Unmanned aerial vehicle group forest fire ground-air cooperative task allocation method

Abstract

The invention discloses an unmanned aerial vehicle group forest fire ground-air cooperative task allocation method. The method comprises the steps that S1, a digital elevation model, a fire point thermodynamic diagram, obstacle distribution and airspace control information of a fire area are acquired; defining a to-be-executed task allocation set and an unmanned aerial vehicle cluster; s2, using cruising ability maximization and wind resistance maximization as double optimization targets of a non-dominated sorting genetic algorithm; s3, calculating a dual-objective function value of each unmanned aerial vehicle; s4, performing chromosome numerical coding on the unmanned aerial vehicles in the unmanned aerial vehicle group with optimized performance, performing iterative optimization by adopting a non-dominated sorting genetic algorithm for simulating binary crossover and polynomial variation, and searching a Pareto optimal solution set; s5, each unmanned aerial vehicle is allocated to each task; and S6, matching the multiple unmanned aerial vehicle groups with the three tasks, and calculating the maximum contribution of each unmanned aerial vehicle to each task so as to form an optimal solution for task allocation of the unmanned aerial vehicles.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of low-altitude cooperative task allocation of unmanned aerial vehicle group in forest fire, and particularly relates to a method for combining unmanned aerial vehicle groups. BACKGROUND

[0002] With the progress of science and technology, unmanned aerial vehicles are increasingly widely used in various fields. The use of unmanned aerial vehicles in forest fires can greatly improve work efficiency and reduce casualties. Nowadays, people place unmanned aerial vehicles above the fire to monitor the fire situation. Such technology can detect the trend and spread speed of the fire at the first moment. However, due to too many unknown factors and complex situations when a fire occurs, multiple unmanned aerial vehicles need to work cooperatively, and the combination scheme needs to be optimized to maximize the benefits of the unmanned aerial vehicle group. Various unmanned aerial vehicle distribution methods need to be considered, and each unmanned aerial vehicle has different working ranges and working areas. The combination method of each unmanned aerial vehicle needs to be considered in various situations.

[0003] The existing task allocation method mainly uses a single-objective optimization model or subjective parameter setting to combine unmanned aerial vehicles. The calculation based on a certain target may lead to inaccurate results, and the subjective weight parameter is not objective enough, which has a large error and ignores the mutual restriction and influence of multiple values in the case of fire. The shortcoming is that the fine calculation of unmanned aerial vehicle distribution and combination is not accurate enough, which leads to a large deviation from the actual situation.

[0004] The optimized non-dominated sorting genetic algorithm combines binary crossover and polynomial mutation two calculation methods for iteration. The results are more objective and the combination methods are more through the calculation of the previous part and the introduction of new parameters. The Pareto optimal solution is obtained through continuous iteration, which makes the adaptability stronger. The coalition algorithm using the greedy strategy can maximize the ability of the unmanned aerial vehicle and complete the task targetedly. SUMMARY

[0005] To achieve the above purpose, the application provides a forest fire ground-air cooperative task allocation method of unmanned aerial vehicle group.

[0006] The method adopts the following technical scheme:

[0007] S1, obtain the digital elevation model of the fire area, the fire point thermal map, the obstacle distribution, and the airspace control information; define a task allocation set to be executed, and the task types include positioning tasks, detection tasks, and fire extinguishing tasks; define an unmanned aerial vehicle cluster, and each unmanned aerial vehicle in the unmanned aerial vehicle cluster has different load capacity and state attributes;

[0008] S2, maximize the endurance and the wind resistance of the UAV as the dual optimization objective of the non-dominated sorting genetic algorithm, establish the air-ground cooperative task allocation model, while meeting the task timing, UAV capability, airspace safety, task uniqueness multiple constraints, form a functional hybrid UAV group;

[0009] S3, for the non-dominated sorting genetic algorithm of each UAV in the UAV group Dual optimization objective function value, form a UAV objective function set, consisting of multiple groups of performance optimized UAV groups;

[0010] S4, the UAVs in the performance optimized UAV group are encoded with chromosome numerical value, and the non-dominated sorting genetic algorithm based on simulated binary crossover and polynomial mutation is used for iterative optimization, combined with the crowding degree and the dual optimization objective function value, to find the Pareto optimal solution set of the UAV group combination mode;

[0011] S5, the different UAV groups in the formed Pareto optimal solution set contain multiple UAVs with different efficiency values for positioning tasks, detection tasks and fire extinguishing tasks, and each UAV is allocated to each task.

