Multi-uav task allocation method based on random difference creative search algorithm

Abstract

The application discloses a multi-unmanned aerial vehicle task allocation method based on a random differentiated creative search algorithm, which comprises the following steps: firstly, modeling a task allocation scene according to an actual agricultural environment, establishing constraint conditions and a target function, and establishing a multi-unmanned aerial vehicle task allocation model for farmland plant protection; secondly, completing population initialization based on a random walk strategy; thirdly, establishing an iteration model based on an optimization team performance model, and cyclically iterating an optimization process until an iteration end condition is met; and fourthly, solving an optimal allocation of a multi-unmanned aerial vehicle to a sub-region of a task to be performed, outputting the allocation scheme, and realizing multi-unmanned aerial vehicle farmland plant protection task allocation. The method of the application utilizes the random walk strategy to optimize in the algorithm initialization stage, enhances the global search capability, and quickly obtains an optimal task allocation scheme through iteration solving, thus having obvious advantages in global search and solving speed, and providing an innovative solution for efficient allocation of multi-unmanned aerial vehicles in complex plant protection tasks.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of multi-unmanned aerial vehicle task allocation, and particularly relates to a multi-unmanned aerial vehicle task allocation method based on a random differentiated creative search algorithm. BACKGROUND

[0002] Unmanned aerial vehicles have the characteristics of strong autonomy, high flexibility, low cost, and good portability. They can autonomously navigate and avoid obstacles, quickly deploy and perform diverse tasks such as agricultural spraying, disaster relief, and environmental monitoring. Compared with single unmanned aerial vehicles, multi-unmanned aerial vehicle systems have the advantages of cooperative operation, efficient coverage, flexible response, and enhanced reliability. Through the cooperative work of multiple unmanned aerial vehicles, large areas can be quickly covered, and the efficiency of data collection and task execution can be improved. In the field of multi-unmanned aerial vehicle task allocation, the application of optimization algorithms has become a research hotspot. These algorithms aim to determine the optimal solution for task allocation in multi-unmanned aerial vehicle systems through precise calculation, in order to achieve specific goals such as minimizing total flight time, maximizing task completion efficiency, and balancing the load of each unmanned aerial vehicle. Classic optimization algorithms such as genetic algorithms, particle swarm optimization algorithms, and ant colony algorithms have excellent global search capabilities and can effectively solve complex problems to find optimal or near-optimal solutions.

[0003] However, despite the significant advantages of these optimization algorithms in multi-unmanned aerial vehicle task allocation, they still face a series of challenges. First, the computational complexity of the algorithm increases exponentially with the number of unmanned aerial vehicles and tasks, which puts higher requirements on computing power. Second, multi-unmanned aerial vehicle task allocation needs to consider multiple constraints such as flight time, energy consumption, capability size, and task scenarios, which are intertwined and increase the complexity of task allocation. In addition, optimization algorithms are prone to fall into local optimal solutions rather than global optimal solutions during the search process, which may lead to suboptimal task allocation schemes and reduce system efficiency and effectiveness, especially in complex and multi-peak search spaces. In addition, most current multi-unmanned aerial vehicle task allocation methods mainly focus on combat task planning and target search, and have few applications in plant protection. Moreover, they do not consider the constraints of real-world environmental factors and unmanned aerial vehicle capabilities. Therefore, the present application proposes a differentiated creative search optimization algorithm based on a random walk strategy to solve a better task allocation scheme, reduce resource waste, and improve task efficiency according to the constraints of task scenarios and unmanned aerial vehicle capabilities. SUMMARY

[0004] To solve the above technical problems, the present application provides a multi-unmanned aerial vehicle task allocation method based on a random differentiated creative search algorithm, which draws on the foraging strategies of animals in nature, combines long jumps with short steps, balances global and local search, and reduces the risk of falling into local optimal solutions.

[0005] The technical scheme adopted by the present application is as follows:

[0006] S1, model the task allocation scene according to the actual agricultural environment, establish the constraint conditions and the objective function, and establish a multi-unmanned aerial vehicle task allocation model for farmland plant protection;

[0007] S2, initialize the population based on the random walk strategy, that is, set the initialization parameters of the differentiated creative search algorithm;

[0008] The initialization parameters include: task area, unmanned aerial vehicle capacity, unmanned aerial vehicle drug carrying capacity, drug requirement per unit of light grade pest area, pest grade, population size, maximum iteration number, number of high performers ngs, and probability constant.

[0009] The population size is the number of candidate solutions.

[0010] S3, establish an iteration model based on the optimized team performance model, that is, simulate the performance of three different members in the team to establish an optimized iteration method, and perform cyclic iteration on the optimization process until the iteration end condition is met, and then go to step S4;

[0011] The three different members in the simulated team include: high performers, average performers, and low performers.

[0012] The high performers are used to create different solutions and explore different areas to prevent premature determination of suboptimal solutions; the average performers are used to refine solutions in promising areas; and the low performers are used to increase team diversity, promote exploration, and increase the idea pool.

[0013] Based on the iteration model, the optimization process is cyclically iterated, and first, the initial parameters of the optimization problem and the iteration end condition are set.

[0014] The initial parameters of the optimization problem include: the speed of new knowledge acquisition qKR, the social influence factor, and the differentiated knowledge acquisition process; and the iteration end condition is that the maximum iteration number is met.

