A pheromone-inspired multi-moving target search method for swarm drones

Abstract

The present invention discloses a pheromone-inspired swarm UAV multi-motion target search method, comprising the following steps: establishing a swarm UAV multi-dynamic target search model; obtaining the UAV search status for dynamic targets based on dynamic task division of distance; judging the UAV search status, if the UAV is in a roaming state, performing a roaming search based on the confidence region pheromone, if it is in a collaborative search state, performing a collaborative search based on a probability-restricted particle swarm algorithm. The present invention is based on the dynamic target confidence region, introduces confidence pheromones, so that the UAV can cover more areas faster, thereby improving the target detection rate of the roaming UAV; for dynamic targets, introduces the probability values ​​of the global optimum and the individual historical optimum, and improves the calculation method of the individual historical optimum, comprehensively considers the target signal model and the distance between the individual UAVs, sets the fitness function of the PSO, and improves the search efficiency.

Description

Technical Field

[0001] The invention relates to a drone search method, and in particular to a pheromone-inspired multi-motion target search method for a group of drones. Background Art

[0002] Drones play an important role in both modern civil and military fields, and are widely used in post-disaster search and rescue, wild fire monitoring, enemy reconnaissance, etc. In many application fields, drones include the process of target search, such as searching for survivors at the disaster site, checking for fire sources in the wild, and reconnaissance of enemy positions on the battlefield. Target search often requires high efficiency to reduce losses or maximize benefits, and swarm drones have superior performance in search efficiency, flexibility, scalability, and robustness. Swarm drones are a type of swarm robot system, and research on them has become an important direction for the development of automation technology.

[0003] Swarm drones are inspired by the group behavior of social organisms in nature. This unmanned system has no control center and is a distributed system. Compared with systems with control centers, distributed systems have the advantages of small computational complexity and high reliability. In existing literature, the models of target search problems mainly include: probability function model and target signal model. The probability function model is mainly used to search for lost targets. Sensors are usually modeled as binary variables with two states: "detected" or "not detected", and the corresponding target search state is "target found" or "target not found". However, in some target search scenarios, the final lost location of the target may not be determined, and during the search process, the sensor may detect the target signal but cannot determine the specific location of the target, requiring further search. In the target search of the target signal model, static target search is mainly studied, and the target signal model assumes that the search subject can directly distinguish the target based on the target signal, which further limits its scope of application. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a pheromone-inspired multi-moving target search method for a swarm of unmanned aerial vehicles with a simple algorithm and high search efficiency.

[0005] The technical solution of the present invention to solve the above problem is: a pheromone-inspired multi-motion target search method for a group of drones, comprising the following steps:

[0006] Step 1: Establish a multi-dynamic target search model for a swarm of drones;

[0007] Step 2: Perform dynamic task division based on distance on the UAV to obtain the search status of the UAV for the dynamic target;

[0008] Step 3: Determine the search status of the UAV for dynamic targets. If the UAV is in a roaming state, proceed to step 4; if it is in a collaborative search state, proceed to step 5.

[0009] Step 4: Conduct roaming search based on pheromones in the confidence area;

[0010] Step 5: Conduct collaborative search based on probability-constrained particle swarm algorithm.

[0011] The beneficial effects of the present invention are:

[0012] 1. The present invention improves the target signal model, assuming that UAVs cannot distinguish signals of different targets to adapt to the current development of sensors, and based on this, improves the cooperative grouping method of UAVs, so as to be more widely adapted to the functions of existing sensors, making the target signal detection process in practical applications easier to achieve.

[0013] 2. The present invention is based on the dynamic target confidence region and introduces confidence pheromones, so that the drone can cover more areas faster, thereby improving the target detection rate of the roaming drone.

[0014] 3. Aiming at dynamic targets, the present invention introduces the probability values ​​of global optimum and individual historical optimum on the basis of PSO, improves the calculation method of individual historical optimum, comprehensively considers the target signal model and the distance between individual drones, and sets the fitness function of PSO, which can prevent the drone from falling into the local optimum during the search process, thereby improving the efficiency of drone search for targets.

[0015] 4. The present invention draws on the boundary scanning obstacle avoidance strategy proposed in previous work, discretizes it, and adapts it to the grid map, thereby maintaining high efficiency and search breadth of obstacle avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 Schematic diagram of the multi-dynamic target search model for a swarm of UAVs established in the present invention.

[0018] Figure 3 Schematic diagram of drone state migration.

[0019] Figure 4 Schematic diagram of the confidence region, where (a) represents the confidence region corresponding to a single position, and (b) represents the confidence region corresponding to the motion trajectory.

[0020] Figure 5 Schematic diagram of the maximum confidence region, where (a) represents the maximum confidence region of a single UAV, and (b) represents the maximum confidence region of multiple UAVs.

[0021] Figure 6 This is the correspondence diagram between the farthest reachable grid and the nearest undetectable grid.

[0022] Figure 7 It is a schematic diagram of initializing the search environment and the search object in the embodiment.

[0023] Figure 8 Schematic diagram of the confidence region pheromone from the perspective of a group of drones in the embodiment.

[0024] Fig. 9 It is a schematic diagram of a certain UAV detecting a target signal for the first time in the embodiment.

[0025] Fig.10 This is a schematic diagram of a group of drones sensing an obstacle for the first time in an embodiment.

[0026] Fig.11 This is a schematic diagram of a sub-coalition consisting of five drones searching for a target in the embodiment.

[0027] Fig.12 This is a schematic diagram of five drones forming a sub-alliance and the drones entering a collaborative search state in the embodiment.

[0028] Fig.13 This is a schematic diagram of the sixth drone joining the sub-alliance and entering the collaborative search state in the embodiment.

[0029] Fig.14 Schematic diagram of an embodiment in which, after searching for targets in the local area, drones disband their sub-alliances and disperse to leave the area.

[0030] Fig.15 Schematic diagram of the completion of a swarm of drone search mission in an embodiment.

[0031] Fig.16 Schematic diagram of the motion trajectories of all targets and all drones during the target search process in the embodiment.

[0032] Fig.17 This is a search performance comparison chart. DETAILED DESCRIPTION

[0033] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0034] like Figure 1 As shown, a pheromone-inspired multi-motion target search method for a group of drones includes the following steps:

[0035] Step 1: Establish a multi-dynamic target search model for a swarm of drones.

[0036] The established multi-dynamic target search model of swarm UAVs includes:

[0037] Mission environment: In reality, drones are in three-dimensional space, which makes the drone planning problem complicated. It is believed that the movement of drones can be decoupled into two movements in the two-dimensional plane and the vertical direction. Therefore, in order to facilitate the discussion of the mission area, a two-dimensional area with a constant height of the drone is considered; in the grid area with a length × width of Lv × Lh, a single grid is represented by G (lo，la) Indicated by, where 1≤lo≤Lv, 1≤la≤Lh, the grid side length is ΔL0, and the grid center (lo, la) is used to represent the position of the grid, denoted by G (lo，la) .

[0038] Task object: The object is represented by the set A = UAV∪T∪B; where the set UAV = {uav i |i=1,2,...,n V ; 10≤ n V ≤100} represents drone, n V represents the total number of drones; set T = {tar j |j=1,2,...,n T ;n T ≥1} indicates the target, n T Represents the total number of targets; set B = {obs m |m=1,2,...,n B ;n B ≥1} indicates an obstacle, n B Represents the total number of obstacles; at time t, the UAV i 、Target tar j 、Obstacles m The positions of uav i|t , tar j|t 、obs m|t ;UAV i1 UAV i2 The Euclidean distance between i1，i2|t =||uav i1|t -uav i2|t ||, i1 = 1, 2, ..., n V , i2=1,2,...,n V , and i1≠i2; UAV i with target tar j The Euclidean distance between i，j|t =||uav i|t -tar j|t ||; UAV i With obstacles obs m The Euclidean distance betweeni，m|t =||uav i|t -obs m|t ||; target tar j With obstacles obs m The Euclidean distance between j，m|t =||tar j|t -obs m|t ||.

