Multi-unmanned aerial vehicle task allocation method based on improved distribution estimation algorithm

Abstract

The invention discloses a multi-unmanned aerial vehicle (UAV) task allocation method based on an improved distribution estimation algorithm, which adopts a mathematical threshold model in the generation probability part of the existing distribution estimation algorithm, and achieves a faster convergence effect in the subsequent population iteration by generating a probability model capable of showing the current capability of the UAV, and in addition, the probability model of the current capability of the UAV can be further improved. According to the method, two local search strategies are fused, the problem of falling into a local optimal solution possibly occurring in the optimization process is effectively solved, the local structure of an existing solution can be broken through to explore a new better solution by exchanging non-adjacent segments in a path, resource waste caused by waiting time can be effectively reduced through insertion operation, and the optimization efficiency is improved. Therefore, the local optimization effect is achieved, the solving quality is improved, and the flight cost of the unmanned aerial vehicle is reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of unmanned aerial vehicles, and in particular relates to a multi-unmanned aerial vehicle task allocation method based on an improved distribution estimation algorithm. Background Art

[0002] In recent years, UAV technology has achieved rapid development. With its low cost, no risk of casualties and convenient operation, it has shown broad application prospects in the military and civilian fields. With the continuous development of technology, the mission execution capability of a single UAV is getting stronger and stronger. However, considering the increasingly complex mission requirements in the future environment, the single-machine mission execution mode is difficult to meet future military needs. Therefore, the efficient use of UAV clusters to achieve complex task allocation has become an important development direction of UAV technology.

[0003] Existing research focuses on single-type task allocation, that is, each drone can only complete one type of task. Although this is effective, it limits the flexibility of drones in handling tasks in complex environments, resulting in inefficient completion of tasks. In contrast, breaking down the same work into multiple types of tasks, clearly setting the priorities between these tasks, and then assigning them to drone clusters for parallel processing not only ensures the orderliness and efficiency of task execution, but also improves overall efficiency by optimizing resource allocation and reducing waiting time. When dealing with complex and changeable emergency tasks such as military operations and natural disaster relief, drone clusters can rely on this allocation model to give full play to their advantages in collaborative operations and ensure that tasks are completed smoothly with higher efficiency and higher safety.

[0004] The multi-UAV task allocation problem is usually divided into two categories: centralized allocation and distributed allocation. The corresponding algorithms are mainly divided into optimization algorithms and heuristic algorithms. Among them, the optimization algorithm is committed to finding the optimal solution to the problem by constructing and solving mathematical models. Its main method is the integer programming method (IP) for accurate problem solving. In addition, the heuristic algorithm usually adopts a more adaptable strategy, focusing on reasonably locating the satisfactory solution that is approximately optimal under certain constraints. Domestic and foreign scholars have proposed many heuristic algorithms to solve this problem. Li et al. proposed an improved dual-chromosome encoding genetic algorithm, which uses chromosomes to represent the task allocation scheme and then uses genetic operations such as selection and crossover to improve the particularity of the problem, such as designing adaptive crossover rate and mutation rate. The algorithm has a conflict mediation mechanism, which can significantly improve population optimization and convergence performance. Cai et al. proposed a method for automated allocation of collaborative tasks of reconnaissance UAVs based on improved discrete particle swarms. The method first optimizes the collaborative task instruction request of reconnaissance UAVs, and then constructs a combinatorial optimization model for automated allocation of collaborative tasks of reconnaissance UAVs. By introducing random global and local search strategies, the discrete particle swarm algorithm is improved to solve the combinatorial optimization model of collaborative task automation allocation.