[0015] After one iteration, the local optimal solution, that is, the individual position, is updated, and it is determined whether the maximum iteration number is met. If not, step S3 is repeated for cyclic iteration, and the global optimal solution, that is, the optimal position, obtained in each iteration in the solution space is recorded after the cycle starts, until the iteration end condition is met, and then go to step S4.

[0016] S4, solve the optimal allocation of the sub-area of the task to be performed by the multi-unmanned aerial vehicle, output the allocation scheme, and realize the multi-unmanned aerial vehicle farmland plant protection task allocation.

[0017] Furthermore, step S1 is specifically as follows:

[0018] S11. Model the task allocation scenario based on the actual agricultural environment.

[0019] There are m sub-regions of tasks to be performed, A = {A1,...,A2}. m}, a set of n drones U = {U1,...,U n Each sub-region has a different size, represented by S = {S1,...,S}. m} represents the area size of each region. The severity of pests and diseases varies in each region. L = {1, 1.5, 2} represents the severity level of the region, where 1 is a mild severity, 1.5 is a moderate severity, and 2 is a severe severity.

[0020] Define x ij For U drones i Assigned to sub-area A of pending tasks j The plant protection task is performed, where i∈n,j∈m, and each UAV is... i It can only be assigned to one task sub-region;

[0021] Define y j The number of drones allocated to each sub-region of the task to be performed, where j∈m, and the sum of the number of drones allocated to each sub-region of the task to be performed equals the total number of drones planned.

[0022] Use T j Indicates the sub-region A of tasks to be performed. j Total task time;

[0023] Define Q j For the drug quantity requirement of the j-th sub-region of pending tasks, q i To determine the payload capacity of the i-th drone, the number of drones allocated to each task sub-area must meet the payload requirements of that area.

[0024] Define k as the amount of pesticide required for each unit of minor pest and disease sub-area. Combine the plot size and the severity of the pest and disease to calculate the total pesticide requirement for each sub-area of ​​the task.

[0025] Define C i Let be the processing capacity per unit area of ​​the i-th drone in the sub-area of ​​the task to be performed under the level of minor pest and disease infestation. The processing capacity of the drone varies for areas with different levels of pest and disease infestation.

[0026] S12. Based on the principle of minimizing total operation time, establish the objective function and its constraints, and establish a multi-UAV task allocation model for farmland plant protection;

[0027]

[0028] wherein, Z represents the total time of all unmanned aerial vehicles completing the plant protection work, L j represents the disease and pest level of the to-be-worked task sub-area A j .

[0029] Further, the step S2 is specifically as follows:

[0030] S21, define the initial to-be-worked task sub-area size S, the disease and pest severity level L, the respective unmanned aerial vehicle processing capacity C and the pesticide carrying size q, the pesticide amount k required for each unit of slight level disease and pest area, the population number NP, and the maximum iteration number NFE max of the population;

[0031] S22, initialize the population position;

[0032] In the difference creative search optimization algorithm, the optimization process starts from a set of candidate solutions randomly generated between the upper bound and the lower bound of the optimization problem.

[0033] x i,d = LB d + U(0, 1) * (UB d -LB d )

[0034] wherein, x i,d represents the element at the dth position of x i , x i represents the task sub-area to which the unmanned aerial vehicle U d is allocated at the dth position. LB and UB represent the lower bound and the upper bound of the solution space, U(0, 1) represents the uniform distribution in the interval (0, 1), LB d and UB d respectively represent the lower bound and the upper bound of the dth dimension specified by the optimization problem. The initial population X is expressed in matrix form, specifically as follows:

[0035]

[0036] wherein, NP represents the number of candidate solutions, and D represents the dimension of the optimization problem.

[0037] S23, re-initialize the population position by using the random walk strategy;

[0038] For the population X initialized in step S22, the population is re-initialized using the random walk strategy, and the expression is as follows:

[0039]

[0040] wherein, Xt denotes the current position, a denotes the step size scaling factor, denotes the Hadamard product, β denotes the stability parameter of the Levy distribution, and Levy(β) denotes the generated Levy step. u ~ N(0, σ 2 ) denotes that the random variable u follows a normal distribution with mean 0 and variance σ 2 . v ~ N(0, 1) denotes that the random variable v follows a standard normal distribution with mean 0 and variance 1. σ denotes a scaling factor related to the parameter of the Levy distribution. Γ denotes the gamma function.

[0041] S24, get the number of high performers, which is 20% of the total population size and at least 6, and set the probability constant Pc according to the actual situation.

[0042] Further, the step S3 is specifically as follows:

[0043] S31, start iteration, update the initial parameters required for the iteration population individuals, and calculate the social influence factor;

[0044] Updating the initial parameters required for the iteration population individuals means setting the initial parameters of the optimization problem, including: the speed of acquiring new knowledge qKR, the social influence factor, and the differentiated knowledge acquisition process.

[0045] Among them, in a team environment, the social cognition of an individual is influenced by peers, reflecting the degree of influence of the social dynamics of the team on individual views, i.e. the social influence factor λ t , the calculation expression is as follows:

[0046]

[0047] Among them, λ t coefficient value is high, indicating that the individual is more susceptible to peer pressure and social influence, and the coefficient value is low, indicating that the individual is more independent; NFE t denotes the current number of function iterations, and NFE max denotes the maximum number of function iterations.

