Unmanned aerial vehicle cluster ground attack task allocation method and system based on improved particle swarm algorithm

Abstract

The invention discloses an unmanned aerial vehicle cluster ground attack task allocation method and system based on an improved particle swarm algorithm, belongs to the technical field of intelligent optimization algorithms, and aims at solving the problem of unmanned aerial vehicle cluster ground attack task allocation. And analyzing two types of constraints including the target type constraint and the unmanned aerial vehicle type constraint, and establishing a mathematical model for maximizing the attack cost-effectiveness ratio. The method comprises the following specific steps: step 1, establishing a target function for distributing and establishing a target attack revenue and a target attack cost to the ground attack task, wherein the target function is used for maximizing the target attack revenue and minimizing the target attack cost; 2, analyzing constraint conditions of ground attack task allocation; 3, integrating the target function and the constraint condition of the ground attack task allocation, and establishing a mathematical model of the ground attack task allocation; and 4, solving the mathematical model established in the step 3 by using an improved particle swarm algorithm to obtain a final distribution result.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent optimization algorithms, and in particular relates to a method and system for allocating ground attack tasks of a cluster of unmanned aerial vehicles based on an improved particle swarm algorithm. Background Art

[0002] With the rapid development of network technology, control technology and artificial intelligence technology, major armies around the world have conducted a lot of basic theoretical research on the clustered use of drones, and have used this to promote the implementation of relevant combat concepts. Drone swarm combat is supported by communication networks, and a large number of low-cost, large-scale drones with certain autonomous capabilities are used to build a combat system in a self-organizing form. It uses the characteristics of unified spatial relationship density and dispersion, unified combat action planning and suddenness, and unified unidirectional and multi-directional attack directions to carry out continuous and parallel reconnaissance and detection, electronic interference, and firepower strikes on enemy targets.

[0003] Drone swarm combat has its own specific biological principles. In fact, swarm behavior is a universal behavior in the biological world. Wildebeests always migrate in groups, ants go out in droves to forage, and fish patrol in groups. In these swarms, individual perception / action capabilities are relatively simple, but through complex collaborative behaviors, the overall capabilities can emerge. Drone swarms draw on this "autonomy + collaboration" behavior mode, and through complex interactive behaviors between a large number of simple individuals, they achieve a combat effect of winning by quantity and collaborative energy gathering.

[0004] Drone swarms can perform a variety of tasks, generally speaking, including offensive and defensive tasks. Among them, offensive tasks mainly include seizing air superiority and ground penetration attacks; defensive tasks include defensive air interception, key point air defense, etc.; in addition, drone swarms can also perform support and guarantee tasks such as air early warning detection, electronic interference, and communication relay. In these tasks, the requirements for the autonomy level of individual drones are different. Generally speaking, the requirements for the autonomy level of air combat missions, ground attack missions, and support and guarantee missions decrease in turn. Due to the differences in the autonomy level of drones, at present, drone swarms are less able to carry out air combat missions, and offensive missions are mainly to perform ground / sea attack missions.

[0005] When drone swarms are performing ground attack missions, mission planning is the key to achieving decision-making and control of drone swarms. Mission planning for drone swarms’ ground attack includes pre-planning and online re-planning. Pre-planning is a plan made in advance before the mission execution begins based on the analysis and judgment of the combat situation. Online re-planning is when the combat situation changes due to unexpected events on the battlefield during the mission execution. The drone swarm needs to reconstruct and adjust the plan based on the changed situation. Mission planning for drone swarms’ ground attack is a whole that includes task allocation, route planning, communication link planning, and sensor planning. Task allocation is the key and difficult part.

