UAV swarm resource scheduling method based on improved wolf pack algorithm

Abstract

The present invention discloses a method for unmanned aerial vehicle (UAV) swarm resource scheduling based on an improved wolf pack algorithm. First, a mathematical model for UAV swarm resource scheduling is established. Then, the algorithm is improved from aspects such as initializing the wolf pack with chaotic mapping, simulating the exploration wolf's wandering step length with dynamic Lévy random motion, designing a dynamic distance determination factor and an adaptive running step length, simulating the wolf pack's siege behavior with Brownian motion, and designing a wolf pack update mechanism based on the degree of difference. Finally, the improved wolf pack algorithm is used to optimize and solve the UAV resource scheduling model, and the optimal UAV swarm resource scheduling scheme is obtained. The present invention solves the problems of slow algorithm solving speed, easy obtaining of local optimal solutions, and insufficient utilization of UAV swarm resources existing in the process of UAV swarm resource scheduling.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a method for resource scheduling of unmanned aerial vehicle swarms. Background Technique

[0002] Unmanned aerial vehicle swarm resource scheduling is a process in the confrontation between multiple unmanned aerial vehicles and multiple targets. It comprehensively evaluates indicators such as the spatial situation and performance of unmanned aerial vehicles and targets, takes constraints such as scheduling benefits and costs, obtains an optimal unmanned aerial vehicle resource scheduling plan with the best comprehensive evaluation, and schedules unmanned aerial vehicles to targets.

[0003] To study the problem of unmanned aerial vehicle swarm resource scheduling, it is necessary to model it first, construct an objective function, and then obtain the optimal resource scheduling plan through computer analysis and calculation. Modeling involves evaluating indicators such as the spatial situation and performance of unmanned aerial vehicles. The spatial situation includes relative angle, relative height, relative speed, relative distance, and relative energy; the performance includes maneuverability, attack ability, detection ability, control ability, survival ability, range ability, and electronic countermeasure ability. It can be seen that the problem of unmanned aerial vehicle resource scheduling is very complex. Without changing the basic computing power of the computer, it takes a lot of time to find the optimal solution of the objective function of unmanned aerial vehicle swarm resource scheduling. Therefore, using an algorithm to solve this problem can obtain the optimal solution in the shortest time. Summary of the Invention

[0004] In order to solve the technical problems mentioned in the above background technique, the present invention proposes a method for unmanned aerial vehicle swarm resource scheduling based on an improved wolf pack algorithm, which solves the problems of slow solution speed, easy to obtain local optimal solutions, and inability to fully utilize unmanned aerial vehicle resources in the optimization of the algorithm for solving unmanned aerial vehicle swarm resource scheduling problems.

[0005] In order to achieve the above technical objectives, the technical solution of the present invention is as follows:

[0006] A method for unmanned aerial vehicle swarm resource scheduling based on an improved wolf pack algorithm includes the following steps:

[0007] (1) Establish an unmanned aerial vehicle swarm resource scheduling model according to the spatial situation and performance;

[0008] (2) Improve the wolf pack algorithm and use it to optimize and solve the problem of unmanned aerial vehicle swarm resource scheduling; the improved wolf pack optimization algorithm includes initializing the wolf pack by chaotic mapping, simulating the wandering step length of exploring wolves using dynamic Lévy random motion, designing a dynamic distance determination factor and an adaptive running step length, simulating the wolf pack siege behavior using Brownian motion, and designing a wolf pack update mechanism based on the degree of difference.

[0009] Further, in step (1), first, the comprehensive advantage A of the UAV over the target and the comprehensive advantage B of the target over the UAV are obtained based on the space situation and performance; then, based on the comprehensive advantage A of the UAV over the target and the comprehensive advantage B of the target over the UAV, the expressions for the benefits and costs of the UAV against the target are obtained; finally, a resource scheduling model for the UAV swarm is established with the goals of minimizing benefits and costs.

[0010] Further, the resource scheduling model for the UAV swarm is as follows:

[0011]

[0012] where Z is the objective function; x ij ={0, 1}, x ij =1 indicates that UAV i confronts target j, and x ij =0 indicates that UAV i does not strike target j; W ij , F ij respectively represent the benefit coefficient and cost coefficient of UAV i against target j; J i , J j respectively represent the values of the UAV and the target; m and n respectively represent the number of UAVs and targets participating in the scheduling; MaxNum i represents the maximum number of targets that a single UAV can confront simultaneously.

[0013] Further, the expressions for the benefit coefficient W and the cost coefficient F are as follows:

[0014]

[0015] where sgn(x) is the sign function, 0≤W≤1, 0≤F≤1.