[0039] Mission: Within the search range, drones search for targets through mutual cooperation; Under the premise of avoiding collision, drones search for target positions by relying on detected target signals, information exchange, and their own movement; When all drones have arrived within the range d0 near all the targets to be searched, the target search mission is completed.

[0040] UAV: UAV i The velocity at time t is denoted as Vv i|t , whose size satisfies 0≤||Vv i|t ||≤V vm , V vm Represents the maximum value of the UAV speed, which is a scalar; and the location of the UAV is the center of the grid; considering the maximum communication radius r com , maximum obstacle perception radius r obs , maximum target detection radius r tar , the drone can realize the following functions: Dv i1，i2|t ≤r com When the UAVs communicate with each other; Dvt i，j|t ≤r tar When the target signal is detected; Dvb i，m|t ≤r obs Obtain the obstacle position.

[0041] Target: target tar j The velocity at time t is denoted as Vt j|t , whose size satisfies 0<||Vt j|t ||≤V tm , V tm It represents the maximum value of the target speed, which is a scalar; and the target location is arbitrary; when the group of drones searches for a target, the drone detects the surrounding target signals through sensors. The target signal and distance satisfy a function that is interfered by the environment, which is described as the target signal detection function and is expressed by formula (1):

[0042]

[0043] Among them I i，|t Indicates the UAV at time t i The detected target signal; η is the environmental interference; G(Dvti，j|t ) represents the relationship between the target signal and the distance in an interference-free environment; Dvt ij|t It exists objectively but the drone cannot know it. When there are multiple targets within the drone's target detection range, the weaker target signal will be covered by the stronger target signal, and the drone cannot distinguish the signals of different targets.

[0044] Assumption 1: The relationship between the parameters satisfies: V tm <V vm <<r tar ≈r obs <<r com , and are all known;

[0045] The speed in the present invention is similar to the speed in the particle swarm algorithm, which represents the change in the position of the drone in each iteration, and can also be understood as the distance moved per unit time. The maximum speed V tm and V vm The units are the same as the distance.

[0046] Assumption 2: All UAVs participating in the target search are isomorphic, and all the targets being searched are of the same type.

[0047] Figure 2 The following is a schematic diagram of part of the search area, which is similar to the UAV i Distance less than V vm The UAV can reach the grid of G. Since the target can be at any location, only when the grid G (lo，la) All points within and uav i Distance less than r tar When , it is recorded as detectable.

[0048] Step 2: Perform dynamic task division on the UAV based on distance to obtain the UAV's search status for dynamic targets.

[0049] When a group of drones searches for a target, their task is to search for all targets in the task area, and searching for targets in a local area is a subtask, and the number of subtasks is at most n. T , drones that complete the same subtask form a sub-alliance, and the sub-alliance at time t is recorded as TU J|t , where 0≤J≤n T , the serial number J is arranged from large to small according to the strongest target signal detected in the sub-alliance. The state of the drones that make up the sub-alliance is called the collaborative search state. The drones form a sub-alliance in the following way:

[0050] Drones obtain target signals through two ways, so drones that can obtain target signals are divided into two categories: Class I drones that detect target signals by their own sensors and Class II drones that obtain target signals through communication between drones. Class I drones are recorded as type_I and Class II drones are recorded as type_II. First, class I drones are grouped and the drones with the strongest target signals in the local area are selected. i0 , when other Class I UAVs are nearby i With UAV i0 If equation (2) is satisfied, then i0 Forming a sub-alliance TU 1|t The remaining Class I drones that do not satisfy formula (2) continue to select the drone with the strongest target signal. Then search for the remaining drones nearby. The drones whose relationship meets formula (2) form a sub-alliance, and this cycle repeats until all Class I drones form a sub-alliance; then, Class II drones join the sub-alliance of the nearest Class I drone that can communicate;

[0051]

[0052] After the subtasks are assigned, in order to avoid wasting drone resources due to too many drones in the same sub-alliance, a closed-loop adjustment is further introduced to evaluate the resource allocation of each sub-alliance: J >N max , the sub-alliance selects N max UAVs are the individuals of the sub-alliance, where N max The maximum number of sub-alliance drones allowed; the principle of selection is: priority is given to Class I drones, among Class I drones, priority is given to drones with strong target signals, and among Class II drones, priority is given to drones that are close to communication drones.

[0053] An example of task division is shown in Table 1. max =6,n T =10, first the uav with the largest target signal 19 And the uav that satisfies formula (2) 17 Form a sub-alliance, uav 16 A sub-alliance is formed, and then the Class II drones that communicate with the Class I drones join the corresponding sub-alliance. 19 、uav 17 With uav6, uav 13 、uav 18 ;uav 16 Compatible with uav1, uav2, uav7, uav9, uav10, uav 12 Sub-alliances are formed, but the total number of partial sub-alliances is N J >Nmax , according to the principle of merit 10 Exit the sub-alliance. Finally, get the sub-alliance TU 1|t ={uav 19 , uav 17 , uav 13 ,uav6,uav 18} and TU 2|t ={uav 16 , uav1, uav2, uav7, uav9, uav 12}, so that the final N J ≤N max and get

[0054] Table 1

[0055]

[0056] When a group of drones searches for multiple targets, they need to quickly cover the mission area and approach the target as quickly as possible after detecting the target signal. When a drone searches for a target, it will initiate a declaration of searching for the target until all targets are searched and the search mission is completed. Therefore, the search of drones for dynamic targets is divided into three states: roaming status (RS) for quickly covering the search area, collaborative status (CS) for forming a sub-alliance, and task completion status (TS); these three states can be expressed as Figure 3 The finite state machine (Finite State Machine) shown.

[0057] Step 3: Determine the search status of the UAV for dynamic targets. If the UAV is in roaming state, proceed to step 4; if it is in collaborative search state, proceed to step 5; if the UAV is in search mission completion state, the search mission is completed and all operations are ended.

[0058] Step 4: Conduct roaming search based on confidence area pheromones.

[0059] The specific process of step 4 is as follows:

[0060] (4-1) The main purpose of roaming search is to enable the drone to quickly detect every corner of the search area. After the drone detects an area, it does not need to detect it again within a certain period of time. For this reason, this part borrows the confidence region and proposes a confidence pheromone related to the confidence region.

[0061] To facilitate subsequent discussion, assume that the function δ(D) is expressed as follows:

[0062]

[0063] Where n r is the number of columns of matrix D, n l is the number of rows in matrix D, b ζ，ξ is the value of the ζth row and ξth column in D;

[0064] function The expression is as follows:

[0065]

[0066] in:

[0067]

[0068]

[0069] (4-2) Define the confidence region and pheromone for dynamic target search:

[0070] Definition 1: The area where the target is determined to be non-existent is called the confidence area, denoted by C z , the non-confidence region is express;

[0071] like Figure 4 As shown in (a), UAV i At time t0, the drone reaches position P1 in the grid area, but no target signal is detected. The detection range of the drone is the confidence area. At time t0+Δt, the drone leaves P1, where Δt is the time increment. Since the target speed is limited, a certain circular confidence area is still maintained around P1. The distance ΔU from the edge of the confidence area to P1 satisfies the following formula:

[0072]

[0073] like Figure 4 As shown in (b), uav i When reaching P′1 at time (t1+Δt), if Δt is small enough, the confidence regions corresponding to P1 and P′1 will form a continuous confidence region; on the continuous motion trajectory, a single UAV will form a Figure 5 Confidence region shown in (a).

