An improved immune multi-agent UAV task allocation method

Abstract

An improved immune multi-agent UAV task allocation method, which relates to the field of UAV decision-making methods, is to solve the problem of inaccurate UAV task allocation and poor task allocation effect of the current UAV task allocation method. The present invention introduces the neighborhood operation mechanism in the multi-agent system, and through cloning, excitation, inhibition, and self-learning of the optimal antibody in the neighborhood, while retaining the fast convergence speed characteristics of the artificial immune method, it improves the ability of the traditional artificial immune method to jump out of the local optimum. Secondly, for the problem that the artificial immune method is easy to fall into the local optimum, this method not only improves the evolutionary operation mechanism, but also introduces the 2-opt optimization operator, which greatly improves the convergence and convergence accuracy of the method. The present invention accurately allocates tasks and has good task allocation effect.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle decision making methods. Background Art

[0002] Unmanned aerial systems such as drones have extremely strong maneuverability. When performing tasks in dangerous areas, they do not need to rely on onboard personnel. They are not only convenient and intelligent to use, but also can reduce the cost of pilot training. Therefore, they are widely used in various application scenarios in various fields such as reconnaissance, strike, surveillance, search and rescue. With the increasing popularity of drone applications, research issues such as autonomous algorithms, intelligent control algorithms, and collaborative planning algorithms for drones are becoming increasingly prominent. In particular, the ability of drones to perform tasks in dry, harsh, or dangerous environments has attracted more and more scientific research staff to conduct research on drone flight formation methods, task allocation methods, trajectory planning methods, and performance evaluation methods. The current metaheuristic method for drone task allocation is still immature in its optimization ability and cannot reach the optimal solution. The weak optimization ability leads to inaccurate drone task allocation and poor task allocation results. Summary of the invention

[0003] The present invention aims to solve the problem that the optimization ability of the current UAV task allocation method is still immature and cannot reach the optimal solution. The weak optimization ability leads to inaccurate UAV task allocation and poor task allocation effect, thereby providing an improved immune multi-agent UAV task allocation method.

[0004] An improved immune multi-agent UAV task allocation method is provided, wherein an immune UAV task allocation model is established in an airborne system of a UAV to which a task is to be allocated, wherein the UAV task immune allocation model is used to output a UAV flight task and send it to the airborne system of the UAV for execution;

[0005] The drone onboard system performs flight operations according to the drone flight mission sent by the drone mission allocation model,

[0006] The UAV task allocation model generates and outputs UAV flight tasks, which specifically includes the following steps:

[0007] Step 1, iterate the UAV task allocation model times, where times is a positive integer;

[0008] Each iteration of the UAV task allocation model includes the following steps:

[0009] Step 1: Initialize the antibody agent and generate the memory population N in the drone task allocation model best , then execute steps one and two;

[0010] Step 12: Perform neighborhood cloning on each antibody agent in step 11 to obtain the antibody agent after the neighborhood cloning process, and then execute step 13;

[0011] Step 13: Perform neighborhood suppression processing on each antibody agent after neighborhood cloning obtained in step 12 to obtain an antibody agent after neighborhood suppression processing, and then execute step 14;

[0012] Step 14: Perform population crossover processing on the antibody agent obtained after neighborhood suppression processing in step 14 to obtain the antibody agent after population crossover processing, and then execute step 15;

[0013] Step 15: Perform self-learning processing on the antibody agent after cross-processing of each population obtained in step 14 to obtain the antibody agent after self-learning processing, and then execute step 16;

[0014] Step 16: Perform partial 2-opt optimization on each antibody agent after self-learning obtained in step 15, and store the first N antibody agents into the memory population N generated in step 11. best , N is a positive integer, completing one iteration;

[0015] Step 2: After completing the preset times iterations, the UAV task allocation model obtains the optimal antibody Agent, and uses the optimal antibody Agent as the optimal task allocation result;

[0016] Step 3: The UAV task allocation model sends the optimal task allocation result obtained in step 2 to the UAV onboard system to complete a multi-agent UAV task allocation.