[0016] Further, the expression for the comprehensive advantage A of the UAV over the target is as follows:

[0017] A = α1S1 + α2S2

[0018] where α1 and α2 are the weight coefficients of each index and the sum of the coefficients is 1, and S1 and S2 respectively represent the space situation advantage and performance advantage of the UAV relative to the target:

[0019] S1 = ω1t1 + ω2t2 + ω3t3 + ω4t4 + ω5t5

[0020] where t1, t2, t3, t4, and t5 respectively represent the angular advantage, altitude advantage, speed advantage, distance advantage, and energy advantage of the UAV relative to the target, and ω1, ω2, ω3, ω4, and ω5 respectively represent the weight coefficients of each index and the sum of the coefficients is 1;

[0021] S2 = η1c1 + η2c2 + η3c3 + η4c4 + η5c5 + η6c6 + η7c7

[0022] Among them, c1, c2, c3, c4, c5, c6, c7 represent the maneuverability advantage, attack ability advantage, detection ability advantage, operation ability advantage, survival ability advantage, range ability advantage, and electronic countermeasure ability advantage of the UAV relative to the target in sequence. η1, η2, η3, η4, η5, η6, η7 represent the weight coefficients of each index in sequence, and the sum of the coefficients is 1;

[0023] Similarly, use the above method to calculate the comprehensive advantage B of the target against the UAV.

[0024] Furthermore, the specific process of step (2) is as follows:

[0025] (a) Artificial wolf position encoding

[0026] Take the objective function of the UAV swarm resource scheduling model established in step (1) as the objective function optimized by the wolf pack algorithm. According to the characteristics of UAV swarm resource scheduling, the position encoding of artificial wolf i is {x i1 , x i2 , ···, x ij , ···, x iL}, 1 ≤ x ij ≤ L, where L represents the artificial wolf encoding length calculated according to the number of UAVs and targets participating in the scheduling. The position of the artificial wolf represents the UAV swarm resource scheduling plan, and x ij = s means that UAV j confronts target s;

[0027] (b) Initializing the wolf pack by the chaos mapping method

[0028] Determine the number a_initial of the initial wolf pack. Generate a 1×L random number (x1, x2, ···, x j , ···, x L ) through rand(1, L), where rand(1, L) represents generating a 1×L random number uniformly distributed between (0, 1); then substitute x1, x2, ···, x j , ···, x L into the Logistic chaos mapping formula in sequence and iterate a_initial - 1 times, that is, generate a_initial groups of 1×L-dimensional vectors in the chaotic state; for one group of vectors (x1, x2, ···, x j , ···, x L ), for each dimension element x1, x2, ···, x j , ···, x LSubstitute into the following formula in sequence, and finally obtain the 1×L-dimensional vectors X1, X2, ···, X j , ···, X L , which is the position of the initial random wolves. Substitute all a_initial groups of 1×L-dimensional vectors in the chaotic state into the following formula in the above method, and then generate a_initial initial wolf packs:

[0029]

[0030] Among them, means rounding to an integer;

[0031] (c) Generate the leading wolf

[0032] Calculate the objective function value of each wolf in the initial wolf pack and sort them. The wolf with the optimal objective function value is the leading wolf;

[0033] (d) Exploration wolf wandering behavior based on dynamic Lévy flight

[0034] Select a_explore wolves with the optimal objective function values except the leading wolf as exploration wolves. The exploration wolf group executes the wandering behavior. Exploration wolf i senses the prey odor concentration at the current position, that is, calculates the target fitness value y i at the current position. If y i is greater than the prey odor concentration y lead at the position of the leading wolf, then exploration wolf i replaces the leading wolf to become the new leading wolf and transfers to the summoning behavior in step (e); if y i ≤y lead , then exploration wolf i randomly selects h directions and advances step a steps in each direction. After recording the prey odor concentration at the end point of each direction, it returns to the original position, selects the direction with the largest prey odor concentration and greater than the prey odor concentration at the current exploration wolf position among the h directions and advances step a steps. Repeat the above process until the exploration wolf wandering times t reach the set value t max or the position of the leading wolf is updated, then transfer to the summoning behavior in step (e);

[0035] (e) Summoning behavior based on dynamic distance determination factor and adaptive running step length

[0036] All wolves except the leading wolf are regarded as fierce wolves. The leading wolf executes the summoning behavior. The fierce wolves run towards the leading wolf at a step length of step b . If during the raid, the prey odor concentration y i perceived by fierce wolf i > y lead , then this fierce wolf replaces the leading wolf to become the new leading wolf and initiates the summoning behavior again, otherwise this fierce wolf continues to run towards the leading wolf until the distance d between it and the leading wolfi,lead ≤ d near , where d near is the judgment distance, and then switch to the siege behavior;