[0006] There have been many studies on the problem of UAV swarm ground attack task allocation, mainly focusing on modeling and solving. Generally speaking, the classic models of UAV swarm ground attack task allocation mainly include the Multiple Travelling Salesman Problem (MTSP), Mixed Integer Linear Programming (MILP), Cooperative Multiple Task Assignment Problem (CMTAP), Hybrid Binary Programming (HBP), etc. The UAV swarm ground attack task allocation model is established as a classic model, and the corresponding fast heuristic method can be used to solve it, which can obtain a better allocation result at a lower time cost. The solution to the UAV swarm ground attack task allocation model mainly includes optimization methods, intelligent optimization methods and market mechanism methods. Among them, the optimization methods include the Hungarian algorithm, branch and bound algorithm and Lagrangian relaxation method. These methods have high requirements on the model and can only solve specific forms of models. Intelligent optimization methods are various algorithms designed to simulate human intelligence, biological group activity characteristics or natural laws, including genetic algorithms (GA), ant colony optimization (ACO), artificial bee colony algorithms (ABC) and wolf pack algorithms (WPA). The advantage of these algorithms is that they do not require high model forms, but the disadvantage is that the timeliness and stability of the algorithms are poor. Market mechanism methods mainly include contract network protocols and auction algorithms. These algorithms essentially use a distributed solution architecture. The algorithm calculation amount is usually not large, but it relies on frequent information interaction and negotiation decision-making. Moreover, since the generated compromise solution is a compromise solution involving all parties, the optimality of the solution may be affected to a certain extent.

[0007] In short, as a hot topic in the research of military operations and UAV combat applications, the allocation of ground attack tasks for UAV swarms is also an issue that urgently needs to be resolved. Summary of the invention

[0008] In order to solve the problem of UAV cluster ground attack task allocation, the present invention starts from the two aspects of attack benefit and attack cost, considers the constraints such as target coverage, target damage effect, UAV attack threshold and UAV range in ground attack to establish a mathematical model, and adopts the improved particle swarm optimization (IPSO) including encoding and decoding rules, constraint processing rules and iterative evolution rules to solve the model, and achieves better solution effect.

[0009] In a first aspect, the present invention provides a method for optimizing blasting design parameters based on deep learning and deep neural networks, the optimization method comprising the following steps:

[0010] Step 1: For ground attack task allocation, establish an objective function to maximize the target attack benefit and minimize the target attack cost;

[0011] Step 2: Analyze the constraints of ground attack task allocation;

[0012] Step 3: Integrate the objective function and constraints of ground attack task allocation to establish a mathematical model for ground attack task allocation;

[0013] Step 4: Use the improved particle swarm algorithm to solve the mathematical model established in step 3 to obtain the final allocation result.

[0014] Furthermore, the objective function established in step 1 is as follows:

[0015] O=S / Q (6)

[0016] Among them, S represents the target attack benefit, and its expression is as follows:

[0017]

[0018] In the above, x mn represents the decision variable, d mn Indicates drone U m To target T n The probability of damage;

[0019] Q represents the target attack cost, which is expressed as follows:

[0020]

[0021] In the above formula, For UAV m The value of the variable related to the voyage cost, vU m is the platform value of the mth UAV, f(e1,e2) is a function of the two variables e1 and e2, and f(e1,e2)=e1·e2 is defined. for

[0022]

[0023] Among them, c m For UAV m The actual range of executing the target attack mission, cUm is the maximum range of the mth UAV.

[0024] Furthermore, the specific steps of step 2 include:

[0025] Step 201: Target coverage is approximately, for any target T n , there should be at least one drone attacking it, that is:

[0026]

[0027] Step 202: Target damage effect constraint requires that the damage effect of all targets after being attacked reaches a certain threshold, that is:

[0028]

[0029] Step 203: The drone attack threshold constraint has a certain upper limit on the number of target attacks, namely:

[0030]

[0031] In the above formula, hT n The threshold of damage effect that needs to be achieved for the nth target;

[0032] Step 204: The actual range of the drone during the target attack mission cannot exceed the maximum range of the drone, that is,

[0033] c m ≤cU m , 1≤m≤M (10).

[0034] Furthermore, the mathematical model of ground attack distribution established by integrating the objective function and constraints in step 3 is:

[0035]

[0036] In formula (11), the first and second constraints are target constraints, and the third and fourth constraints are platform constraints.

[0037] Furthermore, in step 4, according to the characteristics of the UAV cluster ground attack task allocation, the encoding and decoding rules, constraint processing rules and particle updating rules of the particle swarm algorithm are designed to solve the mathematical model.

[0038] The specific steps of solving the problem using the improved particle swarm algorithm are as follows:

[0039] Step 401: Initialize the particle swarm. Each particle is an M*N matrix, where the rows and columns represent the drone and the target respectively, and the elements in the matrix are real numbers that meet the coding rules, i.e. (0, 1.2].