[0005] However, since the multi-UAV task allocation problem often involves multiple UAVs, multiple tasks and complex constraints, this will result in a huge search space, making traditional algorithms inefficient in solving this problem and difficult to find a satisfactory solution within a reasonable time; at the same time, these algorithms are prone to fall into local optimality and cannot jump out of the current search area to explore possible global optimal solutions, which limits the search ability for the global optimal solution. Summary of the invention

[0006] In view of this, the present invention provides a multi-UAV task allocation method based on an improved distribution estimation algorithm, which can find the global optimal solution or a satisfactory solution close to the global optimal solution in a complex search space as the final multi-UAV task allocation solution while realizing task sequence constraints.

[0007] The present invention provides a multi-UAV task allocation method based on an improved distribution estimation algorithm, comprising the following steps:

[0008] Step 1: Collect the drones And the task set S = {s 1,1 ,s 1,2 ,s 1,3 …,s a,b ,s a,2 ,s a,3} location data and order data are preprocessed, where a is the number of the target point, b is the type number of the task in the target point a, and sa,b is the task number, For N v drones;

[0009] Step 2: Evaluate the capabilities of each drone to obtain the initial probability matrix P representing the drone's mission execution capability l ;

[0010] Step 3: Initialize the temporary task set Z to temporarily store the currently assignable tasks. According to the initial probability matrix P of the drone obtained in step 2, l Assign the tasks in Z to the drones. When the tasks in Z are partially assigned, Z is updated until Z is an empty set, completing the task assignment.

[0011] Step 4: Generate individual codes. The individual codes are ordered sequences of mission numbers for each drone. The following formula is used to calculate the individual fitness value:

[0012]

[0013] in, is the flight distance of the kth UAV from mission i to mission j, is the length of time the kth drone needs to wait in place;

[0014] Step 5: transform the individual code generated in step 4, and use the local search method to replace the original individual code with an individual code with a smaller fitness value, thereby completing the first update of the individual code;

[0015] Step 6: Determine the movable task from the individual code, insert the movable task into the optional position of other UAVs through the insertion operation to form a new individual code, wherein the movable task refers to the task that does not change the execution order of the task in the target point after the task is moved, and the optional position refers to the idle position in the individual code that does not change the execution order of the task in the target point after being replaced by the movable task; then calculate the individual fitness value of the new individual code, use the individual code with a smaller fitness value to replace the original individual code, and complete the update of the individual code again;

[0016] Step 7. Generate an initial population based on individual codes, with a population size of M. After updating the probability model based on the EDA algorithm, reallocate tasks based on the updated probability to obtain a new generation of population, and then calculate the fitness value of the new generation of population. When the change in the fitness value of the population is less than the threshold, the iterative evolution is completed, and the current population is the obtained multi-UAV task allocation plan.

[0017] Furthermore, the initial probability matrix P in step 2 l The calculation method is:

[0018] Step 2.1: Calculate the capability K of UAV k to complete task j using the following formula: k,j :

[0019]

[0020] Among them, d(k,g) is the distance between UAV k and task g, and d(k,j) is the distance between UAV k and task j;

[0021] Step 2.2: Use the following formula to convert K k,j Translated into capability threshold: θ k,j =1-K k,j ;

[0022] Step 2.3: Use the following formula to calculate the probability of the drone performing the task, where st j is the possibility of the UAV detecting task j at the target point, The probability that UAV k detects task j at the target point and executes it is:

[0023]

[0024] Using response thresholds Generate an initial probability matrix P that represents the UAV's mission performance capability l .

[0025] Furthermore, in step 3, the method of updating Z after the task part in Z is partially allocated is:

[0026] Number the task in the temporary task set Z as s a,b The tasks where b cannot be divided by B are recorded as the first type of tasks. l After the first-class task is assigned to the UAV, the position of the first-class task in the temporary task set Z will be occupied by the next-level task in the target point where the first-class task is located, so as to complete the update of the temporary task set;

[0027] Number the task in the temporary task set Z as s a,b The tasks whose b is divisible by B are recorded as the second type of tasks. l After the second type of task is assigned to the UAV, the position of the second type of task in the temporary task set Z is deleted.