[0074] Corollary 1: The drone detection target is divided into three cases: Case 1: The distance near the drone is less than or equal to r tar There are obstacles or mission environment boundaries within the range; Case 2: The distance near the drone is less than or equal to 2r tar There are other drones within the range; Case 3: The distance near the drone is less than or equal to rtar There are no obstacles or task environment boundaries within the range, and the distance is less than or equal to 2r tar There are no other drones within the range; the maximum confidence area of ​​Case 1 and Case 2 combined is greater than the maximum confidence area of ​​Case 3. The confidence area detected by the Case 3 drone is recorded as C z，1 The confidence region where the sum of Case 1 and Case 2 is greater than that of Case 3 is recorded as C z，2 ;like Figure 5 As shown in (a), the confidence region obtained by a single drone detection in Case 3 of Inference 1 is denoted as C z，1 ;like Figure 5 As shown in (b), in Case 2 of Corollary 1, the confidence region that is larger than the individual detections of the two drones is recorded as C z，2 .

[0075] In the mission environment, the confidence region is represented by pheromone; the drone stores the pheromone strength at different locations. Indicates uav i The pheromone at position P0 stored at time t, the pheromone value range in the confidence region is [0, τ m ],τ m is the maximum value of pheromone in the confidence region, that is, confidence region C z，2 The pheromone value of the confidence area takes the minimum value in the grid where the drone is located. Confidence Region C z，2 The maximum value of pheromone The pheromone intensity in the non-confidence region is recorded as

[0076] Figure 4 In (a), at time t0, the UAV i Arrived at P1, the pheromone With P1 as the center, r tar The circle with radius C z，1 ; The distance between any point P0 and P1 in the search area is d p ≥0, the pheromone of P0 is:

[0077]

[0078] From formula (7) and formula (8), we can see that under the premise of satisfying the properties of the confidence region and the range of pheromone values, C z，1 The distance of edge movement and time Δt, the pheromone intensity at point P0 and the distance d p They are linearly correlated, so when the drone is far away from P1, the pheromone change and time Δt also satisfy the linear relationship:

[0079]

[0080] Where ρ is the pheromone enhancement parameter; when time P1 becomes the confidence region C z，2 , the pheromone at this point becomes the maximum value From this we can get the pheromone enhancement parameter ρ = τ m V tm / r tar ; When the confidence region C z，1 The grid pheromone increases to τ m When , it becomes the confidence region C z，2 .

[0081] for Figure 4 The pheromone change of P0 in (a) satisfies the following formula:

[0082]

[0083] In the search area, in order to save the storage space of the drone, the drone only stores the pheromone of one point in each grid; i|t Indicates uav i The pheromone strength of all grids stored at time t, PH i|t It is expressed as formula (11):

[0084]

[0085] Any element of the matrix Indicates uav i The grid G ​​stored at time t (lo，la) pheromones.

[0086] The target can be anywhere in the search area. Therefore, only when the entire grid is within the drone detection area and the drone does not detect the target signal can it be determined that the grid has no target and belongs to the confidence area.

[0087] Here the detectable grid is defined:

[0088] Definition 2: When a grid G (lo，la) Farthest point and UAV i Distance||(G (lo，la) +0.5ΔL·δ(G (lo，la) -uav i ))-uav i ||≤r tar , this grid is called the detectable grid, denoted as DG, uav i The set of all detectable grids is denoted as DG i ; On the contrary, ||(G (lo，la) +0.5ΔL·δ(G (lo，la)-uav i ))-uav i ||>r tar , G (lo，la) It is called an undetectable grid, denoted as UG; if the target undetectable grid has a connected detectable grid, it is called the nearest undetectable grid, denoted as NUG.

[0089] For UAV i , any detectable grid The pheromone satisfies the following formula:

[0090]

[0091] ΔE is an intermediate variable; a discrete iteration method is adopted in time. From formula (10), it can be known that any grid G (lo，la) The iterative relationship of pheromone satisfies the following formula:

[0092]

[0093] For region C z，2 , when there is a grid Make G (lo，la) When the following formula is met, G (lo，la) This becomes a non-confidence region:

[0094]

[0095] ΔH is the intermediate variable;

[0096] When the grid

[0097] For any UAV i The stored grid pheromone is updated 5 times in each iteration. The updating process of pheromone in each iteration is as follows:

[0098] Step 1: Calculate and update all grid pheromones according to formula (13);

[0099] Step 2: Calculate and update the pheromones of all detectable grids according to formula (12);

[0100] Step 3: CUAV i ={uav I1 , uav I2 ,...,uav In} indicates that it can be used with uav i A collection of communicating drones; UAV i Get CUAV i The pheromones stored by all drones in the , and all the grid pheromones stored by them are updated as follows:

[0101]

[0102] Step4: If uav i In collaborative search state, for sub-alliance Any drone in TU J Type_I drones are ype_II UAV is gather All grids in become untrusted areas, and pheromones are updated as follows:

[0103]

[0104] Step 5: The pheromones of all grids within the search range are updated as follows:

[0105]

[0106] (4-3) Set the roaming search speed:

[0107] Definition 3: When a grid G (lo，la) UAV i Distance||G (lo，la) -uav i ||≤V vm , this grid is called the reachable grid, denoted as RG, uav i The set of all reachable grids is denoted as RG i ; In the UAV i If a reachable grid is the farthest reachable grid from the UAV in a certain angular direction, then the reachable grid is called the farthest reachable grid. Figure 6 , if V vm =3ΔL, the grids numbered a~p are the furthest reachable grids (FRG).

[0108] There is a corresponding relationship between the farthest reachable grid and the nearest undetectable grid. Assuming that a farthest reachable grid is the farthest reachable grid in the angle interval (θ1, θ2), if the angle of the vector formed by the drone pointing to the farthest point of a nearest undetectable grid is within the interval (θ1, θ2), then the nearest undetectable grid corresponds to the farthest reachable grid; for example Figure 6 In the example, a and a1, b and b1, b2, b3, etc.

[0109] In the roaming state, in order to quickly detect more areas, the drone will move at the maximum speed in the direction after obtaining the speed direction according to the pheromone, and select the farthest reachable grid in the speed direction as the next location; after updating the pheromone, any drone in the roaming state uav i The pheromone in the grid that can be detected by the surrounding targets will satisfy equation (12), which is consistent with the uav i The pheromones of grids with the same distance are equal; however, the pheromones of undetectable grids will be different even if they are at the same distance from the drone. i The next flight direction is determined based on the pheromone intensity of the nearest undetectable grid.

[0110] uav i The farthest reachable grid is denoted as FG i，k , where 1≤k≤k i，re , k represents the kth farthest reachable grid clockwise from the positive half axis of x, k i，re FG is the maximum reachable grid number; i，k The pheromone of the corresponding nearest undetectable grid is denoted as Fph i，k，n , where 0≤n≤n i，gr , n represents the nth nearest undetectable grid clockwise, n i，gr For FG i，k The number of the nearest undetectable grids, uav i The maximum pheromone corresponding to the kth most reachable grid

[0111] uav i Get FG i，k and Fph i，k，max After that, according to the roulette method update speed, the roulette method is expressed by formula (18); the random number r a ~U(0,1), when Q(K-1)<r a <Q(K), uav i Select FG i，K As the next expected position, and the next expected speed Vve i|(t+1) Satisfies formula (19), where FG i，K FG i，K Position, α represents the kinematic inertia coefficient:

[0112]

[0113] V i|(t+1) =αVv i|t +(1-α)(FG i，K -uav i|t ) (19)

[0114] qk Indicates that the drone selects the farthest reachable grid FG i，k As the probability of the next step location grid.