[0040] Step 402 retains the current individual optimal and global optimal solutions. According to the decoding rules and constraint processing rules, feasible solutions are obtained and sorted. The best solution among all feasible solutions is the global optimal solution, and all feasible solutions correspond to the individual optimal solutions of all particles under the current number of iterations.

[0041] Step 403: Particle swarm evolution. Entering the next iteration, the update formula of the position matrix X and the velocity matrix U of each particle is updated using the proposed particle update rule.

[0042] Step 404: Update individual optimal solutions and global optimal solutions. Determine whether the feasible solution corresponding to each particle under the current number of iterations is better than the corresponding individual optimal solution. If so, update it. Determine whether the optimal solution corresponding to the feasible solution of all particles under the current number of iterations is better than the global optimal solution. If so, update it.

[0043] Furthermore, when the particle swarm algorithm is used to solve the mathematical model, the encoding and decoding rules use real number encoding, using the matrix G = (g mn ) M×N represents a population, where g mn The value range is (0,1.2], and the decoding method mainly adopts the rounding down method, that is, for the 1st to Nth columns of the matrix G, Then let x mn =1.

[0044] Furthermore, the decision matrix obtained after encoding and decoding needs to be processed with corresponding constraints. The specific operations are as follows:

[0045] (1) Determine whether all drones meet the constraint conditions and record the serial numbers of all drones that do not meet the constraint conditions;

[0046] (2) The drones that do not meet the constraints are processed in order from low to high, and the expected attack targets of the drone with the current sequence number are judged and cancelled according to the execution order until the third and fourth constraints of the mathematical model are met;

[0047] (3) Determine whether all targets satisfy the first and second constraints of the mathematical model, and record the serial numbers of all targets that do not meet the constraints;

[0048] (4) The targets that do not meet the constraints are processed in order from low to high, and the current target is judged from top to bottom to add drones for execution until the first and second constraints of the mathematical model are met; if the third or fourth constraint of the mathematical model is no longer met after adding a certain drone, the current drone is not added, and the subsequent drones are judged;

[0049] After the decision matrix transformation, the penalty function method is further adopted. If the target number of a solution violating the constraint condition is δ (0≤δ≤N), the corresponding objective function is divided by δ+1.

[0050] Furthermore, the movement of particles mainly depends on the current position and velocity of the particles. The update formulas of the position matrix X and velocity matrix U are as follows:

[0051]

[0052] Among them, X and U are the position and velocity matrices before updating, X' and U' are the position and velocity matrices after updating, X G is the global optimal position under the current number of iterations, X L is the local optimal position under the current number of iterations, w is the inertia coefficient, c1 and c2 are learning factors, and rand1 and rand2 are random numbers in the range of (0,1).

[0053] In a first aspect, the present invention provides a UAV cluster ground attack task allocation system based on an improved particle swarm algorithm, the system comprising:

[0054] The first building block is to establish the objective function of the UAV swarm ground attack mission based on battlefield environment information;

[0055] The second building block analyzes the constraints of target coverage, target damage effect, drone attack threshold and drone range, and establishes a mathematical model for the allocation of drone swarm ground attack tasks in combination with the objective function;

[0056] The processing module is used to solve the mathematical model using an improved particle swarm algorithm.

[0057] Compared with the prior art, the present invention has the following advantages:

[0058] 1. The present invention comprehensively considers the target attack benefit and target attack cost, analyzes two types of constraints, namely target constraints and drone constraints, and establishes a mathematical model that maximizes the attack cost-effectiveness ratio. The improved particle swarm algorithm is mainly used to solve the model. According to the discrete and multi-constraint characteristics of the model, the algorithm's encoding and decoding rules and constraint processing rules are designed respectively. Multiple groups of simulation experiments show that the proposed algorithm can effectively solve the ground attack task allocation model under multi-constraint conditions.

[0059] 2. Compared with the firefly algorithm and the artificial bee colony algorithm, the improved particle swarm algorithm designed by the present invention has a stronger optimization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is the iterative evolution of the IPSO algorithm.

[0061] Figure 2 Iterative evolution of each comparison algorithm.