[0028] Furthermore, the It is equivalent to the flight speed of the drone, that is, at the same waiting time If the drone keeps the original speed v k The flight distance is As shown below:

[0029] Furthermore, when the local search method in step 5 is a 2-opt operation, the individual code generated in step 4 is transformed, and the original individual code is replaced with an individual code having a smaller fitness value by using the local search method, including:

[0030] Step 5.1: For each individual code generated in step 4, exchange two adjacent task numbers in the ordered sequence of task numbers one by one while ensuring that the order of task execution is not disrupted;

[0031] Step 5.2, recalculate the individual fitness value;

[0032] Step 5.3: If the new individual fitness value is smaller than the original individual fitness value, the exchanged individual code is used to replace the original individual code; otherwise, the original individual code remains unchanged.

[0033] Furthermore, the method of updating the probability model based on the EDA algorithm in step 7 is:

[0034] Step 7.1, compare the fitness value of each individual code in the population, and select N individual codes with fitness values ​​less than the threshold as dominant individuals;

[0035] Step 7.2: Statistically calculate the task s among N dominant individuals a,b The number of times assigned to each drone is calculated as follows: Divide the number of times by the number of dominant individuals N to get the number of tasks s among N dominant individuals. a,b The probability of being assigned to each drone is is the task allocation matrix of the qth dominant individual at the lth iteration;

[0036] Step 7.3: Select the learning rate α and update the probability model P using the following formula l (x):

[0037]

[0038] Beneficial effects:

[0039] The present invention adopts a mathematical threshold model in the generation probability part of the existing distribution estimation algorithm (EDA), and achieves a faster convergence effect in subsequent population iterations by generating a probability model that can show the current capabilities of the drone. In addition, the present invention integrates two local search strategies on the basis of EDA to effectively deal with the problem of falling into a local optimal solution that may occur during the optimization process. By exchanging non-adjacent segments in the path, the local structure of the existing solution can be broken to explore a new and better solution. The insertion operation can effectively reduce the waste of resources caused by the waiting time, thereby achieving the effect of local optimization, improving the solution quality, and reducing the flight cost of the drone. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of a directed graph used to describe the multi-UAV task allocation process.

[0041] Figure 2 A schematic diagram of the processing flow of a multi-UAV task allocation method based on an improved distribution estimation algorithm provided by the present invention.

[0042] Figure 3 A schematic diagram of a first updating mechanism for a temporary task point set in a multi-UAV task allocation method based on an improved distribution estimation algorithm provided by the present invention.

[0043] Figure 4 A schematic diagram of a Gantt chart formed after tasks are fully allocated in a multi-UAV task allocation method based on an improved distribution estimation algorithm provided by the present invention.

[0044] Figure 5 Schematic diagram of the results of UAV task allocation after the number of tasks increases.

[0045] Figure 6 Schematic diagram of the best results of drone task allocation.

[0046] Figure 7 A schematic diagram of insertion operation in a multi-UAV task allocation method based on an improved distribution estimation algorithm provided by the present invention. DETAILED DESCRIPTION

[0047] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0048] The Estimation of distribution algorithm (EDA) can describe the distribution of candidate solutions in the search space by constructing a probability model, and use this model to guide the search process so that EDA can explore the search space more comprehensively, thereby improving the search efficiency and reducing the risk of falling into the local optimum. Therefore, the present invention simultaneously considers the priority relationship of different task types being executed in the context of multi-UAV task allocation, and realizes a multi-UAV task allocation method on the basis of improving the EDA algorithm, which effectively improves the solution efficiency. The improvements of the present invention to the traditional EDA algorithm include: the traditional EDA algorithm mostly uses a uniformly distributed probability model when generating the initial probability, while the present invention chooses to use the UAV's own capabilities to generate a probability model to ensure the reasonable allocation of tasks and improve the convergence speed; in addition, 2-opt operations and insertion operations are added to the traditional EDA algorithm to further increase the search space and avoid falling into the local optimum.