[0048] Then traverse each individual in the population, and each individual performs divergent thinking, convergent thinking, and increases team diversity according to its own identity, and before performing divergent thinking, convergent thinking, and increasing team diversity, each individual first performs differentiated knowledge acquisition.

[0049] Among them, differentiated knowledge acquisition DKA is aimed at the different potential of human knowledge acquisition, and the speed of acquiring new knowledge qKR. η i,t _qkr denotes the quantitative knowledge acquisition rate of the individual at the tth iteration, and the expression is as follows:

[0050]

[0051] where, denotes the coefficient value of the individual at the tth iteration, and the calculation expression is as follows:

[0052]

[0053] where, R i,t denotes the rank of the ith individual at the beginning of the tth iteration.

[0054] The action expression of the process of differential knowledge acquisition on each individual is as follows:

[0055] j rand = randint(1, D)

[0056]

[0057] where, v i,d denotes the element of the vdposition of the trial vector v i,t , v i,t denotes the trial vector v i at the tth iteration. η i,t _qkr denotes the quantified knowledge acquisition rate of the individual at the tth iteration. j rand denotes an integer randomly selected from 1 to D.

[0058] S32, judge whether the individual of the population is a high performer, if yes, then perform the acquisition of differential knowledge, and then execute the behavior of divergent thinking, and after execution, enter step S35, if not, enter step S33;

[0059] When the individual is a high performer, first perform the acquisition of differential knowledge, and then execute the behavior of divergent thinking, that is, inject the concept of divergent thinking into the generation of the trial vector v i,d , and the expression of the divergent thinking strategy is as follows:

[0060] v i,d = x r1,d + Lk(a, s)

[0061] where, x r1,d denotes the dth position of the individual randomly selected from {1, 2,..., NP}, and x r1 ≠ x i ≠ x best , x i denotes the current individual, x bestThe best individual in the current iteration is represented. Lk (α, σ) represents an additional term from the Linnik distribution, which is used to encapsulate imagination, and α and σ represent the control parameters of the Linnik distribution, and α = 0.618 and σ = 0.05.

[0062] S33, determine whether the individual of the population is a low performer, if so, acquire differentiated knowledge, then perform the operation of increasing the diversity of the team, and then enter step S35, if not, the individual is a general performer, and enter step S34;

[0063] When the current iteration traverses the last individual of the population, if the generated random number is less than the probability constant Pc, the individual of the population is a low performer, first acquire differentiated knowledge, then perform the operation of increasing the diversity of the team, and the expression of the new member is as follows:

[0064] v i,d = LB + U (0, 1) × (UB-LB)

[0065] Wherein, U (0, 1) represents a uniform distribution in the interval (0, 1), and LB and UB represent the lower bound and upper bound specified by the optimization problem.

[0066] S34, the individual is a general performer, first acquire differentiated knowledge, then perform the operation of converging thinking, and then enter step S35;

[0067] When the execution conditions of the high performer and the low performer are not met, the individual is a general performer, first acquire differentiated knowledge, then perform the operation of converging thinking, and the expression of the converging thinking strategy is as follows:

[0068] v i,d = w × x best,d + λ t × (x r2,d -x i,d ) + ω i,t × (x r1,d -x i,d )

[0069] Wherein, x best,d represents the dth position in the best individual in the current iteration. W represents the weight of the best performance, and the value is 1. X r1,d represents the dth position of the individual randomly selected from {1, 2,..., NP}, and x r1 ≠ x i ≠ x best , which represent three different individuals, x i represents the current individual. X r2,d represents the dth position of the randomly selected individual from {ngs+1,..., NP}, and x r2 ≠ xr1 ≠x i ≠x best , which represent four different individuals. The coefficient ω represents the learning intensity state of the individual. ω i,t represents the efficiency value of ω of the individual at the tth iteration, expressed as follows:

[0070] ω i,t ~ U(0, 1)

[0071] S35, the boundary processing mechanism is executed to ensure that the generated idea remains within the feasibility and reality range;

[0072] If the idea proposed by the individual exceeds the actual limit of the dth position, the boundary processing mechanism will correct the value to the nearest maximum or minimum value within the range allowed by the data set, and the calculation formula is as follows:

[0073]

[0074] where, v i,d represents the element of the trial element v i,t at the dth position, v i,t represents the trial element v i at the tth iteration. LB d and UB d respectively represent the lower bound and upper bound of the dth dimension specified by the optimization problem.

[0075] S36, the individual position is updated, and it is determined whether the maximum iteration number is reached. If not, return to step S31 for the next iteration until the maximum iteration number is reached, end the iteration loop, and enter step S4.

[0076] The beneficial effects of this invention are as follows: First, the method of this invention models the task allocation scenario based on the actual agricultural environment, establishes constraints and its objective function, and builds a multi-UAV task allocation model for farmland plant protection. Then, it initializes the population based on a random walk strategy, and subsequently establishes an iterative model based on an optimized team performance model. The optimization process is iteratively repeated until the iteration termination condition is met, at which point the optimal allocation of the multi-UAV task sub-regions is obtained, and this allocation scheme is output, realizing multi-UAV farmland plant protection task allocation. This invention utilizes a random walk strategy for optimization during the algorithm initialization phase, enhancing global search capabilities. It also rapidly obtains the optimal task allocation scheme through iterative solutions, exhibiting significant advantages in global search and solution speed. This provides an innovative solution for the efficient allocation of multiple UAVs in complex plant protection tasks. Specifically designed for the agricultural plant protection field, this invention constructs a task allocation model that comprehensively considers pest and disease severity, plot size, and UAV performance, optimizing UAV resource allocation, improving the efficiency of plant protection operations, and reducing resource consumption. The improved differential creative search optimization algorithm based on random walk strategy proposed in this invention draws on the foraging strategies of animals in nature. By combining long jumps with short steps, it achieves a balance between global and local search, reducing the risk of getting trapped in local optima. Attached Figure Description

[0077] Figure 1 This is a flowchart of a multi-UAV task allocation method based on a randomized differentiated creative search algorithm according to the present invention.