[0037] (f) Siege behavior based on Brownian motion

[0038] All wolves except the alpha wolf search the area near the alpha wolf, that is, move in a random direction by step c steps. If the prey odor concentration at the moved position is greater than that at the original position, update the position of the wolf. After the siege behavior ends, update the wolf with the optimal prey odor concentration in the wolf pack at this time as the alpha wolf, and go to step (g);

[0039] (g) Wolf pack update mechanism based on difference degree

[0040] Sort the wolf pack according to the objective function value, eliminate the R wolves with smaller objective function values, and then randomly generate a wolf i and judge its difference degree sim i,lead with the alpha wolf lead. If it is higher than the set threshold sim thre , then supplement it into the wolf pack, otherwise discard it. Loop the above process until R new wolves are supplemented into the wolf pack, enhancing the optimization ability of the algorithm on the basis of enriching the diversity of the wolf pack;

[0041] (h) Judge the algorithm termination condition

[0042] If the algorithm reaches the maximum number of iterations it max or the optimal solution meets the optimization accuracy requirement, output the position of the alpha wolf, that is, the optimal UAV swarm resource scheduling scheme, otherwise go to step (d).

[0043] Furthermore, in step (d), let the exploration wolf's random walk step size step a follow the Levy distribution, and the exploration wolf position update formula is as follows:

[0044] x i+1 = Θ(x i , ceil(α·levy))

[0045] where α is the step size scaling factor, levy represents the random step size of Levy motion, ceil represents taking the integer upper bound, and Θ represents for the wolf x i , randomly select a pair of positions and swap the numbers at their positions, and loop this process ceil(α·levy) times to get the latest position x i+1 of the exploration wolf;

[0046] Design a dynamic step size scaling factor α(k) that changes with the number of iterations:

[0047]

[0048] Among them, D max represents the maximum number of iterations, k is the current number of iterations, and α(k) ∈ (0, 1).

[0049] Furthermore, in step (e), define the distance d between the fierce wolf i and the leader wolf lead i,lead as follows:

[0050]

[0051] Among them, represents the exclusive OR operation;

[0052] Determine the distance d near as follows:

[0053]

[0054] Among them, ceil represents taking the upper integer bound, and ω is the distance determination factor;

[0055] Design a dynamic distance determination factor ω(k) that changes with the number of iterations:

[0056]

[0057] Among them, D max represents the maximum number of iterations, k represents the current number of iterations, and ω(k) ∈ (1, L + 1).

[0058] Furthermore, in step (f), design the siege step size step c :

[0059]

[0060] Among them, step c follows a normal distribution with a mean of 0 and a variance of T - S. T and S respectively represent the upper and lower bounds of time within a certain time period of Brownian random motion. randn generates a random number with a normal distribution with a mean of 0 and a variance of 1, and ceil represents taking the upper integer bound.

[0061] Furthermore, in step (f), define the difference degree sim i1 , x i2 , ···, x ij , ···, x iL} between the random wolf {x lead,1 , x lead,2 , ···, x lead,j , ···, x lead,L} and the leader wolf {x i,lead :

[0062]

[0063] Among them, ⊕ represents the exclusive OR operation, and 0 ≤ sim i,lead ≤ 1.

[0064] Beneficial effects brought by adopting the above technical solution:

[0065] The present invention uses an improved wolf pack algorithm to optimize and solve the UAV resource scheduling model, and obtains the optimal UAV group resource scheduling scheme, thereby solving the problems of slow algorithm solving speed, easy to obtain local optimization solutions, and insufficient utilization of UAV group resources existing in the UAV group resource scheduling process. Description of the Drawings

[0066] Figure 1 is the method flow chart of the present invention;

[0067] Figure 2 is the change curve of the step size scaling factor with the number of iterations in the present invention;

[0068] Figure 3 is the change curve of the distance determination factor with the number of iterations in the present invention. Detailed Embodiment

[0069] The following will combine the drawings to elaborate on the technical solution of the present invention in detail.

[0070] The present invention designs a UAV group resource scheduling method based on an improved wolf pack algorithm, as Figure 1 shown, and the specific steps are as follows:

[0071] 1. Establish a mathematical model for UAV group resource scheduling:

[0072] The UAV group resource scheduling problem is to comprehensively evaluate the spatial situation and performance of UAVs and targets according to multi-UAV multi-target confrontation, pursue the maximum confrontation benefit and the minimum cost, and give the optimal scheduling scheme of UAVs to targets. The following establishes a UAV group resource scheduling model.

