Dynamic task planning method based on hybrid genetic algorithm

Abstract

The invention discloses a dynamic task planning method based on a hybrid genetic algorithm, belongs to the technical field of unmanned aerial vehicle dynamic task planning, and is used for realizing dynamic task distribution in a search and reconnaissance task scene. The method comprises the following steps: pre-planning according to task information and unmanned aerial vehicle capability information by using a hybrid genetic algorithm to obtain a pre-planning scheme; the hybrid genetic algorithm is formed by combining an original genetic algorithm and a variable neighborhood search algorithm, the basic genetic algorithm is used for ensuring the exploration capability of the algorithm, and the variable neighborhood search algorithm is used for ensuring the development capability of the algorithm; loading the dynamic task information, and sorting the dynamic tasks according to the occurrence time of the dynamic task information; then, a dynamic allocation mechanism is utilized to allocate dynamic tasks, and four dynamic adjustment strategies are designed for different task scenes, namely a random allocation strategy, an allocation strategy based on clustering center allocation, a nearby allocation strategy and an allocation strategy based on fuel surplus; and finally, the unmanned aerial vehicle updates the task sequence.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle dynamic task planning, and in particular relates to a dynamic task planning method based on a hybrid genetic algorithm, which is used to realize dynamic task allocation in a search and reconnaissance mission scenario. Background Art

[0002] With the rapid development of science and technology, the application of drones and unmanned vehicles has also been greatly expanded. Task allocation is the key to the problem of multiple unmanned platforms performing multiple tasks. In traditional task allocation problems, tasks are completed after one planning, but in actual application scenarios, most tasks are dynamically changing. Therefore, in addition to pre-planning at the beginning of the task, dynamic reallocation of tasks is also required. In the dynamic task planning problem, the types of dynamically changing tasks mainly include the following categories. The first category is that new tasks appear one after another over time, such as search and reconnaissance tasks. The second category is that the task is a dynamically moving target, and the position of the target changes over time, such as the task of tracking a dynamic target. The third category is that the demand for the task changes dynamically, and the demand for the task increases linearly or nonlinearly over time, such as fire rescue tasks.

[0003] There are many planning algorithms for dynamic task allocation problems, such as particle swarm algorithm, ant colony algorithm, genetic algorithm, A* algorithm, etc. Each algorithm has its own characteristics and scope of application. For example, the data storage structure of the map directly affects the selection of the algorithm. The present invention designs four dynamic adjustment strategies for dynamic tasks in different situations, namely, random allocation strategy, allocation strategy based on cluster center allocation, nearest allocation strategy, and allocation strategy based on the remaining fuel amount. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a dynamic task planning method based on a hybrid genetic algorithm, which is used to realize dynamic task allocation in search and reconnaissance mission scenarios. The present invention first uses a hybrid genetic algorithm to perform pre-planning according to task information and drone capability information to obtain a pre-planning scheme. The hybrid genetic algorithm is a combination of an original genetic algorithm and a variable neighborhood search algorithm. The basic genetic algorithm is used to ensure the exploration ability of the algorithm, and the variable neighborhood search algorithm is used to ensure the development ability of the algorithm; then the dynamic task information is loaded, and the dynamic tasks are sorted according to the order of appearance of the dynamic task information; then the dynamic allocation mechanism is used to allocate the dynamic tasks.

[0005] A dynamic task planning method, comprising:

[0006] Step 1: Use the hybrid genetic algorithm to pre-plan according to the mission information and UAV capability information to obtain the pre-planning scheme, which is as follows:

[0007] First, initialize and generate the initial population; the initialized individuals are expressed by the following formula:

[0008] X=[x1,x2,…,x I ,x I+1 ,x I+2 ,…,x I+N-1 ]

[0009] Where I represents the total number of tasks, N represents the total number of drones; the number of times the task sequence is segmented is N-1; x1, x2, …, x I represents the task, and its value range is [1,I]; x I+1 ,x I+2 ,…,x I+N -1 indicates the split position of the task, and the value range is [1, I-1];

[0010] Then, individual selection is performed according to the constructed selection function; then the evolutionary iteration phase is entered, and when the iteration termination condition is met, the pre-planning scheme is output;