[0078] Figure 2 This is an iterative flowchart of the iterative model in an embodiment of the present invention.

[0079] Figure 3 This is a convergence curve of the randomized differentiated creative search algorithm of the present invention in an embodiment of the present invention.

[0080] Figure 4 This is a comparison of the convergence curves of the differentiated creative search algorithm and the random differentiated creative search algorithm in an embodiment of the present invention. Detailed Implementation

[0081] The method of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0082] like Figure 1 The flowchart of a multi-UAV task allocation method based on the Random Differentiated Creative Search Algorithm (DCS-levy) of the present invention is shown below, with specific steps as follows:

[0083] S1. Model the task allocation scenario based on the actual agricultural environment, establish the constraints and objective function, and build a multi-UAV task allocation model for farmland plant protection.

[0084] S2, population initialization is completed based on a random walk strategy, that is, initialization parameters of a differentiated creative search algorithm (DCS) are set;

[0085] The initialization parameters include a task area, a size of a UAV capability, a size of a pesticide amount that can be carried by the UAV, a pesticide amount required for each unit of a slight grade of a pest and disease area, a grade of the pest and disease, a population size, a maximum number of iterations, a number of high performers (ngs), and a probability constant.

[0086] The population size is a number of candidate solutions.

[0087] S3, an iteration model is established based on an optimized team performance model, that is, performances of three different members in a team are simulated to establish an optimized iteration method, and a loop iteration is performed on an optimization process until an iteration end condition is met, and then the method proceeds to step S4;

[0088] The three different members in the simulated team include a high performer, a general performer, and a low performer.

[0089] The high performer is used to create different solutions and explore different areas to prevent a suboptimal solution from being determined too early, the general performer is used to refine a solution in a promising field, and the low performer is used to increase team diversity, promote exploration, and increase an idea pool.

[0090] The loop iteration is performed on the optimization process based on the iteration model, and first, initial parameters of an optimization problem and an iteration end condition are set.

[0091] The initial parameters of the optimization problem include a speed of new knowledge acquisition (qKR), a social influence factor, and a differentiated knowledge acquisition process, and the iteration end condition is that a maximum number of iterations is met.

[0092] After one iteration is completed, a local optimal solution, that is, an individual position, is updated, and it is determined whether the maximum number of iterations is met, if not, step S3 is repeated to perform a loop iteration, and after the loop starts, a global optimal solution, that is, an optimal position, obtained in each iteration in a solution space is recorded until the iteration end condition is met, and then the method proceeds to step S4.

[0093] S4, an optimal allocation of a subarea of a task to be performed by multiple UAVs is obtained, and the allocation scheme is output to implement allocation of a farmland plant protection task by the multiple UAVs.

[0094] In this embodiment, step S1 is specifically as follows:

[0095] S11, a task allocation scene is modeled according to an actual agricultural environment;

[0096] In this embodiment, m = 4 task sub-areas A = {A1,...,A4} are set, n = 10 unmanned aerial vehicle sets U = {U1,...,U 10} are set, each sub-area has different sizes, the area sizes of each area are represented by S = {100, 200, 150, 250}, the pest and disease severity of each area is different, and the severity levels of the areas are represented by L = {1, 1.5, 1.5, 2}, 1 is a light degree, 1.5 is a medium degree, and 2 is a severe degree;

[0097] Define x ij as the unmanned aerial vehicle U i assigned to the task sub-area A j to be worked on, where i ∈ n, j ∈ m, and each unmanned aerial vehicle U i can only be assigned to one task sub-area;

[0098] Define y j as the number of unmanned aerial vehicles assigned to each task sub-area to be worked on, where j ∈ m, and the sum of the number of unmanned aerial vehicles assigned to each task sub-area to be worked on is equal to the total number of unmanned aerial vehicles planned;

[0099] Use T j to represent the total working time of the task sub-area A j to be worked on;

[0100] Define Q j as the pesticide demand of the jth task sub-area to be worked on, q i as the pesticide carrying capacity of the ith unmanned aerial vehicle, and the number of unmanned aerial vehicles assigned to each task sub-area to be worked on must meet the pesticide requirement of the area.

[0101] Define k to represent the amount of pesticide required per unit of light level pest and disease task sub-area, and calculate the comprehensive pesticide demand of each task sub-area to be worked on in combination with the size of the plot and the severity level of the pest and disease.

[0102] Define C i as the unit area processing capacity of the ith unmanned aerial vehicle in the light pest and disease level task sub-area, and the processing capacity of the unmanned aerial vehicle for areas with different pest and disease levels is different.