[0115] Step 5: Conduct collaborative search based on probability-constrained particle swarm algorithm.

[0116] The global optimality in the collaborative state is the optimal one in the sub-alliance, which is called the semi-global optimality and is denoted as sgb J|t The individual historical best is the historical best after the drone individual joins the sub-alliance, denoted as pb i|t ;

[0117] Classic PSO algorithms are mostly used for static target search. The fitness function value at the same position at different times remains unchanged, so the individual historical optimum and the global optimum will always maintain their credibility. However, in dynamic target search, the fitness function value at the same position changes with time, and the reliability of the individual historical optimum and the global optimum for system optimization gradually decreases with iteration. Therefore, probability values ​​are introduced into the updates of semi-global optimum and individual historical optimum respectively, and the probability value is used to limit the inheritance of the optimal in particle swarm optimization PSO. At the same time, the calculation method of individual historical optimum is improved, and probability-restricted particle swarm optimization PFPSO is proposed.

[0118] Assume that at time t the UAV i The fitness function of the location is expressed as f(uav i|t ), the update method of the global optimum and individual historical optimum of PFPSO is as follows:

[0119] Semi-global optimality introduces the global optimal probability value σ j , whose initial value is σ0, where 0<σ0<1; for any sub-coalition TU J|t , among which the Class I UAV with the strongest target signal is uav J , T.U. J|t The corresponding semi-global optimal update process is as follows:

[0120] Step 1: Determine whether there is a sub-alliance at time (t-1) Among them, the Class I drone has the strongest target signal If it exists, assign a uniformly distributed random number r between [0, 1] σ , and execute Step2; otherwise sgb J|t =uav J|t , σ J =σ0, the update ends;

[0121] Step 2: If the condition holds, f() is the fitness function expressed by formula (21), sgb J|t =uavJ|t , σ J =σ0, the update is completed; otherwise, execute Step3;

[0122] Step 3: End Update; for The semi-global optimum.

[0123] Individual historical optimality introduces the individual historical optimal probability value μ i , whose initial value is μ0, where 0<μ0<1; for any UAV in the collaborative search state i , whose individual historical optimal pb i|t The update process is as follows:

[0124] Step 11: Determine when the drone joins the sub-alliance. If it joins the sub-alliance at time t, update pb according to formula (20) i|t , and assign μ i =μ0, the update ends; otherwise, a random number r uniformly distributed between [0, 1] is assigned μ , and execute Step 12;

[0125] Step 12: If the condition (f(uav i|t )≥f(pb i|(t-1) ))||(f(uav i|t )<f(pb i|(t-1) )&r μ ≥μ i ) holds true, update pb according to formula (20) i|t , and assign μ i =μ0, the update is completed; otherwise, execute Step 13;

[0126] Step 13: pb i|t =pb i|(t-1) , μ i =μ i μ0, update completed.

[0127] UAVs in the collaborative search state are divided into two categories: type_I and type_II. For any type I UAV in a sub-alliance, the best individual history still selects the best position in the individual history; and for any type II UAV in a sub-alliance i If uav i The target information is communicated from the UAV Take the best individual history of the drone as the midpoint of the two; i The individual historical optimality of is updated according to formula (20):

[0128]

[0129] Different types of UAVs in collaborative search states have different fitness function calculation methods; in any sub-coalition, if the UAV i is type_I, the fitness function value is the target signal detected by the sensor; if the UAV i For type_II, For its nearest type_I communication drone, the fitness function value is uavi and The opposite of distance, that is:

[0130]

[0131] In summary, the expected speed calculation of PSO with probability constraints is shown in formula (22):

[0132]

[0133] Where c1 and c2 are cognitive coefficients and social coefficients respectively, r1 and r2 are random numbers that obey U(0,1), ω is the inertia weight, α is the inertia link considering kinematics, and pb i|t for UAV i The individual historical optimum at time t, sgb J|t for UAV i Sub-alliance TU J|t The semi-global optimality of Vvp i|(t+1) is an intermediate variable.

[0134] Regardless of whether the drone is in a collaborative state or a roaming state, it is calculating the expected speed Vve i|(t+1) Then consider obstacle avoidance; based on the boundary scanning obstacle avoidance strategy, it is improved into an improved boundary scanning obstacle avoidance IBSOA adapted to the grid environment. When the drone avoids obstacles, the perceived obstacles are divided into two situations: continuous obstacles and discontinuous obstacles. The obstacle avoidance directions are divided into counterclockwise obstacle avoidance and clockwise obstacle avoidance. The drone comprehensively considers the perceived obstacle situation and the historical obstacle avoidance direction to calculate the next speed.

[0135] Due to the rasterization of the map, the grid with obstacles is called the obstacle grid; IBSOA ignores obstacles and comprehensively considers the obstacle avoidance direction and the obstacle grid to avoid obstacles; if the drone is in the roaming state and the distance near it is less than min(r tar , r obs ) range, start to consider obstacle avoidance; if the drone is in a collaborative state and the distance near it is less than Vv m If there is an obstacle within the range, obstacle avoidance begins. The obstacle avoidance process of IBSOA is represented by the following two points:

[0136] 1. UAV i In roaming state, and the distance near it is less than min(r tar , r obs ), <Vve i|(t+1) > There are obstacles in the direction, uav i Start obstacle avoidance; uav i The set of all the farthest reachable grids at time t is denoted as FRG i|t , if there is a grid G (lo，la) ∈FRG i|t , meet the UAV location uav i|t Point to grid position G (lo，la) The ray is at the distance uav i|t Less than min(r tar , r obs ) passes through the obstacle grid, then the grid G (lo，la) From FRG i|t When all grids that meet the above conditions have been removed from FRG i|t After culling, the final grid set FOG is obtained i|t ;uav i According to its own obstacle avoidance direction flag parameter λ i , determines the speed direction after considering obstacle avoidance, λ i The initial value is 0;

[0137] When i = 0, indicating that the obstacle was not avoided in the previous step, and the grid set FOG i|t Select Vve i|(t+1) The grid with the smallest angle difference Gn i As a UAV i The grid to be reached next;

[0138] When i = 1, from the grid set FOG i|t Select the counterclockwise direction with Vve i|(t+1) Angle difference grid Gn i As a UAV i The grid to be reached next;

[0139] When i = -1, from the grid set FOG i|t Select the clockwise direction with Vve i|(t+1) The grid with the smallest angle difference Gn i As a UAV i The grid to be reached next;

[0140] Then according to the grid Gn iAnd equation (23), solve the next step speed Vv i|(t+1) , where Gn i Gn i Position; after obtaining the velocity, update λ according to formula (24) i , the calculation of obstacle avoidance speed ends:

[0141]

[0142] λ i =sgn(F( <Vv i|(t+1) >- <Vve i|(t+1) >)) (24)

[0143] 2. UAV i In the cooperative state, and the distance near it is less than Vv m Within the range, <Vve i|(t+1) > There are obstacles in the direction, uav i Start obstacle avoidance; uav i The set of all reachable grids at time t is denoted as RG i|t If G (lo，la) ∈RG i|t Satisfy, grid position G (lo，la) UAV location i|t The connecting line passes through the barrier grid, and the grid is connected from RG i|t Eliminate, when all grids that meet the above conditions have been removed from RG i|t After culling, we get the final grid set OG i|t ; According to the obstacle avoidance direction sign parameter λ i , determines the speed direction after considering obstacle avoidance, λ i The initial value is 0;

[0144] When i = 0, indicating that the obstacle was not avoided in the previous step, and the grid set OG i|t Select and (uav i|t +Vve i|(t+1) ) The nearest accessible grid without obstacles is grdn i As a UAV i The grid to be reached next;

[0145] When i = 1, from the grid set OG i|t Select the counterclockwise direction with (uav i|t +Vve i|(t+1) ) The nearest accessible grid without obstacles is grdn i As a UAV i The grid to be reached next;

[0146] When i = -1, from the grid set OG i|t Select the clockwise direction with (uav i|t +Vve i|(t+1) ) The nearest accessible grid without obstacles is grdn i As a UAV i The grid to be reached next;

[0147] Then according to the grid grdn i And equation (25), solve the next step speed Vv i|(t+1) , where grdn i for grdn i After obtaining the position and velocity, update λ according to formula (26) i , calculation of obstacle avoidance speed ends;

[0148]

[0149] λ i =sgn(F( <Vv i|(t+1) >- <Vve i|(t+1) >)) (26)

[0150] According to formula (23) or (25), the next step speed Vv is calculated i|(t+1) Then, the drone position is updated according to formula (27) and it is determined whether the target is found:

[0151] uav i|(t+1) =uav i|t +Vv i|(t+1) (27).