[0017] The present invention achieves beneficial effects and significant progress: the present invention realizes the improvement of the traditional artificial immune method. First, in order to avoid the interaction between populations being too random, the present invention introduces the neighborhood operation mechanism in the multi-agent system. Through cloning, excitation, inhibition, and self-learning of the optimal antibody in the neighborhood, the fast convergence speed of the artificial immune method is retained, while improving the ability of the artificial immune algorithm to jump out of the local optimum. Secondly, for the problem that the artificial immune method is prone to fall into the local optimum, this method not only improves the evolutionary operation mechanism, but also introduces the 2-opt optimization operator, which greatly improves the convergence and convergence accuracy of the method. The improved immune multi-agent method is used to solve the task allocation problem in a large range. The simulation results show that the method still has the ability to continue to evolve downward through long-term iterations in the offline planning scenario, and the ability to converge to the global optimal solution is relatively good. The results of multiple experiments in the offline planning scenario show that the probability of this method converging to the global optimal solution is relatively high, and it has good optimization characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the overall flow of the task allocation process of the present invention;

[0019] Figure 2 It is a top view diagram of the task allocation result;

[0020] Figure 3 It is a side view schematic diagram of the task allocation result;

[0021] Figure 4 This is a schematic diagram of the optimal fitness evolution simulation;

[0022] Figure 5 Schematic diagram of average fitness evolution simulation. DETAILED DESCRIPTION

[0023] Specific implementation method 1. Combination Figures 1 to 5 This embodiment describes an improved immune multi-agent UAV task allocation method.

[0024] Based on the concepts of artificial immune algorithm and multi-agent system, the present invention proposes an improved immune multi-agent algorithm for solving the task allocation problem. By limiting some operations within the neighborhood, the ability of the traditional immune algorithm to jump out of the local optimum is improved, while the faster convergence speed of the traditional immune algorithm in the early stage is also maintained, and it is applicable to both dynamic and static task allocation problems. The implementation steps of the present invention in one iteration are as follows:

[0025] 1. Generation of initial antibody agents and memory populations

[0026] The concept of memory population in the clonal selection algorithm is used to save the optimal part of the Agent in each evolution process. Due to the limited size of the memory Agent group, it must be updated and eliminated in real time in each iteration. The specific steps are as follows:

[0027] ① Initialize the size of the memory agent group to N, select the k antibody agents with the best fitness in the current antibody agent grid, and try to add them to the memory agent group. If the memory agent group has not reached the preset size, allow this addition and directly add the k antibody agents to the agent group;

[0028] ② If the memory agent group is full, or the number of antibody agents that can be accommodated is less than k, an elimination operation is performed to compare the fitness of all antibody agents in the current memory agent group with the new k antibody agents, and eliminate the antibody agents with the lowest fitness that exceed the size of the memory agent group and remove them from the memory agent group.

[0029] The overall process can be expressed as Eq.

[0030] N best =SelectN(N best ,L) (1)

[0031] Where N best represents the memory agent group, and SelectN(·) represents the function of selecting the first N agents with the best fitness.

[0032] 2. Neighborhood cloning operation

[0033] The present invention keeps the cloning process in the neighborhood. For each antibody agent, the agent with the best fitness among several agents in its neighborhood and itself is selected, and after a certain degree of multiplication and mutation, the antibody agent with the best fitness is selected to replace the original antibody agent. The neighborhood cloning operation can be expressed as formula.

[0034] A i,j =Select(Mutation(Clone(Select({A i,j ,Neigh.A i,j ,})))) (2)

[0035] Among them, Select(·) represents the operation of selecting the best agent, Mutation(·) represents the mutation operation, and Clone(·) represents the cloning operation. For all the initial antibody agents initialized in step 1, the neighborhood cloning operation is performed to generate a new population of the next generation. After completing the neighborhood cloning step, the next step is performed on the population.