[0011] Among them, the evolutionary iteration stage includes: crossover operation, mutation operation and neighborhood change operation;

[0012] The crossover operation first generates a crossover position, and then exchanges the genes corresponding to the crossover position; wherein the individual task number position and the split position are independently crossover operated;

[0013] The mutation operation first randomly selects two mutation individuals; then generates mutation sites; finally, exchanges the genes on the mutation sites, and then replaces the genes in their respective codes that are the same as the genes after the mutation sites are mutated with the original genes on the mutation sites; different numbers of mutation sites are set according to the number of tasks;

[0014] Step 2: Load dynamic task information and sort the dynamic tasks according to the time when the dynamic task information appears;

[0015] Step 3: Dynamic tasks are assigned and the UAV then updates the task sequence.

[0016] Preferably, the method of performing individual selection according to the constructed selection function includes: performing individual evaluation according to the constructed evaluation function; the optimization target is the total distance flown by the UAV to complete the task, that is:

[0017] f(X)=l1+l2+…+l n +…+l N

[0018] Among them, l nIndicates the flight distance required for the nth UAV to complete the mission. The UAV needs to meet its own endurance time constraint during the mission. The solution formula for the time required for the UAV to complete the mission is as follows:

[0019] t n = l n / v n ,n∈{1,2,…,N}

[0020] Among them, t n represents the time required for the nth UAV to complete the task, v n represents the flight speed of the nth drone; if the individual does not meet the above drone endurance constraint, the individual will be discarded during the evolution process;

[0021] Then, the individual selection is performed according to the constructed selection function; the process of solving the individual fitness value is shown in the following formula:

[0022]

[0023] in, represents the fitness value of the pth individual in the gth generation, represents the objective function value of the p-th individual in the g-th generation; and Respectively represent the minimum and maximum objective function values ​​of the g-th generation;

[0024] The selection of individuals is based on the following formula:

[0025]

[0026] in, represents the pth individual of the gth generation, 1 and 0 represent retaining and discarding this individual respectively, and fv represents the adaptive value threshold.

[0027] Preferably, when selecting individuals, in the early stage of evolution, the fitness value is set larger to ensure that the selected individuals are better; in the later stage of evolution, the fitness value is adjusted smaller to ensure that more individuals are selected. The adjustment process is shown in the following formula:

[0028]

[0029] Here, g represents the number of iterations.

[0030] Preferably, the neighborhood changing operation includes exchange, shift and reversal operations;

[0031] The exchange method comprises: generating an exchange position, exchanging genes at two exchange positions, and then obtaining a new individual;

[0032] The method of shifting comprises: generating a gene position 1 to be shifted and a position 2 to be shifted by the gene position; moving the gene at the gene position 1 to the back of the gene at the gene position 2;

[0033] The reversal method includes: generating a reversal interval, and performing a reversal operation on the genes in the reversal interval.

[0034] Preferably, a random allocation strategy is used to allocate dynamic tasks, specifically: when a new task appears, it is randomly allocated to a drone, and the drone later updates its own task list.

[0035] Preferably, the dynamic tasks are allocated based on the allocation strategy of cluster centers, specifically: first calculate the task center of each drone; then, when a new task appears, assign it to the nearest cluster; finally, the drones in this cluster update their own task lists.

[0036] Preferably, the dynamic tasks are assigned based on the closest distance allocation strategy, specifically: first calculate the distance between each drone and the location where the new task appears; then find the drone with the smallest distance and assign the new task to this drone; finally, this drone updates its own task list

[0037] Preferably, the dynamic tasks are allocated based on the allocation strategy of the remaining fuel, specifically: first, the remaining endurance time of each UAV is calculated according to the time when the dynamic task occurs; then, the UAV with the largest remaining endurance time is obtained and the new task is allocated to this UAV; finally, this UAV updates its own task list.

