the distance between two drones as the weight of the connected edge. Drones with a higher degree mean that they interact more with other drones and have a greater safety risk. The average strength index (AS) of drones is calculated according to formula (30). To quantify the interconnection degree between adjacent drones, the average clustering coefficient (ACC) of nodes is calculated according to formulas (31) and (32).

[0127] Step 5: Calculate the comprehensive safety assessment value based on principal component analysis.

[0128] In this example, there are 80 drones and 8 evaluation metrics, i.e., n=80, p=8. The original data matrix can be represented as X=(x ab ) 80×8 a = 1, 2, ..., 80; b = 1, 2, ..., 8. The original evaluation data are standardized according to formulas (33)-(35).

[0129] The correlation matrix of the evaluation data matrix is ​​R = (r bk ) 8×8 b = 1, 2, ..., 8; k = 1, 2, ..., 8, calculate r according to formula (36) bk .

[0130] From the characteristic equation |λ p -R|=0, yielding 8 eigenvalues ​​λ. g (g = 1, 2, ..., 8), sort λ ​​from largest to smallest as λ1 ≥ λ2 ≥ ... ≥ λ8 ≥ 0, and calculate the eigenvector L corresponding to each eigenvalue. g .

[0131] According to formula (37), the standardized evaluation index variables are converted into principal components. When the value is 0.85, select the number of principal components; finally, calculate the comprehensive safety assessment score according to formula (38).

[0132] Step Six: Determine the Threshold for High-Risk Drones

[0133] Selecting μ = 0.7, the squared prediction error threshold SPE is obtained according to formulas (41)-(45). μ =0.13, threshold of the Hotelling statistic The squared prediction error of the residual space statistic SPE and the Hotling statistic T are calculated according to formulas (39) and (40). 2 , respectively Figure 4 , Figure 5 As shown in the figure, the dashed lines represent the statistical thresholds.

[0134] Step 7: Identify high-risk drones

[0135] When drone i satisfies or SPE i SPE μ If a drone is identified as high-risk, it is considered a high-risk drone; otherwise, it is considered a low-risk drone. Based on the squared prediction error and Hotelling statistic calculated in step six, the number of high-risk drones in the cluster is 39, accounting for 48.75% of the total number of drones, indicating a significantly higher risk of conflict.

Claims

1. A method for UAV swarm conflict detection and security performance assessment based on cooperative artificial potential fields, characterized in that: The steps are as follows: Step 1: Establish a complex network model for drone swarms Step 2: Detect UAV conflicts based on cooperative artificial potential fields Based on the traditional artificial potential field method, an approach rate influence factor for the UAV is added: AR ij =cos(α ij )V i +cos(α ji )V j Where, χ i Let angle be the heading angle of drone i. ij Let α be the azimuth angle of UAV j relative to i. ij Let α be the acute angle formed between the heading of UAV i and the line connecting the two UAVs; similarly, we can obtain α. ji The calculation formula is: Where, χ j Let angle be the heading angle of the drone j. ij Let α be the azimuth angle of UAV i relative to j. ji Let J be the acute angle formed between the heading of UAV j and the line connecting the two UAVs. Considering the approach rate factor, the improved formulas for calculating the attractive force, repulsive force, and resultant force are as follows: in, Let d be the distance between UAV i and the target point. ij Let i be the distance between drone i and drone j. Let k be the azimuth angle of UAV i relative to the target point. AR l is the proximity coefficient. ij These are the elements in the adjacency matrix; when AR is positive, the repulsive force increases, and vice versa. Step 3: Update the drone's flight status The position of UAV i at time (k+1) is calculated by the following kinematic equations: X i (k+1)=X i (k)+V i (k)Δt (13) Where t is the step time, X i With V i These are the position and velocity vectors of drone i, respectively; The calculation of the heading angle of UAV i at time (k+1) is transformed into the heading deflection steer. i calculate: x i (k+1)=x i (k)+steer i (16) Where, ω max The maximum turning rate; due to the inherent performance limitations of the UAV, the angle deflection should not exceed the product of the maximum turning rate and the step size. i ∈(-ω max Δt,ω max Δt); The velocity of drone i at time (k+1) is calculated by the following formula: in, χ represents the weight of drone i. i (k+1) is the heading angle of UAV i at time (k+1), a max V is the maximum acceleration of the drone. max This is the maximum speed of the drone; Step 4: Establish a security assessment index system for drone swarms The aforementioned drone swarm security assessment index system includes the approach rate index (ARI), drone potential energy index (EU), maximum potential energy index (MEU), repulsive force index (RF), maximum repulsive force index (MRF), node degree index (DG), average strength index (AS), and average clustering coefficient index (ACC). Step 5: Calculate the comprehensive safety assessment value based on principal component analysis. To achieve rapid processing of evaluation data and analyze the structural relationships among various evaluation indicators, principal component analysis is used. Through linear transformation, the original high-dimensional evaluation data is combined into a new set of uncorrelated comprehensive evaluation values. The original evaluation data matrix is ​​represented as X = (x ab ) n×p ,a=1,2,…,n; b=1,2,…,p,x ab This represents the value of the b-th evaluation indicator for the a-th sample, where n is the total number of samples and p is the total number of evaluation indicators; the original evaluation data is standardized as follows: The correlation matrix R of the original evaluation data matrix is ​​(r bk ) p×p ,b=1,2,…,p;k=1,2,…,p,r bk The correlation coefficient between evaluation index b and evaluation index k: Where, r bb =1,r bk =r kb ,a=1,2,…,n; b=1,2,…,p; k=1,2,…,p; From the characteristic equation |λ p -R|=0, thus obtaining p eigenvalues ​​λ g (g = 1, 2, ..., p), sort λ ​​from largest to smallest as λ1 ≥ λ2 ≥ ... ≥ λ p ≥0 represents the variance of the principal components, and its value indicates the extent to which each principal component contributes to describing the object being evaluated; each eigenvalue corresponds to an eigenvector L. g =(l g1 ,l g2 ,…,l gp ), g = 1, 2, ..., p; Convert standardized evaluation index variables into principal components: SAF g =l g1 z1+l g2 z2+…+l gp z p ,g=1,2,…,p Among them, SAF p becomes the p-th principal component; the principal component analysis method selects k principal components (k < p) as few as possible for comprehensive evaluation, and the value of k is determined by the variance contribution rate and takes values of 0.85, 0.9 or 0.95; The overall security assessment value is: Step Six: Determine the Threshold for High-Risk Drones Calculate the squared prediction error (SPE) and Hotelling statistic (T) using principal component eigenvectors. 2 : SPE=||x-x r || 2 (39) T 2 =x T SAFΛ -1 SAF T x (40) Where SAF is the principal component matrix, Λ is the diagonal matrix of eigenvalues ​​of the principal components, and x is the original evaluation value. r The evaluation value is the result of reconstruction using principal component analysis. SPE and T 2 The threshold is defined as: in, c represents the eigenvalues ​​of the covariance matrix. μ To satisfy the confidence limit of the standard normal distribution, 1-μ is the confidence level, F μ (δ,τ-δ) is the F-distribution with δ as the first degree of freedom and τ-δ as the second degree of freedom; Step 7: Identify high-risk drones When a certain drone i satisfies or SPE i SPE μ When a drone is identified as a high-risk drone, the presence of a high-risk drone in the drone swarm reflects the airspace safety situation.