[0062] Figure 3 Iterative evolution of each comparison algorithm. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0064] This embodiment specifically provides a method for allocating ground attack tasks of a drone cluster based on an improved particle swarm algorithm, which is introduced in detail from the following aspects.

[0065] 1. Definition of Task Allocation Problem Elements and Their Attributes

[0066] In the problem of UAV swarm ground attack task reallocation, the problem elements mainly include two categories: UAVs and targets. To simplify the problem, it is assumed that all UAVs start from a certain initial position and attack ground targets. The platform value and task execution capability are different, and the change of the altitude layer of the UAV is not considered, that is, the UAV always moves at the same altitude layer. It is assumed that the ground targets are all fixed targets, mainly including enemy command posts, important roads and bridges, radar positions and missile launch positions, and after preliminary reconnaissance and detection, we have mastered the relevant target information.

[0067] (1) The set of UAVs is UAV = {U1, U2, …, U M}, where M is the number of drones. m (1≤m≤M), let its initial position coordinates be (xU m ,yUm ), the platform value is vU m , the task execution threshold is hU m , the longest range is cU m .

[0068] (2) The target set is Target = {T1, T2, ..., T N}, where N is the number of targets. For target T n (1≤n≤N), let its position coordinates be (xT n ,yT n ), the target value is vT n , the target expected damage effect threshold is hT n .

[0069] UAV m To target T n The probability of damage is d mn , the probability of all drones damaging all targets forms a matrix D = (d mn ) M×N ; Target T n UAV m The threat size is r nm , the threat size of all targets to all drones constitutes a matrix R = (r nm ) N×M .

[0070] The decision matrix of the UAV swarm ground attack task allocation model is X = (x mn ) M×N , the decision variable x mn Defined as

[0071]

[0072] 2. Modeling of ground attack task allocation problem

[0073] The ground attack task allocation model mainly includes two aspects: constraints and objective functions. First, the objective function of the task allocation model is analyzed. The principle of any type of combat action is to kill the enemy to the greatest extent while preserving oneself as much as possible. Therefore, the objective function of ground attack task allocation mainly includes two aspects: one is to maximize the target attack benefit, and the other is to minimize the target attack cost.

[0074] (1) Target attack benefit. The target attack benefit is related to the target value and the probability of damage to the target after being attacked. The calculation formula for the target attack benefit is defined as:

[0075]

[0076] in, Target T n The probability of being damaged by one or more drones is defined as The calculation formula is:

[0077]

[0078] (3) Target attack cost. The target attack cost is related to the value of the UAV platform, the threat level of the target to the UAV, the actual range of the UAV mission, and the maximum range of the UAV. The calculation formula for the target attack cost Q is defined as:

[0079]

[0080] in, For UAV m The value of the variable related to the voyage cost, vU m is the platform value of the mth UAV, f(e1,e2) is a function of the two variables e1 and e2, and f(e1,e2)=e1·e2 is defined. for

[0081]

[0082] Among them, c m For UAV m The actual range of executing the target attack mission, cUm is the maximum range of the mth UAV.

[0083] In summary, the objective function O of the ground attack task allocation model is defined as:

[0084] O=S / Q (6)

[0085] Then, the constraints of the task allocation model are analyzed. The design of the constraints mainly considers the following aspects: the target must be attacked, the target of the drone attack has a certain upper limit, and the target attack must achieve a certain damage effect, as follows:

[0086] (1) Target coverage constraint: To ensure that all discovered targets are attacked, for any target T n , there should be at least one drone attacking it, that is:

[0087]

[0088] (2) Target damage effect constraint: To ensure the purpose of damaging all targets, the damage effect of all targets after being attacked must reach a certain threshold, that is:

[0089]

[0090] In the above formula, hT nThe damage threshold that needs to be reached for the nth target.

[0091] (3) Drone attack threshold constraint: Due to the limitation of payload, drones cannot attack an unlimited number of targets. There is a certain upper limit on the number of targets to be attacked, namely:

[0092]

[0093] (4) UAV range constraint: Due to the limitation of fuel, the actual range of the UAV in performing the target attack mission cannot exceed the maximum range of the UAV, that is:

[0094] c m ≤cU m , 1≤m≤M (10).

