Heterogeneous UAV collaborative mission planning method under heterogeneous carrier platform delivery mode
By building a combined optimization model and adaptive optimization algorithm, the problem of the combined role of factors in the coordinated task allocation of heterogeneous drones is solved, the safety and efficiency of task execution are improved, and the diversity of heterogeneous carrier platforms and drones is adapted to the .
Patent Information
- Application Number
- CN202411088696.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-09
AI Technical Summary
The existing heterogeneous drone collaborative mission allocation method ignores multiple factors in considering the drone departure and recovery process, and fails to effectively deal with the no-fly zone and the unattacked risk domain, resulting in mission execution efficiency and security Restricted.
A combination optimization model was constructed, combining teaching and learning optimization algorithm (AMTLBO) with grouped multi-teacher teaching, adaptive learning rate and self-motivation mechanism, with risk punishment and program execution time as the objective functions, considering heterogeneous carrier platform and unmanned The heterogeneity of the aircraft is optimized to avoid the risk of drones approaching the unattacked target.
It improves the solution performance of the collaborative mission planning of drones, reduces the risk cost of task allocation, enhances the safety and efficiency of task execution, and adapts to the diversity of heterogeneous carrier platforms and drones.
Smart Images

Figure CN118625836B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of unmanned aerial vehicle task allocation and relates to a method for collaborative task planning of heterogeneous unmanned aerial vehicles in a heterogeneous carrier platform delivery mode. Background Art
[0002] Heterogeneous UAV collaboration technology has been widely used in modern warfare. Many inventions have emerged in this field to provide solutions for various UAV collaborative task allocation problems, but the current inventions still have many shortcomings. For example, due to the limited flight endurance of UAVs, the combat range is generally limited. Although the Chinese patent "CN115471110A A multi-heterogeneous UAV system collaborative task allocation method based on conditional probability and considering multi-objective evolutionary algorithm" solves the various constraints of UAVs during task execution, it ignores various factors in the UAV departure and recovery process. Although the Chinese invention "CN115755963B A UAV swarm collaborative task planning method considering vehicle delivery mode" considers factors such as UAV dispatch recovery and delivery vehicles, the UAVs performing the task are isomorphic. Although the Chinese invention "CN114417735B A multi-UAV collaborative task planning method in cross-regional joint operations" not only considers heterogeneous UAVs but also dispatch recovery factors compared to the above two inventions, this invention ignores environmental factors such as no-fly zones.
[0003] Based on the above three comparative inventions, the present invention expands the scope of UAV collaborative mission planning by utilizing various airborne or shipborne platforms such as large transport aircraft and aircraft carriers. In addition, it considers safe flight factors such as no-fly zones in the environment and risk domains of unattacked targets that are generally ignored by existing inventions. Under the combined effect of the above factors, the model constructed by the present invention places higher requirements on the optimization performance of the solution method. To this end, the present invention proposes an adaptive self-motivation teaching and learning method to solve this problem. Summary of the invention
[0004] In order to solve the above technical problems, the present invention first constructs a combinatorial optimization model in the context of heterogeneous carrier platforms delivering drones, and then proposes an improved teaching and learning optimization algorithm to solve the combinatorial optimization model. In the present invention, the carrier platforms for delivering drones include various types, such as ships at sea, land airports or large transport aircraft in the air, that is, this problem includes multiple heterogeneous airports. In addition, drones are also heterogeneous, and different types of drones have different functions to cope with various mission requirements from the target.
[0005] The present invention also considers that the areas around the unattacked targets are threatening, and the drone will bear additional risk costs in the process of approaching these areas. In addition, the present invention considers the no-fly zones in the environment caused by enemy fire. Figure 2It is an example diagram of the background of the present invention. In the above background, a combinatorial optimization model with risk penalty and solution execution time as the objective function and various constraints is constructed. Based on the combinatorial optimization model, an adaptive self-motivated teaching-learning-based optimization algorithm (AMTLBO) is proposed to solve the problem, which includes grouped multi-teacher teaching, adaptive learning rate and self-motivation mechanisms.
[0006] In order to achieve the above object, the technical solution adopted by the present invention is:
[0007] A method for planning a collaborative task of heterogeneous unmanned aerial vehicles under a heterogeneous carrier platform delivery mode, wherein the specific steps of the method for planning a collaborative task of heterogeneous unmanned aerial vehicles under the heterogeneous carrier platform delivery mode are as follows: Step 1) Analyze the characteristics of the problem, including carrier platform factor analysis, unmanned aerial vehicle factor analysis, environmental factor analysis, target factor analysis and task intention analysis. Step 2) Clarify the optimization variables, objective functions and constraints, and establish a combined optimization model. Step 3) Design an adaptive self-motivated teaching and learning algorithm: First, construct a suitable discrete solution code according to the characteristics of the problem, randomly initialize the discrete solution through resources and task requirements, and then obtain the initial population; then, obtain the distance between two points in the obstacle environment through the visual graph method, and calculate the population objective function value in combination with the track adjustment method and the logical unlocking method; then, arrange the population in ascending order according to the population objective function value, and calculate the median and mean of the population objective function value for size comparison, and select different update methods according to the judgment results: when the median is greater than the mean value, execute the teaching stage population update method; when the median is less than the mean value, execute the learning stage population update method; then, execute the self-motivation mechanism to further optimize the population. Step 4) Problem solving and evaluation: First, set appropriate algorithm parameters, then solve the problem using the method designed in step 3) until the maximum number of iterations is reached, and select the optimal solution for evaluation. This includes the following steps:
[0008] Step 1: Analyze the characteristics of the problem, as follows:
[0009] Step 1-1: Analysis of carrier platform elements;
[0010] Various types of heterogeneous delivery platforms are considered in the present invention, such as aircraft carriers, large transport aircraft, and land platforms. Denoted as the airport set, where the superscript type of P 1 ,...,type a represents the type of carrier platform P, and a represents the number of carrier platform types. For example, when it includes sea-based platforms, land-based platforms, and air-based platforms, a=3. Respectively represent the corresponding type1 ,...,type a The number of different types of carrier platforms. In the present invention, the UAV flies out from a heterogeneous carrier platform to perform a mission and flies back to the heterogeneous carrier platform after the mission is completed.