2. The method according to claim 1, characterized in that: Step two further includes: when two conflicting drones have the same target point, they will fall into a state of attraction and repulsion, thus generating a local minimum point, and will never be able to reach the target. This phenomenon is called the "conflict dilemma". To solve this problem, the two conflicting drones cooperate, and the right-of-way of one drone is given priority before the conflict is resolved. The drone with priority does not need to take collision avoidance maneuvers. Only the other drones need to adjust their speed and heading, thereby effectively reducing the amount of computation and drone maneuvers. Prioritize drones that are closer to the intersection of their original flight paths, and calculate α for each group of conflicting drones. ij With α ji Thus, the priority matrix M is obtained: The formula for calculating repulsive force is then improved to:

3. The method according to claim 1, characterized in that: The specific process of step four is as follows: (1) Approach Rate Index (ARI): The higher the approach rate between two UAVs, the higher the risk of conflict; the approach rate evaluation index for UAV i is the maximum approach rate between UAV i and surrounding UAVs: AR ij =cos(α ij )V i +cos(α ji )V j Where, χ i Let angle be the heading angle of drone i. ij Let χ be the azimuth angle of UAV j relative to UAV i. j Let angle be the heading angle of the drone j. ji Let be the azimuth angle of UAV i relative to UAV j; (2) Potential Energy Index (EU) of UAV: ​​The potential energy generated by all UAVs within the detection range of UAV i is equal to the sum of the potential energy generated by the UAVs surrounding UAV i. Among them, E ij EU represents the potential energy generated by drone j on drone i. i This represents the energy of drone i, and num is the total number of drones within drone i's detection range. (3) Maximum potential energy index MEU: when EU i =EU j To differentiate conflict risks, a maximum potential energy value index is introduced; when the maximum potential energy value of UAV i is greater than that of UAV j, i.e., when the maximum potential energy value of MEU is greater than that of UAV j, the maximum potential energy value of UAV i is greater than that of UAV j. i MEU j At that time, the risk of conflict for drone i is greater than that for drone j; (4) Repulsion force index RF: The greater the repulsion force experienced by the UAV, the higher its risk of conflict; the formula for calculating the repulsion force index RF of UAV i is as follows: (5) Maximum Repulsive Force Index (MRF): Similar to the maximum potential energy index (MEU), the maximum repulsive force index is defined as follows: when RF... i =RF j At that time, the maximum repulsive force is greater, and the risk of conflict is higher: (6) Node degree index DG: The degree of drone i is the number of connected edges connected to it; the weighted degree of drone i is the sum of the weights of the connected edges connected to it; the reciprocal of the distance between two drones is used as the weight of the connected edges; the higher the degree of the drone, the more it interacts with other drones, and the greater the security risk. Where, ω ij Let ω be the weight of the edge connecting drone i to drone j. ij =1 / d ij , k i and These are the degree and weighted degree of drone i, respectively; (7) Average strength index AS: average strength Indicates the proximity of drone i to other drones: (8) Average Clustering Coefficient (ACC): Complex networks exhibit obvious clustering characteristics, and the clustering coefficient (ACC) is high. ω (i) The degree of interconnectivity between adjacent drones was quantified; Among them, c ω (i) represents the clustering coefficient of drone i; the network average clustering coefficient. Quantify the density of all drones; in, N is the number of drones in the network, when The drones in the network are isolated from each other, eliminating the risk of conflict. There is at least one connecting edge between the two drones in the network, which poses a high risk of conflict.

Images