[0095] The mathematical model of ground attack task allocation is established by integrating the objective function and constraints of the model:

[0096]

[0097] In formula (11), the first and second constraints are target constraints, and the third and fourth constraints are platform constraints.

[0098] 3. Ground attack task allocation based on improved particle swarm algorithm

[0099] In order to solve the ground attack task allocation model shown in formula (11), the IPSO algorithm is mainly used. The particle swarm algorithm is a global optimization algorithm designed to simulate the foraging behavior of biological groups. It has the advantages of fewer parameters, simple implementation, and strong optimization ability. The particle swarm algorithm is mainly used to solve continuous domain optimization problems, but it can also be discretized to solve discrete domain optimization problems. According to the characteristics of the problem of ground attack task allocation of drone clusters, this paper designs the algorithm's encoding and decoding rules, constraint processing rules, and particle update rules to solve formula (11).

[0100] 3.1 Algorithm encoding and decoding rules

[0101] When using the IPSO algorithm to solve the problem shown in formula (11), the design of the encoding and decoding rules is very important. The encoding rules of the algorithm emphasize simplicity and efficiency, that is, using less storage space to effectively represent the decision variables of the problem, and the encoding can adapt to the algorithm's iterative operators without more conversion operations; the decoding rules of the algorithm emphasize that the mapping to the problem solution space is unique and the mapping process is simple, which can ensure that the iterative operators in the encoding space are independent of the solution space. According to the constraints and other characteristics of the model, the encoding method used in this paper is real number encoding, using the matrix G = (g mn ) M×Nrepresents a population, where g mn The value range is (0,1.2], and the decoding method mainly adopts the rounding down method, that is, for the 1st to Nth columns of the matrix G, Then let x mn =1.

[0102] Taking M=4 and N=5 as an example, a code G is randomly generated, and the corresponding solution X obtained by using the above decoding rule is as follows.

[0103]

[0104] According to the above decoding rules, the population can be decoded into a decision matrix. Such a decision matrix may not satisfy the constraints in equation (11) and needs to be processed. In fact, the calculation of the UAV range will also be affected by the execution order of the attack task. The decimal part of the elements with values ​​greater than or equal to 1 in each row is sorted in ascending order to determine the execution order of the attack task. Taking the above G as an example, the second row represents the second UAV U2, and the decimal parts of the elements greater than or equal to 1 are 0.1441, 0.1896 and 0.1001, corresponding to T1, T4 and T5 respectively. Since 0.1001<0.1441<0.1896, the execution order of U2's attack task is T5→T1→T4.

[0105] 3.2 Algorithm Constraint Processing Rules

[0106] After decoding, the decision matrix is ​​obtained, but it cannot guarantee that the constraints in formula (11) are satisfied, and corresponding constraint processing is required. For multi-constraint optimization problems, the constraint processing rules mainly include penalty function method, decision matrix transformation method, multi-objective optimization method, feasibility rule method and ε constraint method, which need to be selected according to the characteristics of the problem to be solved. Considering that the feasible solution accounts for a relatively small proportion of the solution space, this paper mainly adopts a method combining the decision matrix transformation method and the penalty function method.

[0107] The decision matrix transformation method is to change the values ​​of decision variables in the decision matrix so that the decision matrix can meet all constraints. In order to meet the UAV class constraints and target class constraints in formula (11), the following constraint processing operations are mainly performed.

[0108] (1) Determine whether all drones meet the constraint conditions and record the serial numbers of all drones that do not meet the constraint conditions;

[0109] (2) The drones that do not meet the constraints are processed in order from low to high, and the expected attack targets of the drone with the current sequence number are judged and cancelled according to the execution order until the third and fourth constraints of the mathematical model are met;

[0110] (3) Determine whether all targets satisfy the first and second constraints of the mathematical model, and record the serial numbers of all targets that do not meet the constraints;

[0111] (4) The targets that do not meet the constraints are processed in order from low to high, and the current target is judged from top to bottom to add drones for execution until the first and second constraints of the mathematical model are met; if the third or fourth constraint of the mathematical model is no longer met after adding a certain drone, the current drone is not added, but the subsequent drones are judged.