[0011] Step 1-2: UAV element analysis, as follows:
[0012] For this task system, the UAV (Unmanned Aerial Vehicle UAV) is required to have different task execution capabilities, so the UAV of the present invention is a heterogeneous UAV with different functions, such as a reconnaissance UAV S with only reconnaissance capability (performing classification and evaluation tasks), a bombing UAV M with only attack capability (performing attack tasks), and a combat UAV C that can perform both reconnaissance and attack tasks, as shown in Table 1. Indicates that U i represents the i-th drone; N U Represents the total number of drones. Figure 3 (a), (b) and (c) are reconnaissance aircraft, bomber and fighter respectively.
[0013] Table 1. UAV types and functions
[0014]
[0015] Step 1-3: Environmental factor analysis, as follows:
[0016] The present invention assumes that there are some prohibited airspaces on the battlefield, including radars, mountains, air defense strike zones, etc. All no-fly zones are simplified into convex polygons and circles. UAVs must not appear in the prohibited airspace during flight. and collection Respectively represent the circular obstacles and convex polygonal obstacle sets on the battlefield, where N c and N b are the total number of circular obstacles and polygonal obstacles respectively. Then the total set of obstacles can be defined as: Finally, the present invention appropriately expands the no-fly zone in the battlefield and uses the UAV as a particle in the battlefield environment to simplify the flight process of the UAV.
[0017] Steps 1-4: Target element analysis, as follows:
[0018] The target considered in the present invention is stationary, and it can also be assumed that the target is a particle on a two-dimensional plane. T target, use collection Indicates that T i represents the target numbered i. For each target Ti , it is necessary to perform the classification task (C task), attack task (A task) and evaluation task (V task) with priority restrictions, that is, the task set of each target is {C, A, V}. In addition, in the present invention, there is an area of 'approaching unattacked target-induced risk' around the unattacked target, referred to as AUTIR (the approaching unattacked target-induced risk), which is an important attribute of the unattacked target.
[0019] Steps 1-5: Task intent analysis, as follows:
[0020] During the execution of the mission, it is hoped that the drone will complete the designated mission in the shortest time. Therefore, the shortest execution time of the entire plan is taken as one of the objective functions. In addition, the present invention takes into account the risks caused by approaching unattacked targets, so that drones try to avoid approaching unattacked targets as much as possible during the execution of the mission. In traditional drone collaborative task allocation, since this threat is not taken into account, there are situations where drones pass through the above-mentioned threat airspace. For example, an aircraft is to start from point 1 to perform a mission to point 3, but the straight line path will pass through the threat area of the unattacked target point 2. This situation will increase the risk cost of the entire task allocation plan. In order to avoid the above-mentioned risk costs, the present invention will provide a suitable task allocation plan from the task planning level to avoid the situation where the drone route passes through the threat area of the unattacked target. Note that once the target is attacked, the airspace around the target is no longer considered a threat. Figure 4 It is a supplementary illustration of AUTIR. The present invention introduces the threat function in step 2-2 as another objective function in the combined optimization model to reduce the additional risk cost in the task allocation scheme.
[0021] Step 2: Construct a combined optimization model, as follows:
[0022] Step 2-1: Abstract the multi-UAV collaborative mission scenario into a network:
[0023] The network is the basis for modeling mission planning problems. The location points of airports and targets are defined as nodes of the network. K represents the node set, then E = {(K i ,K j )|K i ,K j ∈K,i,j=0,1,…,N K , i≠j} represents the set of edges, which together form the network CG={K,E}. i represents the i-th network node; K j represents the jth network node; i and j both represent the network node serial number; N K Indicates the total number of network nodes.
[0024] Step 2-2: Establish the objective function of the combinatorial optimization problem through the network:
[0025] First, consider the solution execution time evaluation function F 1 , as shown in formula (1):
[0026]
[0027] in, u represents the drone number, U represents the drone number set, is a decision variable. If UAV U u From K i To K j Execute task type k, then indicates returning to the carrier platform, k = 1, 2, 3 respectively indicate executing tasks of type C, A, and V; otherwise, in It's a drone u From K i To K j The estimated path length, Represents UAV u speed.
[0028] In addition, the closer the drone is to the target, the greater the risk it suffers. Therefore, the penalty function term is expressed as formula (2):
[0029]
[0030] Among them, N T Indicates the number of targets; N U Indicates the number of drones; N K represents the number of nodes; l represents the number of the target; The target T l Give UAV u From K i To K j The threat penalty brought by , at time t, is as shown in formula (3).
[0031]
[0032] Where r represents T l to (K i ,K j ), R threatRepresents the threat radius of the target. Note that when considering threat penalties, the present invention only considers targets that have not been attacked, and targets that have been attacked are considered to have no threat. Therefore, the threat penalties for the same path at different times are different. In addition, for simplicity, time t in equation (3) is not updated in real time. For example, if the drone is at t 0 Moments from V i Departure, t 2 Time to arrive at V j . The target T on the path l In t 1 ∈[t 0 ,t 2 ] is attacked, then don't consider T l Threats to this path.
[0033] Finally, the objective function of the combined optimization model of the present invention is as shown in formula (4), where α represents the penalty coefficient.