[0036] 3. Neighborhood suppression operation

[0037] The neighborhood suppression operator suppresses antibodies with high affinity to maintain the diversity of the population. Since the cloning operation of this algorithm is performed in the neighborhood, if an antibody agent is the local optimal agent, then its excellent fitness will make it the parent of all antibody agent clones in its neighborhood. In other words, antibodies in a small range all preserve the characteristics of the local optimal agent, and the affinity is naturally high, so it needs to be suppressed. Therefore, in each neighborhood, the neighborhood suppression operation can be expressed as formula.

[0038]

[0039] Among them B i,jis a randomly generated new antibody agent. For all antibody agents generated in step 2, they are searched in the neighborhood to find high-affinity antibodies for consistency. Since the neighborhood range is small, for each agent, if an agent that is exactly the same as itself is found in the neighborhood, it will be directly replaced with a randomly generated antibody agent to maintain the diversity of the agent group. After completing the neighborhood suppression step, perform the next step on the population.

[0040] 4. Population crossover operation

[0041] The population crossover operation draws on the crossover operation in the artificial immune algorithm, that is, some gene positions of two randomly selected parent antibodies are exchanged according to certain rules with a certain probability, so as to interact with information and generate two new offspring. The objects selected for the crossover operation in this algorithm are different. For each antibody agent, a roulette-based selection method is used to select another parent agent in the entire grid to perform a crossover operation with it, and then the original agent is replaced with the better offspring after the crossover. This operation not only provides each antibody agent, regardless of its quality, with the opportunity to cross, improves the diversity of the population, and conducts a global search, but also has mutual incentives between excellent individuals, ensuring the direction of evolution. The crossover operator uses a partial matching crossover operator, and the population crossover operation can be expressed as a formula.

[0042]

[0043] Roulette(L) represents the roulette selection function for the entire grid L, unifrnd(0,1) represents a random number between 0 and 1 that follows a uniform distribution, times represents the total number of iterations, and i represents the current number of iterations. It can be noted that the probability of the population crossover operation gradually decreases with the increase in the number of iterations. The purpose is to accelerate the convergence of the population and prevent the population from losing too much good information due to excessive crossover. For all antibody agents generated in step 3, a population crossover operation is performed within the entire population. After completing the population crossover step, perform the next step on the population.

[0044] 5. Self-learning operation

[0045] The self-learning operation draws on the self-learning ability of agents in multi-agent systems. In this chapter, multi-agents are antibody agent groups, and the most representative individual is the individual with the best fitness. By cloning and mutating the individual, selecting the best individual after mutation to replace the original best agent, and completing the local search in a small range, the antibody agent group can be further evolved. The self-learning operation can be expressed as Eq.

[0046] A best=Select(Mutation(Clone(A best ))) (5)

[0047] For all antibody agents generated in step 4, they are evaluated, and the best antibody agent is selected for mutation operation, and the best antibody agent after mutation is selected to replace the original agent. After completing the self-learning step, the next step is performed on the population.

[0048] 6. Part 2-opt operator

[0049] The 2-opt optimization algorithm starts from an arbitrary feasible solution, repeatedly searches, and when a cross edge is found, the coding in the middle of the cross edge is reversed. This replacement operation is performed until a local optimum is reached, and finally the local optimum solution is used as the solution of the algorithm. In the task allocation problem of the present invention, although the 2-opt operation cannot be directly performed on the entire antibody, for each target point under the drone, the convergence of the algorithm can be accelerated by partial 2-opt operations.

[0050] For all antibody agents generated in step 5, select the starting and ending positions of some 2-opt operators according to the number of drones and target points, and implement some 2-opt optimization. After completing the optimization step, the top N antibodies with better performance are stored in the memory population, and all steps of one iteration are completed.

[0051] Afterwards, the present invention sets the number of iterations to times, and executes the above six steps in each iteration. When the number of iterations is completed, the output optimal antibody Agent is the optimal task allocation result.

[0052] In this specific implementation, taking the number of drones as 4, the number of target points as 30, and times as 5000 as an example, the simulation results of the implementation of the invention are shown in the attached figure. Figure 4 and Figure 5 shown.