[0112] From the above steps, it can be seen that the decision matrix transformation method can ensure that the UAV class constraints are fully satisfied, but it cannot ensure that the target class constraints are fully satisfied. Therefore, after the decision matrix transformation, the penalty function method is further adopted. If the number of targets that violate the constraints of a solution is δ (0≤δ≤N), then the corresponding objective function is divided by δ+1.

[0113] 3.3 Particle Update Rules

[0114] In the IPSO algorithm, particle movement mainly depends on the current position and velocity of the particle. The update formula of the position matrix and velocity matrix is ​​shown in formula (12).

[0115]

[0116] Among them, X and U are the position and velocity matrices before updating, X' and U' are the position and velocity matrices after updating, X G is the global optimal position under the current number of iterations, X L is the local optimal position under the current number of iterations, w is the inertia coefficient, c1 and c2 are learning factors, and rand1 and rand2 are random numbers in the range of (0,1).

[0117] The specific steps of solving the problem using the improved particle swarm algorithm are as follows:

[0118] Step 401: Initialize the particle swarm. Each particle is an M*N matrix, where the rows and columns represent drones and targets respectively, and the elements in the matrix are real numbers that meet the coding rules, i.e. (0, 1.2].

[0119] Step 402: Keep the current individual optimal and global optimal solutions. Obtain feasible solutions according to the decoding rules and constraint processing rules, and sort the feasible solutions. The best solution among all feasible solutions is the global optimal solution, and all feasible solutions correspond to the individual optimal solutions of all particles under the current number of iterations.

[0120] Step 403: Particle swarm evolution. Entering the next round of iteration, the update formula of the position matrix X and the velocity matrix U of each particle is updated using the proposed particle update rule.

[0121] Step 404: Update individual optimal solutions and global optimal solutions. Determine whether the feasible solution corresponding to each particle under the current number of iterations is better than the corresponding individual optimal solution. If so, update it; determine whether the optimal solution of the feasible solutions corresponding to all particles under the current number of iterations is better than the global optimal solution. If so, update it.

[0122] Simulation Results Analysis

[0123] In order to verify the effectiveness and superiority of the IPSO algorithm in solving the model shown in formula (11), relevant simulation comparison experiments were carried out using MATLAB R2021a on a computer equipped with a core frequency of 2.27GHz processor. Among them, simulation experiment 1 verifies the effectiveness of the IPSO algorithm, that is, the IPSO algorithm can solve a feasible solution that satisfies all constraints, and verifies the superiority of the IPSO algorithm over the Glowworm Swarm Optimization (GSO) and the Artificial Bee Colony (ABC) algorithm in a set of typical scenarios; simulation experiment 2 verifies the superiority of the IPSO algorithm over the GSO algorithm and the ABC algorithm in random scenarios.

[0124] In terms of scene setting, the number of drones in the drone cluster is 30, the initial positions of the drones are randomly distributed in a 100*100 square area, the drone value is a random number in the range of [0.1, 0.7], the threshold for the drone to perform ground attack tasks is a random integer in [3, 6], and the maximum range of the drone is 300, 400, 500, 600 or 700; the number of various targets to be attacked is 40, the target positions are randomly distributed in a 100*100 square area, the target value is a random number in the range of [0.3, 0.9], and the expected probability of target damage is a random number in the range of [0.5, 0.8]; the probability of drone damage to the target is a random number in the range of [0.2, 0.9], and the threat size of the target to the drone is a random number in the range of [0.1, 0.8].

[0125] Simulation experiment 1: The scenario is set as a set of typical scenarios under the above values, as shown in Table 1, which are the initial position, value, threshold for executing ground attack missions and maximum range of the drone; as shown in Table 2, which are the target position, value and expected damage probability. In terms of IPSO algorithm values, the population size is set to 50 and the number of iterations is set to 200.

[0126] Table 1 UAV attributes

[0127]

[0128]

[0129] Table 2 Target attribute table

[0130]

[0131] Under the above parameter input, the IPSO algorithm is used to solve equation (11). Figure 1 As shown in Figure 1, it is the iterative evolution of the IPSO algorithm.

