Unmanned aerial vehicle cluster hunting attack method, product, medium and equipment

Abstract

The invention discloses an unmanned aerial vehicle cluster hunting attack method, a product, a medium and equipment, and relates to the field of unmanned aerial vehicle cluster game. A pure pursuit strategy or a surrounding pursuit strategy is determined to be adopted according to the maximum speed relation between the unmanned aerial vehicle and the target; when a surrounding pursuit strategy is adopted, determining the number of the surrounding unmanned aerial vehicles based on the positions and speeds of the unmanned aerial vehicles and the target, and constructing an Apollonius circle surrounding model; the speed of the trapper unmanned aerial vehicle is decomposed into a surrounding direction and a pursuit direction, and a surrounding pursuit strategy is divided into a surrounding forming stage and a situation maintaining stage; in the surrounding forming stage, the multiple trapper unmanned aerial vehicles are controlled to move in the surrounding direction; in the situation maintenance stage, whether cooperative attacks are carried out is determined according to the distance between the trapper unmanned aerial vehicle and the target, unmanned aerial vehicle advantage situation speed selection is carried out by using a speed obstacle method, and multi-angle saturation attacks of the trapper unmanned aerial vehicle on the target are realized through deep reinforcement learning training. By adopting the method, the target hunting success rate can be improved.

Description

Technical Field

[0001] The present invention relates to the field of drone cluster game technology, and in particular to a drone cluster encirclement and attack method, product, medium and equipment. Background Art

[0002] Drones have the characteristics of light weight, good safety and low cost. With the development of artificial intelligence technology and computer technology, the advantages of drone swarm technology are becoming increasingly apparent. Drone swarm hunting refers to the cooperative execution of encirclement missions by drone swarms in the air, which is an important application of drone swarm combat technology.

[0003] At present, there are mainly two methods based on kinematics and dynamics in the field of drone swarm capture. Among them, the kinematic method is relatively simple but difficult to handle complex situations. Most of them focus on formation changes and the formation of capture situations, and rarely consider close-range confrontation with the target. The dynamic method requires the establishment of a more complex mathematical model, with a large number of parameters and difficult adjustment and optimization. In addition, there is little research on the situation of high-speed targets. Summary of the invention

[0004] The purpose of the present invention is to provide a drone cluster capture and attack method, product, medium and equipment, which is aimed at studying targets with high-speed performance and considering close-range confrontation with the target until the target is eliminated, so as to improve the success rate of the capture.

[0005] To achieve the above object, the present invention provides the following solution.

[0006] On the one hand, the present invention provides a method for attacking a drone cluster, comprising:

[0007] Get the position and speed of the drone and the target;

[0008] Determine different capture strategies based on the maximum speed relationship between the UAV and the target; the capture strategies include pure pursuit strategy and encirclement pursuit strategy;

[0009] When the encirclement and pursuit strategy is adopted, the number of the encircler drones is determined based on the position and speed of the drone and the target, and the Apollonius circle encirclement model is constructed;

[0010] The speed of the hunter drone is decomposed into two directions: encirclement and pursuit, and the encirclement and pursuit strategy is divided into two stages: encirclement formation and situation maintenance;

[0011] In the encirclement formation stage, multiple encircler drones are controlled to move along the encirclement direction based on the Apollonius circle encirclement model to form an encirclement situation around the target;

[0012] Calculate the total occupancy angle of multiple hunter drones, and determine whether to enter the situation maintenance phase based on the total occupancy angle;

[0013] In the situation maintenance phase, whether to conduct coordinated attack with the hunter drone is determined based on the distance between the hunter drone and the target;

[0014] When it is determined to conduct a coordinated attack by the hunter and the UAV, the speed obstacle method is used to select the speed of the UAV in an advantageous situation, and the speed set that can be reached in the hunter's advantageous situation is determined;

[0015] Based on the achievable speed set of the hunter's advantageous situation, deep reinforcement learning training is used to realize the multi-angle saturation attack of the hunter's drone on the target.