[0034] minF=F 1 +αF 2 ; (4)
[0035] Step 2-3: Analyze the constraints, as follows:
[0036] Through the analysis of the actual application process, the following constraints are analyzed:
[0037] (1) Multi-drone collaboration constraint: UAVs need to collaborate to perform all tasks, that is, each task will be assigned to a UAV to perform, as expressed in Equation (5).
[0038]
[0039] Where l represents the number of the target.
[0040] (2) Ammunition load constraint: The amount of ammunition carried by the UAV is limited, so the number of attack missions of the UAV cannot exceed its ammunition load, as shown in Equation (6).
[0041]
[0042] in, Represents UAV u ammunition capacity; Corresponds to When the UAV u From K i To K j Carry out attack missions.
[0043] (3) Priority constraints: Each objective contains three types of tasks (classification task C, attack task A, and evaluation task V).
[0044] These three types of tasks for the same goal must be completed in sequence, as shown in formula (7).
[0045]
[0046] (4) Dispatch and recovery constraints: In the multi-airport problem, any UAV that takes off from any carrier platform must eventually return to the departure carrier platform, as shown in Equation (8).
[0047]
[0048] (5) Node flux conservation: The number of drones entering each node is equal to the number of drones leaving the node, which can be expressed as equation (9).
[0049]
[0050] (6) Collision avoidance constraint: The flight path of the UAV cannot intersect with the no-fly zone. This constraint is shown in formula (10).
[0051]
[0052] in, Represents node K i To Node K j Φ′ represents the obstacle obtained by expanding Φ.
[0053] Through the above steps, a combinatorial optimization model for multi-UAV collaborative task allocation for heterogeneous delivery platforms was constructed.
[0054] Step 3: Design of adaptive self-motivated teaching and learning algorithm, as follows:
[0055] Step 3-1: Scheme generation and population initialization:
[0056] For N T targets and N U In the case of a UAV, the solution generated by the present invention includes 3 rows and 3N T The first row is the target, using numbers 1, 2, ..., N T Indicates the number of the corresponding target. The second line is the mission type of the target, using numbers 1, 2, and 3 to represent classification, attack, and evaluation types of tasks respectively. The third line is the drone, using numbers 1, 2, ..., N U Indicates the number of the drone. Each column is a task assignment, indicating that the drone is going to perform the corresponding task of the target. Figure 5 It is an example diagram of the task allocation scheme of the present invention.
[0057] During the initialization process, first, a resource constraint set is constructed based on the input drone information. The resource constraint set includes a set of drones with reconnaissance capabilities, a set of drones with attack capabilities, and a set of remaining ammunition. Then, for each target's different mission types, a drone is randomly selected from the set of drones that have the ability to perform the mission and assigned to the mission. Finally, all tasks are assigned to drones according to the above method. Figure 6 The process of initializing the task allocation scheme is given.
[0058] The initial solutions obtained above are sorted by the drone number (row 3), and the following is obtained: Figure 5 The task allocation scheme in the form of a given size is called the initial population. It is worth noting that the solution generated by this method must satisfy the task requirement constraints, ammunition load constraints and sortie recovery constraints.
[0059] Step 3-2: Objective function calculation:
[0060] Step 3-2-1: Priority constraint processing method:
[0061] Before calculating the objective function value, the priority constraint needs to be processed first. In order to meet the priority constraint, the present invention draws on the waiting circle method in Dubins to adjust the trajectory. When the priority constraint is not met, the UAV should fly around a circular path with a minimum turning radius outside the target threat range, waiting for other UAVs to perform related pre-tasks. However, there may be a phenomenon where the UAV falls into infinite waiting, in which case the task allocation scheme is said to be deadlocked. When encountering a deadlock scheme, the present invention draws on the logical unlocking method in the Chinese invention CN114417735B 'A method for collaborative task planning of multiple UAVs in cross-regional joint operations' to unlock the task allocation scheme.
[0062] Step 3-2-2: Estimating the distance between two points using the visual graph method:
[0063] After satisfying the priority constraints, the distance between two points in the obstacle environment is estimated by the visual graph method, and the objective function value is calculated in combination with the Dubins track adjustment method. Because the obstacle information in the environment is known in advance, the present invention uses the visual graph method to solve the problem caused by the coupling of path planning and task allocation.
[0064] Step 3-2-2-1: Single obstacle:
[0065] When the drone trajectory between two points passes through a single polygonal obstacle, the present invention connects the vertices of the polygon to the starting point and the end point respectively. The shortest of all the lines connecting the starting point to the end point is the candidate path. If the trajectory passes through a circle, the shortest path is selected by finding the tangent of the circle passing through the starting point and the end point.
[0066] Step 3-2-2-2: Multiple obstacles:
[0067] If there are multiple obstacles between two points, first consider only the obstacle closest to the starting point. Then select the middle point of the candidate path as the new starting point, reconnect the starting point and the end point, and use the above method to continuously update the starting point until there are no obstacles between the starting point and the end point. Then you can connect all points from the starting point to the end point, and choose the shortest path that does not pass through obstacles.
[0068] Step 3-2-3: Calculation of objective function value:
[0069] Calculate the objective function value of the unlocked solution according to equations (1), (2), (3), and (4) in step 2-2.
[0070] Step 3-3: Design of population status determination mechanism:
[0071] Sort the objective function values calculated in step 3-2, and then calculate the median and mean of the objective function values of the population. If the median > mean, use the teaching phase update method of step 3-4-1, otherwise use the learning phase update method of step 3-4-2.