[0132] As shown in Table 3, the final distribution of UAV ground attack tasks solved by the IPSO algorithm is shown in Table 3. The actual damage probabilities of all targets are 0.85, 0.88, 0.86, 0.64, 0.88, 0.87, 0.87, 0.89, 0.88, 0.88, 0.58, 0.84, 0.82, 0.90, 0.85, 0.88, 0.90, 0.85, 0.89, 0.89, 0.88, 0.86, 0.87, 0.90, 0.89, 0.82, 0.90, 0.88, 0.80, 0.88, 0.88, 0.74, 0.83, 0.87, 0.63, 0.81, 0.88, and 0.90, all of which reach the expected damage probability and meet the target class constraints. The actual ranges of all drones are 24.91, 74.35, 76.33, 0, 35.10, 47.99, 67.83, 0, 92.98, 106.04, 74.11, 66.51, 83.64, 13.42, 84.68, 119.32, 13.51, 7.49, 0, 99.27, 41.40, 176.94, 35.35, 63.14, 85.63, 202.21, 85.70, 89.93, 49.74, and 158.44, which are all less than their maximum ranges, and the actual number of missions performed is less than their respective mission thresholds, which meets the drone category constraints.

[0133] Table 3. Results of task allocation generated by IPSO algorithm

[0134]

[0135] The scenario is still set as a set of typical scenarios under the above values, and the number of iterations of the IPSO algorithm, GSO algorithm, and ABC algorithm is set to 200 times. Figure 2 As shown in Figure 1, iterative evolution of the three algorithms: PSO algorithm, GSO algorithm and ABC algorithm. Figure 2It can be seen that in this typical scenario, the optimization ability of the PSO algorithm is stronger than that of the GSO algorithm and the ABC algorithm, and the algorithm convergence is stronger than that of the ABC algorithm.

[0136] Simulation experiment 2: The scenario is set as a set of random scenarios under the above values, all parameters are randomly set, and the IPSO algorithm, GSO algorithm and ABC algorithm are run 20 times respectively. Figure 3 As shown in Figure 2, it is a box-and-whisker plot of the objective function value of each comparison algorithm after running 20 times.

[0137] Among them, the mean value of the objective function value of the IPSO algorithm is 18.2212, the median value of the objective function value is 18.5020, and the variance value of the objective function value is 3.1221; the mean value of the objective function value of the GSO algorithm is 14.9028, the median value of the objective function value is 14.7334, and the variance value of the objective function value is 1.7668; the mean value of the objective function value of the ABC algorithm is 16.2795, the median value of the objective function value is 16.0471, and the variance value of the objective function value is 3.4773.

[0138] It can be seen that the IPSO algorithm has the strongest optimization ability, but its stability is inferior to the GSO algorithm and better than the ABC algorithm.

[0139] Conclusion: Aiming at the problem of allocating ground attack tasks of drone clusters, this paper, based on the analysis of existing research results, comprehensively considers the target attack benefits and target attack costs, and analyzes two types of constraints: target constraints and drone constraints, to establish a mathematical model that maximizes the attack cost-effectiveness ratio. The improved particle swarm algorithm is mainly used to solve the model. According to the discrete and multi-constraint characteristics of the model, the encoding and decoding rules and constraint processing rules of the algorithm are designed respectively. Multiple groups of simulation experiments show that the proposed algorithm can effectively solve the ground attack task allocation model under multi-constraint conditions, and compared with the firefly algorithm and the artificial bee colony algorithm, the improved particle swarm algorithm has a stronger optimization ability and its stability is between the two.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for allocating ground attack tasks of drone clusters based on an improved particle swarm algorithm, characterized in that: The allocation method includes the following steps: Step 1: For ground attack task allocation, establish an objective function to maximize the target attack benefit and minimize the target attack cost; Step 2: Analyze the constraints of ground attack task allocation; Step 3: Integrate the objective function and constraints of ground attack task allocation to establish a mathematical model for ground attack task allocation; Step 4: Use the improved particle swarm algorithm to solve the mathematical model established in step 3 to obtain the final allocation result.

2. According to the method of allocating UAV cluster ground attack tasks based on improved particle swarm algorithm in claim 1, it is characterized in that: The objective function established in step 1 is as follows: O=S / Q (6) Among them, S represents the target attack benefit, and its expression is as follows: In the above, x mn represents the decision variable, d mn Indicates drone U m To target T n The probability of damage; Q represents the target attack cost, which is expressed as follows: In the above formula, For UAV m The value of the variable related to the voyage cost, vU m is the platform value of the mth UAV, f(e1,e2) is a function of the two variables e1 and e2, and f(e1,e2)=e1·e2 is defined. for Among them, c m For UAV m The actual range of executing the target attack mission, cUm is the maximum range of the mth UAV.

