A Multi-UAV Task Allocation Method Based on a Three-Armed Evolutionary Algorithm

Abstract

The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a multi-unmanned aerial vehicle task allocation method based on a three-way evolutionary algorithm, including establishing an uncertain multi-unmanned aerial vehicle task allocation model based on rough sets; using the three-way decision idea to design a three-way evolutionary algorithm to solve the task allocation model and obtain an optimal multi-unmanned aerial vehicle task allocation scheme; the present invention comprehensively considers three uncertain indexes of resource cost, voyage cost, and task reward, constructs a more comprehensive allocation model, and at the same time designs a three-way evolutionary algorithm, which improves the solving speed and solving accuracy of the traditional evolutionary algorithm and also improves in terms of robustness.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a multi-unmanned aerial vehicle task allocation method based on a three-branch evolutionary algorithm. Background Technique

[0002] Unmanned Aerial Vehicles (UAVs) have been widely and successfully applied in various fields. After a long period of development, UAV technology has played a unique role in various fields, but many new problems have also arisen. Due to its resource and capacity limitations, it is difficult for a single UAV to complete complex and long-lasting tasks. In order to achieve multi-UAV cooperation to complete multiple tasks, it is necessary to allocate tasks to multi-UAVs so that each UAV can execute tasks in an orderly manner and maximize the overall efficiency. However, in multi-UAV task allocation, many uncertain problems in the real world are often faced. For example, due to insufficient or incomplete data information, as well as the uncertainty of the tasks themselves, the location information, resource requirement information, and task reward information of the tasks cannot be accurately obtained, which also brings uncertainty to the decision-making of multi-UAV task allocation. Therefore, the problem of uncertain multi-UAV task allocation has become a hot issue in the field of multi-UAV control.

[0003] The uncertain multi-task allocation problem is a typical combinatorial optimization problem. With the increase in the number of UAVs and tasks, problems such as combinatorial explosion will be faced. For the many-to-many situation, that is, each UAV can execute multiple tasks and each task can be jointly executed by multiple UAVs, the complexity of uncertain multi-UAV task allocation is further increased. Traditional evolutionary algorithms have a slow convergence speed, low solution accuracy, and weak robustness, and cannot meet the requirements of uncertain multi-UAV task allocation. Therefore, it is very important to design an uncertain multi-UAV task allocation method that is applicable to the background of complex task allocation and has a fast solution speed, high accuracy, and strong robustness. Summary of the Invention

[0004] To solve the above problems, the present invention provides a multi-UAV task allocation method based on a three-branch evolutionary algorithm, including the following steps:

[0005] S1. In the multi-UAV task allocation scenario, establish a multi-UAV task allocation model and constraint conditions based on resource cost, range cost, and task reward;

[0006] S2. Construct a genetic encoding x and a decoding function g(x): x→TS of the three-branch evolutionary algorithm according to the multi-UAV task allocation model, where TS represents the multi-UAV task allocation scheme;

[0007] S3. Convert the target optimization function ξ(TS) to ξ(g(x)) based on the decoding function g(ξ). The target optimization function ξ(TS) is used to evaluate the allocation scheme. Use the target optimization function ξ(g(c)) as the fitness function of the three - branch evolutionary algorithm and denote it as f(c);

[0008] S4. Use genetic encoding x to generate individuals to obtain the initial population C;

[0009] S5. Subdivide the initial population C into three sub - populations. The three sub - populations respectively adopt three different evolutionary strategies to obtain a new population and calculate the individual fitness;

[0010] S6. Determine whether the termination condition is satisfied. If so, end the algorithm and decode the optimal individual of the current new population to obtain the final multi - UAV task allocation scheme; otherwise, return to step S5 and continue to iterate with the new population.