[0072] Step 3-4: Population updating method design:
[0073] In order to reduce the amount of calculation, the present invention changes the traditional teaching phase and learning phase to no longer execute both phases in one iteration, but to select execution based on the population state. The details are as follows:
[0074] Step 3-4-1: Teaching phase update:
[0075] Step 3-4-1-1: Grouping by multiple teachers:
[0076] First, the present invention adds a multi-teacher mechanism. In swarm intelligence algorithms, the initial iteration usually tends to increase population diversity and try to avoid premature convergence to the local optimum. If there is only one teacher in the teaching stage, the group will converge prematurely and reach the local optimum. In order to avoid this situation, the present invention introduces a multi-teacher mechanism that changes with iteration in the population update in the teaching stage to ensure the diversity of the early group update.
[0077] The multi-teacher mechanism divides the entire class into different learner groups according to the objective function value at the beginning of each iteration. Then, the best performing learner in each learner group is selected as the teacher of that group. As the iterations proceed, the convergence efficiency of the algorithm will be improved by gradually reducing the number of learner groups and improving the quality of teachers. In actual operation, the initial number of teachers is predetermined and modified according to formula (11) throughout the iteration process.
[0078] n t =ceil((1-I / I max )×N teacher ); (11)
[0079] Among them, I represents the current iteration number, I max represents the maximum number of iterations, ceil(·) is the round-up function, N teacher is the initial number of teachers, n t is the number of teachers in the current iteration. The process of selecting teachers starts with formula (11), which determines the number of teachers.
[0080] Next, sort the entire population in ascending order of objective function value.
[0081] Finally, the teacher of each group is determined according to formula (12) and formula (13).
[0082]
[0083] Among them, f teacher,p represents the objective function value of the pth teacher, f best and f worst Represent the best and worst objective function values in the group respectively. q and f q+1 Represents the sorted objective function value, where f q ≤f q+1 If the objective function value of teacher p satisfies formula (13), then learner q is regarded as teacher p, and all learners between it and the next teacher belong to its learning group. If there are repeated teachers, the first learner whose sorting sequence is not repeated is selected as the teacher.
[0084] f q ≤f teacher,p ≤f q+1 (13)
[0085] Through the above steps, different teaching groups and corresponding teachers will be obtained. Through calculation and comparison, the task allocation scheme with the median objective function value in the group is determined, which represents the average level of the group. So far, the multi-teacher grouping process is completed.
[0086] Step 3-4-1-2: Calculation of adaptive teaching factor:
[0087] During the teaching stage, all task allocation schemes in the population except the teacher must converge to the teacher, thereby completing the iterative update of the population. In order to better quantify the speed at which each learner converges to the teacher, the present invention proposes an adaptive teaching factor. In the traditional teaching and learning optimization algorithm, the teaching factor is randomly generated and can be 1 or 2. This method corresponds to the situation where the learner does not learn anything from the teacher and the learner learns everything from the teacher. In the present invention, it is further considered that each learner is affected by his own level and what he learns from the teacher is not the same. In order to evaluate the learner's ability and the gap with the group teacher, the adaptive teaching factor proposed by the present invention is shown in formula (14):
[0088]
[0089] Among them, f q is the qth learner x q The objective function value, f teacher,p is x q The objective function value of the teacher in the group. q Adaptive teaching factor is between [0,2]. Therefore, in the algorithm of the present invention, the teaching factor will change adaptively during the search process. The size of each group can be divided into three categories:
[0090] (1) Since these learners have a poor foundation, they can only try to get closer to the median of the group, that is, to perform a single-point crossover operation with the median of the group.
[0091] (2) These learners are at the average level within the group and are the main beneficiaries of the teaching phase, i.e., they have a single-point crossover operation with the teacher within the group.
[0092] (3) These learners had a good foundation and gained a lot from the teaching process, that is, they conducted two single-point cross operations with the teachers in the group.
[0093] Step 3-4-1-3: Single-point crossover operation:
[0094] After calculating the teaching factor of the learner in step 3-4-1-2, it is necessary to further use the single-point crossover operation to update the task allocation scheme. The specific process is: first, provide two parent schemes for crossover. Parent 1 is the learner to be updated, and Parent 2 is the target that Parent 1 wants to approach. In the present invention, Parent 2 is the teacher or median learner in the group. Then, select a column from Parent 1 as the starting column for the crossover operation, while not changing the target and task type information in the crossover column of Parent 1. Finally, the drones in the crossover column of Parent 1 that perform the same task type in Parent 2 replace the drones in the crossover column of Parent 1 to achieve the approach and convergence from Parent 1 to Parent 2. Replacing all drones from the starting column to the last column of Parent 1 in this way is a complete single-point crossover process.
[0095] Step 3-4-2: Learning phase update:
[0096] The update method of the learning phase of the present invention is not much different from the original algorithm. Compared with the original algorithm, in this phase, learners can freely learn from any learner, whether they are better or worse than themselves (the original algorithm only learns from individuals who are better than themselves). This modification can adapt to complex discrete solution spaces with a large number of local extreme values. Similarly, we use single-point crossover as an operation for a learner to approach another learner.
[0097] Note: During the population update process in steps 3-4-1 and 3-4-2, if the objective function value of the updated solution is better than the original solution, the original solution will be replaced by the updated solution.
[0098] Step 3-5: Self-motivated local search:
[0099] All the improvements mentioned in steps 3-4 are reflected in learning from others. In addition, the algorithm search strategy can also be improved from the perspective of self-motivation, thereby improving the algorithm's global optimization ability. The algorithm proposed in the present invention simulates the optimization process of discrete solutions by simulating the learner's self-motivation and self-improvement learning process. To this end, the algorithm introduces a self-motivation mechanism and uses mutation operations to implement this mechanism. This mechanism enhances the robustness of the algorithm, increases population diversity, and reduces the possibility of the algorithm falling into a local optimum. The mutation operation is an operation that can help a genetic algorithm jump out of the local optimum. In a genetic algorithm, chromosomes are usually mutated according to a given mutation probability to increase disturbances, thereby increasing the possibility of jumping out of the local optimum. There are two mutation methods designed by the present invention to jump out of the local optimum: one is to change the drone that performs the task, and the other is to change the order in which the drones perform tasks. The mutation probability of both is p. m .