[0073] 1) Mathematical description of the spatial situation

[0074] The spatial situation refers to the three-dimensional spatial state information between UAVs and targets, including relative angle, relative height, relative speed, relative distance, and relative energy. The spatial situation advantage function of UAVs for targets is expressed as follows:

[0075] S1 = ω1t1 + ω2t2 + ω3t3 + ω4t4 + ω5t5 (1)

[0076] Among them, t1, t2, t3, t4, t5 represent the angular advantage, altitude advantage, speed advantage, distance advantage, and energy advantage of the UAV relative to the target in sequence, and ω1, ω2, ω3, ω4, ω5 represent the weight coefficients of each index in sequence, and the sum of the coefficients is 1;

[0077] 2) Mathematical description of the performance advantage function

[0078] Performance is an evaluation of the overall confrontation ability of a single UAV against a single target. Its main evaluation indicators are maneuverability, attack ability, detection ability, control ability, survivability, range ability, and electronic countermeasure ability. The performance advantage function of the UAV against the target is expressed as follows:

[0079] S2 = η1c1 + η2c2 + η3c3 + η4c4 + η5c5 + η6c6 + η7c7 (2)

[0080] Among them, c1, c2, c3, c4, c5, c6, c7 represent the maneuverability advantage, attack ability advantage, detection ability advantage, control ability advantage, survivability advantage, range ability advantage, and electronic countermeasure ability advantage of the UAV relative to the target in sequence, and η1, η2, η3, η4, η5, η6, η7 represent the weight coefficients of each index in sequence, and the sum of the coefficients is 1.

[0081] Considering the comprehensive space situation and performance advantage, the comprehensive advantage of the UAV against the target is expressed as follows:

[0082] A = α1S1 + α2S2 (3)

[0083] Among them, S1 and S2 represent the space situation advantage and performance advantage of the UAV relative to the target in sequence, and α1 and α2 are the weight coefficients of each index, and the sum of the coefficients is 1.

[0084] Similarly, the comprehensive advantage B of the target against the UAV can also be calculated using the above method. Thus, based on the comprehensive advantage A of the UAV against the target and the comprehensive advantage B of the target against the UAV, the benefit coefficient W and cost coefficient F of the UAV against the target are expressed as follows:

[0085]

[0086] Among them, sgn(x) is the sign function. When x < 0, the value is -1; when x = 0, the value is 0; when x > 0, the value is 1, 0 ≤ W ≤ 1, and 0 ≤ F ≤ 1.

[0087] The resource scheduling of the UAV swarm is a multi-objective optimization problem. Considering the benefits and costs of the UAV swarm resource scheduling comprehensively, with the goal of maximizing benefits and minimizing costs, the objective function is established as follows:

[0088]

[0089] Among them, x ij ={0, 1}, x ij =1 indicates that UAV i confronts target j, and x ij =0 indicates that UAV i does not strike target j; W ij , F ij respectively represent the benefit coefficient and cost coefficient of UAV i confronting target j; J i , J j respectively represent the values of UAVs and targets, which are related to the spatial tactical status and manufacturing costs of UAVs and targets; m and n respectively represent the numbers of UAVs and targets participating in the scheduling; Equation (7) means that at least one UAV is assigned to each target, and Equation (8) stipulates the maximum number of targets that a single UAV can confront simultaneously.

[0090] 2. Solve the UAV swarm resource scheduling problem using the improved wolf pack algorithm

[0091] Step1 Artificial wolf position encoding

[0092] Take Equation (6) as the objective function optimized by the wolf pack algorithm. According to the characteristics of UAV swarm resource scheduling, the position encoding of artificial wolf i is {x i1 , x i2 , ···, x ij , ···, x iL} (1 ≤ x ij ≤ L), where L represents the artificial wolf encoding length calculated according to the numbers of UAVs and targets participating in the scheduling. The position of the artificial wolf represents the UAV swarm resource scheduling plan, and x ij =s means that our UAV j confronts the enemy target s;

[0093] Step2 Initialize the wolf pack using the chaotic mapping method

[0094] The mathematical definition of the Logistic chaotic mapping is as follows:

[0095] z k+1 =μz k (1 - z k ) (9)

[0096] Among them, z k ∈(0, 1), μ is the control parameter. When 3.5699456 < μ ≤ 4, the Logistic mapping shows a chaotic state. When μ = 4, the Logistic mapping presents typical chaotic characteristics and the chaotic state is the best;