[0103] S12, according to the principle of minimizing the total working time, a target function and its constraint conditions are established, and a multi-unmanned aerial vehicle task allocation model for farmland plant protection is established;

[0104]

[0105] Where Z represents the total time for all unmanned aerial vehicles to complete the plant protection work, L j represents the total working time of the task sub-area A jpests and diseases level.

[0106] In the embodiment, the step S2 is specifically as follows:

[0107] S21, define the initial task sub-area size S to be operated, the pest and disease severity level L, each unmanned processing capacity C={5, 6, 5.5, 6.2, 6, 5.8, 6.1, 5.9, 6, 5.7} and the pesticide carrying size q={10, 12, 15, 10, 14, 16, 13, 11, 14, 15}, the pesticide required for each unit of mild level pest and disease area k=0.1, the population number NP=30, and the maximum number of iterations NFE max =500;

[0108] S22, initialize the population position;

[0109] In the difference creative search optimization algorithm, the optimization process starts from a set of candidate solutions randomly generated between the upper and lower bounds of the optimization problem.

[0110] x i,d =LB d +U(0,1)×(UB d -LB d )

[0111] Where x i,d represents the element at the dth position (dimension) of x i , x i represents the task sub-area allocated to the dth position by the unmanned aerial vehicle U d . LB and UB represent the lower and upper bounds of the solution space, U(0,1) represents the uniform distribution in the interval (0,1), LB d and UB d represent the lower and upper bounds of the dth dimension specified by the optimization problem. The initial population X is expressed in matrix form, specifically as follows:

[0112]

[0113] Where NP represents the number of candidate solutions (referred to as population size), and D represents the dimension of the optimization problem.

[0114] S23, re-initialize the population position using the random walk strategy;

[0115] For the population X initialized in step S22, the population is re-initialized using the random walk strategy, expressed as follows:

[0116]

[0117] Where X t represents the current position, and α represents the step scaling factor, denotes the Hadamard product (i.e., multiplication of corresponding elements), β denotes the stability parameter of the Levy distribution, which controls the distribution characteristics of the step size, and Levy(β) denotes the generated Levy step size. u ~ N(0, σ 2 ) denotes that the random variable u follows a normal distribution with mean 0 and variance σ 2 . v ~ N(0, 1) denotes that the random variable v follows a standard normal distribution with mean 0 and variance 1. σ denotes a scaling factor related to the parameters of the Levy distribution. Γ denotes the gamma function, which extends the factorial function to the continuous domain.

[0118] S24, the number of high performers is obtained, which is 20% of the total population size and at least 6, and the probability constant Pc=0.5 is set according to the actual situation.

[0119] As Figure 2 shown, in the embodiment, the step S3 is specifically as follows:

[0120] S31, start iteration, update the initial parameters required for the iteration population individuals, and calculate the social influence factor;

[0121] The initial parameters required for updating the iteration population individuals are to set the initial parameters of the optimization problem, including: the speed of acquiring new knowledge (qKR), the social influence factor, and the differentiated knowledge acquisition process.

[0122] Among them, in the team environment, the social cognition of individuals will be affected by peers, reflecting the degree of influence of the social dynamics of the team on individual opinions, that is, the social influence factor λ t , the calculation expression is as follows:

[0123]

[0124] Among them, λ t coefficient value is high, indicating that the individual is more susceptible to peer pressure and social influence, and the coefficient value is low, indicating that the individual is more independent; NFE t represents the current function iteration number, and NFE max represents the maximum number of function iterations.

[0125] Then traverse each individual in the population, and each individual performs divergent thinking, convergent thinking, and increases team diversity according to its own identity, and before performing divergent thinking, convergent thinking, and increasing team diversity, each individual first performs differentiated knowledge acquisition.

[0126] Among them, differentiated knowledge acquisition (DKA) is aimed at the different potential of human knowledge acquisition, and the speed of acquiring new knowledge (qKR). η i,t _qkr represents the quantitative knowledge acquisition rate of the individual at the tthiteration, and the expression is as follows:

[0127]

[0128] where, denotes the coefficient value of the individual at the tth iteration, is calculated as follows:

[0129]

[0130] where, R i,t denotes the rank of the ith individual at the beginning of the tth iteration.

[0131] The effect of the process of differential knowledge acquisition on each individual is expressed as follows:

[0132] j rand = randint(1, D)

[0133]

[0134] where, v i,d denotes the element of the trial vector v i,t at the dth position (dimension), v i,t denotes the trial vector v i at the tth iteration. η i,t _qkr denotes the quantified knowledge acquisition rate of the individual at the tth iteration. j rand denotes an integer randomly selected from 1 to D.

[0135] S32, determine whether the population individual is a high performer, if so, perform differential knowledge acquisition, then perform divergent thinking behavior, after execution, enter step S35, if not, enter step S33;

[0136] High performers produce creative ideas. Divergent thinking involves generating diverse, creative ideas and solutions to problems, which is a key characteristic of high performers and the core of creativity.

[0137] When the individual is a high performer, first perform differential knowledge acquisition, then perform the behavior of divergent thinking, that is, inject the concept of divergent thinking into the generation of the trial vector v i,d , the expression of the divergent thinking strategy is as follows:

[0138] v i,d = x r1,d + Lk(a, s)

[0139] where, x r1,d denotes the dth position of the individual randomly selected from {1, 2,..., NP}, and x r1 ≠ x i ≠ xbest , x i represents the current individual, x best represents the best individual so far. Lk(α,σ) represents the additional term from Linnik distribution to encapsulate imagination, α and σ represent the control parameters of Linnik distribution, and α = 0.618, σ = 0.05.