[0016] Optionally, the method of determining to adopt different capture strategies according to the maximum speed relationship between the drone and the target specifically includes:

[0017] When v max >v T max When , a UAV is selected to execute a pure pursuit strategy until the target is destroyed; max is the maximum speed of the drone; v T max is the target maximum speed;

[0018] When v max ≤v T max When attacking an enemy, select multiple drones to execute the encirclement and pursuit strategy.

[0019] Optionally, the determining the number of the capturer drones based on the positions and speeds of the drones and the target and constructing the Apollonius circle capture model specifically includes:

[0020] Using formula Calculate the maximum speed ratio between the drone and the target;

[0021] According to the maximum speed ratio λ, the formula Calculate the number of hunter drones n; [] means rounding up;

[0022] Using formula Calculate the speed ratio λ between the hunter drone i and the target T i ; where v i is the speed of the hunter drone i; v T is the speed of the target T;

[0023] According to the speed ratio λ i , construct the Apollonius circle capture model Where (x ig ,y ig ) is the coordinate of the capturer UAV i relative to the target T; C i is the Apollonius circle corresponding to the capture drone i; R i C i The radius of .

[0024] Optionally, the calculating of the total occupancy angle of the plurality of hunter drones and determining whether to enter the situation maintenance phase according to the total occupancy angle specifically includes:

[0025] Using the formula θ i =2sin -1 (λ) Calculate the occupancy angle of the hunter drone i;

[0026] Using formula Calculate the angle α of the hunter drone i relative to the target T i ;

[0027] According to θ i and α i , using the formula Calculate the overlap angle ε between the capture drone i and the capture drone i+1 i，i+1 ;

[0028] According to θ i and ε i，i+1 , using the formula Calculate the total occupancy angle θ of all the hunter drones G ;

[0029] When θ G Reach 360° and each Apollonius circle C i When they are tangent or overlapped, it means that the conditions for encirclement and capture are met and the situation maintenance stage begins.

[0030] Optionally, determining whether to perform a coordinated attack by the besieger drone according to the distance between the besieger drone and the target specifically includes:

[0031] Set the distance threshold D, when the distance D between the hunter drone i and the target T iT When it is less than the threshold D, a coordinated attack by the roundup drone is carried out.

[0032] Optionally, the method of selecting the speed of the drone in an advantageous situation by using the speed obstacle method and determining the set of speeds that can be reached in the advantageous situation of the hunter specifically includes:

[0033] Divide the target reachable area W into attack areas Z G,ATK , Invulnerable Area Z G,NATK , obstacle avoidance area Z G,OBS and non-obstacle avoidance area Z G,NOBS ;

[0034] Using formula Z P =WZ G,ATK -Z G,OBS Calculate the area of ​​advantage Z of the roundup drone P ;

[0035] According to the situation area Z P Determine the speed set that can be reached by the predator's advantageous situation

[0036] Optionally, the method of implementing a multi-angle saturation attack of the hunter drone on the target through deep reinforcement learning training based on the achievable speed set of the hunter's advantageous situation specifically includes:

[0037] Select the speed and heading pair with the shortest speed concentration time and closest to the current speed of the hunter UAV i as the guiding speed and heading pair (v guide ,h guide );

[0038] According to the guidance speed and heading (v guide ,h guide ) Design reward functions, including speed reward, enveloping angle reward, collision reward and attack distance reward;

[0039] Using reward function as evaluation index to train deep reinforcement learning of the hunter drone action;

[0040] The optimal action is selected through extensive training, and the desired speed and heading of the hunter drone are output;

[0041] Control the hunter drone according to the desired speed and heading to carry out multi-angle saturation attack on the target.

[0042] On the other hand, the present invention also provides a computer program product, including a computer program, which implements the drone cluster encirclement and attack method when executed by a processor.

[0043] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the drone cluster encirclement and attack method.

[0044] On the other hand, the present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the drone cluster encirclement and attack method.

