A large-scale task allocation method for heterogeneous UAVs based on multi-objective quantum polar bear mechanism
Through the quantum polar bear mechanism and encoding method, combined with non-dominant solution sorting and congestion calculation, the problems of local convergence and insufficient target number in the cooperative task allocation of drones are solved, and efficient task allocation and multi-objectives are achieved for large-scale ground target allocation. optimization.
Patent Information
- Application Number
- CN202211220690.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-10-07
AI Technical Summary
The existing UAV collaborative mission allocation method is prone to local convergence during multi-objective optimization, and the number of UAVs hitting targets is insufficient, making it difficult to deal with the problem of large-scale ground target allocation.
Using the quantum polar bear mechanism, the mapping between the drone mission allocation matrix and quantum position is realized by designing the quantum polar bear coding method, using non-dominant solution sorting and crowding degree calculation, combining seal capture and ice floating drift strategies to update the quantum position, overcome local Convergence problem, optimize task execution efficiency.
The scale of multi-drone strikes on ground targets has been effectively expanded, the efficiency and accuracy of task allocation has been improved, the drawbacks of local convergence have been overcome, and the uniform Pareto front-end solution for multi-objective optimization has been achieved.
Smart Images

Figure CN115657710B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for allocating large-scale operations of heterogeneous unmanned aerial vehicles (UAVs) using a multi-objective quantum polar bear mechanism, and belongs to the field of UAV operations. Background Art
[0002] Unmanned Aerial Vehicle (UAV) is a powered, controllable, unmanned aerial vehicle that can perform a variety of tasks. It has the advantages of low cost, flexibility, easy deployment, and long endurance. It is now widely used in agriculture, search and rescue, environmental monitoring and other fields.
[0003] Multi-UAV collaborative task allocation refers to allocating tasks to multiple UAVs under multiple specific constraints with the purpose of achieving the optimal objective function, thereby determining the optimal task allocation plan. There are many ways to allocate tasks for UAV cluster tasks. According to the control method of UAV task allocation, it can be divided into centralized control, distributed control and hierarchical distributed control. The present invention adopts a centralized control method to perform multi-UAV task allocation.
[0004] According to existing literature, Wang Feng et al. proposed a hybrid variable multi-objective particle swarm optimization method based on co-evolution for the problem of cooperative multi-task allocation of unmanned aerial vehicles in the article "Mixed variable multi-objective particle swarm optimization algorithm based on co-evolution for solving the problem of cooperative multi-task allocation of unmanned aerial vehicles" published in the Journal of Computers (2021, 44(10): 1967-1983). The method takes the minimum total flight distance and the shortest task completion time as the optimization objectives, and constructs a mixed variable multi-constrained model of cooperative multi-task allocation problem of unmanned aerial vehicles. Experimental simulations have proved that this method is superior to other methods in terms of convergence performance. However, the number of military targets attacked by unmanned aerial vehicles in this method is too small to prove the superiority of this method, and the proposed multi-objective particle swarm optimization method has the disadvantage of being prone to local convergence.
[0005] In summary, the above literature has made certain contributions to the tasks of multiple drones, but the number of drone strike targets is too small. Therefore, the present invention designs a large-scale task allocation method for heterogeneous drones with a multi-objective quantum polar bear mechanism, taking task execution efficiency as the objective function, and using the quantum polar bear method to optimize the objective function, overcoming the problem that previous optimization methods are prone to local convergence. At the same time, a quantum polar bear encoding mechanism is designed to achieve the mapping between the position of the quantum polar bear and the task allocation scheme. Summary of the invention
[0006] In order to solve the problem of UAVs attacking large-scale ground targets, the present invention designs a quantum polar bear mechanism to assign tasks to multiple UAVs, takes task execution efficiency as the objective function, and considers multiple constraints to make the task allocation problem more practical.
[0007] The purpose of the present invention is achieved in this way: Step 1, establishing a heterogeneous UAV to large-scale ground target model.
[0008] Assume that there are N drones to attack ground targets, and the set of drones is defined as U = {U1,U2,...,U N}, where UAV U n The attribute set is U n = {v n ,w n ,l n},n=1,2,...,N,v n Indicates drone U n The speed, w n Indicates drone U n Amount of ammunition carried, l n Indicates drone U n The location of the base. Assume that these drones are to attack M ground targets, and the set of ground targets is defined as T = {T1, T2, ..., T M}, where the ground target T m The attribute set is s m Represents the ground target T m The mission value, Represents the ground target T m location, To attack ground targets T m The time required.