[0097] Determine the number a_initial of the initial wolf pack, and generate a 1×L random number (x1, x2, ···, x j , ···, xL ), where rand(1, L) represents generating a 1×L random number uniformly distributed between (0, 1); then x1, x2, ···, x j , ···, x L are successively substituted into Equation (9) and iterated a_initial - 1 times, that is, generating a_initial groups of 1×L - dimensional vectors in the chaotic state; for one group of vectors (x1, x2, ···, x j , ···, x L ), each - dimensional element x1, x2, ···, x j , ···, x L are successively substituted into Equation (10), and finally 1×L - dimensional vectors X1, X2, ···, X j , ···, X L are obtained. This is the position of the initial random wolves. Substituting all a_initial groups of 1×L - dimensional vectors in the chaotic state into Equation (10) according to the above method can generate an initial wolf pack with the quantity of a_initial;

[0098]

[0099] Step3 Generate the alpha wolf

[0100] Calculate the objective - function value of each wolf in the initial wolf pack and sort them. The wolf with the optimal objective - function value is the alpha wolf;

[0101] Step4 Exploration - wolf wandering behavior based on dynamic Lévy flight

[0102] Select a_explore wolves with the optimal objective - function values except the alpha wolf as exploration wolves. The exploration - wolf group performs the wandering behavior. Exploration wolf i senses the prey - odor concentration at the current position, that is, calculates the target fitness value y i at the current position. If y i is greater than the prey - odor concentration y lead at the position of the alpha wolf, then exploration wolf i replaces the alpha wolf to become the new alpha wolf and transfers to the summoning behavior; if y i ≤y lead , then exploration wolf i randomly selects h directions and advances step a steps in each direction. After recording the prey - odor concentration at the end point of each direction, it returns to the original position, selects the direction with the largest prey - odor concentration and greater than the prey - odor concentration at the current exploration - wolf position among the h directions and advances step a steps. Repeat the above steps until the exploration - wolf wandering times t reach the set value t max or the position of the alpha wolf is updated, and then transfer to the summoning behavior;

[0103] The Levy random motion is characterized by a large number of short step lengths accompanied by a small number of long step lengths, and let the exploration wolf's random walk step length be step a obeys the Levy distribution, which helps the exploration wolf expand the search range and jump out of the local optimum with a long step length when it gets stuck in the local optimum during the random walk. The exploration wolf's position update formula is as follows:

[0104] x i+1 = Θ(x i ,ceil(α·levy)) (11)

[0105] where α is the step length scaling factor, levy represents the random step length of the Levy motion, ceil represents taking the integer upper bound of α·levy, and equation (11) means that for the wolf x i = {x i1 ,x i2 ,···,x ij ,···,x iL}, randomly select a pair of positions and swap the numbers at those positions, and repeat this process ceil(α·levy) times to obtain the latest position x i+1 . The Mantegna algorithm is used to simulate the random step length levy of the Levy flight, and its mathematical definition is as follows:

[0106]

[0107] where β takes the constant value of 1.5, u and v respectively follow the normal distributions with expectations of 0 and variances of σ u 2 , σ v 2 , and σ u , σ v are expressed as follows:

[0108]

[0109] σ v = 1 (14)

[0110] where Γ is the standard gamma function.

[0111] Considering that the value of α will affect the optimization effect of the algorithm, in order to better coordinate the global search ability and local search ability of the algorithm, a dynamic step length scaling factor that changes with the number of iterations is designed as shown in equation (15), and the change curve of the step length scaling factor with the number of iterations is as Figure 2 shown.

[0112]

[0113] where D max represents the maximum number of iterations, k is the current number of iterations, and α(k) ∈ (0~1). FromFigure 2 It can be seen that in the initial stage of algorithm iteration, α increases with the increase of the number of iterations, and the algorithm conducts a global large-scale search in the wolf pack space. In the later stage of iteration, as the step size scaling factor gradually decreases, the algorithm gradually turns to conduct a local fine search in the wolf pack space.