[0045] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0046] On the one hand, the present invention adopts a hierarchical multi-situation cooperative encirclement algorithm based on kinematic modal analysis, constructs an encirclement model through the Apollonius circle, and divides the encirclement task into two stages: encirclement formation and encirclement situation maintenance. The corresponding optimal strategy is selected by analyzing the speed ratio and the stage. On the other hand, the present invention adopts a speed obstacle prediction and enhanced cooperative attack algorithm based on dynamic modal analysis, using the speed obstacle method to enable the drone to maintain a favorable attack situation; neural networks and reinforcement learning are used to optimize the attack strategy to achieve multi-angle saturation attack. Compared with the prior art, the present invention has a higher encirclement success rate than other methods, does not require drone formation, and the two modes can be switched. In addition, the drone attack success rate is high, the estimated cost is small, and the robustness is good. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of the drone swarm attack method provided by the present invention;

[0049] Figure 2 A schematic diagram of the Apollonius circle capture model according to an embodiment of the present invention;

[0050] Figure 3 This is a speed decomposition diagram of the hunter drone according to an embodiment of the present invention;

[0051] Figure 4 is a target escape direction diagram of an embodiment of the present invention;

[0052] Figure 5 This is an attack diagram of the UAV's advantageous situation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The present invention aims to provide a drone cluster capture and attack method, product, medium and equipment, which targets high-speed performance targets and considers close-range confrontation with targets until the targets are eliminated, so as to improve the capture success rate.

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] Figure 1 This is a flow chart of the drone swarm attack method provided by the present invention. Figure 1 The present invention provides a method for attacking a drone cluster, comprising:

[0057] Step 1: Get the position and speed of the drone and the target.

[0058] Initialize the capture scene. The scene has no boundaries and no obstacles except the drone and the target. The target in the scene is recorded as T, and the position p of the target T is obtained. T , speed v T and the maximum speed v T max . Let the drone in the scene be i, and get the position p of drone i i , speed v i and the maximum speed v max .

[0059] Step 2: Determine different capture strategies based on the maximum speed relationship between the UAV and the target; the capture strategies include pure pursuit strategy and encirclement pursuit strategy.

[0060] Determine the relationship between the maximum speed of the drone and the target, and select different capture strategies. When the maximum speed of the drone is higher than the maximum speed of the target, a pure pursuit strategy is implemented. When the maximum speed of the drone does not exceed the maximum speed of the target, a surround and pursue strategy is implemented, and the target is eventually destroyed. Specifically, when the maximum speed of the drone is faster than the maximum speed of the target, only one drone needs to execute a pure pursuit strategy to achieve the target attack; when the speed of the drone is slower, generally at least three drones are required to complete the capture, that is, when the maximum speed of the drone is less than or equal to the maximum speed of the target, at least three drones are required to capture the target.

[0061] Therefore, when v max >v T max When v max ≤v T max When , it is necessary to select n drones as the capturer drones (referred to as capturers) to execute the encirclement and pursuit strategy for the target.

[0062] Step 3: When the encirclement and pursuit strategy is adopted, the number of the encircler drones is determined based on the position and speed of the drone and the target, and the Apollonius circle encirclement model is constructed.

[0063] The maximum speeds of the UAV and the target are known, and the maximum speed ratio of the UAV to the target is calculated according to the following formula (1):

[0064]

[0065] According to the maximum speed ratio λ, the number of hunter drones n is calculated using formula (2):

[0066]

[0067] Where [] means rounding up.

[0068] Furthermore, the speed ratio λ between the hunter UAV i and the target T is calculated using formula (3): i :

[0069]

[0070] where v i is the speed of the hunter drone i; v T is the speed of the target T.