[0009] From the above, we can see that there are N drones to attack M ground targets, so the task allocation matrix is X = {x n,m |x n,m ∈{0, 1}} N×M , where x n,m =1 means UAV U n Attack ground targets T m , x n,m =0 means UAV U n Do not attack ground targets T m , n=1,2,...,N,m=1,2,...,M.
[0010] The minimum joint optimization model is as follows: min[f1(X),f2(X)]. f1(X) is the inverse of the total value obtained by the drone attacking the ground target. in, represents the task constraint penalty function, |·| is the absolute value function, represents the weapon constraint function. α1, α2 represent the penalty function weight factors.
[0011] f2(X) is the maximum time it takes for the drone to take off from the base and return to the base, and also represents the end time of the model. α3, α4 represent the penalty function weight factors, For UAV n The time it takes for the drone to take off from the base and return to the base. n Execute tasks T1, T2, ..., T in sequence η ,but Among them, d 0,1 For UAV n Base location n T1 position with ground target The distance between 1,2 is the ground target T1 position T2 position with ground target The distance between η-1,η For ground targets T η-1 Location With ground target T η Location The distance between.
[0012] Step 2: Initialize the quantum position of the quantum polar bear and set the parameters.
[0013] Set the population size to K1 and the maximum number of iterations to K2. In the initial population, the quantum position of the random initial quantum polar bear is: Where S is the maximum dimension of the quantum position vector, and any dimension of all quantum positions is a random number between [0,1]. The position of the quantum polar bear can be obtained through quantum position mapping. The quantum position of the i-th quantum polar bear in the k-th iteration is Then the position of the i-th quantum polar bear in the k-th iteration can be mapped to Where i = 1, 2, ..., K1, k ∈ {1, 2, ..., K2}, The h-th dimensional variable representing the position of the i-th quantum polar bear, is the upper bound of the h-th dimension position vector of the search space, is the lower bound of the position vector of the h-th dimension of the search space.
[0014] Step three, calculate the fitness function value of the quantum polar bear position.
[0015] The present invention designs a quantum polar bear encoding method to achieve the mapping between the position vector of the quantum polar bear and the task allocation matrix. Assuming that N drones want to attack M ground targets, the search dimension of the quantum polar bear in the population is S=M, and the position of the i-th quantum polar bear is It includes an integer part and a decimal part. The integer part represents the drone number sequence. The decimal part is arranged in ascending order. The sorted order represents the target number of the drone attack. Figure 2 The mapping relationship between the position of the quantum polar bear and the task allocation matrix is described using N=5 and M=9 as an example.
[0016] The position of the i-th quantum polar bear in the k-th iteration Mapping to task assignment matrix Substitute it into the model established by the present invention to obtain the fitness function value of the i-th quantum polar bear in the k-th iteration Among them, t = 1 represents the inverse of the total value obtained by the objective function drone attacking ground targets, and t = 2 represents the end time of the objective function model.
[0017] Step 4: Sort the non-dominated solutions according to the fitness function values of all quantum polar bear positions.
[0018] The model established by the present invention belongs to the minimum optimization problem. Therefore, there are two quantum polar bears labeled i1 and i2 whose positions are respectively and If all fitness functions satisfy And there is at least one fitness function Then the quantum polar bear position Dominate the position of the quantum polar bear is a non-dominated solution. On the contrary, if all fitness functions satisfy And there is at least one fitness function Quantum Polar Bear Domination of Quantum Polar Bears is a non-dominated solution. If the above relationship does not exist, then the two quantum polar bears have no domination relationship.
[0019] Traverse each quantum polar bear in the quantum polar bear population Get the position of the i-th quantum polar bear The number of quantum polar bears and quilt Dominated quantum polar bear position set Then count the number of dominated positions of all quantum polar bears, if This means that there is no quantum polar bear position dominating The non-dominated level of is 1; if but The non-dominated level of is 2; and so on, the non-dominated level of each quantum polar bear position is the number of dominated positions plus 1. Finally, all quantum polar bears are sorted in ascending order according to their non-dominated levels.