[0152] Simulation Verification

[0153] In order to verify the performance of the algorithm proposed in the analysis article, several groups of comparative experiments were carried out in the simulation environment of MATLAB 2021a to compare the search efficiency and search energy consumption between different search modes.

[0154] The present invention takes into account the randomness and inertia of the dynamic target motion, and designs the motion model of the target as shown in the following formula:

[0155]

[0156] Here r3 and r4 both obey the uniform distribution U(0,1), and β is the inertia constant of the target motion. i|(t+1) is the random speed; Vte i|(t+1) is the random expected speed of the next step of the target. i|(t+1) Direction Vt m If there are no obstacles within the range, Vti|(t+1) =Vte i|(t+1) , if there is an obstacle, the target is at Vt m Generate a Vt in the direction without obstacles within the range m , random speed in the direction as the next step speed Vt i|(t+1) .

[0157] Referring to the description of the target signal, the details in equation (1) are set as follows:

[0158]

[0159] Where m is the attenuation coefficient of the environment (0<m<1); Q is the constant signal power of the target; η represents Gaussian white noise, which satisfies the standard normal distribution.

[0160] The simulation search process of swarm drones includes the settings of different parameters such as the search subject - drones, the search object - targets, and the search environment.

[0161] Table 2

[0162]

[0163] Figures 7 to 16 The figure shows the process of a group of drones searching for multiple moving targets in the search algorithm mode PCDI according to PFPSO, CAP, IBSOA and DBDTA. v =20, N t =10, dots represent drones, and stars represent targets. Figure 7 The positions of the drone, the target, and the distribution of obstacles at the beginning of the target search are presented; Fig.16 The motion trajectories of all targets and drones during the entire search process. It can be clearly seen that the drone successfully searches for all targets while avoiding obstacles. Figures 8 to 15 It is the confidence area pheromone, drone, obstacle and searched target from the drone's perspective. After the moving target is searched, the target will remain stationary at the location where it was when it was searched.

[0164] Figure 7 In the figure, the search environment and search object are initialized, where the drones are regularly arranged in the area in the lower left corner, and the targets and obstacles are randomly distributed in the area in the upper right corner. The obstacles shown in the figure are stationary, and the motion of each target satisfies the random motion process with motion inertia represented by equation (4.23).

[0165] Figure 8 The figure shows the confidence area pheromone from the perspective of the swarm of drones at the moment; it can be seen from the figure that the swarm of drones is temporarily unable to know the location of the target and obstacles.

[0166] Fig. 9 In the process, all the drones in the group go through a period of roaming search based on the pheromone of the confidence region; a drone detects the target signal for the first time, and then builds a sub-alliance with several nearby drones according to the dynamic task division based on distance; because the target signal is detected, it means that there is a target in the detection area, and part of the confidence region becomes an unconfidence region.

[0167] Fig.10 In the process, the swarm of drones senses an obstacle for the first time. Since the drone that senses the obstacle is in a roaming search state, the drone immediately starts to avoid the obstacle using an improved boundary scanning obstacle avoidance strategy.

[0168] Fig.11 In the example, the sub-coalition of 5 drones is based on the probability-constrained particle swarm algorithm, which continuously approaches the target and finally completes the target search in the local area. After the search is completed for the target in the local area, the sub-coalition will be disbanded, and the drones in it will be transformed back to the roaming search state.

[0169] Fig.12 In the example, 5 drones form a sub-coalition, and the drones enter a collaborative search state. The sub-coalition formed by the drones collaboratively searches for targets in the local area; Fig.13 At the middle moment, the sixth UAV joins the sub-alliance. At this moment, the UAVs that join the sub-alliance enter the collaborative search state and participate in the target search in the local area.

[0170] Fig.14 After searching for the target in the local area, the drones disband the sub-alliance and enter the roaming search state. The drones are inspired by the pheromones in the trust area and disperse to leave the area to search for targets in other non-trust areas.

[0171] Fig.15 In the process, the swarm of drones completes the search for the last moving target based on the PCDI mode. At the same time, all targets are searched and the search mission is completed.

[0172] Fig.16 The figure shows the motion trajectories of all targets and all drones during the target search process. The trajectories connected by five-pointed stars are the target motion trajectories, black objects represent obstacles, and the trajectories formed by dotted lines represent the trajectories of the group of drones when searching for targets. It can be seen from the trajectories in the figure that when the group of drones are searching for multiple moving targets, whether in a collaborative search state or a roaming search state, guided by the improved boundary scanning obstacle avoidance strategy, the drones can successfully avoid obstacles.

[0173] In order to verify the effectiveness of the method proposed in this paper, Figure 1Based on the search framework shown in the figure, the search algorithm mode is designed. IBSOA is used to maintain obstacle avoidance, and DBDTA is used to determine whether the drone needs to join a sub-alliance; different algorithms are used for collaborative search and roaming search, forming four search algorithm modes of PCDI, PSDI, KCDI and KSDI as shown in Table 3. Among them, KCPSO is a kinematic constraint particle swarm optimization algorithm, and SVF is a simplified virtual force model.

[0174] Table 3

[0175]

[0176] When the number of drones is N v =20, 30, 40, 50, 60, 70, 80, 90, 100, use four search modes respectively, search Figure 7 The target in the environment shown in the figure is searched 100 times independently, and the total number of iterations required to search for 10 moving targets under different numbers of drones and different search modes is counted - the number of iterations required to search for the target, and the total energy consumption - the total movement distance of all drones. The results are shown in Table 4.

[0177] Table 4

[0178]

[0179] The comparison of PCDI and KCDI, PSDI and KSDI modes can analyze the advantages and disadvantages of PFPSO. Compared with KCPSO, PFPSO has an average search efficiency increase of 4.05%, an average reduction of 3.08% in total energy consumption, and an average increase of 30.45% in search stability. It can be seen that when there are fewer drones, the advantages of PFPSO are more obvious, but when the number of drones is close to 100, PFPSO has only a slight advantage over KCPSO. PFPSO in the PCDI algorithm mode has more advantages; no matter in which mode, PFPSO shows higher stability. The comparison of PCDI and PSDI, KCDI and KSDI modes can analyze CAP. Compared with SVF, CAP has an average increase of 22.55% in search efficiency, an average reduction of 24.65% in total energy consumption, and an average increase of 74.55% in search stability. Under different total number of drones, the search performance of CAP is significantly improved compared with SVF. In terms of search efficiency, it is improved by about 20% under different total number of drones. In terms of reduction of total energy consumption, the performance is improved as the total number of drones increases. In terms of algorithm search stability, there is a significant improvement under different total number of drones.