[0049] The present invention is directed to N v UAVs are assigned to N t The application scenario is to complete tasks at multiple target points. The target point is the specific location that the drone needs to reach. The tasks to be completed at each target point can be divided into multiple subtasks that have dependencies on each other. For example, the work to be completed at each target point can be divided into three subtasks: classification, response, and verification. These three subtasks have a priority constraint relationship, requiring classification before response tasks, and similarly, response tasks must be performed before verification. The multi-drone task allocation problem can be drawn as follows Figure 1 As shown in the directed graph, each drone needs to perform tasks according to the directed graph. For example, UAV1 first performs task 1 in target point 2, i.e., target2, then performs task 1 in target point 3, and finally performs task 2 in target point 1. When all tasks are completed, the drone returns to the starting point. At the same time, when performing tasks, the drone must ensure that the previous task of the current task in the same target point has been completed, otherwise it needs to wait in place. For example, when executing task 2 in target point 1, it is necessary to ensure that task 1 in target point 1 has been completed.

[0050] According to the description of the above scenario, the multi-UAV task allocation process mainly has the following constraints: each task can only be accessed once; the UAV can successfully complete and end the execution of the task; the UAV must return to the starting node after completing the execution of the task; there is a dependency relationship between the subtasks in each target point; each target point can only execute one subtask at a time, that is, the next subtask can only be executed after the current subtask is completed.

[0051] The present invention defines the following symbols:

[0052]

[0053]

[0054]

[0055] The present invention provides a multi-UAV task allocation method based on an improved distribution estimation algorithm, and its processing flow is as follows: Figure 2 As shown, it mainly includes the following steps:

[0056] Step 1: According to the drone set And the task set S = {s 1,1 ,s 1,2 ,s 1,3 …,s a,b ,s a,2 ,s a,3}’s location data and order data are preprocessed to obtain the initial data including: the starting position, number and speed of the drone, the location and number of the target point, and the location and number of the task, where a is the number of the target point, b is the type number of the task at the target point a, and s a,b The ID of the task.

[0057] For example, b=1 represents a classification task, b=2 represents a response task, and b=3 represents a verification task.

[0058] Step 2: Evaluate the capabilities of each drone to obtain the initial probability matrix P representing the drone's mission execution capability l The specific process includes:

[0059] Step 2.1: Calculate the distance between task j and drone k according to formula (1), where k∈[1, N v ],j∈[1,3N t ].

[0060]

[0061] Step 2.2: Calculate the capability of drone k to complete task j according to formula (2), and then convert it into a threshold value through formula (3). The task can only be executed when it reaches the drone capability threshold:

[0062]

[0063] θ k,j =1-K k,j (3)

[0064] Step 2.3: According to formula (4), the ability of the UAV to perform the task is expressed as probability, where st j is the possibility of the UAV detecting task j at the target point, is the probability that UAV k detects task j at the target point and executes it.

[0065]

[0066] Using response thresholds Generate an initial probability matrix P that can represent the UAV's mission performance capability l .

[0067] Step 3: Initialize the temporary task set Z to temporarily store the currently available tasks to limit the order in which the tasks are assigned and ensure their orderly execution. Initially, due to the task constraints, the task numbers stored in the temporary task set Z are Then according to the initial probability matrix P of the drone obtained in step 2 l Assign the tasks in the temporary task set Z to the drones; when the tasks in Z are partially assigned, Z is updated until Z is an empty set to complete the assignment of all tasks in the task set S.

[0068] Among them, when the tasks in Z are partially assigned, Z is updated until Z is an empty set. The update mechanism adopted is designed according to the priority constraint relationship between different types of subtasks in the target point, which specifically includes the following two cases:

[0069] In the first case, the task number s in the temporary task set Z is a,b The tasks where b cannot be divided by B are recorded as the first-class tasks. The update mechanism of the first-class tasks is given by Figure 3 As shown, according to the initial probability matrix P l After the first-class task is assigned to the UAV, the position of the first-class task in the temporary task set Z will be occupied by the next-level task in the target point where the first-class task is located, so as to complete the update of the temporary task set.