[0020] Step 5: Calculate the crowding degree of the quantum polar bear positions in all non-dominated levels.
[0021] Assume that there are b quantum polar bears in the quantum polar bear set with non-dominated level λ, and they are arranged in ascending order according to the t-th fitness function value, t = 1, 2, After sorting The t-th fitness function value of the quantum polar bear, is the maximum value of the t-th fitness function in the quantum polar bear set, is the minimum value of the tth fitness function in the quantum polar bear set, then The crowdedness of the t-th fitness function in the quantum polar bear when or hour, The first The total crowding value of quantum polar bears is Finally, the b quantum polar bears are arranged in descending order of crowding degree.
[0022] Step 6: Use the seal catching strategy to update the quantum position of the quantum polar bear.
[0023] During the hunting process, the quantum polar bear walks on the Arctic ice to explore potential prey. At this time, it not only observes the situation on the ice, but also the situation under the water. In order to find prey, the quantum polar bear quietly moves to the best location. Once the quantum polar bear basically reaches the attack position or it is discovered by the seal, it attacks the prey at the maximum speed. Quantum polar bears have their own search radius, and they search for prey within their own search radius. The h-dimensional search radius of the i-th quantum polar bear in the k-th iteration is is a random number between [0,0.3]. The h-dimensional quantum rotation angle of the i-th quantum polar bear in the iteration is is a random number between [0,π].
[0024] Use the quantum revolving door to update the h-dimensional quantum position of the i-th quantum polar bear in the seal catching strategy: Through quantum position Mapping to get location Then calculate The fitness function value And discuss by category, if Dominate but Keep the result unchanged; otherwise, The details are as follows:
[0025] Step seven, use the ice drift strategy to update the quantum position of the quantum polar bear.
[0026] If a hungry quantum polar bear cannot find any food in its nearest neighborhood, it will stand on a large and stable ice floe, which will not break for a long time due to the quantum polar bear's heavy weight. The quantum polar bear uses the ice floe to drift as far as possible to where seals gather. In the ice floe drift strategy, the h-dimensional quantum rotation angle of the i-th quantum polar bear is is a random number between [0,1], is [0,φ i ], where φ i is the sum of the Euclidean distances between all quantum polar bears and the i-th quantum polar bear, in,
[0027] Use the quantum revolving door to update the h-dimensional quantum position of the i-th quantum polar bear in the ice drift strategy: Then calculate The fitness function value And discuss by category, if Dominate but Keep the result unchanged, otherwise, The details are as follows:
[0028] Step 8: Use population reproduction and extinction strategies to update the quantum position of the quantum polar bear.
[0029] In the quantum polar bear population, the quantum polar bears may starve to death due to failure to catch prey, or reproduce after successful hunting. This update strategy simulates the harsh conditions of the Arctic. In step 4, after sorting the non-dominated solutions of the quantum polar bears, if there are more than 3 non-dominated levels in the quantum polar bear population, all the quantum polar bears in the last non-dominated level will starve to death, and then the offspring of the quantum polar bears with a non-dominated level of 1 will replace them. If there are only 2 or 1 non-dominated levels in the quantum polar bear population, the quantum polar bears will not carry out population reproduction and extinction strategies.
[0030] Assume that there are more than 3 non-dominated levels in the quantum polar bear population, and there are b quantum polar bears in the quantum polar bear set with non-dominated level 1, and there are b1 quantum polar bears in the last non-dominated level. The h-th dimension variable of the new quantum position of the j-th quantum polar bear is in, And ε takes a random value between [1,b].
[0031] Step nine, determine whether the maximum number of iterations K2 of the quantum polar bear has been reached. If so, terminate the iteration, map the positions of the quantum polar bears with a non-dominated level of 1 into a task allocation matrix and output it; otherwise, set k=k+1 and continue to execute step four.
[0032] Compared with the prior art, the present invention has the following beneficial effects:
[0033] (1) In the prior art, the number of UAVs and ground targets is relatively small. The large-scale task allocation method for heterogeneous UAVs designed in the present invention expands the scale of multi-UAV strikes on ground targets and increases the difficulty of simulation. In future research, it can be expanded to the UAV swarm coordination model.