[0180] The four search algorithm modes of PCDI, PSDI, KCDI, and KSDI all have the same task division. When searching for multiple moving targets in an unknown obstacle environment, the swarm of drones effectively completed the search for multiple targets using the four search algorithm modes. This verifies the effectiveness of distance-based dynamic task division in the search algorithm. Visualizing the average values ​​in Table 4, we get Fig.17 ; The curve shown in the figure shows that regardless of the total number of drones, the algorithm performance of the search algorithm mode is basically ranked from best to worst as PCDI, KCDI, PSDI, and KSDI.

[0181] The present invention improves the target signal model so that the model can be more widely adapted to the existing sensor functions. Based on the improved search model, a distance-based dynamic task division method suitable for the improved target signal model is proposed. The unknown obstacle environment is rasterized, and the confidence region pheromone of the confidence region is proposed based on the confidence region. The reliability of the individual extreme value and the global extreme value of the particle swarm algorithm is analyzed, and a probability-limited particle swarm algorithm is proposed to ensure the reliability of the extreme value.

[0182] The simulation results show that the distance-based dynamic task division is suitable for the task division of swarm drones. After comparing different search algorithm modes, the confidence region pheromone greatly improves the performance of multi-moving target search, and has a significant improvement in search efficiency, energy saving, and search stability; although the probability-restricted particle swarm algorithm has only a slight improvement in search efficiency and energy saving, and even in some cases the total energy consumption is higher than the traditional method, it has a significant improvement in search stability.