[0100] Step 4, problem solving and evaluation:
[0101] Step 4-1: Parameter setting of adaptive self-motivated teaching and learning algorithm:
[0102] According to the characteristics of the problem, set appropriate solution parameters for the adaptive self-motivated teaching and learning algorithm designed in step 3, including the threat penalty coefficient α, the target threat radius R threat 、Maximum iterative algebra iter max , population size n, number of group teaching N teacher and the self-motivation mechanism mutation probability p m .
[0103] Step 4-2: Input the set problem and solve the problem:
[0104] Input the information of drone, carrier platform, environment, target and task. Obtain the optimal task allocation plan through the adaptive self-motivated teaching and learning algorithm designed in step 3, and visualize the task allocation plan.
[0105] Step 4-3: Evaluation of Adaptive Self-Motivated Teaching and Learning Algorithms:
[0106] Evaluate the performance indicators of the algorithm (such as task completion time, etc.) and solution efficiency (computation time).
[0107] The beneficial effects of the present invention are:
[0108] (1) By introducing a carrier platform to deliver and recover drones, the limitation of drone endurance on the collaborative execution of drone tasks is reduced. (2) At the same time, the heterogeneity of carrier platforms and drones is integrated into the collaborative task allocation combinatorial optimization model. (3) The no-fly zone and the risk domain of unattacked targets are considered in the collaborative task allocation problem to improve the safety of drone mission execution. BRIEF DESCRIPTION OF THE DRAWINGS
[0109] Figure 1 It is a flow chart of the steps of the present invention.
[0110] Figure 2 The present invention is based on the background description of the multi-UAV collaborative task allocation example diagram based on carrier platform delivery.
[0111] Figure 3 Schematic diagram of heterogeneous UAVs of the present invention; (a) is a reconnaissance aircraft, (b) is a bomber, and (c) is a fighter.
[0112] Figure 4 This is a diagram explaining the additional risk of approaching an unattacked target according to the present invention.
[0113] Figure 5 This is a schematic diagram of discrete decoding of the present invention.
[0114] Figure 6 This is a schematic diagram of the encoding initialization process of the present invention.
[0115] Figure 7 It is an illustration of the visual range estimation method of the present invention.
[0116] Figure 8 This is a diagram illustrating a single-point crossover according to the present invention.
[0117] Fig. 9 1 is an illustration of the local search (variation) of the present invention; (a) is variation mode 1, which changes the drone that performs the task; (b) is variation mode 2, which changes the order in which the drones perform the tasks.
[0118] Fig.10 This is a visual task allocation solution trajectory diagram in an example of the present invention. DETAILED DESCRIPTION
[0119] The present invention is further described below in conjunction with specific examples.
[0120] A method for collaborative task planning of heterogeneous UAVs in a heterogeneous carrier platform delivery mode comprises the following steps:
[0121] Step 1: Analyze the characteristics of the problem, as follows:
[0122] Table 2 UAV information
[0123]
[0124] Table 2 is the information of the drone, where the second column indicates the type of drone. R is a reconnaissance drone that can perform reconnaissance and assessment tasks, M is a missile-carrying drone that can perform attack tasks. C is a combat drone that can perform reconnaissance, attack and assessment tasks. In addition, Tables 3 and 4 are the carrier platform information and target information respectively. p represents the average speed of the carrier platform, and {1,2} means that the carrier platform is docked with UAV No. 1 and UAV No. 2.
[0125] Table 3 Carrier platform information
[0126]
[0127] Table 4 Target information
[0128] Target 1 2 3 4 5 6 7 8 Location (km) [25,36] [5,20] [47.5,30] [-2,41] [20,15] [5,66] [16,49] [15,5] Target 9 10 11 12 13 14 15 16 Location (km) [40,18] [47,52] [30,19.5] [30,2] [32,58] [-15,25] [35,37] [15,23]
[0129] In summary, we can know type p =2, N U =8,N T =16, N K = 20. In addition to the above problem parameters, the algorithm parameters are set as shown in Table 5.
[0130] Table 5 Algorithm parameter values
[0131]
[0132] Step 2: Construct a combined optimization model and complex network based on the information of the carrier platform, drone, target, etc.:
[0133] minF=F 1 +αF 2
[0134]
[0135] Step 3-1: Integrate input information and construct resource constraint sets through carrier platform and drone information:
[0136] Table 6 Resource constraint set
[0137]
[0138] According to the resource constraint set in Table 6, use the initialization encoding method to generate similar Figure 5 The 50 initial solutions are shown. Figure 6 A coding initialization example process is given. In this example, there are three drones participating in the task, and the resource constraint set is: drones with reconnaissance capabilities [1,3], drones with attack capabilities [2,3], and ammunition quantity [0,4,1]. According to the task requirements, drones with corresponding capabilities are selected to perform. For example, in the second column, the attack task of target 1 needs to be assigned to a drone, which can be randomly selected from the set of drones with attack capabilities. In this example, drone 3 is selected to perform the task, and the ammunition quantity of drone 3 is 1. After completing the task, the ammunition quantity is 0, and it does not have the ability to attack again, so the drone is removed from the set with attack capabilities. Similarly, the other tasks are assigned in sequence, and the initialization process of the scheme is completed.
[0139] Step 3-2: Calculate the objective function values of 50 solutions.
[0140] Step 3-2-1: Unlock the 50 generated initial solutions and turn them into feasible solutions.