[0011] Furthermore, the multi - UAV task allocation model is expressed as:

[0012] maxξ(TS)=-λ1ξ1(TS)-λ2ξ2(TS)+λ3ξ3(TS)

[0013] The constraint conditions are expressed as:

[0014]

[0015] C2:|TS i |≤usn i

[0016] C3:λ1+λ2+λ3=1

[0017] C4:λ1,λ2,λ3>0

[0018] Among them, TS={TS i |i=1,2,…,N} represents the multi - UAV task allocation scheme, TS i represents the task sequence assigned to UAV i, ts i,k ∈TS i represents the k - th task executed by UAV i; ξ(TS) represents the target optimization function with the total revenue of the multi - UAV task allocation scheme as the target, ξ1(TS) represents the total resource cost of executing the multi - UAV task allocation scheme; ξ2(TS) represents the total voyage cost of executing the multi - UAV task allocation scheme; ξ3(TS) represents the total task reward of executing the multi - UAV task allocation scheme; λ1 represents the resource cost index weight, λ2 represents the voyage cost index weight, λ3 represents the task reward index weight, M represents the number of tasks to be executed, N represents the number of UAVs, usn irepresents the number of resource packages owned by UAV i; C1 represents that in the multi-UAV task allocation scheme TS, UAV i executes the k-th task ts i in its task sequence TS i,k (ts i,k ∈{1,…,M}), and the task ts i,k must be one of the existing M tasks; C2 represents that the number of tasks |TS i | executed by UAV i is not greater than the number of resource packages it owns; C3 and C4 are the weight constraints of λ1, λ2, λ3.

[0019] Furthermore, the total resource cost ξ1(TS) of executing the multi-UAV task allocation scheme is expressed as:

[0020]

[0021] where, represents the resource size of the resource package of the i-th UAV after coarsening, represents the minimum value of the possible values of the resource size of the resource packages among all UAVs, represents the maximum value of the possible values of the resource size of the resource packages among all UAVs.

[0022] Furthermore, the total flight range cost ξ2(TS) of executing the multi-UAV task allocation scheme is expressed as:

[0023]

[0024] where, represents the flight range of UAV i from task j′ to task j, and the data form is an interval rough number, j = ts i,k , j′ = ts i,(k-1) , when k - 1 = 0, j′ represents the initial position of the UAV, that is represents the flight range of UAV i from the (k - 1)-th task to the k-th task in its own task sequence; represents the minimum value of the possible values of the flight range required from any position g to any task position h, represents the maximum value of the possible values of the flight range required from any position g to any task position h.

[0025] Furthermore, the total task reward ξ3(TS) of executing the multi-UAV task allocation scheme is expressed as:

[0026]

[0027] where, PS j represents the subjective probability that the j-th task is successfully executed, It represents the task reward feedback after the resource requirements of the j-th task are met. The data form is an interval rough number, which is expressed as: It is an interval rough number operator that roughs the task rewards feedback after the resource requirements of all tasks are met, and the data form is an interval rough number, which is expressed as: It represents the minimum value of all possible task reward values. It represents the maximum value of all possible task reward values.

[0028] Furthermore, the genetic encoding x of the three-way evolutionary algorithm is divided into a resource segment, a division segment, and a task segment for encoding, which is expressed as:

[0029]

[0030] Among them, It represents the resource segment. The length of the resource segment is K - 1, and K represents the total number of resource packages carried by all drones; It represents the division segment. The length of the division segment is M, and M represents the number of tasks; It represents the task segment. The length of the task segment is M - 1. It represents the i-th gene position in the resource segment of the genetic encoding. It represents the j-th gene position in the division segment of the genetic encoding. It represents the k-th gene position in the task segment of the genetic encoding.

[0031] Furthermore, before dividing the initial population C in step S4, calculate the fitness of all individuals in the initial population C and sort them in ascending order. Divide the initial population C into three sub-populations C1, C2, and C3 according to the ratio of 3:5:2, so that it satisfies where x1 represents any individual in population C1, x2 represents any individual in population C2, and x3 represents any individual in population C3.

[0032] Furthermore, during each iteration,

[0033] The evolutionary strategy of sub-population C1 includes: the first mutation operation, that is, according to the mutation probability α1 = 0.5, randomly mutate each gene locus of the individuals in sub-population C1 within the value range to obtain new individuals until |C1| new individuals are generated; the first crossover operation, that is, randomly sample an individual from sub-population C1 and C2 respectively for pairing, and exchange the corresponding gene loci of the two paired individuals according to the crossover probability β1 = 0.5 to obtain new individuals until |C1| new individuals are generated; the first selection operation, that is, calculate the fitness of all new individuals in the first mutation operation and the first crossover operation, sort all individuals in C1, including new individuals and original individuals, by fitness ranking, and select individuals with high fitness in descending order of fitness, so that the size of sub-population C1 is restored to the original size;