Claims

1. An improved immune multi-agent UAV task allocation method, wherein an immune UAV task allocation model is established in the UAV airborne system to which the task is to be allocated, and the UAV task immune allocation model is used to output the UAV flight task and send it to the UAV airborne system for execution; The drone onboard system performs flight operations according to the drone flight mission sent by the drone mission allocation model, Its characteristics are: The UAV task allocation model generates and outputs UAV flight tasks, which specifically includes the following steps: Step 1, iterate the UAV task allocation model times, where times is a positive integer; Each iteration of the UAV task allocation model includes the following steps: Step 1: Initialize the antibody agent and generate the memory population N in the drone task allocation model best , then execute steps one and two; Step 12: Perform neighborhood cloning on each antibody agent in step 11 to obtain the antibody agent after the neighborhood cloning process, and then execute step 13; Step 13: Perform neighborhood suppression processing on each antibody agent after neighborhood cloning obtained in step 12 to obtain an antibody agent after neighborhood suppression processing, and then execute step 14; Step 14: Perform population crossover processing on the antibody agent obtained after neighborhood suppression processing in step 14 to obtain the antibody agent after population crossover processing, and then execute step 15; Step 15: Perform self-learning processing on the antibody agent after cross-processing of each population obtained in step 14 to obtain the antibody agent after self-learning processing, and then execute step 16; Step 16: Perform partial 2-opt optimization on each antibody agent after self-learning obtained in step 15, and store the first N antibody agents into the memory population N generated in step 11. best , N is a positive integer, completing one iteration; Step 2: After completing the preset times iterations, the UAV task allocation model obtains the optimal antibody Agent, and uses the optimal antibody Agent as the optimal task allocation result; Step 3: The UAV task allocation model sends the optimal task allocation result obtained in step 2 to the UAV onboard system to complete a multi-agent UAV task allocation.

2. The improved immune multi-agent UAV task allocation method according to claim 1 is characterized in that In step 1, the memory population N generated in the UAV task allocation model best , used to save the optimal partial antibody Agent in each evolution process.

3. The improved immune multi-agent UAV task allocation method according to claim 2 is characterized in that The memory population N best The size is limited, so in each iteration, the memory population N is updated in real time. best To update and eliminate operations, the specific steps are as follows: Initialize the size of the memory agent group to N, where N is a positive integer. Select the k antibody agents with the best fitness in the current antibody agent grid, where k is a positive integer, and try to add them to the memory population N best If the memory population N best If the preset size is not reached, the addition is allowed, and the k antibody agents are directly added to the memory population N. best middle; If the memory population N best If the memory population is full or the number of antibody agents that can be accommodated is less than k, an elimination operation is performed to replace the current memory population N best The fitness of all antibody agents and the new k antibody agents is compared, and the one with the lowest fitness exceeds the memory population N. best Scaled antibody agent elimination, that is, removing the memory population N best .

4. The improved immune multi-agent UAV task allocation method according to claim 3 is characterized in that In step 1 and 2, the neighborhood cloning process for each antibody agent is performed by the following formula: A i,j =Select(Mutation(Clone(Select({A i,j ,Neigh.A i,j ,})))) achieved; Where: Select(·) represents the operation of selecting the optimal antibody Agent, Mutation(·) represents the mutation operation, and Clone(·) represents the cloning operation.

5. The improved immune multi-agent UAV task allocation method according to claim 1 is characterized in that In step 13, the antibody agent obtained in step 12 after each neighborhood clone is subjected to neighborhood suppression processing by the following formula: achieved; Among them B i,j A new antibody agent is generated randomly.

6. The improved immune multi-agent UAV task allocation method according to claim 1 is characterized in that In step 14, each memory population N best The population crossover process is performed using the following formula: achieved; Roulette(L) represents the roulette selection function for the entire grid L, unifrnd(0,1) represents a random number between 0 and 1 that follows a uniform distribution, times represents the total number of iterations, and i represents the current number of iterations.

7. The improved immune multi-agent UAV task allocation method according to claim 1 is characterized in that In step 15, the antibody agent obtained in step 14 after cross processing is self-learned by the following formula: A best =Select(Mutation(Clone(A best ))) achieved.

Images