[0038] The present invention has the following beneficial effects:

[0039] First, the present invention adopts a hybrid genetic algorithm to perform pre-planning according to the mission information and the UAV capability information to obtain a pre-planning scheme. The hybrid genetic algorithm is a combination of the original genetic algorithm and the variable neighborhood search algorithm. The basic genetic algorithm is used to ensure the algorithm's exploration ability, and the variable neighborhood search algorithm is used to ensure the algorithm's development ability;

[0040] Second, the present invention uses a dynamic allocation mechanism to allocate dynamic tasks. The four dynamic adjustment strategies designed are random allocation strategy, allocation strategy based on cluster center allocation, nearby allocation strategy, and allocation strategy based on the remaining fuel amount, which are used to deal with different dynamic task situations;

[0041] Third, the method of the present invention is used to realize dynamic task allocation in search and reconnaissance mission scenarios. The present invention first uses a hybrid genetic algorithm to perform pre-planning according to the task information and the UAV capability information to obtain a pre-planning scheme. The hybrid genetic algorithm is a combination of the original genetic algorithm and the variable neighborhood search algorithm. The basic genetic algorithm is used to ensure the exploration ability of the algorithm, and the variable neighborhood search algorithm is used to ensure the development ability of the algorithm; then the dynamic task information is loaded, and the dynamic tasks are sorted according to the order of appearance of the dynamic task information; then the dynamic allocation mechanism is used to allocate the dynamic tasks, and four dynamic adjustment strategies are designed for different task scenarios, namely, random allocation strategy, allocation strategy based on cluster center allocation, nearby allocation strategy, and allocation strategy based on the remaining fuel; finally, the UAV updates the task sequence. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of encoding and decoding;

[0043] Figure 2 Schematic diagram of the crossover operator;

[0044] Figure 3 is a schematic diagram of the mutation operator;

[0045] Figure 4 Schematic diagram of the variable neighborhood operator. DETAILED DESCRIPTION

[0046] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0047] A dynamic task planning method based on a hybrid genetic algorithm, the method comprising the following steps:

[0048] Step 1: Use the hybrid genetic algorithm to pre-plan according to the mission information and UAV capability information to obtain the pre-planning scheme, which is as follows:

[0049] First, initialize and generate the initial population; the initialized individuals are expressed by the following formula:

[0050] X=[x1,x2,…,x I ,x I+1 ,x I+2 ,…,x I+N-1 ]

[0051] Where I represents the total number of tasks and N represents the total number of drones. Since the number of drones is N, the number of times the task sequence is segmented is N-1. I Represents a task, and its value range is [1,I]. I+1 ,x I+2 ,…,x I+N-1Indicates the split position of the task, and its value range is [1,I-1].

[0052] For example, the specific encoding and decoding process is as follows Figure 1 As shown in the figure, the split positions are 2, 5, and 7, which means that the task sequence is split after positions x2, x5, and x7, and then 4 task sequences are obtained, which are respectively assigned to 4 drones. Figure 1 As shown, the mission sequence of UAV 1 is 2→4, the mission sequence of UAV 2 is 6→1→7, the mission sequence of UAV 3 is 3→9, and the mission sequence of UAV 4 is 5→8.

[0053] Then, individual evaluation is performed according to the constructed evaluation function; the optimization goal is the total distance that the UAV flies to complete the mission, that is:

[0054] f(X)=l1+l2+…+l n +…+l N

[0055] Among them, l n It represents the flight distance required for the nth drone to complete the mission. The drone needs to meet its own endurance time constraint during the mission. The solution formula for the time required for the drone to complete the mission is as follows:

[0056] t n = l n / v n ,n∈{1,2,…,N}

[0057] Among them, t n represents the time required for the nth UAV to complete the task, v n Indicates the flight speed of the nth drone. If the individual does not meet the above drone endurance constraints, it will be discarded during the evolution process.

[0058] Then, the individual selection is performed according to the constructed selection function; the process of solving the individual fitness value is shown in the following formula:

[0059]

[0060] in, represents the fitness value of the pth individual in the gth generation, Represents the objective function value of the pth individual in the gth generation. and Represent the minimum and maximum objective function values ​​of the g-th generation respectively. ε can be set to 0.001.

[0061] The selection of individuals can be based on the following formula:

[0062]

[0063] in, represents the pth individual of the gth generation, 1 and 0 represent retaining and discarding this individual respectively, and fv represents the adaptive value threshold.