[0034] The evolutionary strategy of sub-population C2 includes: the second mutation operation, that is, randomly sample an individual x′1 from sub-population C1, randomly sample an individual x′2 from sub-population C2, and randomly sample an individual x′3 from sub-population C3, mutate to obtain a mutant individual x new = x′2+(x′3 - x′1), and then correct the mutant individual using the value range to obtain new individuals until |C2| new individuals are generated; the second crossover operation, that is, randomly sample two individuals from sub-population C2 for pairing, and exchange the corresponding gene loci of the two paired individuals according to the crossover probability β2 = 0.2 to obtain new individuals until |C2| new individuals are generated; the second selection operation, that is, calculate the fitness of all new individuals in the second mutation operation and the second crossover operation, sort all individuals in C2 by fitness ranking, and select individuals with high fitness in descending order of fitness, so that the size of sub-population C2 is restored to the original size;

[0035] The evolutionary strategy of sub-population C3 includes: individuals fall asleep according to the sleep probability γ3 = 0.5, that is, individuals do not perform other operations; if individuals do not fall asleep, each gene locus of the individual is randomly mutated within the value range with a mutation probability of α3 = 0.1, and when the fitness value of the mutated individual is greater than the original individual, the new individual replaces the original individual.

[0036] The beneficial effects of the present invention:

[0037] A multi-UAV task allocation method based on a three-way evolutionary algorithm provided by the present invention comprehensively considers three important uncertainty indexes in the multi-UAV task allocation scenario under uncertain environment: resource cost, voyage cost and task reward. According to the situation that one UAV can execute multiple tasks and one task can be jointly completed by multiple UAVs, as well as constraints such as resource packages in uncertain multi-UAV task allocation, an optimization mathematical model of uncertain multi-UAV task allocation based on rough sets is established. Compared with common task allocation optimization models, the model established by the present invention takes more comprehensive considerations, and pertinently uses the rough environment description space to unify the uncertainty problems faced in actual task allocation, and completes the mathematical modeling of uncertain scenarios. Using the three-way decision-making idea, the proposed three-way evolutionary algorithm improves the solution speed and solution accuracy of traditional evolutionary algorithms, and also improves in terms of robustness. In complex uncertain multi-UAV task allocation scenarios, the method proposed in this patent has obvious advantages, which is conducive to reasonably and efficiently completing the decision-making of uncertain multi-UAV task allocation schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is the overall architecture diagram of the method of the present invention;

[0039] Figure 2 is the flowchart of the three-way evolutionary algorithm proposed by the present invention;

[0040] Figure 3 is an example diagram of the genetic coding proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0042] In one embodiment, as Figure 1 shown, an uncertain multi-UAV task allocation scenario is designed: there are a total of N UAVs to cooperate to complete M tasks. The UAVs all start from the warehouse, and the tasks are distributed on a planar map within a certain range; each UAV can execute multiple tasks, and each task can also be executed by multiple UAVs. When a UAV executes a task, it means that the UAV provides its own resource package to the task. Among them, the number of resource packages owned by the i-th UAV is usn i , and the resource size of each of its resource packages is us i , and the resource requirement of the j-th task is When the resource requirement of the j-th task is met, a task reward will be feedback

[0043] In the above uncertain multi-UAV mission allocation scenario, the impacts of three uncertain indicators, namely resource cost, range cost, and mission reward, on the decision-making of the multi-UAV mission allocation scheme are comprehensively considered.

[0044] The representation of uncertain data in the uncertain environment description space includes but is not limited to: linguistic variables, uncertain data groups with multiple possible instances, and interval numbers, etc.; while in the rough environment description space, interval rough numbers are uniformly used to perform data modeling on uncertain data. The uncertain environment description space is uniformly transformed into the rough environment description space by using interval rough numbers, that is, a multi-UAV mission allocation model is constructed based on rough sets, including:

[0045] S1-1: Define the multi-UAV mission allocation scheme TS = {TS i | i = 1, 2, …, N}, where TS i represents the mission sequence executed by the i-th UAV, and ts i,k ∈TS i represents the k-th mission executed by the i-th UAV.