[0071] According to the speed ratio λ i , build Figure 2 The Apollonian circle capture model shown:

[0072]

[0073] Where (x ig ,y ig ) is the coordinate of the capturer UAV i relative to the target T; C i is the Apollonius circle corresponding to the capture drone i; R i C i The radius of Figure 2 As shown, the centers O1, O2, and O3 of the three circles represent the centers of the Apollo circles respectively, P1, P2, and P3 represent the positions of the three drones, T represents the target position, and the angles L1TL2, L2TL3, and L3TL2 represent the capture occupancy angles θ of the drones. Here, the three Apollo circles are tangent to each other, but they may be tangent or intersecting or not intersecting. When they intersect, there will be an overlapping angle, and when they do not intersect, there will be a gap angle.

[0074] Step 4: Decompose the speed of the hunter drone into two directions: encirclement and pursuit, and divide the encirclement and pursuit strategy into two stages: encirclement formation and situation maintenance.

[0075] The speed breakdown of the Stalker drone is shown below: Figure 3 As shown, it is divided into two directions: encirclement and pursuit. The encirclement mission is further divided into two stages: encirclement formation and situation maintenance.

[0076] like Figure 2As shown, the pursuit direction is from the hunter drone to the target, and the speed v of the hunter drone i i The velocity component in the pursuit direction is denoted by v ih ; The encirclement direction is perpendicular to the pursuit direction, and the velocity component in the encirclement direction is recorded as v is .

[0077] In the encirclement formation phase, the pure encirclement strategy is implemented, that is, the encircler drone moves along the encirclement direction. In the situation maintenance phase, the encirclement pursuit strategy is implemented, and the optimal pursuit angle of the drone is calculated through a nonlinear equation group. During the execution of the encirclement pursuit strategy, if the total occupation angle of n encircler drones is less than 320, the encirclement strategy is re-implemented.

[0078] Step 5: In the encirclement formation phase, based on the Apollonius circle encirclement model, multiple encircler drones are controlled to move along the encirclement direction to form an encirclement situation for the target.

[0079] In the encirclement formation phase, the pure encirclement strategy is implemented. It is necessary to control multiple encircler drones along the speed decomposition based on the Apollonius circle encirclement model. Figure 3 Move in the encirclement direction to form an encirclement situation for the target, such as Figure 4 shown. Figure 4 In the figure, u1, u2, and u3 represent three capture drones, T represents the target, the red arrow represents the target's movement direction, and the blue arrow represents the direction of the capture drone. Similar to an artificial potential field, the drone has a repulsive force on the target, and the drone closest to the target has an attractive force. The combined force of the drones is the target's movement direction.

[0080] The initial stage is generally the encirclement formation stage. In the encirclement formation stage, the pure encirclement strategy is implemented, and each encircler drone moves in the encirclement direction. The specific direction is determined by calculating the angle of the drone relative to the target and the overlap angle. The angle of the encircler drone i relative to the target T is α i , i=0, 1, 2...n, the larger i is, the larger the angle is, and counterclockwise is positive. The coverage angle between two adjacent drones is denoted as ε i，i+1 , greater than zero is the clearance angle, less than 0 is the overlap angle. When the overlap angle is less than 0, the encirclement direction of the small-numbered UAV i is away from the large-numbered UAV i+1. When the overlap angle is greater than 0, it is close to the large-numbered UAV i+1. Overlap angle ε i，i+1 The specific calculation is as follows.

[0081] First, use formula (5) to calculate the occupancy angle θ of a single hunter drone i: i :

[0082] θ i =2sin -1 (λ)(5)

[0083] Then, the angle α of the hunter UAV i relative to the target T is calculated using formula (6): i :

[0084]

[0085] According to θ i and α i , the overlap angle ε between the capturer drone i and the capturer drone i+1 can be calculated using formula (7): i，i+1 :

[0086]

[0087] As shown in formula (7), the overlap angle ε between two adjacent hunter drones is i，i+1 The calculation method is to take the angle α of each hunter relative to the target i Arrange the numbers from small to large, and the overlapping angle is the angle difference between the adjacent numbered hunters relative to the target minus half of the occupied angle sum.