Claims

1. A pheromone-inspired multi-motion target search method for a swarm of drones, characterized in that: The following steps are involved: Step 1: Establish a multi-dynamic target search model for a swarm of drones; In the step 1, the established swarm drone multi-dynamic target search model includes: Mission environment: The mission area is considered to be a two-dimensional area with a constant altitude of the drone. In the grid area with a length × width of Lv × Lh, a single grid is represented by G (lo，la) Indicated by, where 1≤lo≤Lv, 1≤la≤Lh, the grid side length is ΔL0, and the grid center (lo, la) is used to represent the position of the grid, denoted by G (lo，la) ; Task object: The object is represented by the set A = UAV∪T∪B; where the set UAV = {uav i |i=1,2,...,n V ; 10≤n V ≤100} represents drone, n V represents the total number of drones; set T = {tar j |j=1,2,...,n T ;n T ≥1} indicates the target, n T Represents the total number of targets; set B = {obs m |m=1,2,...,n B ;n B ≥1} indicates an obstacle, n B Represents the total number of obstacles; at time t, the UAV i 、Target tar j 、Obstacles m The positions of uav i|t , tar j|t 、obs m|t ;UAV i1 UAV i2 The Euclidean distance between i1，i2|t =||uav i1|t -uav i2|t ||, i1=1,2,...,n V , i2=1,2,...,n V , and i1≠i2; UAV i with target tar j The Euclidean distance between i,j|t =||uav i|t -tar j|t ||; UAV i The Euclidean distance between the obstacle obsm and the obstacle is Dvb i，m|t =||uav i|t -obs m|t ||; target tar j With obstacles obs m The Euclidean distance between j，m|t =||tar j|t -obs m|t ||; Mission: Within the search range, drones search for targets through mutual cooperation; Under the premise of avoiding collision, drones search for target positions based on detected target signals, information exchange, and their own movement; When all drones have arrived within the d0 range near all the targets to be searched, the target search mission is completed; UAV: UAV i The velocity at time t is denoted as Vv i|t , whose size satisfies 0≤||Vv i|t ||≤V vm , V vm Represents the maximum value of the UAV speed, which is a scalar; and the location of the UAV is the center of the grid; considering the maximum communication radius r com , maximum obstacle perception radius r obs , maximum target detection radius r tar , the drone can realize the following functions: i1，i2|t ≤r com When the UAVs communicate with each other; Dvt i,j|t ≤r tar When the target signal is detected; Dvb i，m|t ≤r obs Obtain the obstacle position when Target: target tar j The velocity at time t is denoted as Vt j|t , whose size satisfies 0<||Vt j|t ||≤V tm , V tm It represents the maximum value of the target speed, which is a scalar; and the target location is arbitrary; when the group of drones searches for a target, the drone detects the surrounding target signals through sensors. The target signal and distance satisfy a function that is interfered by the environment, which is described as the target signal detection function and is expressed by formula (1): Among them I i，| t represents the UAV at time t i The detected target signal; η is the environmental interference; G(Dvt i，j|t ) represents the relationship between the target signal and the distance in an interference-free environment; Dvt i，j|t Objectively exists but the drone cannot know it. When there are multiple targets within the drone's target detection range, the weaker target signal will be covered by the stronger target signal, and the drone cannot distinguish the signals of different targets; Assumption 1: The relationship between the parameters satisfies: V tm <V vm <<r tar ≈r obs <<r com , and are all known; Assumption 2: All drones participating in the target search are homogeneous, and all the targets being searched are of the same type; Step 2: Perform dynamic task division based on distance on the UAV to obtain the search status of the UAV for the dynamic target; Step 3: Determine the search status of the UAV for dynamic targets. If the UAV is in a roaming state, proceed to step 4; if it is in a collaborative search state, proceed to step 5; if the UAV is in a search mission completion state, the search mission is completed and all operations are terminated; Step 4: Conduct roaming search based on pheromones in the confidence area; The specific process of step 4 is as follows: (4-1) Assume that the function δ(D) is expressed as follows: Where n r is the number of columns of matrix D, n l is the number of rows in matrix D, b ζ，ξ is the value of the ζth row and ξth column in D; function The expression is as follows: in: (4-2) Define the confidence region and pheromone for dynamic target search: Definition 1: The area where the target is determined to be non-existent is called the confidence area, denoted by C z , the non-confidence region is express; UAV i At time t0, the drone reaches position P1 in the grid area, but no target signal is detected. The detection range of the drone is the confidence area. At time t0+Δt, the drone leaves P1, where Δt is the time increment. Since the target speed is limited, a certain circular confidence area is still maintained around P1. The distance ΔU from the edge of the confidence area to P1 satisfies the following formula: uav i When reaching P′1 at time (t0+Δt), if Δt is small enough, the confidence regions corresponding to P1 and P′1 will form a continuous confidence region; Corollary 1: The drone detection target is divided into three cases: Case 1: The distance near the drone is less than or equal to r tar There are obstacles or mission environment boundaries within the range; Case 2: The distance near the drone is less than or equal to 2r tar There are other drones within the range; Case 3: The distance near the drone is less than or equal to r tar There are no obstacles or task environment boundaries within the range, and the distance is less than or equal to 2r tar There are no other drones within the range; the maximum confidence area of ​​Case 1 and Case 2 combined is greater than the maximum confidence area of ​​Case 3. The confidence area detected by the Case 3 drone is recorded as C z，1 The confidence region where the sum of Case 1 and Case 2 is greater than that of Case 3 is recorded as C z，2 ; In the mission environment, the confidence region is represented by pheromone; the drone stores the pheromone strength at different locations. Indicates uav i The pheromone at position P0 stored at time t, the pheromone value range in the confidence region is [0, τ m ],τ m is the maximum value of pheromone in the confidence region, that is, confidence region C z，2 The pheromone value of the confidence area takes the minimum value in the grid where the drone is located. Confidence Region C z，2 The maximum value of pheromone The pheromone intensity in the non-confidence region is recorded as UAV at t0 i Arrived at P1, the pheromone With P1 as the center, r tar The circle with radius C z，1 ; The distance between any point P0 and P1 in the search area is d p ≥0, the pheromone of P0 is: From formula (7) and formula (8), we can see that under the premise of satisfying the properties of the confidence region and the range of pheromone values, C z，1 The distance of edge movement and time Δt, the pheromone intensity at point P0 and the distance d p They are linearly correlated, so when the drone is far away from P1, the pheromone change and time Δt also satisfy the linear relationship: Where ρ is the pheromone enhancement parameter; when time P1 becomes the confidence region C z，2 , the pheromone at this point becomes the maximum value From this we can get the pheromone enhancement parameter ρ = τ m V tm / r tar ; When the confidence region C z，1 The grid pheromone increases to τ m When , it becomes the confidence region C z，2 ; The pheromone change of P0 satisfies the following formula: In the search area, in order to save the storage space of the drone, the drone only stores the pheromone of one point in each grid; i|t Indicates uav i The pheromone strength of all grids stored at time t, PH i|t It is expressed as formula (11): Any element of the matrix Indicates uav i The grid G ​​stored at time t (lo，la) pheromones; The target can be anywhere in the search area. Therefore, only when the entire grid is within the drone detection area and the drone does not detect the target signal can it be judged that the grid has no target and belongs to the confidence area. Here, the detectable grid is defined as follows: Definition 2: When a grid G (lo，la) Farthest point and UAV i Distance||(G (lo,la) +0.5ΔL·δ(G (lo,la) -uav i ))-uav i ||≤r tar , this grid is called the detectable grid, denoted as DG, uav i The set of all detectable grids is denoted as DG i ; On the contrary, ||(G (lo,la) +0.5ΔL·δ(G (lo,la) -uav i ))-uav i ||>r tar , G (lo，la) It is called an undetectable grid, denoted as UG; if the target undetectable grid has a connected detectable grid, it is called the nearest undetectable grid, denoted as NUG; For UAV i , any detectable grid The pheromone satisfies the following formula: ΔE is an intermediate variable; a discrete iteration method is adopted in time. From formula (10), it can be known that any grid G (lo，la) The iterative relationship of pheromone satisfies the following formula: For region C z，2 , when there is a grid Make G (lo，la) When the following formula is met, G (lo，la) This becomes a non-confidence region: ΔH is the intermediate variable; When the grid For any UAV i The stored grid pheromone is updated 5 times in each iteration. The updating process of pheromone in each iteration is as follows: Step 1: Calculate and update all grid pheromones according to formula (13); Step 2: Calculate and update the pheromones of all detectable grids according to formula (12); Step 3: CUAV i ={uav I1 , uav I2 ,...,uav In } indicates that it can be used with uav i A collection of communicating drones; UAV i Get CUAV i The pheromones stored by all drones in the , and all the grid pheromones stored by them are updated as follows: Step4: If uav i In collaborative search state, for sub-alliance Any drone in TU J Type_I drones are Type_Ⅱ UAV is gather All grids in become untrusted areas, and pheromones are updated as follows: Step 5: The pheromones of all grids within the search range are updated as follows: (4-3) Set the roaming search speed: Definition 3: When a grid G (lo，la) UAV i Distance||G (lo,la) -uav i ||≤V vm , this grid is called the reachable grid, denoted as RG, uav i The set of all reachable grids is denoted as RG i ; In the UAV i If a reachable grid is the farthest reachable grid from the UAV in a certain angular direction, then the reachable grid is called the farthest reachable grid. There is a corresponding relationship between the farthest reachable grid and the nearest undetectable grid. Assuming that a farthest reachable grid is the farthest reachable grid in the direction of the angle interval (θ1, θ2), if the angle of the vector formed by the drone pointing to the farthest point of a nearest undetectable grid is within the interval (θ1, θ2), then the nearest undetectable grid corresponds to the farthest reachable grid. In the roaming state, in order to quickly detect more areas, the drone will move at the maximum speed in the direction after obtaining the speed direction according to the pheromone, and select the farthest reachable grid in the speed direction as the next location; after updating the pheromone, any drone in the roaming state uav i The pheromone in the grid that can be detected by the surrounding targets will satisfy equation (12), which is consistent with the uav i The pheromones of grids with the same distance are equal; however, the pheromones of undetectable grids will be different even if they are at the same distance from the drone. i The next flight direction is determined based on the pheromone intensity of the nearest undetectable grid; uav i The farthest reachable grid is denoted as FG i，k , where 1≤k≤k i,re , k represents the kth farthest reachable grid clockwise from the positive half axis of x, k i,re FG is the maximum reachable grid number; i，k The pheromone of the corresponding nearest undetectable grid is denoted as Fph i，k,n , where 0≤n≤n i，gr , n represents the nth nearest undetectable grid clockwise, n i，gr For FG i，k The number of the nearest undetectable grids, uav i The maximum pheromone corresponding to the kth most reachable grid uav i Obtain FG i，k and Fph i,k,max After that, update the speed according to the roulette method, which is expressed by Equation (18); the random number r a ~U(0, 1), when Q(K - 1) < r a < Q(K), uav i Select FG i，K as the next expected position, and the next expected speed Vve i|(t+1) satisfies Equation (19), where FG i，K represents the position of FG i，K and α represents the kinematic inertia coefficient: Ve i|(t+1) =αVv i|t +(1-α)(FG i，K -uav i|t ) (19) q k Indicates that the drone selects the farthest reachable grid FG i，k As the probability of the next step position grid; Step 5: Conduct collaborative search based on probability-constrained particle swarm algorithm; In the step 5: The global optimality in the collaborative state is the optimal one in the sub-alliance, which is called the semi-global optimality and is denoted as sgb J|t The individual historical best is the historical best after the drone individual joins the sub-alliance, denoted as pb i|t ; In dynamic target search, the fitness function value at the same position changes with time. The reliability of individual historical optimal and global optimal for system optimization gradually decreases with iteration. Therefore, probability values ​​are introduced into the update of semi-global optimal and individual historical optimal respectively, and the inheritance of the optimal in particle swarm optimization PSO is limited by probability values. At the same time, the calculation method of individual historical optimal is improved, and probability-restricted particle swarm optimization PFPSO is proposed. Assume that at time t the UAV i The fitness function of the location is expressed as f(uav i|t ), the update method of the global optimum and individual historical optimum of PFPSO is as follows: Semi-global optimality introduces the global optimal probability value σ j , whose initial value is σ0, where 0<σ0<1; for any sub-coalition TU J|t , among which the Class I UAV with the strongest target signal is uav J , T.U. J|t The corresponding semi-global optimal update process is as follows: Step 1: Determine whether there is a sub-alliance at time (t-1) Among them, the Class I drone has the strongest target signal. If it exists, assign a uniformly distributed random number r between [0, 1] σ , and execute Step2; otherwise sgb J|t =uav J|t , σ J =σ0, the update ends; Step 2: If the condition holds true, f() is the fitness function expressed by formula (21), sgb J|t =uav J|t , σ J =σ0, the update is completed; otherwise, execute Step3; Step 3: End Update; for The semi-global optimum of Individual historical optimality introduces the individual historical optimal probability value μi, whose initial value is μ0, where 0<μ0<1; for any UAV in the collaborative search state i , whose individual historical optimal pb i|t The update process is as follows: Step 11: Determine when the drone joins the sub-alliance. If it joins the sub-alliance at time t, update pb according to formula (20) i|t , and assign μ i =μ0, the update ends; otherwise, a random number r uniformly distributed between [0, 1] is assigned μ , and execute Step 12; Step 12: If the condition (f(uav i|t )≥f(pb i|(t-1) ))||(f(uav i|t ) <f(pb i|(t-1) )&r μ ≥μ i ) holds true, update pb according to formula (20) i|t , and assign μ i =μ0, the update is completed; otherwise, execute Step 13; Step 13: pb i|t =pb i|(t-1) , μ i =μ i μ0, update completed; UAVs in the collaborative search state are divided into two categories: type_I and type_II. For any type I UAV in a sub-alliance, the best individual history still selects the best position in the individual history; and for any type II UAV in a sub-alliance i If uav i The target information is communicated from the UAV Take the best individual history of the drone as the midpoint of the two; i The individual historical optimality of is updated according to formula (20): Different types of UAVs in collaborative search states have different fitness function calculation methods; in any sub-coalition, if the UAV i is type_I, the fitness function value is the target signal detected by the sensor; if the UAV i For type_II, For its nearest type_I communication drone, the fitness function value is uav i and The opposite of distance, that is: In summary, the expected speed calculation of PSO with probability constraints is shown in formula (22): Where c1 and c2 are cognitive coefficients and social coefficients respectively, r1 and r2 are random numbers that obey U(0,1), ω is the inertia weight, α is the inertia link considering kinematics, and pb i|t for UAV i The individual historical optimum at time t, sgb J|t for UAV i Sub-alliance TU J|t The semi-global optimality of Vvp i|(t+1) is an intermediate variable.