[0034] (2) The model established by the present invention belongs to a multi-objective optimization problem, which is less studied in the prior art. Compared with the previous single-objective multi-UAV coordination problem, it greatly increases the simulation difficulty. When optimizing one objective function, the other objective function established in the model often deteriorates. This is in line with the idea that problems cannot be considered one-sidedly in real life. Therefore, the present invention has practical research significance.
[0035] (3) The present invention designs a multi-objective quantum polar bear mechanism method and a quantum-coded polar bear quantum position evolution mechanism, and obtains a new quantum polar bear mechanism method, which overcomes the drawbacks of the previous methods that are prone to local convergence and improves the optimization rate of the evolution mechanism. In addition, the non-dominated solution sorting method and the crowding degree calculation method are used to sort the population, and finally a uniform Pareto front solution can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 Schematic diagram of the large-scale task allocation method for heterogeneous drones based on the quantum polar bear mechanism designed for the present invention.
[0037] Figure 2 The mapping relationship between the position of the quantum polar bear and the task allocation matrix is described using N=5 and M=9 as an example.
[0038] Figure 3 Comparison diagram of non-dominated solutions obtained by MOQPBO, MOQKH and MONGO methods.
[0039] Figure 4 Comparison diagram of non-dominated solutions obtained by MOQPBO, MOQKH and MONGO methods.
[0040] Figure 5 Comparison diagram of non-dominated solutions obtained by MOQPBO, MOQKH and MONGO methods. DETAILED DESCRIPTION
[0041] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0042] Step 1: Establish a heterogeneous UAV-to-large-scale ground target model.
[0043] Assume that there are N drones to attack ground targets, and the set of drones is defined as U = {U1,U2,…,U N}, where UAV U n The attribute set is U n = {v n ,w n ,l n},n=1,2,…,N,v n Indicates drone U n The speed, w n Indicates drone U n Amount of ammunition carried, l n Indicates drone U n The location of the base. Assume that these drones are to attack M ground targets, and the set of ground targets is defined as T = {T1, T2, …, T M}, where the ground target T m The attribute set is m=1,2,...,M,s m Represents the ground target T m The mission value, Represents the ground target T m location, To attack ground targets T m The time required.
[0044] From the above, we can see that there are N drones to attack M ground targets, so the task allocation matrix is X = {x n,m |x n,m ∈{0, 1}} N×M , where x n,m =1 means UAV U n Attack ground targets T m , x n,m =0 means UAV U n Do not attack ground targets T m , n=1,2,...,N,m=1,2,...,M.
[0045] The minimum joint optimization model is as follows: min[f1(X),f2(X)]. f1(X) is the inverse of the total value obtained by the drone attacking the ground target. in, represents the task constraint penalty function, |·| is the absolute value function, represents the weapon constraint function. α1, α2 represent the penalty function weight factors.
[0046] f2(X) is the maximum time it takes for the drone to take off from the base and return to the base, and also represents the end time of the model. α3, α4 represent the penalty function weight factors, For UAV n The time it takes for the drone to take off from the base and return to the base. n Execute tasks T1, T2, ..., T in sequence. η ,but Among them, d 0,1 For UAV n Base location n T1 position with ground target The distance between 1,2 is the ground target T1 position T2 position with ground target The distance between η-1,η For ground targets T η-1 Location With ground target T η Location The distance between.
[0047] Step 2: Initialize the quantum position of the quantum polar bear and set the parameters.
[0048] Set the population size to K1 and the maximum number of iterations to K2. In the initial population, the quantum position of the random initial quantum polar bear is: Where S is the maximum dimension of the quantum position vector, and any dimension of all quantum positions is a random number between [0,1]. The position of the quantum polar bear can be obtained through quantum position mapping. The quantum position of the i-th quantum polar bear in the k-th iteration is Then the position of the i-th quantum polar bear in the k-th iteration can be mapped to Where i = 1, 2, ..., K1, k ∈ {1, 2, ..., K2}, The h-th dimensional variable representing the position of the i-th quantum polar bear, is the upper bound of the h-th dimension position vector of the search space, is the lower bound of the position vector of the h-th dimension of the search space.
[0049] Step three, calculate the fitness function value of the quantum polar bear position.