[0088] Figure 3 Medium i,i+1 -ε i-1,i Represents the difference in overlap angles. If it is greater than zero, it means that the overlap angle between UAVs i and i+1 is smaller or the gap angle is larger than that between UAVs i and i-1. That is, in terms of polar angle relative to the target, UAV i is closer to UAV i-1, so UAV i chooses to move in the direction of i+1 (i.e. Figure 3 The pursuit direction is the direction of the target v ih .

[0089] Step 6: Calculate the total occupancy angle of multiple hunter drones, and determine whether to enter the situation maintenance phase based on the total occupancy angle.

[0090] According to θ i and ε i，i+1 , the total occupancy angle θ of all the hunter drones can be calculated using formula (8): G , when the total occupied angle reaches 360°, it is considered as encirclement and enters the situation maintenance stage. The calculation is as follows:

[0091]

[0092] When θ G Reach 360° and each Apollonius circle C i When they are tangent or overlapped, it means that the conditions for encirclement and capture are met and the situation maintenance stage begins.

[0093] In the method of the present invention, when the speed of the drone is faster than the target, a pure pursuit strategy is executed; when the speed of the drone is slower than the target, a mixed encirclement and pursuit strategy is executed. In the situation maintenance stage, the encirclement and pursuit strategy is executed, and the speed of the drone is divided into two parts: encirclement and pursuit. The encirclement is maintained and gradually approached until an attack is possible. In the situation maintenance stage, if the total occupied angle is too small (less than 320°), it is necessary to re-enter the encirclement formation stage to prevent the target from escaping.

[0094] Step 7: During the situation maintenance phase, determine whether to conduct a coordinated attack with the hunter drone based on the distance between the hunter drone and the target.

[0095] Both the drone and the target have a maximum attack distance limit. The attack range is a fan-shaped area with a span of 90 degrees centered on the flight direction. Set the distance threshold D, which is generally set slightly larger than the drone's attack range. Divide the drone's speed into two parts: encirclement and pursuit. The pursuit direction is as follows: Figure 3 As shown in the figure, the encirclement direction and encirclement formation stage are selected in the same way. By judging the distance between the drone and the target, when the distance D between the encircler drone i and the target T is iT When it is less than the threshold D, it switches to the close-range coordinated attack stage and conducts a coordinated attack with the capturer drone. In addition, in order to maintain the encirclement situation at this stage, when the total occupied angle is less than 320°, it switches back to the encirclement strategy and repeats step 5.

[0096] Step 8: When it is determined to conduct a coordinated attack with the hunter and the drone, the speed obstacle method is used to select the speed of the drone's advantageous situation and determine the speed set that the hunter can reach in the advantageous situation.

[0097] In the coordinated attack phase, a simplified UAV motion model is established as follows:

[0098]

[0099] where v i ,w i Respectively represent the speed and angular velocity of UAV i; a v ,a w Represent acceleration and angular acceleration respectively; (x i ,y i ) is the position coordinate of UAV i; h i is the heading of UAV i; v min ,v max Respectively represent the minimum speed and maximum speed; a v,min ,a v,max Respectively represent the minimum acceleration and maximum acceleration; a w,min ,a w,max represent the minimum angular acceleration and the maximum angular acceleration respectively.

[0100] The constant turn rate and velocity (CTRV) model is used to predict the target trajectory, and the target’s reachable area W is divided into attack areas Z through unscented Kalman filter (UKF) filtering. G,ATK , Invulnerable Area Z G,NATK , obstacle avoidance area Z G,OBS and non-obstacle avoidance area Z G,NOBS .like Figure 5 As shown, the target-related area is divided into the drone reachable area W and the attack area Z G,ATK , Invulnerable Area Z G,NATK , obstacle avoidance area Z G,OBS , non-obstacle avoidance area Z G,NOBS . Figure 5 The red area is the attack area Z G,ATK , which is centered at P and has d attack A 90-degree sector with a radius of d attack Indicates the attack distance. The small circle indicates the obstacle avoidance area Z G,OBS , d safe Indicates the safe distance. The rest is the drone’s advantageous situation area, that is, the advantageous situation area is the intersection of the target’s invulnerable area and the non-obstacle avoidance area.