[0140] S33, determine whether the individual of the population is a low performer, if yes, acquire the differentiated knowledge, then execute the operation of increasing team diversity, and enter step S35, if not, the individual is a general performer, enter step S34;

[0141] Low performers improve team diversity. A team with constantly changing members will generate more diverse ideas, replacing underperforming members with new members.

[0142] When the iteration of this time traverses to the last individual of the population, if the generated random number is less than the probability constant Pc, the individual of the population is a low performer, first acquire the differentiated knowledge, then execute the operation of increasing team diversity, and the expression of generating a new member is as follows:

[0143] v i,d = LB + U(0, 1) × (UB - LB)

[0144] Wherein, U(0, 1) represents the uniform distribution in the interval (0, 1), LB and UB represent the lower bound and upper bound specified by the optimization problem.

[0145] S34, the individual is a general performer, first acquire the differentiated knowledge, then execute the operation of thinking convergence, and enter step S35;

[0146] General performers will perform thinking convergence behavior, which will refine ideas into solutions.

[0147] When the execution conditions of high performers and low performers are not met, it is a general performer, first acquire the differentiated knowledge, then execute the thinking convergence behavior, which depends on the knowledge base of the best performer and combines the random contribution from two different team members brought by the current individual, therefore, it promotes the diversity and convergence of the generated solution. The expression of thinking convergence strategy is as follows:

[0148] v i,d = w × x best,d + λ t × (x r2,d - x i,d ) + ω i,t × (x r1,d - x i,d )

[0149] Wherein, xbest,d represents the dth position in the best individual in the current iteration. w represents the best cognitive weight and has a value of 1. x r1,d represents the dth position of a randomly selected individual from {1, 2,..., NP} and x r1 ≠ x i ≠ x best , which represent three different individuals, x i represents the current individual. x r2,d represents the dth position of a randomly selected individual from {ngs+1,..., NP} and x r2 ≠ x r1 ≠ x i ≠ x best , which represent four different individuals.

[0150] The coefficient ω represents the learning intensity state of the individual.

[0151] It takes into account the individual's potential to acquire knowledge, which can be influenced by factors such as mood, level of interest, and environment. This coefficient is like a fascinating mimic, reflecting the inherent complexity of interpersonal communication and collaboration. It encapsulates the essence of the mysterious human mind, where emotions and thoughts change like a kaleidoscope.

[0152] ω i,t represents the efficiency value of ω for the individual at the tth iteration, expressed as follows:

[0153] ω i,t ~ U(0, 1)

[0154] S35, a boundary handling mechanism is executed to ensure that the generated ideas remain within the feasibility and reality range;

[0155] If the individual's idea exceeds the actual limit of the dth position, the boundary handling mechanism will correct the value to the nearest maximum or minimum within the range allowed by the dataset, calculated as follows:

[0156]

[0157] where v i,d represents the trial element v i,t at the dth position (dimension), v i,t represents the trial element v i at the tth iteration. LB d and UB d represent the lower and upper bounds of the dth dimension as specified by the optimization problem, respectively.

[0158] S36, updating the individual position, and then determining whether the maximum iteration number is reached, if not, returning to step S31 for the next iteration until the maximum iteration number is reached, ending the iteration loop, and entering step S4.

[0159] In this embodiment, if the constraint condition established in step S12 is not satisfied, the fitness value and is 6, and is determined according to the actual situation.

[0160] In this embodiment, the total iteration number of the optimization model is NFE max = 500, the initial population number NP = 30, and the finally obtained task allocation scheme is as follows:

[0161] According to the task scene and parameter setting of this embodiment, the optimal fitness value obtained by calculation is 3.913043, and the multi-unmanned aerial vehicle task allocation scheme Best_solusion is expressed as follows:

[0162] Best_solusion = [4 1 4 4 4 2 3 3 4 2]

[0163] The optimal allocation matrix assignment_matrix is (sub-region x unmanned aerial vehicle number), and is expressed as follows:

[0164]

[0165] The fitness value convergence curve of the improved algorithm of the method of the present application is shown in Figure 3 The convergence curve of the algorithm before improvement (DCS) and the convergence curve of the algorithm after improvement (DCS-levy) are compared as shown in Figure 4 It can be seen that the fitness value is lower after improvement, and it is less likely to fall into a local optimal solution than before improvement, and the solving speed is faster. The simulation results prove that the method of the present application can realize the optimal allocation of multi-unmanned aerial vehicle tasks.

[0166] In summary, the method of this invention provides an innovative solution for efficient and energy-saving allocation of multiple UAVs in agricultural plant protection tasks, possessing significant practical application value and theoretical research significance. Compared to existing task allocation methods, the method of this invention employs a differential creative search optimization algorithm, a novel and high-performance optimization strategy compared to existing particle swarm optimization and genetic algorithms. Applying it to the field of agricultural plant protection not only broadens its application scope but also brings innovative solutions to this field. The method of this invention is designed based on an improved random walk strategy. This strategy utilizes the long jump characteristic to quickly explore a broad search space in the early stages of the algorithm, effectively improving global search capabilities and reducing the possibility of getting trapped in local optima. Combining the long jump and short step strategies, the algorithm rapidly conducts extensive exploration in the early stages and performs refined searching in the later stages, thereby accelerating the overall convergence speed. The method of this invention is more efficient in solving complex multi-UAV task allocation problems, can adapt to diverse application scenarios, and has the potential for widespread application in other fields.