[0050] The present invention designs a quantum polar bear encoding method to achieve the mapping between the position vector of the quantum polar bear and the task allocation matrix. Assuming that N drones want to attack M ground targets, the search dimension of the quantum polar bear in the population is S=M, and the position of the i-th quantum polar bear is It includes an integer part and a decimal part. The integer part represents the drone number sequence. The decimal part is arranged in ascending order. The sorted order represents the target number of the drone attack. Figure 2 The mapping relationship between the position of the quantum polar bear and the task allocation matrix is described using N=5 and M=9 as an example.
[0051] The position of the i-th quantum polar bear in the k-th iteration Mapping to task assignment matrix Substitute it into the model established by the present invention to obtain the fitness function value of the i-th quantum polar bear in the k-th iteration Among them, t = 1 represents the inverse of the total value obtained by the objective function drone attacking ground targets, and t = 2 represents the end time of the objective function model.
[0052] Step 4: Sort the non-dominated solutions according to the fitness function values of all quantum polar bear positions.
[0053] The model established by the present invention belongs to the minimum optimization problem. Therefore, there are two quantum polar bears labeled i1 and i2 whose positions are respectively and If all fitness functions satisfy And there is at least one fitness function Then the quantum polar bear position Dominate the position of the quantum polar bear is a non-dominated solution. On the contrary, if all fitness functions satisfy And there is at least one fitness function Quantum Polar Bear Domination of Quantum Polar Bears is a non-dominated solution. If the above relationship does not exist, then the two quantum polar bears have no domination relationship.
[0054] Traverse each quantum polar bear in the quantum polar bear population Get the position of the i-th quantum polar bear The number of quantum polar bears and quilt Dominated quantum polar bear position set Then count the number of dominated positions of all quantum polar bears, if This means that there is no quantum polar bear position dominating The non-dominated level of is 1; if but The non-dominated level of is 2; and so on, the non-dominated level of each quantum polar bear position is the number of dominated positions plus 1. Finally, all quantum polar bears are sorted in ascending order according to their non-dominated levels.
[0055] Step 5: Calculate the crowding degree of the quantum polar bear positions in all non-dominated levels.
[0056] Assume that there are b quantum polar bears in the quantum polar bear set with non-dominated level λ, and they are arranged in ascending order according to the t-th fitness function value, t = 1, 2, After sorting The t-th fitness function value of the quantum polar bear, is the maximum value of the t-th fitness function in the quantum polar bear set, is the minimum value of the tth fitness function in the quantum polar bear set, then The crowdedness of the t-th fitness function in the quantum polar bear when or hour, The first The total crowding value of quantum polar bears is Finally, the b quantum polar bears are arranged in descending order of crowding degree.
[0057] Step 6: Use the seal catching strategy to update the quantum position of the quantum polar bear.
[0058] During the hunting process, the quantum polar bear walks on the Arctic ice to explore potential prey. At this time, it not only observes the situation on the ice, but also the situation under the water. In order to find prey, the quantum polar bear quietly moves to the best location. Once the quantum polar bear basically reaches the attack position or it is discovered by the seal, it attacks the prey at the maximum speed. Quantum polar bears have their own search radius, and they search for prey within their own search radius. The h-dimensional search radius of the i-th quantum polar bear in the k-th iteration is is a random number between [0,0.3]. The h-dimensional quantum rotation angle of the i-th quantum polar bear in the iteration is is a random number between [0,π].
[0059] Use the quantum revolving door to update the h-dimensional quantum position of the i-th quantum polar bear in the seal catching strategy: Through quantum position Mapping to get location Then calculate The fitness function value And discuss by category, if Dominate but Keep the result unchanged; otherwise, The details are as follows:
[0060] Step seven, use the ice drift strategy to update the quantum position of the quantum polar bear.
[0061] If a hungry quantum polar bear cannot find any food in its nearest neighborhood, it will stand on a large and stable ice floe, which will not break for a long time due to the quantum polar bear's heavy weight. The quantum polar bear uses the ice floe to drift as far as possible to where seals gather. In the ice floe drift strategy, the h-dimensional quantum rotation angle of the i-th quantum polar bear is is a random number between [0,1], is [0,φ i ], where φ i is the sum of the Euclidean distances between all quantum polar bears and the i-th quantum polar bear, in,
[0062] Use the quantum revolving door to update the h-dimensional quantum position of the i-th quantum polar bear in the ice drift strategy: Then calculate The fitness function value And discuss by category, if Dominate but Keep the result unchanged, otherwise, The details are as follows:
[0063] Step 8: Use population reproduction and extinction strategies to update the quantum position of the quantum polar bear.