[0101] The present invention uses the speed obstacle method to calculate the advantageous situation reachable area of ​​the hunter UAV and determine the advantageous situation reachable speed set of the hunter. The specific calculation formula is as follows.

[0102] Formula (10) is used to calculate the target reachable location set:

[0103]

[0104] where p T represents the target position vector; t is the time, t0 and t f Represent the initial time and the end time respectively, because the speed barrier method will have a time interval based on the current time, t f -t0 represents the time interval; Δt is the detection cycle of the airborne sensor; then p T (tf), p T (t0) means t f and the target position vector at time t0. Other time-related variables have similar meanings; v T (l) represents the speed at the lth moment from the current moment; N represents the total number of moments, N = (t f -t0) / Δt.

[0105] Calculate attack zone Z G,ATK :

[0106]

[0107] where p U represents the drone position vector; PA and PB represent the two sides of the sector area corresponding to the target attack area; is the Minkowski vector and operator; the symbol From t0 to t f Take the union of all moments; D(P,d attack , PA, PB) means that P is the center of the circle and d attack is the radius, with PA and PB as the two sides.

[0108] Calculate obstacle avoidance area Z G,OBS :

[0109]

[0110] in Indicates that P is the center of the circle and the radius area.

[0111] Calculate the reachable area W of the drone:

[0112]

[0113] in Indicates that P is the center of the circle and the radius area.

[0114] Calculate the dominant situation area Z P :

[0115] Z P =WZ G,ATK -Z G,OBS (14)

[0116] The hunter drone approaching the advantageous situation area can only ensure that it is not attacked by the target and has a certain threat situation to the target. It is also necessary to further select the drone speed v to construct the following hunter advantageous situation reachable speed set:

[0117]

[0118] Step 9: Based on the achievable speed set of the hunter’s advantageous situation, deep reinforcement learning training is used to realize the multi-angle saturation attack of the hunter’s UAV on the target.

[0119] The step 9 specifically includes:

[0120] Step 9.1: Select the speed and heading pair with the shortest time of speed concentration in the predator's advantageous situation and closest to the current speed of the predator UAV i as the guiding speed and heading pair (vguide ,h guide ).

[0121] Step 9.2: According to the guidance speed and heading pair (v guide ,h guide ) Design reward functions, including speed reward, enveloping angle reward, collision reward and attack distance reward.

[0122] In the coordinated attack phase, the input is the situation of both sides and the speed v that the UAV can reach in the advantageous situation. The PPO algorithm is used to calculate the expected speed and heading, which include:

[0123] State space design: The state space is a collection of UAV state information, which is used to describe the UAV's own motion state and store and update motion-related information, including the UAV's speed, heading, and position.

[0124] Action space design: The action space is a set of speeds and directions of the UAV, which is used to describe the achievable flight speed and direction of the UAV, and output the expected speed and heading of the UAV.

[0125] Reward function design: The reward function evaluates the quality of an action and is used to select the best action, including advantage situation speed guidance reward, attack distance reward, encirclement reward, and collision reward.

[0126] Among them, the advantage situation speed guidance reward:

[0127]

[0128] In order to guide the UAV to quickly approach the advantageous situation area and maintain the advantageous situation, the speed v and heading h pair with the shortest time of the speed concentration that can be reached by the predator in the advantageous situation and closest to the current speed of the UAV are selected as the guide, and the speed similar to the guiding speed is rewarded. k1 is the adjustment coefficient, v guide is the guidance speed, h guide It refers to the navigation direction.

[0129] Attack Range Bonus:

[0130] r2=-k2D iT (17)

[0131] When the achievable speed set of the predator's advantage situation is not empty, in order to encourage the drone to approach the target and carry out close-range attack behavior, the capture distance is penalized, and k2 is the adjustment coefficient.