[0046] S1-2: Define that when implementing the multi-UAV mission allocation scheme, the total resource cost ξ1(TS) faced by all UAVs when executing their respective mission sequences, and the calculation formula is:

[0047]

[0048] Among them, represents the resource size of the resource package of the i-th UAV after roughing, and the data form is interval rough number, specifically indicating that the actual value of must be within the interval but is more likely to be between represents the minimum possible value of the resource size of the resource packages of all UAVs, and represents the maximum possible value of the resource size of the resource packages of all UAVs.

[0049] The resource cost for the i-th UAV to execute the j-th mission is determined by the resource package used when executing the mission. The resource size of the resource package of UAV i is a definite value us i , and in order to unify the data description space, the resource size of the resource package of UAV i is roughed as:

[0050]

[0051] S1-3: Define the total voyage cost ξ2(TS) faced by all UAVs when executing their respective task sequences during the execution of the multi-UAV task allocation plan. The calculation formula is as follows:

[0052]

[0053] Among them, represents the voyage of UAV i from the position of task j ′ to the position of task j. The data form is an interval rough number, specifically When k - 1 = 0, j ′ represents the initial position of the UAV, that is represents the voyage of UAV i from the (k - 1)-th task in its own task sequence to the k-th task in its own task sequence; roughen each section of the voyage of all UAVs, and its data form is represented as an interval rough number, specifically represents the voyage of the UAV from any position g to any task h, represents the minimum value of the possible values of the voyage required from any position g to any task position h, represents the maximum value of the possible values of the voyage required from any position g to any task position h. The voyage cost faced by UAV i when executing task j is determined by the distance from the position of the UAV before executing task j to the position where task j is located.

[0054] S1-4: Define the total task reward ξ3(TS) of all successfully executed tasks when executing the multi-UAV task allocation plan. The calculation formula is as follows:

[0055]

[0056] When the resource requirements of task j are met, obtain its task reward PS j represents the subjective probability that task j is successfully executed, is an interval rough number operator, indicating that a constant multiplies an interval rough number, and the result is still an interval rough number. represents the total resources provided by all UAVs executing task j for task j. The resource size required for task j to be successfully executed is uncertain data, represented by an interval rough number whose subjective probability density function is k ′ represents the estimated sensitivity; the subjective probability that the total resources provided by the UAV sequence executing task j for task j are greater than the resource requirements of task j is is expressed as follows:

[0057]

[0058] S1-5: Based on the constraints of the resources owned by the UAVs and the requirements of the mission assignment scenario, construct an uncertain multi-UAV mission assignment model, expressed as:

[0059] maxξ(TS)=-λ1ξ1(TS)-λ2ξ2(TS)+λ3ξ3(TS)

[0060]

[0061] C2:|TS i |≤usn i

[0062] C3:λ1+λ2+λ3=1

[0063] C4:λ1,λ2,λ3>0

[0064] C1 represents that in the multi-UAV mission assignment scheme TS, UAV i executes the k-th task ts in its task sequence TS i in i,k (ts i,k ∈{1,…,M}), and the task ts i,k must be one of the existing M tasks; C2 represents that the number of tasks |TS i | executed by UAV i is not greater than the number of resource packages it owns; C3 and C4 are the weight constraints of λ1, λ2, and λ3.

[0065] In one embodiment, construct a three-branch evolutionary algorithm to solve the uncertain multi-UAV mission assignment model, including:

[0066] S2-1: First, according to the scenario of multi-UAV mission assignment, that is, each UAV can execute multiple tasks, and each task can be executed cooperatively by multiple UAVs, design the genetic encoding of the three-branch evolutionary algorithm as The decoding function is g(x), g:x→TS.

[0067] Specifically, the genetic encoding of the multi-UAV mission assignment scheme is divided into three parts. They are the resource segment The length of the resource segment is K-1, and K represents the total number of resource packages carried by all UAVs; the division segment The length of the division segment is M, and M represents the number of tasks; the task segment The length of the task segment is M-1. represents the i-th gene position in the resource segment of the genetic encoding, represents the j-th gene position in the division segment of the genetic encoding, represents the k-th gene position in the task segment of the genetic encoding.