[0064] In the quantum polar bear population, the quantum polar bears may starve to death due to failure to catch prey, or reproduce after successful hunting. This update strategy simulates the harsh conditions of the Arctic. In step 4, after sorting the non-dominated solutions of the quantum polar bears, if there are more than 3 non-dominated levels in the quantum polar bear population, all the quantum polar bears in the last non-dominated level will starve to death, and then the offspring of the quantum polar bears with a non-dominated level of 1 will replace them. If there are only 2 or 1 non-dominated levels in the quantum polar bear population, the quantum polar bears will not carry out population reproduction and extinction strategies.
[0065] Assume that there are more than 3 non-dominated levels in the quantum polar bear population, and there are b quantum polar bears in the quantum polar bear set with non-dominated level 1, and there are b1 quantum polar bears in the last non-dominated level. The h-th dimension variable of the new quantum position of the j-th quantum polar bear is in, And ε takes a random value between [1,b].
[0066] Step nine, determine whether the maximum number of iterations K2 of the quantum polar bear has been reached. If so, terminate the iteration, map the positions of the quantum polar bears with a non-dominated level of 1 into a task allocation matrix and output it; otherwise, set k=k+1 and continue to execute step four.
[0067] The multi-objective quantum polar bear optimization method is denoted as MOQPBO, the multi-objective quantum krill method is denoted as MOQKH, and the multi-objective polar bear optimization method is denoted as MONGO. In order to verify the performance of the large-scale task allocation method of heterogeneous UAVs based on the multi-objective quantum polar bear mechanism, the present invention conducted three sets of simulation experimental tests. Set the population size to K1 = 100, the maximum number of iterations K2 = 500, α1 = 1, α2 = 1, α3 = 20, and α4 = 5. Table 1 gives the locations of the five bases.
[0068] Table 1 Base location
[0069]
[0070] Table 2 gives the attribute set of the drone.
[0071] Table 2. Attribute set of drones
[0072]
[0073]
[0074] Table 3 gives the target information
[0075] Table 3 Target information
[0076]
[0077] In the first set of experiments, N = 4 UAVs performed M = 9 ground target tasks. The attribute information of the UAVs is U1~U4 in Table 2, and the attribute information of the ground targets is T1~T9 in Table 3. Figure 3 It is a comparison chart of non-dominated solutions obtained by MOQPBO method, MOQKH method and MONGO method. It can be seen from the figure that the optimal solution generated by MOQPBO method can dominate the optimal solutions generated by MOQKH method and MONGO method, and the number of optimal solutions is significantly more than that of the other two methods. The task allocation schemes corresponding to some non-dominated solutions of MOQPBO method are shown in Table 4.
[0078] Table 4 Task allocation schemes corresponding to some individuals in the Pareto front end
[0079]
[0080] In the second set of experiments, N = 12 UAVs performed M = 30 ground target tasks. The attribute information of the UAVs is U1~U 12 The attribute information of the ground target is T1~T in Table 3 30 . Figure 4 It is a comparison chart of non-dominated solutions obtained by MOQPBO method, MOQKH method and MONGO method. It can be seen from the figure that the optimal solution generated by MOQPBO method can dominate the optimal solutions generated by MOQKH method and MONGO method, and the number of optimal solutions is significantly more than that of the other two methods. The task allocation schemes corresponding to some non-dominated solutions of MOQPBO method are shown in Table 5.
[0081] Table 5 Task allocation schemes corresponding to some individuals in the Pareto front end
[0082]
[0083] In the third group of experiments, N = 15 UAVs performed M = 55 ground target tasks. The attribute information of the UAVs is U1~U 15 The attribute information of the ground target is T1~T in Table 3 55 . Figure 3 It is a comparison chart of non-dominated solutions obtained by MOQPBO method, MOQKH method and MONGO method. It can be seen from the figure that the optimal solution generated by MOQPBO method can dominate the optimal solutions generated by MOQKH method and MONGO method, and the number of optimal solutions is significantly more than that of the other two methods. The task allocation schemes corresponding to some non-dominated solutions of MOQPBO method are shown in Table 6.