[0114] Step5 Summoning behavior based on the dynamic distance determination factor and the adaptive running step

[0115] All wolves except the lead wolf are regarded as fierce wolves. The lead wolf executes the summoning behavior, and the fierce wolves run towards the lead wolf at a step size of step b . If during the rush, the prey odor concentration y i sensed by fierce wolf i lead > y i,lead , then this fierce wolf replaces the lead wolf to become the new lead wolf and initiates the summoning behavior again. Otherwise, this fierce wolf continues to run towards the lead wolf until the distance d near (d near is the determination distance) between it and the lead wolf is less than or equal to d i,lead . After that, it turns to the siege behavior. The distance d

[0116]

[0117] between fierce wolf i and the lead wolf lead is defined as follows: where near represents the exclusive OR operation, where the numbers at the corresponding positions are the same as 0 and different as 1; the determination distance d

[0118]

[0119] is expressed as follows: where represents taking the integer upper bound of Figure 3 . Considering that the value of ω will affect the optimization effect of the algorithm, in order to better coordinate the local search ability and the global search ability of the algorithm, a dynamic distance determination factor that changes with the number of iterations is designed as shown in Equation (18). The change curve of the distance determination factor with the number of iterations is as

[0120]

[0121] shown. max where D Figure 3 represents the maximum number of iterations, k represents the current number of iterations, and ω(k) ∈ (1 ~ L + 1). It can be seen from near that in the initial stage of algorithm iteration, the value of the distance determination factor ω is relatively large, that is, d nearGradually increasing, the algorithm begins to conduct a fine search of the wolf pack space. In the later stage of the algorithm iteration, the gradually increasing distance determination factor enlarges the running range of the fierce wolves so that the algorithm can jump out of the local optimum;

[0122] Meanwhile, to accelerate the convergence speed of the algorithm, an adaptive running step length step of the fierce wolf i is designed b as follows:

[0123]

[0124] It can be seen from Equation (19) that when the fierce wolf is far from the lead wolf, the fierce wolf runs towards the lead wolf with a larger step length. When the fierce wolf gradually approaches the lead wolf, it moves towards the lead wolf with a smaller step length;

[0125] Step6 Siege behavior based on Brownian motion

[0126] All wolves except the lead wolf search the area near the lead wolf, that is, move step in a random direction c steps. If the prey odor concentration at the moved position is greater than that at the original position, update the position of the wolf. After the siege behavior ends, update the wolf with the optimal prey odor concentration in the wolf pack at this time as the lead wolf, and go to Step7;

[0127] The random motion pattern of Brownian motion is similar to the siege behavior of the wolf pack, and the change range of the random step length is very small. Use Brownian motion to simulate the siege behavior of the wolf pack and design the siege step length step c as follows:

[0128]

[0129] where step c obeys the normal distribution with a mean of 0 and a variance of T - S. T and S represent the upper and lower bounds of time in a certain time period of Brownian random motion in turn. randn generates a normal distribution random number with a mean of 0 and a variance of 1, denotes taking the integer upper bound of;

[0130] Step7 Wolf pack update mechanism based on the degree of difference

[0131] Sort the wolf pack according to the objective function value, eliminate the R wolves with smaller objective function values, and then randomly generate a wolf i and judge its degree of difference sim i,lead with the lead wolf lead. If it is higher than the set threshold sim thre , then supplement it into the wolf pack, otherwise discard it. Loop the above steps until R new wolves are supplemented into the wolf pack to enhance the optimization ability of the algorithm on the basis of enriching the diversity of the wolf pack;

[0132] Random wolf {x i1,x i2 ,···,x ij ,···,x iL}, and the difference degree from the alpha wolf {x lead,1 ,x lead,2 ,···,x lead,j ,···,x lead,L} is defined as follows:

[0133]

[0134] Wherein, represents the exclusive OR operation, where the numbers at the corresponding positions are the same as 0 and different as 1;

[0135] Step8 Judgment Algorithm Termination Condition

[0136] If the algorithm reaches the maximum number of iterations it max or the optimal solution meets the optimization accuracy requirement, then output the position of the alpha wolf, that is, the optimal UAV swarm resource scheduling scheme, otherwise go to Step4.

[0137] The embodiments are only for illustrating the technical idea of the present invention, and the protection scope of the present invention cannot be limited thereby. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the present invention.

Claims

1. An unmanned aerial vehicle swarm resource scheduling method based on an improved wolf pack algorithm, characterized in that, It includes the following steps: (1) Establish a resource scheduling model for the UAV swarm according to the space situation and performance; (2) Improve the wolf pack algorithm and use it to optimize and solve the resource scheduling problem of the UAV swarm; the improved wolf pack optimization algorithm includes initializing the wolf pack by chaotic mapping, simulating the wandering step length of the exploring wolf using dynamic Lévy random motion, designing a dynamic distance determination factor and an adaptive running step length, simulating the siege behavior of the wolf pack using Brownian motion, and designing a wolf pack update mechanism based on the degree of difference; In step (1), first obtain the comprehensive advantage A of the UAV over the target and the comprehensive advantage B of the target over the UAV according to the space situation and performance; then, based on the comprehensive advantage A of the UAV over the target and the comprehensive advantage B of the target over the UAV, obtain the expressions for the benefit and cost of the UAV to the target; finally, establish a resource scheduling model for the UAV swarm with the goal of minimizing the benefit and cost; The resource scheduling model for the UAV swarm is as follows: Among them, Z is the objective function; x ij ={0,1}, x ij =1 indicates that UAV i counteracts target j, and x ij =0 indicates that UAV i does not strike target j; W ij , F ij respectively represent the benefit coefficient and cost coefficient of UAV i counteracting target j; J i , J j respectively represent the values of UAVs and targets; m and n respectively represent the number of UAVs and targets participating in the scheduling; MaxNum i represents the maximum number of targets that a single UAV can counteract simultaneously; The expressions for the benefit coefficient W and the cost coefficient F are as follows: where sgn(x) is the sign function, 0 ≤ W ≤ 1, 0 ≤ F ≤ 1.