[0167] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1.A multi-UAV task allocation method based on random differential creative search algorithm, the specific steps are as follows: S1, according to the actual agricultural environment, the task allocation scene is modeled, the constraint condition and the objective function are established, and the multi-UAV task allocation model for farmland plant protection is established; S2, population initialization is completed based on random walk strategy, that is, the initialization parameters of differential creative search algorithm are set; The initialization parameters include: Task area, UAV capacity, UAV carrying capacity, drug requirement per unit of light level pest area, pest level, population size, maximum iteration number, number of high performers ngs and probability constant; Wherein, the population size is the number of candidate solutions; S3, an iteration model is established based on the optimization team performance model, that is, the performance of three different members in the simulated team is simulated to establish an optimization iteration method, and the optimization process is iterated until the iteration end condition is met, and then step S4 is turned to; The three different members in the simulated team include: high performers, general performers and low performers; Wherein, the high performers are used to create different solutions and explore different areas to prevent premature determination of suboptimal solutions; the general performers are used to improve the solutions in promising areas; the low performers are used to increase team diversity, promote exploration and increase the idea pool; The optimization process is iterated based on the iteration model, first the initial parameters of the optimization problem and the iteration end condition are set; Wherein, the initial parameters of the optimization problem include: the speed of new knowledge acquisition qKR, the social influence factor and the differential knowledge acquisition process; the iteration end condition is that the maximum iteration number is met; After one iteration, the local optimal solution, that is, the individual position, is updated, and it is judged whether the maximum iteration number is met, if not, step S3 is repeated for cyclic iteration, and the global optimal solution in the solution space, that is, the optimal position, is recorded after the cycle starts, until the iteration end condition is met, and then step S4 is turned to; S4, the optimal allocation of the multi-UAV to the sub-area of the task to be performed is solved, and the allocation scheme is output, realizing the multi-UAV task allocation for farmland plant protection. 2.The multi-UAV task allocation method based on random differential creative search algorithm according to claim 1, wherein, The step S1 is specifically as follows: S11, the task allocation scene is modeled according to the actual agricultural environment; There are m sub-regions A = {A1,..., A m} to be worked, n unmanned aerial vehicle sets U = {U1,..., U n}, each sub-region has different size, and the area size of each region is represented by S = {S1,..., S m}, the severity of each region is different, and the severity level of the region is represented by L = {1, 1.5, 2}, 1 is light, 1.5 is medium, and 2 is severe; Definition x ij To the UAV U i Allocated to the task sub-area A j The plant protection task performed, where i∈n,j∈m, and each UAV U i Can only be assigned to one task sub-area; Definition y j a number of UAVs allocated to each task sub-area to be worked, wherein j∈m, and the sum of the number of UAVs allocated to each task sub-area to be worked is equal to the total number of UAVs planned. Using T j to indicate the total working time of the task sub-area A j to be worked Definition Q j The amount of medicine required for the jth sub-region of the task to be performed, q i The amount of medicine carried by the ith unmanned aerial vehicle, the number of unmanned aerial vehicles allocated to each sub-region of the task to be performed must meet the medicine requirement of the region; Define k to represent the drug requirement per unit of light level pest area to be worked, and calculate the comprehensive drug demand of each sub-area to be worked by combining the size of the field and the severity level of the pest; Definition C i Ci is the processing capacity of the i-th UAV in the unit area of the sub-region of the task to be operated at the light pest level. The processing capacity of the UAV is different for different pest levels. S12, according to the principle of minimizing the total working time, the objective function and its constraint condition are established, and the multi-UAV task allocation model for farmland plant protection is established; Wherein, Z represents the total time of all unmanned aerial vehicles completing the plant protection work, L j represents the pest level of the to-be-worked sub-region A j . 3.The multi-UAV task allocation method based on random differential creative search algorithm of claim 1, wherein, The step S2 is specifically as follows: S21, define initial to-be-operated task sub-region area size S, pest and disease severity level L, each unmanned processing capacity C and pesticide carrying size q, pesticide amount required per unit of light level pest and disease area k, population number NP, and maximum iteration number of population NFE max ; S22, initialize the population position; In the differential creative search optimization algorithm, the optimization process starts from a set of candidate solutions generated randomly between the upper and lower bounds of the optimization problem; x i,d = LB d + U(0,1) x (UB d - LB d ) where x i,d represents the element at the dth position of x i , x i represents the dth position of the UAV U d assigned to the sub-region of the task; LB and UB represent the lower bound and the upper bound of the solution space, U(0, 1) represents the uniform distribution on the interval (0, 1), LB d and UB d respectively represent the lower bound and the upper bound of the dth dimension specified by the optimization problem; the initial population X is represented in the form of a matrix, as follows: Wherein, NP represents the number of candidate solutions, and D represents the dimension of the optimization problem; S23, the population position is initialized again using the random walk strategy; For the population X initialized in step S22, the population is initialized again using the random walk strategy, and the expression is as follows: where X t represents the current position, and a represents a step scaling factor, represents a Hadamard product, β represents a stability parameter of a Levy distribution, and Levy(β) represents a generated Levy step; u ~ N(0, σ 2 ) represents that a random variable u follows a normal distribution with a mean of 0 and a variance of σ 2 ; v ~ N(0, 1) represents that a random variable v follows a standard normal distribution with a mean of 0 and a variance of 1; and σ represents a scaling factor related to a parameter of a Levy distribution; and Γ represents a gamma function. S24, get the number of high performers, the number is 20% of the total population size and at least 6, and set the probability constant Pc according to the actual situation. 