[0084] Table 6 Task allocation schemes corresponding to some individuals in the Pareto front end
[0085]
Claims
1. A large-scale task allocation method for heterogeneous UAVs based on a multi-objective quantum polar bear mechanism, characterized in that: Here are the steps: Step 1: Establish a heterogeneous UAV-to-large-scale ground target model; There are N drones that want to attack ground targets. The set of drones is defined as U = {U1, U2, ..., U N }, where UAV U n The attribute set is U n = {v n ,w n ,l n },n=1,2,…,N,v n Indicates drone U n The speed, w n Indicates drone U n Amount of ammunition carried, l n Indicates drone U n The location of the base; the drone needs to attack M ground targets, and the set of ground targets is defined as T = {T1, T2, ..., T M }, where the ground target T m The attribute set is m=1,2,...,M,s m Represents the ground target T m The mission value, Represents the ground target T m location, To attack ground targets T m time required; if N drones want to attack M ground targets, the task allocation matrix is X = {x n,m |x n,m ∈{0, 1}} N×M , where x n,m =1 means UAV U n Attack ground targetsT m , x n,m =0 means UAV U n Do not attack ground targets T m , n=1,2,...,N,m=1,2,...,M; The minimum joint optimization model is as follows: min[f1(X),f2(X)], where f1(X) is the inverse of the total value obtained by the drone attacking the ground target. in, represents the task constraint penalty function, |·| is the absolute value function, represents the weapon constraint function; α1, α2 represent the penalty function weight factors; f2(X) is the maximum time it takes for the drone to take off from the base and return to the base, α3, α4 represent the penalty function weight factors, For UAV n The time it takes to take off from the base and return to the base; assuming that the UAV U n Execute tasks T1, T2, ..., T in sequence. η ,but Among them, d 0,1 For UAV n Base location n T1 position with ground target The distance between 1,2 is the ground target T1 position T2 position with ground target The distance between η-1,η For ground targets T η-1 Location With ground target T η Location The distance between Step 2: Initialize the quantum position of the quantum polar bear and set the parameters; Step 3, calculate the fitness function value of the quantum polar bear position; N drones want to attack M ground targets, then the search dimension of the quantum polar bears in the population is S = M, and the position of the i-th quantum polar bear is It includes an integer part and a decimal part. The integer part represents the drone number sequence. The decimal part is sorted in ascending order. The sorted order represents the target number of the drone attack. The position of the i-th quantum polar bear in the k-th iteration is Mapping to task assignment matrix Then the fitness function value of the i-th quantum polar bear in the k-th iteration is Among them, t = 1 represents the inverse of the total value obtained by the objective function UAV attacking the ground target, and t = 2 represents the end time of the objective function model; Step 4: sort the non-dominated solutions according to the fitness function values of all quantum polar bear positions; Step 5, calculate the crowdedness of the quantum polar bear positions in all non-dominated levels; Step 6: Use the seal catching strategy to update the quantum position of the quantum polar bear; The h-dimensional search radius of the i-th quantum polar bear in the k-th iteration is is a random number between [0,0.3]; in the seal capture strategy The h-dimensional quantum rotation angle of the i-th quantum polar bear in the iteration is is a random number between [0,π]; Use the quantum revolving door to update the h-dimensional quantum position of the i-th quantum polar bear in the seal catching strategy: Through quantum position Mapping location Then calculate The fitness function value And discuss by category, if Dominate but Keep the result unchanged; otherwise, The details are as follows: Step 7: Use the ice drift strategy to update the quantum position of the quantum polar bear; In the floating ice drift strategy, the h-dimensional quantum rotation angle of the i-th quantum polar bear is is a random number between [0,1], is [0,φ i ], where φ i is the sum of the Euclidean distances between all quantum polar bears and the i-th quantum polar bear, in, Use the quantum revolving door to update the h-dimensional quantum position of the i-th quantum polar bear in the ice drift strategy: Then calculate The fitness function value t=1,2, and discuss by category, if Dominate but Keep the result unchanged, otherwise, let The details are as follows: Step 8: Use population reproduction and extinction strategies to update the quantum position of the quantum polar bear; In the quantum polar bear population, the quantum polar bears may starve to death because they