2. The method for scheduling the resources of an unmanned aerial vehicle swarm based on an improved wolf pack algorithm according to claim 1, wherein The expression for the comprehensive advantage A of the UAV over the target is as follows: A = α1S1 + α2S2 where α1 and α2 are the weight coefficients of each index and the sum of the coefficients is 1, and S1 and S2 respectively represent the space situation advantage and performance advantage of the UAV relative to the target: S1 = ω1t1 + ω2t2 + ω3t3 + ω4t4 + ω5t5 where t1, t2, t3, t4, and t5 represent the angular advantage, altitude advantage, speed advantage, distance advantage, and energy advantage of the UAV relative to the target in turn, and ω1, ω2, ω3, ω4, and ω5 represent the weight coefficients of each index in turn and the sum of the coefficients is 1; S2 = η1c1 + η2c2 + η3c3 + η4c4 + η5c5 + η6c6 + η7c7 where c1, c2, c3, c4, c5, c6, and c7 represent the maneuverability advantage, attack ability advantage, detection ability advantage, operation ability advantage, survival ability advantage, range ability advantage, and electronic countermeasure ability advantage of the UAV relative to the target in turn, and η1, η2, η3, η4, η5, η6, and η7 represent the weight coefficients of each index in turn and the sum of the coefficients is 1; Similarly, use the above method to calculate the comprehensive advantage B of the target over the UAV.

3. The method for resource scheduling of an unmanned aerial vehicle swarm based on an improved wolf pack algorithm according to claim 1, wherein The specific process of step (2) is as follows: (a) Artificial wolf position encoding Take the objective function of the UAV swarm resource scheduling model established in step (1) as the objective function optimized by the wolf pack algorithm. According to the characteristics of UAV swarm resource scheduling, the position encoding of artificial wolf i is {x i1 , x i2 , ···, x ij , ···, x iL}, 1 ≤ x ij ≤ L, where L represents the artificial wolf coding length calculated according to the number of UAVs and targets participating in the scheduling. The position of the artificial wolf represents the UAV swarm resource scheduling scheme, and x ij = s means that UAV j confronts target s; (b) Initializing the wolf pack by the chaotic mapping method Determine the number \(a_{initial}\) of the initial wolf pack, and generate a 1×L random number \((x_1, x_2, \cdots, x j , \cdots, x L ) through rand(1,L), where rand(1,L) represents generating a 1×L random number uniformly distributed between (0, 1); then substitute \(x_1, x_2, \cdots, x j , \cdots, x L into the Logistic chaotic mapping formula in sequence and iterate \(a_{initial}-1\) times, that is, generate \(a_{initial}\) groups of 1×L-dimensional vectors in the chaotic state; for one group of vectors \((x_1, x_2, \cdots, x j , \cdots, x L ), substitute each dimensional element \(x_1, x_2, \cdots, x j , \cdots, x L into the following formula in sequence, and finally obtain a 1×L-dimensional vector \(X_1, X_2, \cdots, X j , \cdots, X L , which is the position of the initial random wolves. Substitute all \(a_{initial}\) groups of 1×L-dimensional vectors in the chaotic state into the following formula according to the above method, that is, generate an initial wolf pack with the number of \(a_{initial}\): Among them, means rounding off to an integer; (c) Generating the lead wolf Calculate the objective function value of each wolf in the initial wolf pack and sort them. The wolf with the optimal objective function value is the lead wolf; (d) Exploring wolf wandering behavior based on dynamic Lévy flight Select a_explore wolves with the optimal objective function value except the alpha wolf as exploring wolves. The exploring wolf group performs a wandering behavior. The exploring wolf i senses the prey odor concentration at the current position, that is, calculates the objective fitness value y at the current position i , if y i is greater than the prey odor concentration y lead at the position of the alpha wolf, then the exploring wolf i replaces the alpha wolf as the new alpha wolf and transfers to the summoning behavior in step (e); if y i ≤y lead , then the exploring wolf i randomly selects h directions and advances step a steps in each direction. After recording the prey odor concentration at the end point of each direction, it returns to the original position, selects the direction with the largest prey odor concentration among the h directions and greater than the prey odor concentration at the current position of the exploring wolf and advances step a steps. Repeat the above process until the wandering times t of the exploring wolf reaches the set value t max or the position of the alpha wolf is updated, then transfer to the summoning behavior in step (e); (e) Summoning behavior based on the dynamic distance determination factor and the adaptive running step length All wolves except the alpha wolf are regarded as fierce wolves. The alpha wolf performs the summoning behavior, and the fierce wolves run towards the alpha wolf at a step size of step b . If during the raid, the prey odor concentration y i sensed by fierce wolf i is > y lead , then this fierce wolf replaces the alpha wolf as the new alpha wolf and initiates the summoning behavior again. Otherwise, the fierce wolf continues to run towards the alpha wolf until the distance d i,lead ≤ d near , where d near is the judgment distance, and then it turns into the siege behavior; (f) Siege behavior based on Brownian motion All wolves except the alpha wolf search the area near the alpha wolf, that is, move in a random direction by step c steps. If the prey odor concentration at the moved position is greater than that at the original position, update the wolf's position. After the siege behavior ends, update the wolf with the optimal prey odor concentration in the wolf pack at this time as the alpha wolf, and go to step (g); (g) Wolf pack update mechanism based on the degree of difference Sort the wolf pack according to the objective function value, eliminate the R wolves with smaller objective function values, then randomly generate a wolf i and judge its difference degree sim with the leading wolf lead i,lead , if it is higher than the set threshold sim thre , then supplement it into the wolf pack, otherwise discard it. Loop the above process until R new wolves are supplemented into the wolf pack, enhancing the optimization ability of the algorithm on the basis of enriching the diversity of the wolf pack; (h) Judging the algorithm termination condition If the algorithm reaches the maximum number of iterations it max or the optimal solution meets the optimization accuracy requirement, then output the position of the leading wolf, that is, the optimal UAV swarm resource scheduling scheme; otherwise, go to step (d).