[0141] Step 3-2-2: Use the visualization method to calculate the distance between all two points in the task allocation solution. Figure 7In the example, we need to reach point b from point a. By using the method described in the manual, we first consider only the first circular obstacle to obtain new starting points c and d, and then use them as starting points to consider the second polygonal obstacle to obtain new starting points e and f. When e is connected to b, it no longer passes through the obstacle and does not need to select a new starting point. When f is connected to b, it still needs to pass through the circular obstacle, so we continue to look for a new starting point g. Finally, we select the shorter of the two paths adfgb and aceb as the distance from a to b.
[0142] Step 3-2-3: Then combine the waiting circle trajectory adjustment method and equations (1), (2), (3), and (4) in step 2-2 to calculate the objective function value of the unlocked solution.
[0143] Step 3-3: Sort the objective function values and calculate the median and mean of the population objective function. Then calculate the current state of the population using the median and mean. If the median > mean, use the teaching phase update method of step 3-4-1, otherwise use the learning phase update method of step 3-4-2.
[0144] Step 3-4: Update the population according to the population status obtained in step 3-3. In actual operation, various limiting factors need to be considered. Figure 8 The example illustrates the crossover operation in detail: First, if the crossover column is an attack mission, the drones in the current crossover column are removed from the set of drones with remaining attack capabilities in parent 1. Second, all columns of parent 2 are traversed to obtain the set of all drones that perform attack missions in that parent, and then the intersection is taken. Finally, a drone is randomly selected from the intersection to replace the drone in the crossover column of parent 1.
[0145] Step 3-5: According to the probability of self-motivated mutation, two local search operations are performed randomly in sequence. Fig. 9 Give examples of two local search operations. Fig. 9 In (a), the drones performing the tasks in the third column are changed according to the resource constraint set. Fig. 9 In (b), the order of columns 4-6 is changed, thereby changing the order in which the drones perform tasks.
[0146] Step 4: Algebraic iter according to the maximum iteration max = 100, and execute steps 3-2 to 3-5 in a loop. Finally, the optimal task allocation scheme is output as shown in Table 7. In addition, according to Table 7, Fig.10 The visual task allocation solution trajectory diagram is shown.
[0147] Table 7 Task allocation scheme
[0148]
[0149] Finally, the algorithm of the present invention is used to calculate the three different groups of preliminary solutions of the above numerical cases 10 times each, and the average values are shown in Table 8. Fbest is the objective function value, F 1 is the task plan execution time, F 2 is the threat penalty for not attacking the target. The smaller F in the table 2 This shows that the optimal solution obtained by using the algorithm of the present invention is less subject to threat penalties, and the present invention ensures the rationality and security of the generated task allocation solution.
[0150] Table 8 Algorithm performance evaluation
[0151] index Fbest <![CDATA[F 1 ]]> <![CDATA[F 2 ]]> CPU Time result 1262.25 1187.95 74.30 10558.11s
[0152] The present invention relates to the problem of multi-UAV collaborative task allocation enhanced by a carrier platform. A collaborative task allocation style based on the delivery of UAVs by a carrier platform is proposed. The delivery and recovery process of the carrier platform for UAVs is taken into account, and the risk of UAVs that are not performing missions approaching but not hitting targets is considered. A multi-UAV collaborative task allocation model under the carrier platform delivery model is constructed. In order to solve the model, an adaptive self-motivated teaching and learning optimization algorithm is developed. The algorithm first adapts the continuous teaching and learning optimization algorithm to discrete problems, that is, introduces discrete coding update methods such as crossover and mutation, and uses these mechanisms to replace the original mathematical operations, thereby obtaining an optimization method for discretized problems. Then, multi-teacher group teaching, adaptive learning rate, population state judgment, and local search of self-motivated mechanisms are introduced. The introduction of the above mechanisms enables the algorithm to have stronger exploration capabilities and ensures the algorithm's optimization ability.
[0153] The above-mentioned examples only express the implementation modes of the present invention, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.
Claims
1. A method for collaborative mission planning of heterogeneous UAVs in a heterogeneous carrier platform delivery mode, characterized in that: The steps of the heterogeneous UAV collaborative task planning method in the heterogeneous carrier platform delivery mode are as follows: Step 1) Analyze the characteristics of the problem, including analysis of the launch platform, drone, environment, target, and mission intent. The details are as follows: Step 1-1: Analysis of carrier platform elements; Will Denoted as the airport set, where the superscripts of P are type1,...,type a Indicates the type of carrier platform P, a indicates the number of carrier platform types, and the subscript Respectively represent the corresponding type1,...,type a The number of different launch platforms; UAVs fly out from heterogeneous launch platforms to perform missions and return to the same platforms after completing their missions; Step 1-2: Drone element analysis: UAVs are heterogeneous drones with different functions, including reconnaissance drones S with only reconnaissance capabilities, bombing drones M with only attack capabilities, and combat drones C that perform reconnaissance and attack missions. Indicates that U i represents the i-th drone; N U Indicates the total number of drones; Step 1-3: Environmental Factor Analysis: Assume that there is a prohibited airspace on the battlefield, and simplify all no-fly zones into convex polygons and circles. UAVs must not appear in the prohibited airspace during flight. and collection Represents the circular obstacles and convex polygonal obstacle sets on the battlefield, where N c and N b are the total number of circular obstacles and convex polygonal obstacles respectively; the total set of obstacles is defined as: Simplify the flight process by treating drones as point masses in a battlefield environment; Steps 1-4: Target element analysis: Assume there are N T target, using a collection Indicates that T i Represents the target numbered i; for each target T i , it is necessary to perform classification tasks, attack tasks and evaluation tasks with priority restrictions, that is, the task set of each target is {C, A, V}, where C represents the classification task, A represents the attack task, and V represents the evaluation task; In addition, there is an area of 'risk caused by proximity to unattacked targets' around unattacked targets; Steps 1-5: Task Intent Analysis: Minimizing the execution time of the entire solution is used as one of the