fail to catch prey, or reproduce after successfully hunting. This update strategy simulates the harsh conditions in the Arctic. In step 4, after sorting the non-dominated solutions of the quantum polar bears, if there are more than 3 non-dominated levels in the quantum polar bear population, all the quantum polar bears in the last non-dominated level will starve to death, and then the offspring of the quantum polar bears with a non-dominated level of 1 will replace them. If there are only 2 or 1 non-dominated levels in the quantum polar bear population, the quantum polar bears will not carry out population reproduction and extinction strategies. Assume that there are more than 3 non-dominated levels in the quantum polar bear population, and there are b quantum polar bears in the quantum polar bear set with non-dominated level 1, and there are b1 quantum polar bears in the last non-dominated level; the h-th dimension variable of the new quantum position of the j-th quantum polar bear is Where j = 1, 2, ..., b1, And ε takes random values between [1,b]; Step nine, determine whether the maximum number of iterations K2 of the quantum polar bear has been reached. If so, terminate the iteration, map the positions of the quantum polar bears with a non-dominated level of 1 into a task allocation matrix and output it; otherwise, set k=k+1 and continue to execute step four.
2. According to the multi-objective quantum polar bear mechanism heterogeneous UAV large-scale task allocation method of claim 1, it is characterized in that: Step 2 specifically includes: setting the population size to K1 and the maximum number of iterations to K2; in the initial population, the quantum position of the random initial quantum polar bear, the first generation initial quantum position of the i-th quantum polar bear is Where S is the maximum dimension of the quantum position vector. Any dimension of all quantum positions is a random number between [0,1]. The position of the quantum polar bear can be obtained through quantum position mapping. The quantum position of the i-th quantum polar bear in the k-th iteration is Then the position of the i-th quantum polar bear in the k-th iteration is mapped to Where i = 1, 2, ..., K1, k ∈ {1, 2, ..., K2}, The h-th dimensional variable representing the position of the i-th quantum polar bear, is the upper bound of the h-th dimension position vector of the search space, is the lower bound of the position vector of the h-th dimension of the search space.
3. According to the multi-objective quantum polar bear mechanism heterogeneous UAV large-scale task allocation method of claim 2, it is characterized in that: Step 4 specifically includes: There are two quantum polar bears labeled i1 and i2 whose positions are and If all fitness functions satisfy And there is at least one fitness function Then the quantum polar bear position Dominate the position of the quantum polar bear is a non-dominated solution; if all fitness functions satisfy And there is at least one fitness function Quantum Polar Bear Domination of Quantum Polar Bears is a non-dominated solution; if the above relationship does not exist, then these two quantum polar bears have no domination relationship; Traverse each quantum polar bear in the quantum polar bear population Get the position of the i-th quantum polar bear The number of quantum polar bears and quilt Dominated quantum polar bear position set P i k ; Then count the number of dominated positions of all quantum polar bears, if This means that there is no quantum polar bear position dominating The non-dominated level of is 1; if but The non-dominated level of is 2; and so on, the non-dominated level of each quantum polar bear position is the number of dominated positions plus 1; finally, all quantum polar bears are sorted in ascending order according to their non-dominated levels.
4. According to the multi-objective quantum polar bear mechanism heterogeneous UAV large-scale task allocation method of claim 3, it is characterized in that: Step 5 specifically includes: there are b quantum polar bears in the quantum polar bear set with a non-dominated level of λ, and they are arranged in ascending order according to the t-th fitness function value, t = 1, 2, After sorting The t-th fitness function value of the quantum polar bear, F t max is the maximum value of the tth fitness function in the quantum polar bear set, F t min is the minimum value of the t-th fitness function in the quantum polar bear set, then the crowdedness of the t-th fitness function in the i-th quantum polar bear is when or hour, The first The total crowding value of the quantum polar bears is Finally, the b quantum polar bears are arranged in descending order of crowding degree.
Citation Information
Patent Citations
Multi-unmanned aerial vehicle task allocation method based on multi-target quantum krill group mechanism
CN112926825A