4. The method for scheduling the resources of a swarm of drones based on the improved wolf pack algorithm according to claim 3, wherein, In step (d), let the exploration wolf's random walk step size step a follow the Lévy distribution, and the exploration wolf position update formula is as follows: x i+1 = Θ(x i , ceil(α·levy)) Among them, α is the step size scaling factor, levy represents the random step size of Lévy motion, ceil represents taking the integer upper bound, and Θ represents the wolf x i , randomly select a pair of positions and swap the numbers at these positions, and perform this process cyclically ceil(α·levy) times to obtain the latest position x of the exploring wolf i+1 ; Design a dynamic step size scaling factor α(k) that changes with the number of iterations: Among them, D max represents the maximum number of iterations, k is the current number of iterations, and α(k) ∈ (0, 1).

5. The method for scheduling resources of an unmanned aerial vehicle swarm based on an improved wolf pack algorithm according to claim 3, wherein In step (e), define the distance d between the fierce wolf i and the lead wolf lead as follows: i,lead as follows: Among them, represents an exclusive OR operation; Determination distance d near It is expressed as follows: where ceil represents taking the integer upper bound, and ω is the distance determination factor; Design a dynamic distance determination factor ω(k) that changes with the number of iterations: Among them, D max represents the maximum number of iterations, k represents the current number of iterations, and ω(k) ∈ (1, L + 1).

6. The method for scheduling resources of an unmanned aerial vehicle swarm based on an improved wolf pack algorithm according to claim 3, wherein, In step (f), design the siege step size step c : where step c obeys a normal distribution with a mean of 0 and a variance of T - S. T and S respectively represent the upper and lower bounds of time within a certain time period of Brownian random motion. randn generates a random number with a normal distribution having a mean of 0 and a variance of 1, and ceil represents taking the integer upper bound.

7. The method for scheduling the resources of an unmanned aerial vehicle swarm based on an improved wolf pack algorithm according to claim 3, wherein, In step (f), define the difference degree sim i1 , x i2 , ···, x ij , ···, x iL} between the random wolves {x lead,1 , x lead,2 , ···, x lead,j , ···, x lead,L} and the lead wolf {x i,lead : Among them, represents the exclusive OR operation, 0 ≤ sim i,lead ≤ 1.

Images