2. The pheromone-inspired swarm drone multi-moving target search method according to claim 1 is characterized in that: In step 2, when the group of drones searches for targets, their task is to search for all targets in the task area, and searching for targets in a local area is a subtask, and the number of subtasks is at most n. T , drones that complete the same subtask form a sub-alliance, and the sub-alliance at time t is recorded as TU J|t , where 0≤J≤n T , the serial number J is arranged from large to small according to the strongest target signal detected in the sub-alliance. The state of the drones that make up the sub-alliance is called the collaborative search state. The drones form a sub-alliance in the following way: UAVs obtain target signals through two ways, so they are divided into two categories: Category I UAVs that detect target signals by their own sensors and Category II UAVs that obtain target signals through communication between UAVs. Category I UAVs are denoted as type_I and Category II UAVs are denoted as type_II. First, Category I UAVs are grouped and the UAVs with the strongest target signals in the local area are selected. i0 , when other Class I UAVs are nearby i With UAV i0 If equation (2) is satisfied, then i0 Forming a sub-alliance TU 1|t The remaining Class I drones that do not satisfy formula (2) continue to select the drone with the strongest target signal. Then search for the remaining drones nearby. The drones whose relationship meets formula (2) form a sub-alliance, and this cycle repeats until all Class I drones form a sub-alliance; then, Class II drones join the sub-alliance of the nearest Class I drone that can communicate; 3. The pheromone-inspired swarm drone multi-moving target search method according to claim 2 is characterized in that: In step 2, after the subtasks are assigned, in order to avoid wasting drone resources due to too many drones in the same sub-alliance, a closed-loop adjustment is further introduced to evaluate the resource allocation of each sub-alliance: when the number of drones in a sub-alliance is N J >N max , the sub-alliance selects N max UAVs are the individuals of the sub-alliance, where N max The maximum number of sub-alliance drones allowed; the principle of selection is: Class I drones are given priority, among Class I drones, those with strong target signals are given priority, and among Class II drones, those that are close to the communication drone are given priority; The search of UAVs for dynamic targets is divided into three states: the roaming state that quickly covers the search area, the collaborative state that forms a sub-alliance, and the search mission completion state.

4. The pheromone-inspired swarm drone multi-moving target search method according to claim 3 is characterized in that: In step 4 or step 5, whether the drone is in cooperative state or roaming state, the expected speed Vve is calculated. i|(t+1) Then consider obstacle avoidance; based on the boundary scanning obstacle avoidance strategy, it is improved into an improved boundary scanning obstacle avoidance IBSOA adapted to the grid environment. When the drone avoids obstacles, the perceived obstacles are divided into two situations: continuous obstacles and discontinuous obstacles. The obstacle avoidance direction is divided into counterclockwise obstacle avoidance and clockwise obstacle avoidance. The drone comprehensively considers the perceived obstacle situation and the historical obstacle avoidance direction to calculate the next speed; Due to the rasterization of the map, the grids with obstacles are called obstacle grids; IBSOA ignores obstacles and comprehensively considers the obstacle avoidance direction and obstacle grid to avoid obstacles; if the drone is in roaming state and the distance near it is less than min(r tar , t obs ) range, start to consider obstacle avoidance; if the drone is in a collaborative state and the distance near it is less than Vv m If there is an obstacle within the range, obstacle avoidance begins. The obstacle avoidance process of IBSOA is represented by the following two points:

1. UAV i In roaming state, and the distance near it is less than min(r tar , r obs ), <Vve i|(t+1) > There are obstacles in the direction, uav i Start obstacle avoidance; uav i The set of all the farthest reachable grids at time t is denoted as FRG i|t , if there is a grid G (lo，la) ∈FRG i|t , meet the UAV location uav i|t Point to grid position G (lo，la) The ray is at the distance uav i|t Less than min(r tar , r obs ) passes through the obstacle grid, then the grid G (lo，la) From FRG i|t When all grids that meet the above conditions have been removed from FRG i|t After culling, the final grid set FOG is obtained i|t ;uav i According to its own obstacle avoidance direction flag parameter λ i , determines the speed direction after considering obstacle avoidance, λ i The initial value is 0; When i = 0, indicating that the obstacle was not avoided in the previous step, and the grid set FOG i| Select Vve from t i|(t+1) The grid with the smallest angle difference Gn i As a UAV i The grid to be reached next; When i = 1, from the grid set FOG i|t Select the counterclockwise direction with Vve i|(t+1) Angle difference grid Gn i As a UAV i The grid to be reached next; When i = -1, from the grid set FOG i|t Select the clockwise direction with Vve i|(t+1) The grid with the smallest angle difference Gn i As a UAV i The grid to be reached next; Then according to the grid Gn i And equation (23), solve the next step speed Vv i|(t+1) , where Gn i Gn i Position; after obtaining the velocity, update λ according to formula (24) i , the calculation of obstacle avoidance speed ends: l i =sgn(F( <Vv i|(t+1) >- <Vve i|(t+1) >)) (24) 2. UAV i In the cooperative state, and in its vicinity, the distance is less than Vv m Within the range, <Vve i|(t+1) > There are obstacles in the direction, uav i Start obstacle avoidance; uav i The set of all reachable grids at time t is denoted as RG i|t ; if G (lo，la) ∈RG i|t Satisfy, grid position G (lo，la) UAV location i|t The connecting line passes through the barrier grid, and the grid is connected from RG i|t Eliminate, when all grids that meet the above conditions have been removed from RG i|t After culling, we get the final grid set OG i|t ; According to the obstacle avoidance direction sign parameter λ i , determines the speed direction after considering obstacle avoidance, λ i The initial value is 0; When i = 0, indicating that the obstacle was not avoided in the previous step, and the grid set OG i|t Select and (uav i|t +Vve i|(t+1) ) The nearest accessible grid without obstacles is grdn i As a UAV i The grid to be reached next; When i = 1, from the grid set OG i|t Select the counterclockwise direction with (uav i|t +Vve i|(t+1) ) The nearest accessible grid without obstacles is grdn i As a UAV i The grid to be reached next; When i = -1, from the grid set OG i|t Select the clockwise direction with (uav i|t +Vve i|(t+1) ) The nearest accessible grid without obstacles is grdn i As a UAV i The grid to be reached next; Then according to the grid grdn i And equation (25), solve the next step speed Vv i|(t+1) , where grdn i for grdn i After obtaining the position and velocity, update λ according to formula (26) i , calculation of obstacle avoidance speed ends; l i =sgn(F( <Vv i|(t+1) >- <Vve i|(t+1) >)) (26) According to formula (23) or (25), the next step speed Vv is calculated i|(t+1) Then, the drone position is updated according to formula (27) and it is determined whether the target is found: to do i|(t+1) =water i|t +Vv i|(t+1) (27).

Images