[0070] For example, taking three target points and three drones as an example, the initial temporary task set Z is {s 1,1 ,s 2,1 ,s 3,1}, where tasks can be assigned; where task s 2,1 For the first type of task, according to the initial probability matrix, the task s 2,1 After being assigned to the drone, the mission s 2,1 The position in Z will be determined by task s 2,1 The next level task in the target point is occupied by task s 2,2 Occupy tasks 2,1 ’s original location, thereby achieving the effect of temporarily updating the task set.

[0071] In the second case, the task number s in the temporary task set Z is a,bThe tasks whose b is divisible by B are recorded as the second type of tasks. l After the second type of task is assigned to the UAV, the position of the second type of task in the temporary task set Z is deleted.

[0072] Step 4: After the task allocation is completed, an individual code will be generated. The individual code is the ordered sequence of the task numbers of each drone. The individual fitness value is calculated using formula (5):

[0073]

[0074] in, is the flight distance of the kth UAV from task i to task j, which can be calculated by formula (1); is the length of time the kth UAV needs to wait in place. When the kth UAV needs to perform task j, the kth UAV needs to wait in place because task i has not been completed.

[0075] The length of time the drone waits in place corresponds to the idle position in the individual code, which can be equivalent to the flight speed of the drone, that is, at the same waiting time If the drone keeps the original speed v k The flight distance is As shown in formula (6):

[0076]

[0077] Specifically, when the tasks are fully assigned, we will get Figure 4 In the Gantt chart shown, the same color represents three tasks at the same target point, and the blank space in the chart is the length of time the drone needs to wait for the previous task to be completed. For example, if the UAV UAVA in the figure wants to perform task 8, it needs to wait for the UAV B Complete Mission 7, which is the previous level mission with the same target point as Mission 8.

[0078] Step 5: transform the individual code generated in step 4, and then use the local search method to use the individual code with a smaller fitness value to replace the original individual code, completing the first update of the individual code.

[0079] The increase in the number of tasks will lead to the following changes in the tasks assigned to drones: Figure 5 As shown, relative to Figure 6 As for the distribution results in Figure 5 This will increase the flight cost of the drone. In order to effectively solve this problem, the present invention adopts a local search method to eliminate unnecessary redundant path segments, thereby significantly reducing the total flight distance of the drone. When the local search method is a 2-opt operation, the following steps are included:

[0080] Step 5.1: For each individual code generated in step 4, exchange the two adjacent task numbers in the ordered sequence of task numbers one by one, while ensuring that the task execution order is not destroyed, that is, the tasks under the same target point cannot be exchanged. For example, UAV A The initial individual code is “12,0,9,5,,8”. The first exchange is to exchange 12 and 0 to get “0,12,9,5,8,” the second exchange is to exchange 0 and 9 to get “12,9,0,5,,,8,” the third exchange is to exchange 9 and 5 to get “12,0,5,9,,8,” and the fourth exchange is to exchange 5 and 8 to get “12,0,9,,,8,5.”

[0081] Step 5.2, recalculate the individual fitness value according to step 4;

[0082] Step 5.3: If the new individual fitness value is smaller than the original individual fitness value, that is, d ij +d (i+1)(j+1) <d i(i+1) +d j(j+1) , the exchanged individual code is used to replace the original individual code; otherwise, the original individual code remains unchanged.

[0083] Step 6: Determine the movable task from the individual code, and insert the movable task into the optional position of other UAVs through the insertion operation to form a new individual code, wherein the movable task refers to the task that does not change the execution order of the task in the target point after the task is moved, and the optional position refers to the idle position in the individual code that does not change the execution order of the task in the target point after being replaced by the movable task; then calculate the individual fitness value of the new individual code, and use the individual code with a smaller fitness value to replace the original individual code to complete the update of the individual code again.