[0132] Encirclement Rewards:

[0133]

[0134] In order to carry out multi-angle saturation attack on the target, the angle difference between UAVs j and k is not greater than π, thus forming a trap situation. k3 is the adjustment coefficient; min means taking the minimum value; UAVs j and k are the left and right adjacent UAVs of UAV i, respectively, and γ i,j and γ i,k They represent the angles of UAV j and UAV k relative to the polar angle of target T respectively.

[0135] Collision Rewards:

[0136] r4=C(19)

[0137] Penalties are imposed in the event of a collision.

[0138] In summary, the total step length reward is:

[0139]

[0140] Step 9.3: Use the reward function as the evaluation metric to perform deep reinforcement learning training of the hunter drone action.

[0141] The network structure of deep reinforcement learning training is designed as a decentralized actor executing a centralized critic training mode to output the expected speed and heading.

[0142] Specifically, the PPO algorithm is used to perform learning and training through a dual network of distributed execution and centralized training, namely the policy network and the value network. The optimal expected speed and direction are output by evaluating the reward function. The action space comes from the set of speeds that can be reached in the predator's advantage situation.

[0143] The input of the policy network is the state observation characteristics of the capture drone, and it outputs the mean and standard deviation of the Gaussian distribution of the expected heading and expected speed, thereby characterizing the continuous action. The input of the value network is the joint observation information of the capture drone cluster, and it outputs the estimated value of the state value, which is used to evaluate the quality of the action.

[0144] Step 9.4: Select the optimal action through extensive training and output the desired speed and heading of the hunter drone.

[0145] The total step length reward r of the drone is used as a value network evaluation indicator during the training process to evaluate the quality of the action. The optimal action is selected through a large amount of training, and the desired speed and heading are finally output.

[0146] Step 9.5: Control the hunter drone to carry out a multi-angle saturation attack on the target according to the desired speed and heading.

[0147] In the coordinated attack phase, determine whether the target is destroyed. If the attack fails and the target escapes, the above steps need to be repeated. If the drone is fast, a pure pursuit strategy is selected. When the distance threshold is reached, an attack strategy can be adopted. Otherwise, steps 3 to 9 are executed.

[0148] The invention proposes a method for drone cluster encirclement and attack based on hybrid modal analysis. Based on kinematic mode, the encirclement model is constructed using the Apollonius circle to determine the number of encircled drones without formation control. At the same time, based on the dynamic mode, the speed obstacle method and deep reinforcement learning are used to realize the attack on the surrounded target. The method has the advantages of simple encirclement model, high encirclement success rate, good robustness and good safety.

[0149] In some embodiments, the present invention also provides a computer program product, including a computer program, which implements the drone cluster encirclement and attack method when executed by a processor.

[0150] In some embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the drone cluster encirclement and attack method when executed by a processor.

[0151] In some embodiments, the present invention also provides a computer device, which includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Wherein, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Wherein, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the drone cluster encirclement attack method can be implemented.

[0152] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A drone cluster encirclement and attack method, characterized in that: include: Get the position and speed of the drone and the target; Different capture strategies are determined based on the maximum speed relationship between the drone and the target; The encirclement and capture strategy includes a pure pursuit strategy and an encirclement and pursuit strategy; When the encirclement and pursuit strategy is adopted, the number of the encircler drones is determined based on the position and speed of the drone and the target, and the Apollonius circle encirclement model is constructed; The speed of the hunter drone is decomposed into two directions: encirclement and pursuit, and the encirclement and pursuit strategy is divided into two stages: encirclement formation and situation maintenance; In the encirclement formation stage, multiple encircler drones are controlled to move along the encirclement direction based on the Apollonius circle encirclement model to form an encirclement situation around the target; Calculate the total occupancy angle of multiple hunter drones, and determine whether to enter the situation maintenance phase based on the total occupancy angle; In the situation maintenance phase, whether to conduct coordinated attack with the hunter drone is determined based on the distance between the hunter drone and the target; When it is determined to conduct a coordinated attack by the hunter and the UAV, the speed obstacle method is used to select the speed of the UAV in an advantageous situation, and the speed set that can be reached in the hunter's advantageous situation is determined; Based on the achievable speed set of the hunter's advantageous situation, deep reinforcement learning training is used to realize the multi-angle saturation attack of the hunter's drone on the target.