[0068] The decoding steps are as Figure 3As shown below, the specific process is as follows:

[0069] (1) Resource segment decoding: For a total of K resource packets, first number the resource packets of all drones. As Figure 3 shown, sort them in ascending order of the numbers to obtain an ordered resource packet sequence; in Figure 3 each gene position of the resource segment in the genetic encoding corresponds to a specific value A in a box. The value A of the i-th gene position x i m = A means that in the decoding process, the A-th resource packet is extracted from the ordered resource packet sequence for the i-th time. According to the ordered resource packet sequence, the number of the resource packet extracted for the i-th time is obtained. After each extraction, re-sort the remaining resource packets in ascending order of the numbers to obtain a new ordered resource packet sequence with a length reduced by one. Before the i-th extraction, i - 1 resource packets have been extracted. Then, for the i-th extraction, only from the remaining K - i + 1 resource packets can be extracted. Therefore the value range of the corresponding value A is Z represents the set of integers. For K resource packets, only after K - 1 extractions and the last remaining one resource packet can form a "resource packet sequence of x" with a length of K, that is, the resource packet sequence after decoding the resource segment. Therefore, the length of the resource segment is K - 1.

[0070] Specifically, as Figure 3 shown, it means that when extracting the resource packet for the first time, the third resource packet in the ordered resource packet sequence is selected and placed at the first position of the "resource packet sequence of x", and at the same time, the encoding of this resource packet is obtained as 3; it means that when extracting the resource packet for the second time, after excluding the resource packet with the number 3 that has been extracted, the second resource packet in the re-sorted ordered resource packet sequence is selected and placed at the second position of the "resource packet sequence of x", and at the same time, the encoding of this resource packet is obtained as 2; until the K - 1 extractions are completed, the remaining resource packet with the number 5 is placed at the last position of the "resource packet sequence of x" to obtain the resource packet sequence after decoding the resource segment.

[0071] (2) Partition segment decoding: Given that the number of tasks is M, in addition to considering the situation of distributing all resource packets to M tasks, the situation where there are remaining resource packets also needs to be considered. Then, the resource packet sequence needs to be divided into M + 1 parts. To divide the resource packet sequence with a length of K into M + 1 parts, M partition points are required. Therefore, the length of the partition segment is M. The first M parts are distributed to the corresponding M tasks. There may be parts with 0 resource packets, indicating that no resource packets are allocated to the corresponding tasks, that is, the corresponding tasks are not executed; the (M + 1)-th part is not distributed to any task. If the number of resource packets in this part is 0, it means that all resource packets have been distributed to the tasks and there are no remaining resource packets. In Figure 3Each gene position in the divided segment of the genetic code corresponds to a specific value B in a box. The j-th gene position in the divided segment of the genetic code The meaning of the value B in the decoding process is that the division point of the j-th division is the B-th division position of the "resource packet sequence of x". In front of the first resource packet in the "resource packet sequence of x" is the 0-th division position, and behind it is the 1-st division position. In front of the second resource packet is the 1-st division position, and behind it is the 2-nd division position. As can be seen from Figure 3 The 1-st division position is behind the first resource packet and in front of the second resource packet. And so on. There are a total of K + 1 division positions in the "resource packet sequence of x" with a length of K. Therefore The corresponding value range of B is Z represents the set of integers. After M divisions, the resource packet sequence is divided into M + 1 parts to obtain the "resource packet division of x", that is, the resource packet division result after decoding the divided segment.

[0072] (3) Task segment decoding: For M tasks, first number all the tasks and sort them in ascending order according to the numbers to obtain an ordered task sequence. In Figure 3 Each gene position in the task segment of the genetic code corresponds to a specific value C in a box. The k-th gene position in the task segment of the genetic code The meaning of the value C in the decoding process is to extract the C-th task from the ordered task sequence for the k-th time. According to the ordered task sequence, the number of the task extracted for the k-th time is obtained. After each extraction, the remaining tasks are sorted in ascending order again to obtain a new ordered task sequence with a length reduced by one. Before the k-th extraction, k - 1 tasks have been extracted. Then for the k-th time, it can only be extracted from the remaining M - k + 1 tasks. Therefore The corresponding value range of C is Z represents the set of integers. For M tasks, only after M - 1 extractions and the last remaining task, a "task sequence of x" with a length of M can be formed, thus determining the order of task execution. Therefore, the length of the task segment is M - 1.