[0084] Since the present invention relates to the order of task execution, the problem of waiting time for drones is unavoidable. However, if the drone waits too long, the overall flight efficiency will be reduced and the cost will increase significantly. Therefore, the present invention introduces a task scheduling strategy, namely, an insertion operation.

[0085] For example, Figure 7 As shown in the figure, there are 3 waiting times in the tasks assigned to all UAVs, that is, there are 3 idle positions, and Task 2 is a movable task. Therefore, without changing the order of task execution, Task 2 is inserted into the UAV A The UAV is located at an idle position. A The cost of UAV remains the same B The cost will be reduced, and then the total cost will be reduced, which effectively reduces the waiting time and improves the efficiency of task execution.

[0086] Step 7. Generate an initial population based on individual codes, with a population size of M. After updating the probability model based on the EDA algorithm, reallocate the tasks according to the updated probability to obtain a new generation of population, and then calculate the fitness value of the new generation of population. When the fitness value of the population no longer changes, it is considered that the algorithm has converged and the iterative evolution is completed. The current population is the obtained multi-UAV task allocation plan.

[0087] The specific steps are as follows:

[0088] Step 7.1: Compare the fitness value of each individual code in the population, and select N individual codes whose fitness values ​​are less than the threshold as dominant individuals.

[0089] Step 7.2: Statistically calculate the task s among N dominant individuals a,b The number of times assigned to each drone is calculated as follows: Divide the number of times by the number of dominant individuals N to get the number of tasks s among N dominant individuals. a,b The probability of being assigned to each drone is

[0090] Specifically, since there are N dominant individuals, there will be N allocation tables as shown in Table 1. By counting the task s in the N tables, a,b Assigned to UAVs A 、UAV B 、UAV C The number of times, that is Then divide it by the number of dominant individuals N to get the number of tasks s among N dominant individuals. a,b The probability assigned to each UAV is then used to update the probability model.

[0091] For example, if there are currently 10 dominant individuals, and s 1,1 The number of times assigned to drone A is 8, and the number of times assigned to drone B and drone C are both 1, then through Available tasks 1,1 The probabilities assigned to drones A, B, and C are 0.8, 0.1, and 0.1 respectively. Using them for subsequent updates will make task s 1,1 The probability of being assigned to drone A is greater, which means it is easier to obtain a better fitness value.

[0092] Taking three drones and three target points as an example, the situation where the task of one dominant individual is assigned to the drone is listed, as shown in Table 1.

[0093] Table 1 Example of drone allocation

[0094]

[0095]

[0096] Step 7.3: Select an appropriate learning rate α and update the probability model P using formula (7) l (x), where l refers to the number of iterations;

[0097]

[0098] in, is the task assignment matrix of the qth dominant individual at the lth iteration.

[0099] Step 7.4: Reassign tasks based on the updated probability, repeat the iteration, and calculate the fitness value of the population for each generation. When the fitness value of the population hardly changes, the algorithm is considered to have converged and the iteration is stopped.