[0064] The selection operation is designed to be adjusted according to the number of iterations of the algorithm. In the early stage of evolution, the fitness value is set to be larger to ensure that the selected individuals are better. In the later stage of evolution, the fitness value is appropriately reduced to ensure that more individuals are selected. The adjustment process is shown in the following formula:

[0065]

[0066] Here, g represents the number of iterations.

[0067] Then it enters the evolutionary iteration phase, and when the termination condition is met, the pre-planning solution is output. The evolutionary iteration phase mainly includes three types of evolutionary operators: crossover operator, mutation operator, and variable neighborhood operation operator.

[0068] The crossover operation first generates the crossover position, and then exchanges the genes corresponding to the crossover position. The difference is that in the encoding of the present invention, the individual task number position and the split position should be independently crossover mutation operation. In view of the characteristics of this problem, the crossover operator designed is as follows Figure 2 shown.

[0069] When this crossover operator performs individual crossover operations, the original individuals retain their respective task sequences and exchange their segmentation positions. Figure 2 As shown, the segmentation position sequence of individual one is 257, and the segmentation position sequence of individual two is 348. After performing this crossover operation, the segmentation position sequence of the new individual one is 348, and the segmentation position sequence of the new individual two is 257.

[0070] The mutation operation first randomly selects two mutant individuals. Then, a mutation position is generated. Finally, the genes on the mutation position are exchanged, and the genes in their respective codes that are the same as the genes after the mutation position is mutated are replaced with the original genes on the mutation position. According to the characteristics of the problem, a parameter adaptive mutation operation is designed. Different numbers of mutation positions are set according to the number of tasks. The designed mutation operator is as follows: Figure 3 shown.

[0071] exist Figure 3 In the example, the case where the number of variant positions is 1 is illustrated. First, the variant position c=2 is generated, and then the genes at gene position 2 of individual one and individual two are exchanged, that is, 4 at gene position 2 of individual one and 1 at gene position 2 of individual two are exchanged, and finally, the genes with the same genes after the variant position mutation in their respective codes are changed to the original genes at the variant position.

[0072] The variable neighborhood operators include swap, shift and reversal operations. The number of gene positions operated is adaptively adjusted with the problem size, i.e., the encoding length. The designed variable neighborhood operators are as follows: Figure 4 shown.

[0073] First, let's explain swap(1,1) in the swap operator. The first step of the swap(1,1) operation is to generate the swap bit, such as Figure 4 As shown, the two exchange positions are 2 and 8. The second step of the swap(1,1) operation: exchange genes 4 and 5 at the two exchange positions, and then obtain a new individual. The operations of swap(1,2) and swap(2,1) are similar to swap(1,1), where 2 represents 2 gene positions, that is, 1 gene position is exchanged with 2 gene positions, or 2 gene positions are exchanged with 1 gene position.

[0074] Then, the shift(1,0) in the shift operator is explained. The first step of the shift(1,0) operation is to generate a gene position to be shifted and a position where the gene position will move. Figure 4 As shown, the generated gene position is 2, and this gene position will be moved to the back of gene position 8. The second step of the shift(1,0) operation: move gene 4 on gene position 2 to the back of gene 5 on gene position 8. The operation of shift(2,0) is similar to shift(1,0), where 2 means shifting 2 gene positions.

[0075] Finally, the reversal operator is explained. The first step of the reversal operation is to generate a reversal interval, such as Figure 4 As shown, the reversal interval is 2 to 8. The second step of the reversal operation: the genes in the reversal interval are inverted, that is, the genes in the original reversal interval are 4617395, and after the inversion operation, they are 5937164.

[0076] Step 2: Load dynamic task information and sort the dynamic tasks according to the time when the dynamic task information appears;

[0077] Load dynamic task information and sort the dynamic tasks according to the time when the dynamic task information appears; the dynamic task information is expressed by the following formula:

[0078]

[0079] in, represents the appearance time of new task j, that is, how long after the start time the new task appears. j and jrepresents the horizontal and vertical coordinates of dynamic task j, and J represents the number of dynamic tasks.

[0080] Dynamic tasks are sorted in ascending order based on their appearance time.

[0081] Step 3: Use the dynamic allocation mechanism to allocate dynamic tasks, and then the drone updates the task sequence.