[0073] Specifically, as Figure 3 shown It means that when extracting tasks for the first time, the second resource packet in the ordered task sequence is selected and ranked first in the "task sequence of x", and at the same time, the code of this resource packet is obtained as 2; When performing the second extraction task, after excluding the resource package numbered 2 that has been extracted, select the third resource package in the ordered resource package sequence after re - sorting, place it in the second position of the "task sequence of x", and at the same time obtain the encoding of this resource package as 4; until the (M - 1) - th extraction is completed, place the remaining one task at the last position of the "task sequence of x" to obtain the task sequence after decoding the task segment.

[0074] (4) Decoding combination: Divide the resource package sequence obtained from the resource segment into M + 1 parts, take the first M parts and correspond them one by one with the tasks in the task sequence obtained from the task segment, combine to obtain the resource package sequence assigned to each task, and then according to the one - to - one correspondence between the resource package and the drone, in the order of task execution, assign the tasks to the drones in turn, and finally obtain the multi - drone task assignment scheme TS.

[0075] S2 - 2: Construct the objective optimization function ξ(g(x)) based on the decoding function g(x), use the objective optimization function ξ(g(x)) as the fitness function of the three - branch evolutionary algorithm, and denote it as f(x);

[0076] S2 - 3: Generate individuals using genetic encoding x to obtain the initial population C;

[0077] S2 - 4: Subdivide the initial population C into three sub - populations, and the three sub - populations adopt three different evolutionary strategies respectively to obtain a new population and calculate the individual fitness;

[0078] Specifically, apply the "divide - conquer - efficiency" idea of three - way decision to the evolutionary algorithm.

[0079] "Divide": Calculate the fitness of all individuals in the initial population C, sort them from low to high according to the fitness, and divide the initial population C into three sub - populations C1, C2, C3 according to the ratio of 3:5:2, so that it satisfies and where x1 represents any individual in population C1, x2 represents any individual in population C2, and x3 represents any individual in population C3. These two conditions mean that all individuals in population C3 are superior to all individuals in population C2, and all individuals in population C2 are superior to all individuals in population C1.

[0080] "Conquer": At each iteration,

[0081] The evolutionary strategy of sub-population C1 includes: the first mutation operation, that is, according to the mutation probability α1 = 0.5, randomly mutate each gene locus of the individuals in sub-population C1 within the value range to obtain new individuals. Each gene locus belongs to a segment (resource segment, division segment or task segment), and each gene locus within each segment has a corresponding value range. Like the above decoding process, until |C1| new individuals are generated; the first crossover operation, that is, randomly sample an individual from sub-population C1 and C2 respectively for pairing, and exchange the corresponding gene loci of the two paired individuals according to the crossover probability β1 = 0.5 to obtain new individuals, until |C1| new individuals are generated; the first selection operation, that is, calculate the fitness of all new individuals in the first mutation operation and the first crossover operation, sort all individuals in C1 by fitness ranking, including new individuals and original individuals, and select individuals with high fitness in descending order of fitness, so that the size of sub-population C1 is restored to the original size;

[0082] The evolutionary strategy of sub-population C2 includes: the second mutation operation, that is, randomly sample an individual x′1 in sub-population C1, randomly sample an individual x′2 in sub-population C2, and randomly sample an individual x′3 in sub-population C3, and mutate to obtain a mutated individual x new = x′2+(x′3 - x′1), and then correct the mutated individual using the value range to obtain new individuals, until |C2| new individuals are generated; the second crossover operation, that is, randomly sample two individuals from sub-population C2 for pairing, and exchange the corresponding gene loci of the two paired individuals according to the crossover probability β2 = 0.2 to obtain new individuals, until |C2| new individuals are generated; the second selection operation, that is, calculate the fitness of all new individuals in the second mutation operation and the second crossover operation, sort all individuals in C2 by fitness ranking, including new individuals and original individuals, and select individuals with high fitness in descending order of fitness, so that the size of sub-population C2 is restored to the original size;

[0083] The evolutionary strategy of sub-population C3 includes: individuals fall into a dormant state according to the dormant probability γ3 = 0.5, that is, individuals do not perform other operations; if an individual does not fall into a dormant state, each gene locus of the individual is randomly mutated within the value range with a mutation probability of α3 = 0.1. When the fitness value of the mutated individual is greater than the original individual, the new individual replaces the original individual.