3. The method for allocating ground attack tasks of drone clusters based on improved particle swarm algorithm according to claim 2 is characterized in that: The specific steps of step 2 include: Step 201: Target coverage is approximately, for any target T n , there should be at least one drone attacking it, that is: Step 202: Target damage effect constraint requires that the damage effect of all targets after being attacked reaches a certain threshold, that is: In the above formula, hT n The threshold of damage effect that needs to be achieved for the nth target; Step 203: The drone attack threshold constraint has a certain upper limit on the number of target attacks, namely: Step 204: The actual range of the drone during the target attack mission cannot exceed the maximum range of the drone, that is, c m ≤cU m ,1≤m≤M (10)。 4. The method for allocating ground attack tasks of drone clusters based on improved particle swarm algorithm according to claim 3 is characterized in that: In step 3, the mathematical model of ground attack distribution is established by integrating the objective function and constraints: max O=S / Q In formula (11), the first and second constraints are target constraints, and the third and fourth constraints are platform constraints.

5. The method for allocating ground attack tasks of drone clusters based on improved particle swarm algorithm according to claim 4 is characterized in that: In step 4, according to the characteristics of the UAV cluster ground attack task allocation, the encoding and decoding rules, constraint processing rules and particle updating rules of the particle swarm algorithm are designed to solve the mathematical model.

6. The method for allocating ground attack tasks of drone clusters based on improved particle swarm algorithm according to claim 5 is characterized in that: When the particle swarm algorithm is used to solve the mathematical model, the encoding and decoding rules use real number encoding, using the matrix G = (g mn ) M×N represents a population, where g mn The value range is (0,1.2], and the decoding method mainly adopts the rounding down method, that is, for the 1st to Nth columns of the matrix G, Then let x mn =1.

7. The method for allocating ground attack tasks of drone clusters based on improved particle swarm algorithm according to claim 6 is characterized in that: The decision matrix obtained after encoding and decoding needs to be processed with corresponding constraints. The specific operations are as follows: (1) Determine whether all drones meet the constraint conditions and record the serial numbers of all drones that do not meet the constraint conditions; (2) The drones that do not meet the constraints are processed in order from low to high, and the expected attack targets of the drone with the current sequence number are judged and cancelled according to the execution order until the third and fourth constraints of the mathematical model are met; (3) Determine whether all targets satisfy the first and second constraints of the mathematical model, and record the serial numbers of all targets that do not meet the constraints; (4) Process the targets that do not meet the constraints in order from low to high, and judge the current target from top to bottom to increase the number of drones to execute until the first and second constraints of the mathematical model are met; If the third or fourth constraint of the mathematical model is no longer satisfied after adding a certain drone, the current drone will not be added, but the subsequent drones will continue to be judged; After the decision matrix transformation, the penalty function method is further adopted. If the target number of a solution violating the constraint condition is δ (0≤δ≤N), the corresponding objective function is divided by δ+1.

8. The method for allocating ground attack tasks of drone clusters based on improved particle swarm algorithm according to claim 6 is characterized in that: The movement of particles mainly depends on the current position and velocity of the particles. The update formulas of the position matrix X and velocity matrix U are as follows: Among them, X and U are the position and velocity matrices before updating, X' and U' are the position and velocity matrices after updating, X G is the global optimal position under the current number of iterations, X L is the local optimal position under the current number of iterations, w is the inertia coefficient, c1 and c2 are learning factors, and rand1 and rand2 are random numbers in the range of (0,1).

9. A system for allocating ground attack tasks of drone clusters based on an improved particle swarm algorithm according to any one of claims 1 to 8, the system comprising: The first building block is to establish the objective function of the UAV swarm ground attack mission based on battlefield environment information; The second building block analyzes the constraints of target coverage, target damage effect, drone attack threshold and drone range, and establishes a mathematical model for the allocation of drone swarm ground attack tasks in combination with the objective function; The processing module is used to solve the mathematical model using an improved particle swarm algorithm.

Images