4.The method of claim 1, wherein, The step S3 is specifically as follows: S31, start iteration, update the initial parameters required for the iteration population individuals, and calculate the social influence factor; The initial parameters required for updating the iteration population individuals are setting the initial parameters of the optimization problem, including: the speed of acquiring new knowledge qKR, the social influence factor, and the differentiated knowledge acquisition process; wherein, in a team environment, the social cognition of an individual is influenced by peers, reflecting the degree of influence of the social dynamics of the team on the individual's opinion, i.e. the social influence factor λ t The calculation expression is as follows: where λ t A high coefficient value indicates that the individual is more susceptible to peer pressure and social influence, and a low coefficient value indicates that the individual is more independent; NFE t represents the current number of function iterations, NFE max represents the maximum number of function iterations; Then traverse each individual in the population, and each individual performs divergent thinking, convergent thinking, and increases team diversity according to its own identity, and before performing divergent thinking, convergent thinking, and increasing team diversity, each individual first performs differentiated knowledge acquisition; where the differential knowledge acquisition DKA is directed to the different potential of human beings to acquire knowledge, while focusing on the speed qKRof new knowledge acquisition; η i,t _qkrrepresents the quantified knowledge acquisition rate of the individual at the tthiteration, expressed as follows: wherein, represents the individual's value at the tth iteration coefficient value, the calculation expression is as follows: wherein R i,t denotes the rank of the i-th individual at the beginning of the t-th iteration; The process of differentiated knowledge acquisition has the following expression for each individual: j rand = randint(1, D) where v i,d represents the element v i,t at the dth position, v i,t represents the trial vector v i at the tth iteration; η i,t _qkr represents the individual's quantified knowledge acquisition rate at the tth iteration; j rand represents an integer randomly selected from 1 to D; S32, determine whether the population individual is a high performer, if so, perform differentiated knowledge acquisition, then execute the divergent thinking behavior, and enter step S35 after execution, if not, enter step S33; When the individual is a high performer, then the acquisition of differentiated knowledge is performed first, and then the act of divergent thinking, i.e. the injection of divergent thinking concepts into the generation of the trial vectors v i,d The divergent thinking strategy expression is as follows: v i,d = x r1,d + Lk(a, s) where x r1,d represents the dth position of the individual randomly selected from {1, 2,..., NP}, and x r1 ≠ x i ≠ x best , x i represents the current individual, x best represents the currently best individual; Lk(a, s) represents an additional term from Linnik distribution for encapsulating imagination, a and s represent control parameters of Linnik distribution, and a = 0.618, s = 0.05; S33, determine whether the population individual is a low performer, if so, perform differentiated knowledge acquisition, then execute the operation of increasing team diversity and enter step S35, if not, the individual is a general performer, enter step S34; When the iteration traverses to the last individual in the population, if the generated random number is less than the probability constant Pc, the population individual is a low performer, first performs differentiated knowledge acquisition and then executes the operation of increasing team diversity, and the expression of generating a new member is as follows: v i,d = LB + U(0,1) x (UB - LB) Wherein, U(0, 1) represents uniform distribution in the interval (0, 1), and LB and UB represent the lower bound and upper bound specified by the optimization problem; S34, the individual is a general performer, first performs differentiated knowledge acquisition, then executes the convergent thinking operation and enters step S35; When the execution conditions of high performers and low performers are not met, it is a general performer, first performs differentiated knowledge acquisition, then executes the convergent thinking behavior, and the expression of the convergent thinking strategy is as follows: v i,d = w x x best,d + λ t x (x r2,d - x i,d ) + ω i,t x (x r1,d - x i,d ) where x best,d represents the dth position in the best individual in the current iteration; w represents the best cognitive weight and has a value of 1 ; x r1,d represents the dth position of a randomly selected individual from {1, 2,..., NP} and x r1 ≠ x i ≠ x best represent three different individuals, x i represents the current individual; x r2,d represents the dth position of a randomly selected individual from {ngs+1,..., NP} and x r2 ≠ x r1 ≠ x i ≠ x best represent four different individuals; the coefficient ω represents the learning intensity state of the individual; ω i,t represents the efficiency value of ω for the individual at the tth iteration, expressed as follows: ω i,t ~ U(0,1) S35, execute the boundary processing mechanism to ensure that the generated ideas remain within the feasibility and reality range; If the idea proposed by the individual exceeds the actual limit of the dth position, the boundary processing mechanism will correct the value to the nearest maximum or minimum within the range allowed by the data set, and the calculation formula is as follows: where v i,d represents the trial vector v i,t at the dth position, v i,t represents the trial vector v i at the tth iteration; LB d and UB d denote the lower and upper bounds of the dth dimension as specified by the optimization problem, respectively. S36, update the individual position, and then determine whether it is the maximum iteration number, if not, return to step S31 for the next iteration until the maximum iteration number is met, end the iteration loop, and enter step S4.

Images