[0084] "Effect": After the evolution in the two stages of "division" and "governance", a new population is obtained. The fitness relationship between sub-populations has changed, and the fitness concentration within each sub-population is evolving in a non-decreasing manner.

[0085] S2-5: Determine whether the termination condition is met. If so, end the algorithm and decode the optimal individual of the current new population to obtain the final multi-UAV mission allocation plan; otherwise, return to step S5 for continued iteration.

[0086] In one embodiment, as Figure 2 shown, the specific process of the three-branch evolutionary algorithm is as follows:

[0087] S3-1: Initialize the population and algorithm parameters, and calculate the fitness of the population.

[0088] S3-2: Determine whether the termination condition is met. If so, end the algorithm to obtain the optimal solution; if not, execute step S3-3.

[0089] S3-3: Divide the population into three sub-populations (small populations) C1, C2, and C3.

[0090] S3-4: The sub-population C1 sequentially performs the first mutation operation, the first crossover operation, and the first selection operation.

[0091] S3-5: The sub-population C2 sequentially performs the second mutation operation, the second crossover operation, and the second selection operation.

[0092] S3-6: The sub-population C3 sequentially performs the third mutation operation and the third selection operation.

[0093] S3-7: Obtain a new population according to S3-4, S3-5, and S3-6 and calculate its fitness, then return to step S3-2.

[0094] In the present invention, unless otherwise clearly specified and defined, terms such as "installation", "setting", "connection", "fixation", "rotation", etc. shall be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components or the interaction relationship between two components. Unless otherwise clearly defined, for those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0095] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-UAV task allocation method based on a three-branch evolutionary algorithm, characterized in that It includes the following steps: S1. In the multi-UAV mission allocation scenario, establish a multi-UAV mission allocation model and constraint conditions based on resource cost, range cost, and mission reward; The multi-UAV mission allocation model is expressed as: maxξ(TS)=-λ1ξ1(TS)-λ2ξ2(TS)+λ3ξ3(TS) The constraint conditions are expressed as: C2:|TS i |≤usn i C3:λ1+λ2+λ3=1 C4:λ1,λ2,λ3>0 Among them, TS = {TS i | i = 1, 2, …, N} represents the multi-UAV mission allocation scheme, and TS i represents the task sequence assigned to UAV i, and ts i,k ∈TS i represents the k-th task executed by UAV i; ξ(TS) represents the objective optimization function with the total revenue of the multi-UAV mission allocation scheme as the goal, ξ1(TS) represents the total resource cost of executing the multi-UAV mission allocation scheme; ξ2(TS) represents the total voyage cost of executing the multi-UAV mission allocation scheme; ξ3(TS) represents the total task reward of executing the multi-UAV mission allocation scheme; λ1 represents the resource cost index weight, λ2 represents the voyage cost index weight, λ3 represents the task reward index weight, M represents the number of tasks to be executed, N represents the number of UAVs, and usn i represents the number of resource packages owned by UAV i; C1 represents that in the multi-UAV mission allocation scheme TS, UAV i executes the k-th task ts i in its task sequence TS i,k , where ts i,k ∈{1, …, M}, and the task ts i,k must be one of the existing M tasks; C2 represents that the number of tasks |TS i | executed by UAV i is not greater than the number of resource packages it owns; C3 and C4 are the weight constraints of λ1, λ2, λ3. The total resource cost ξ1(TS) for executing the multi-UAV mission allocation scheme is expressed as: Among them, represents the resource size of the resource package of the i-th UAV after roughening, represents the minimum value of the possible values of the resource size of the resource package among all UAVs, represents the maximum value of the possible values of the resource size of the resource package among all UAVs; The total range cost ξ2(TS) for executing the multi-UAV mission allocation scheme is expressed as: Among them, represents the flight distance of UAV i from mission j′ to mission j, and the data form is interval rough number, where j = tsi ,k , j′ = tsi ,(k-1) , when k - 1 = 0, j′ represents the initial position of the UAV, that is represents the flight distance of UAV i from the (k - 1)-th mission to the k-th mission in its own mission sequence; represents the minimum value of the possible values of the flight distance required from any position g to any mission position h, represents the maximum value of the possible values of the flight distance required from any position g to mission position h; The total mission reward ξ3(TS) for executing the multi-UAV mission allocation scheme is expressed as: Among them, PS j represents the subjective probability that the j-th task is successfully executed, represents the task reward feedback after the resource requirements of the j-th task are met, and the data form is an interval rough number, is an interval rough number operator, represents the minimum value of all possible task reward values, represents the maximum value of all possible task reward values; S2. Construct the genetic encoding x and decoding function g(x) of the three-branch evolutionary algorithm according to the multi-UAV mission allocation model: x→TS, where TS represents the multi-UAV mission allocation scheme; The genetic encoding x of the three-branch evolutionary algorithm is divided into a resource segment, a division segment, and a mission segment for encoding, and is expressed as: Among them, represents the resource segment, and the length of the resource segment is K - 1, where K represents the total number of resource packets carried by all drones; represents the division segment, and the length of the division segment is M, where M represents the number of tasks; represents the task segment, and the length of the task segment is M - 1; represents the i-th gene position of the resource segment in the genetic encoding, represents the j-th gene position of the division segment in the genetic encoding, represents the k-th gene position of the task segment in the genetic encoding; S3. Convert the objective optimization function ξ(TS) to ξ(g(x)) based on the decoding function g(x), and use the converted objective optimization function ξ(g(x)) as the fitness function of the three-branch evolutionary algorithm, and denote it as f(x); S4. Generate individuals using the genetic encoding x to obtain the initial population C; S5. Subdivide the initial population C into three subpopulations, and the three subpopulations respectively adopt three different evolutionary strategies to obtain a new population and calculate the individual fitness; S6. Determine whether the termination condition is satisfied. If so, end the algorithm and decode the optimal individual of the current new population to obtain the final multi-UAV mission allocation scheme; otherwise, return to step S5 and continue to iterate using the new population.