[0082] Dynamic allocation mechanism is used to allocate dynamic tasks, and then the UAV updates the task sequence. Four dynamic adjustment strategies are designed for different mission scenarios, namely random allocation strategy, allocation strategy based on cluster center allocation, nearest allocation strategy, and allocation strategy based on remaining fuel.

[0083] Next, the detailed process of the four dynamic adjustment strategies is described:

[0084] (1) Random assignment strategy (DS = 1): The main strategy is to randomly assign a new task to a drone when it appears, and the drone can update its task list later. The random assignment strategy can prevent the general assignment strategy from being applicable only to a certain type of special case due to its fixed assignment rules.

[0085] The specific steps are as follows:

[0086] 1) Traverse all dynamic tasks;

[0087] 2) Generate a random number Uj = rand(N), where Uj represents the number of the drone assigned to the dynamic task j, rand() is a function for generating random numbers, and rand(N) represents generating a random number. The value range of the random number is [1, N].

[0088] 3) The drone updates the mission sequence.

[0089] (2) Allocation strategy based on cluster center allocation (DS=2): The main strategy is to first calculate the task center of each drone. Then, when a new task appears, it is assigned to the nearest cluster. Finally, the drones in this cluster update their task lists. The allocation strategy based on cluster center allocation is suitable for scenarios where tasks are more dispersed. After the drone completes the new task, it can quickly take care of the original task.

[0090] The specific steps are as follows:

[0091] 1) Traverse the mission sequences of all drones;

[0092] 2) Obtain the cluster center of each drone’s mission sequence. The cluster center C of drone n is n =(x n ,y n ) is solved as follows: x n and n Respectively represent the horizontal and vertical coordinates of the cluster center, x i and i I and n represents the length of the task list of drone n;

[0093] 3) Traverse all dynamic tasks;

[0094] 4) Solve the Euclidean distance between the cluster centers of each dynamic task and UAV task sequence;

[0095] 5) Get the minimum distance and assign the dynamic task to this UAV;

[0096] 6) The drone updates the mission sequence.

[0097] (3) Allocation strategy based on the closest distance (DS=3): The main strategy is to first calculate the distance between each UAV and the location where the new task appears. Then, find the UAV with the smallest distance and assign the new task to this UAV. Finally, this UAV can update its own task list. The allocation strategy based on the closest distance is suitable for situations where dynamically appearing tasks have a higher priority. This strategy can ensure that the UAV can respond quickly when a dynamic task appears.

[0098] The specific steps are as follows:

[0099] 1) Traverse all dynamic tasks;

[0100] 2) Traverse all drones;

[0101] 3) Calculate the current position of each UAV based on the occurrence time of the dynamic task;

[0102] 4) Solve the Euclidean distance between the dynamic task and the current position of the drone and obtain the minimum distance;

[0103] 5) Assign dynamic tasks to this drone;

[0104] 6) The drone updates the mission sequence.

[0105] (4) Allocation strategy based on remaining fuel (DS=4): The main strategy is to first calculate the remaining flight time of each UAV based on the time when the dynamic task appears. Then, find the UAV with the largest remaining flight time and assign the new task to this UAV. Finally, this UAV can update its own task list. The allocation strategy based on remaining fuel is suitable for situations where the original task may not be completed after the dynamic task appears and the task is assigned. This strategy can ensure that the task is completed to the greatest extent overall.

[0106] The specific steps are as follows:

[0107] 1) Traverse all dynamic tasks;

[0108] 2) Traverse all drones;

[0109] 3) According to the occurrence time of dynamic tasks, calculate the remaining fuel of each UAV and obtain the maximum remaining fuel;

[0110] 4) Assign dynamic tasks to this UAV;

[0111] 5) The drone updates the mission sequence.