objective functions. A task allocation plan is given at the mission planning level to prevent the drone's route from passing through the threat area of the target that is not attacked. When the target is attacked, the airspace around the target is no longer considered a threat. A threat function is introduced as another objective function in the combinatorial optimization model. Step 2) Define the optimization variables, objective function, and constraints, and establish a combinatorial optimization model; the details are as follows: Step 2-1: Abstract the multi-UAV collaborative mission scenario into a network: The location points of the airport and the target are defined as nodes of the network, and K is used to represent the node set. Then E={(K i ,K j )|K i ,K j ∈K,i,j=0,1,…,N K ,i≠j} represents the set of edges, which together constitute the network CG={K,E}; where K i represents the i-th network node; K j Indicates the jth network node; i and j both represent the network node sequence number; N K Indicates the total number of network nodes; Step 2-2: Establish the objective function of the combined optimization model for multi-UAV collaborative task allocation through the network: The objective function includes the solution execution time evaluation function F1 and the penalty function term F2. The objective function of the combined optimization model is shown in formula (4), where α represents the penalty coefficient; minF=F1+αF2; (4) Step 2-3: Analyze the constraints of the combinatorial optimization model for multi-UAV collaborative task allocation, as follows: The constraints include multi-machine coordination constraints, ammunition load constraints, priority constraints, sortie recovery constraints, node flux conservation, and collision avoidance constraints; Step 3) Design an adaptive self-motivated teaching and learning algorithm: First, randomly initialize the discrete solution to obtain the initial population. Then, use the visibility graph method to obtain the distance between two points in the obstacle environment, and combine the track adjustment method and the logical unlocking method to calculate the population objective function value. Next, sort the population in ascending order according to the population objective function value, calculate the median and mean of the population objective function value, and determine the population update method. Finally, implement the self-motivation mechanism to further optimize the population. The details are as follows: Step 3-1: Generate solutions and initialize the population: For N T targets and N U In the case of a UAV, the generated solution contains 3 rows and 3N T Columns; the first row is the target, using numbers 1, 2, ..., N T Indicates the number of the corresponding target; the second line is the mission type of the target, using numbers 1, 2, and 3 to represent classification, attack, and evaluation type tasks respectively; the third line is the drone, using numbers 1, 2, ..., N U Indicates the number of the drone; each column is a task assignment, indicating that the drone is to perform the corresponding task of the target; During the initialization process, a resource constraint set is first constructed based on the input drone information. The resource constraint set includes a set of drones with reconnaissance capabilities, a set of drones with attack capabilities, and a set of remaining ammunition. Then, for each target's different mission type, a drone is randomly selected from the set of drones capable of performing that mission and assigned to that mission. Finally, all tasks are assigned to drones according to the above method. The drones are sorted by number to obtain a task allocation plan. The initial solution set of a given size generated by this method is called the initial population. Step 3-2: Objective function calculation: Step 3-2-1: Priority constraint processing; Step 3-2-2: Estimating the distance between two points using the visual graph method: After satisfying the priority constraints, the distance between two points in the obstacle environment is estimated by the visibility graph method, and the objective function value is calculated; Step 3-2-3: Calculation of objective function value: Calculate the objective function value of the unlocked solution according to step 2-2; Step 3-3: Design of population status determination mechanism: Sort the objective function values calculated in step 3-2, then calculate the median and mean of the objective function values of the population, compare the median and mean, and select different update methods based on the judgment results: if the median is greater than the mean, use the teaching phase population update method in step 3-4-1; if the median is less than the mean, use the learning phase population update method in step 3-4-2; Step 3-4: Design population update method: In one iteration, the teaching phase or the learning phase is selected according to the population state: Step 3-4-1: Teaching stage update; Step 3-4-2: Learning phase update; Steps 3-5: Self-motivated local search; Step 4) Problem Solving and Evaluation: First, set the parameters of the adaptive self-motivated teaching and learning algorithm. Then, use the adaptive self-motivated teaching and learning algorithm designed in step 3) to solve the problem until the maximum number of iterations is reached. The optimal solution is selected and evaluated as follows: Step 4-1: Parameter setting of the adaptive self-motivated teaching and learning algorithm: According to the characteristics of the problem, the solution parameters of the adaptive self-motivated teaching and learning algorithm designed in step 3 are set, including the threat penalty coefficient α, the target threat radius R threat , maximum iteration algebra iter max , population size n, number of group teaching N teacher and the self-motivation mechanism mutation probability p m ; Step 4-2: Input the set problem and solve the problem: Input drone information, carrier platform information, environment information, target and task information; obtain the optimal task allocation plan through the adaptive self-motivated teaching and learning algorithm designed in step 3, and visualize the task allocation plan.
2. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 1 is characterized in that: The step 2-2 is specifically as follows: The evaluation function F1 of the program execution time is shown in formula (1): in, u represents the drone number, and U represents the set of drone numbers; is a decision variable, if UAV U u From K i to K j Execute task type k, then k=0 means returning to the carrier platform, k=1, 2, 3 means executing tasks of type C, A and V respectively; otherwise, in It is a drone U u From K i to K j The estimated path length, Represents UAV u speed; The penalty function F2 is shown in formula (2): Among them, N T Indicates the number of targets; N U Indicates the number of drones; N K Indicates the number of nodes; l indicates the target number; is the target T l Give UAV U u From K i to K j The threat penalty brought by, at time t, the threat function is shown in formula (3); Where r represents T l to (K i ,K j ), R threat Represents the threat radius of the target. When considering threat penalties, only targets that have not been attacked are considered, and attacked targets are considered to have no threat. Therefore, the threat penalties of the same path at different times are different.
3. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 1 is characterized in that: The steps 2-3 are specifically as follows: (1) Multi-drone collaboration constraint: UAVs need to collaborate to perform all tasks, as expressed in Equation (5); Where l represents the target number; (2) Ammunition load constraint: The number of attack missions of a UAV cannot exceed its ammunition load, as shown in Equation (6); in, Represents UAV u ammunition capacity; Corresponding to formula (1) When k=2, the UAV U u From K i to K j Carry out attack missions; (3) Priority constraint: Each target contains three types of tasks: classification task C, attack task A, and evaluation task V. These three types of tasks for the same target are completed in sequence, as shown in Equation (7). (4) Dispatch and recovery constraints: In the multi-airport problem, any UAV that takes off from any carrier platform must eventually return to the departure carrier platform, as shown in Equation (8); (5) Node flux conservation: the number of drones entering each node is equal to the number of drones leaving the node, which can be expressed as equation (9); (6) Collision avoidance constraint: The flight path of the UAV cannot intersect with the no-fly zone, as shown in Equation (10); in, Represents node K i To node K j The estimated path of Φ′ is obtained by expanding Φ.
4. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 1, characterized in that: In step 3: In step 3-2-1, the priority constraint processing method is: Priority constraints are processed before calculating the objective function value. The waiting circle method is used to adjust the trajectory. When the priority constraints are not met, the UAV flies around a circular path with the minimum turning radius outside the target threat range, waiting for other UAVs to perform related pre-tasks. If a UAV is stuck in an infinite waiting state, the task allocation scheme is considered locked, and a logical unlocking method is used to unlock the task allocation scheme. In step 3-2-2, the distance between two points is estimated using the visual graph method as follows: Step 3-2-2-1: Single obstacle: When the drone trajectory between two points passes through a single convex polygon obstacle, the vertices of the convex polygon are connected to the starting point and the end point respectively; the shortest of all the lines connecting the starting point to the end point is the candidate path; if the trajectory passes through a circle, the shortest path is selected by finding the tangent line of the circle passing through the starting point and the end point; Step 3-2-2-2: Multiple obstacles: If there are multiple obstacles between two points, first consider only the obstacle closest to the starting point; then select the middle point of the candidate path as the new starting point, reconnect the starting point and the end point, and continuously update the starting point using the above method until there are no obstacles between the starting point and the end point. Then connect all points from the starting point to the end point and choose the shortest path that does not pass through obstacles.
5. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 1, characterized in that: In the steps 3-4: In the above step 3-4-1, the teaching phase update is specifically as follows: Step 3-4-1-1: Grouping by multiple teachers: An iteration-dependent multi-teacher mechanism is introduced in the population update of the teaching phase. This multi-teacher mechanism divides the entire class into different learner groups according to the objective function value at the beginning of each iteration. The best-performing learner in each learner group is selected as the teacher of that group. The initial number of teachers is predetermined and modified according to formula (11) throughout the iteration process. n t =ceil((1-I / I max )×N teacher ); (11) Among them, I represents the current iteration number, I max Indicates the maximum number of iterations, ceil(·) is the rounding up function, N teacher is the initial number of teachers, n t is the number of teachers in the current iteration; Sort the entire population in ascending order of objective function value; Determine the teachers of each group according to formula (12) and formula (13); f q ≤f teacher,p ≤f q+1 (13) Among them, f teacher,p represents the objective function value of the p-th teacher, f best and f worst Represent the best and worst objective function values in the group respectively; f q and f q+1 Represents the sorted objective function value, where f q ≤f q+1 If the objective function value of teacher p satisfies formula (13), then learner q is regarded as teacher p, and all learners between it and the next teacher belong to its learning group. If there are repeated teachers, the first learner whose sorting sequence is not repeated is selected as the teacher. Through the above steps, different teaching groups and corresponding teachers are obtained; through calculation and comparison, the task allocation scheme with the median objective function value within the group is determined, which represents the average level of the group, and the multi-teacher grouping process is completed; Step 3-4-1-2: Calculation of adaptive teaching factor: In order to evaluate the learner's ability and the gap with the group teacher, the proposed adaptive teaching factor is shown in formula (14): Among them, f q is the qth learner x q The objective function value, f teacher,p is x q The objective function value of the teacher in the group; the learner x q Adaptive teaching factor Between [0, 2], the teaching factor will change adaptively during the search process; Step 3-4-1-3: Single-point crossover operation: After calculating the learner's teaching factor in step 3-4-1-2, use the single-point crossover operation to update the task allocation plan.
6. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 5, characterized in that: In the steps 3-4-1-2, according to the adaptive teaching factor The size of each group is divided into three categories: (1) This type of learner has a poor foundation and performs a single-point crossover operation with the median of the group; (2) This type of learner is at the average level within the group and conducts a single-point crossover operation with the teacher in the group; (3) This type of learners has a good foundation and conducts two single-point cross operations with the teacher in the group.
7. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 5, characterized in that: In the step 3-4-1-3, the specific process of the single-point crossover operation is as follows: first, two parent schemes are provided for crossover, parent 1 is the learner to be updated, parent 2 is the target to be approached by parent 1, and parent 2 is the teacher or median learner in the group; then, a column is selected from parent 1 as the starting column for the crossover operation, while the target and task type information in the crossover column of parent 1 is not changed; finally, the drones performing the same task type in parent 2 replace the drones in the crossover column of parent 1 to achieve approach and convergence from parent 1 to parent 2; replacing all drones from the starting column to the last column of parent 1 in this way is a complete single-point crossover process.
8. The method for planning collaborative tasks of heterogeneous UAVs in a heterogeneous carrier platform delivery mode according to claim 1, characterized in that: In the steps 3-4, a self-motivation mechanism can also be introduced through mutation operation.