2. The multi-UAV mission assignment method based on a three-branch evolutionary algorithm according to claim 1, wherein, After obtaining the initial population C in step S4, calculate the fitness of all individuals in the initial population C and sort them in ascending order. Divide the initial population C into three subpopulations C1, C2, and C3 according to the ratio of 3:5:2, so that it satisfies Where x1 represents any individual in population C1, x2 represents any individual in population C2, and x3 represents any individual in population C3.

3. A multi-UAV mission allocation method based on a three-branch evolutionary algorithm according to claim 2, characterized in that During each iteration, The evolutionary strategy of the subpopulation C1 includes: the first mutation operation, that is, according to the mutation probability α1 = 0.5, randomly mutate each gene position of the individuals in the subpopulation C1 within the value range to obtain new individuals until |C1| new individuals are generated; the first crossover operation, that is, randomly sample an individual from each of the subpopulations C1 and C2 for pairing, and exchange the corresponding gene positions of the paired two individuals according to the crossover probability = 1 = 0.5 to obtain new individuals until |C1| new individuals are generated; the first selection operation, that is, calculate the fitness of all new individuals in the first mutation operation and the first crossover operation, sort all individuals in C1, including new individuals and original individuals, in the order of fitness, and select individuals with high fitness in descending order of fitness, so that the size of the subpopulation C1 is restored to the original size; The evolutionary strategy of sub-population C2 includes: a second mutation operation, that is, randomly sampling an individual x′1 from sub-population C1, randomly sampling an individual x′2 from sub-population C2, and randomly sampling an individual x′3 from sub-population C3, and mutating to obtain a mutated individual x new = x′2+(x′3 - x′1), and then using the value range to correct the mutated individual to obtain a new individual until |C2| new individuals are generated; a second crossover operation, that is, randomly sampling two individuals from sub-population C2 for pairing, and exchanging the corresponding gene positions of the two paired individuals according to the crossover probability β2 = 0.2 to obtain new individuals until |C2| new individuals are generated; a second selection operation, that is, calculating the fitness of all new individuals in the second mutation operation and the second crossover operation, sorting all individuals in C2 by fitness ranking, and selecting individuals with high fitness in descending order of fitness so that the size of sub-population C2 is restored to the original size; The evolutionary strategy of sub-population C3 includes: individuals fall into a dormant state according to the dormant probability γ3 = 0.5, that is, the individuals do not perform other operations; if an individual does not fall into a dormant state, each gene locus of the individual undergoes random mutation within the value range with a mutation probability of α3 = 0.

1. When the fitness value of the mutated individual is greater than that of the original individual, the new individual replaces the original individual.

Images