[0112] 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 dynamic task planning method, characterized in that: include: Step 1: Use the hybrid genetic algorithm to pre-plan according to the mission information and UAV capability information to obtain the pre-planning scheme, which is as follows: First, initialize and generate the initial population; the initialized individuals are expressed by the following formula: X=[x1,x2,…,x I , x I+1 , x I+2 ,…,x I+N-1 ] Where I represents the total number of tasks, N represents the total number of drones; the number of times the task sequence is segmented is N-1; x1, x2, …, x I represents the task, and its value range is [1,I]; x I+1 ,x I+2 ,…,x I+N -1 Indicates the split position of the task, and its value range is [1, I-1]; Then, individual selection is performed according to the constructed selection function; then the evolutionary iteration phase is entered, and when the iteration termination condition is met, the pre-planning solution is output; Among them, the evolutionary iteration stage includes: crossover operation, mutation operation and neighborhood change operation; The crossover operation first generates a crossover position, and then exchanges the genes corresponding to the crossover position; wherein the individual task number position and the split position are independently crossover operated; The mutation operation first randomly selects two mutation individuals; then generates mutation sites; finally, exchanges the genes on the mutation sites, and then replaces the genes in their respective codes that are the same as the genes after the mutation sites are mutated with the original genes on the mutation sites; different numbers of mutation sites are set according to the number of tasks; Step 2: Load dynamic task information and sort the dynamic tasks according to the time when the dynamic task information appears; Step 3: Dynamic tasks are assigned and the UAV then updates the task sequence.

2. A dynamic task planning method as claimed in claim 1, characterized in that: The method for performing individual selection according to the constructed selection function includes: performing individual evaluation according to the constructed evaluation function; the optimization target is the total distance of the flight of the UAV to complete the task, that is: f(X)=l1+l2+…+l n +…+l N Among them, l n Indicates the flight distance required for the nth UAV to complete the mission. The UAV needs to meet its own endurance time constraint during the mission. The solution formula for the time required for the UAV to complete the mission is as follows: t n =l n / v n ,n∈{1,2,…,N} Among them, t n represents the time required for the nth UAV to complete the task, v n represents the flight speed of the nth drone; if the individual does not meet the above drone endurance constraint, the individual will be discarded during the evolution process; Then, the individual selection is performed according to the constructed selection function; the process of solving the individual fitness value is shown in the following formula: in, represents the fitness value of the pth individual in the gth generation, represents the objective function value of the p-th individual in the g-th generation; and Respectively represent the minimum and maximum objective function values ​​of the g-th generation; The selection of individuals is based on the following formula: in, represents the pth individual of the gth generation, 1 and 0 represent retaining and discarding this individual respectively, and fv represents the adaptive value threshold.

3. A dynamic task planning method as claimed in claim 2, characterized in that: When selecting individuals, in the early stage of evolution, the fitness value is set larger to ensure that the selected individuals are better; in the later stage of evolution, the fitness value is adjusted smaller to ensure that more individuals are selected. The adjustment process is shown in the following formula: Here, g represents the number of iterations.

4. A dynamic task planning method as claimed in claim 1, characterized in that: Neighborhood-changing operations include swap, shift, and reverse operations; The exchange method comprises: generating an exchange position, exchanging genes at two exchange positions, and then obtaining a new individual; The method of shifting comprises: generating a gene position 1 to be shifted and a position 2 to be shifted by the gene position; moving the gene at the gene position 1 to the back of the gene at the gene position 2; The reversal method includes: generating a reversal interval, and performing a reversal operation on the genes in the reversal interval.

5. A dynamic task planning method as claimed in claim 1, characterized in that: Dynamic tasks are assigned using a random assignment strategy. Specifically, when a new task appears, it is randomly assigned to a drone, which then updates its own task list.

6. A dynamic task planning method as claimed in claim 1, characterized in that: Dynamic tasks are assigned based on the cluster center allocation strategy, specifically: first calculate the task center of each drone; then, when a new task appears, assign it to the nearest cluster; finally, the drones in this cluster update their own task lists.

7. A dynamic task planning method as claimed in claim 1, characterized in that: Dynamic tasks are assigned based on the nearest distance allocation strategy, specifically: first calculate the distance between each UAV and the location where the new task appears; then, find the UAV with the smallest distance and assign the new task to this UAV; finally, this UAV updates its own task list.

8. A dynamic task planning method as claimed in claim 1, characterized in that: Dynamic tasks are allocated based on the allocation strategy of the remaining fuel. Specifically, the remaining endurance of each UAV is calculated according to the time when the dynamic task occurs. Then, the UAV with the largest remaining endurance is found and the new task is allocated to this UAV. Finally, this UAV updates its own task list.

Images