2. The drone swarm attack method according to claim 1, characterized in that: The different capture strategies are determined based on the maximum speed relationship between the drone and the target, specifically including: When v max >v Tmax When , a UAV is selected to execute a pure pursuit strategy until the target is destroyed; max is the maximum speed of the drone; v Tmax is the target maximum speed; When v max ≤v Tmax When attacking an enemy, select multiple drones to execute the encirclement and pursuit strategy.

3. The drone swarm attack method according to claim 2, characterized in that: The method of determining the number of the capturer drones based on the positions and speeds of the drones and the target and constructing the Apollonius circle capture model specifically includes: Using formula Calculate the maximum speed ratio between the drone and the target; According to the maximum speed ratio λ, the formula Calculate the number of hunter drones n; [] means rounding up; Using formula Calculate the speed ratio λ between the hunter drone i and the target T i ; where v i is the speed of the hunter drone i; v T is the speed of the target T; According to the speed ratio λ i , construct the Apollonius circle capture model Where (x ig ,y ig ) is the coordinate of the capturer UAV i relative to the target T; C i is the Apollonius circle corresponding to the capture drone i; R i C i The radius of .

4. The drone swarm attack method according to claim 3 is characterized in that: The calculating of the total occupancy angles of the multiple hunter drones and determining whether to enter the situation maintenance phase according to the total occupancy angles specifically includes: Using the formula θ i =2sin -1 (λ) Calculate the occupancy angle of the hunter drone i; Using formula Calculate the angle α of the hunter drone i relative to the target T i ; According to θ i and α i , using the formula Calculate the overlap angle ε between the capture drone i and the capture drone i+1 i，i+1 ; According to θ i and ε i，i+1 , using the formula Calculate the total occupancy angle θ of all the hunter drones G ; When θ G Reach 360° and each Apollonian circle C i When they are tangent or overlapped, it means that the conditions for encirclement and capture are met and the situation maintenance stage begins.

5. The drone swarm attack method according to claim 4, characterized in that: The determining whether to conduct a coordinated attack by the besieger drone according to the distance between the besieger drone and the target specifically includes: Set the distance threshold D, when the distance D between the hunter drone i and the target T iT When it is less than the threshold D, a coordinated attack by the roundup drone is carried out.

6. The drone swarm attack method according to claim 5, characterized in that: The method of using the speed obstacle method to select the speed of the drone in an advantageous situation and determine the speed set that can be reached in the advantageous situation of the hunter specifically includes: Divide the target reachable area W into attack areas Z G,ATK , Invulnerable Area Z G,NATK , obstacle avoidance area Z G,OBS and non-obstacle avoidance area Z G,NOBS ; Using formula Z P =WZ G,ATK -Z G,OBS Calculate the area of ​​advantage Z of the roundup drone P ; According to the situation area Z P Determine the speed set that can be reached by the predator's advantageous situation 7. The drone swarm attack method according to claim 6, characterized in that: The method based on the achievable speed set of the predator's advantageous situation and deep reinforcement learning training is used to realize the multi-angle saturation attack of the predator's drone on the target, specifically including: Select the speed and heading pair with the shortest speed concentration time and closest to the current speed of the hunter UAV i as the guiding speed and heading pair (v guide ,h guide ); According to the guidance speed and heading (v guide ,h guide ) Design reward functions, including speed reward, enveloping angle reward, collision reward and attack distance reward; Using reward function as evaluation index to train deep reinforcement learning of the hunter drone action; The optimal action is selected through extensive training, and the desired speed and heading of the hunter drone are output; Control the hunter drone according to the desired speed and heading to carry out multi-angle saturation attack on the target.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the drone cluster encirclement and attack method described in any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the drone cluster encirclement and attack method described in any one of claims 1 to 7 is implemented.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the drone cluster encirclement and attack method described in any one of claims 1 to 7.

Images