[0100] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-UAV task allocation method based on improved distribution estimation algorithm, It is characterized in that The following steps are involved: Step 1: Collect the drones And the task set S = {s 1,1 ,s 1,2 ,s 1,3 …,s a,b ,s a,2 ,s a,3 } location data and order data are preprocessed, where a is the number of the target point, b is the type number of the task in the target point a, and s a,b is the task number, For N v drones; Step 2: Evaluate the capabilities of each drone to obtain the initial probability matrix P representing the drone's mission execution capability l ; Step 3: Initialize the temporary task set Z to temporarily store the currently assignable tasks. According to the initial probability matrix P of the drone obtained in step 2, l Assign the tasks in Z to the drones. When the tasks in Z are partially assigned, Z is updated until Z is an empty set, completing the task assignment. Step 4: Generate individual codes. The individual codes are ordered sequences of mission numbers for each drone. The following formula is used to calculate the individual fitness value: in, is the flight distance of the kth UAV from mission i to mission j, is the length of time the kth drone needs to wait in place; Step 5: transform the individual code generated in step 4, and use the local search method to replace the original individual code with an individual code with a smaller fitness value, thereby completing the first update of the individual code; Step 6: Determine the movable task from the individual code, insert the movable task into the optional position of other UAVs through the insertion operation to form a new individual code, wherein the movable task refers to the task that does not change the execution order of the task in the target point after the task is moved, and the optional position refers to the idle position in the individual code that does not change the execution order of the task in the target point after being replaced by the movable task; then calculate the individual fitness value of the new individual code, use the individual code with a smaller fitness value to replace the original individual code, and complete the update of the individual code again; Step 7. Generate an initial population based on individual codes, with a population size of M. After updating the probability model based on the EDA algorithm, reallocate tasks based on the updated probability to obtain a new generation of population, and then calculate the fitness value of the new generation of population. When the change in the fitness value of the population is less than the threshold, the iterative evolution is completed, and the current population is the obtained multi-UAV task allocation plan.

2. The multi-UAV task allocation method according to claim 1, It is characterized in that The initial probability matrix P in step 2 l The calculation method is: Step 2.1: Calculate the capability K of UAV k to complete task j using the following formula: k,j : Where d(k,g) is the distance between UAV k and task g, and d(k,j) is the distance between UAV k and task j; Step 2.2: Use the following formula to convert K k,j Translated into capability threshold: θ k,j =1-K k,j ; Step 2.3: Use the following formula to calculate the probability of the drone performing the task, where st j is the possibility that the UAV detects task j at the target point, The probability that UAV k detects task j at the target point and executes it is: Using response thresholds Generate an initial probability matrix P that represents the UAV's mission performance capability l .

3. The multi-UAV task allocation method according to claim 1, It is characterized in that In step 3, the method for updating Z after the task part in Z is assigned is as follows: Number the task in the temporary task set Z as s a,b The tasks where b cannot be divided by B are recorded as the first type of tasks. l After the first-class task is assigned to the UAV, the position of the first-class task in the temporary task set Z will be occupied by the next-level task in the target point where the first-class task is located, so as to complete the update of the temporary task set; Number the task in the temporary task set Z as s a,b The tasks whose b is divisible by B are recorded as the second type of tasks. l After the second type of task is assigned to the UAV, the position of the second type of task in the temporary task set Z is deleted.

4. The multi-UAV task allocation method according to claim 1, It is characterized in that Said It is equivalent to the flight speed of the drone, that is, at the same waiting time If the drone keeps the original speed v k The flight distance is As shown below:

5. The multi-UAV task allocation method according to claim 1, It is characterized in that When the local search method in step 5 is a 2-opt operation, the individual codes generated in step 4 are transformed to replace the original individual codes with individual codes having a smaller fitness value by using the local search method, including: Step 5.1: For each individual code generated in step 4, exchange two adjacent task numbers in the ordered sequence of task numbers one by one while ensuring that the order of task execution is not disrupted; Step 5.2, recalculate the individual fitness value; Step 5.3: If the new individual fitness value is smaller than the original individual fitness value, the exchanged individual code is used to replace the original individual code; otherwise, the original individual code remains unchanged.

6. The multi-UAV task allocation method according to claim 1, It is characterized in that The method of updating the probability model based on the EDA algorithm in step 7 is: Step 7.1, compare the fitness value of each individual code in the population, and select N individual codes with fitness values ​​less than the threshold as dominant individuals; Step 7.2: Statistically calculate the task s among N dominant individuals a,b The number of times assigned to each drone is calculated as follows: Divide the number of times by the number of dominant individuals N to get the number of tasks s among N dominant individuals. a,b The probability of being assigned to each drone is is the task allocation matrix of the qth dominant individual at the lth iteration; Step 7.3: Select the learning rate α and update the probability model P using the following formula l (x):

Images