[0012] S6, using the coalition formation algorithm based on the greedy strategy, the multiple UAV groups obtained by calculating the Pareto optimal solution are matched with the three tasks, and the positioning task, detection task and fire extinguishing task are established as cooperative alliances, and the maximum contribution of each UAV to each task is calculated, thereby forming the optimal solution of the UAV task allocation

[0013] Further, the air-ground cooperative task allocation model in step S2 is defined as follows:

[0014] The formula of the objective function for maximizing the endurance is:

[0015]

[0016] Wherein, is the endurance; is the flight speed; is the fuel consumption rate per unit time; is the lift; is the resistance; is the initial weight; is the final weight

[0017] The formula of the objective function for maximizing the wind resistance is:

[0018]

[0019]

[0020] Wherein, for wind resistance ability; denotes the total number of wind force levels; denotes the first measurement of a certain wind force level under the nth wind force condition; denotes the nth test; is the true aerodynamic force of the nth measurement under the nth wind force condition; is a prediction model, which predicts the aerodynamic force according to the environmental conditions at the time, with respect to the input values; is a feature vector, which represents the environmental and UAV state parameters related to the aerodynamic force at the nth measurement under the nth wind force level; denotes the wind speed; denotes the wind direction; denotes the UAV airspeed; denotes the angle of attack; denotes the side slip angle; is a potential linear coefficient specific to each wind force condition;

[0021] The above constraints include:

[0022] Task uniqueness: ensure that each task is executed by at most one UAV;

[0023] UAV capability: ensure that the resource requirements of each task are within the UAV's bearing capacity, while not exceeding the UAV's load capacity;

[0024] airspace safety: at any time , the three-dimensional spatial distance , between any two UAVs must be greater than the minimum safety distance ;

[0025] Task order: for task pairs that have execution order requirements , it must be ensured that the completion time of task is earlier than the start time of task .

[0026] ​​​​​​Further, in step S4, the genetic algorithm uses a segmented chromosome numerical coding method with a double objective function as an index. The first half of the chromosome is a task allocation sequence, each gene position represents a task, and the gene value is an integer representing the number of the UAV assigned to execute the task. The second half is a sequence of task allocation order for each UAV, each gene position represents a task, and the gene value is a real number. When executing the task, the tasks are sorted in ascending order according to the real number value, and the sorted task order is the global execution order of all tasks. A part of the task allocation sequence uses uniform crossover, and another part uses exchange mutation. A part of the task order sequence uses sequential crossover, and another part uses inversion mutation. The non-dominated sorting and crowding calculation in the non-dominated sorting genetic algorithm framework are used for individual selection, and an external optimal set is used to save the Pareto optimal solution obtained each time.

[0027] Further, the first half uses an integer coding method to encode the UAV, and the total number of tasks is , each gene locus corresponds to a task , and the number of the UAV assigned to execute the task is , and if the task is not assigned, it is assigned a value of 0. The second half uses a real number coding method, and each gene locus corresponds to a task , and the gene value of each task is a uniformly distributed real random number in the interval . When executing the task, the tasks are sorted in ascending order according to the real random number, and the global finite execution order of all tasks is determined, thereby analyzing the individual task order sequence of each UAV.

[0028] Further, the first half of the integer type task allocation sequence uses uniform crossover and exchange mutation. In uniform crossover, each chromosome position is randomly filled with 0 or 1. When this position is 1, the offspring gene value inherits the parent gene value at this position. When this position is 0, the offspring gene value inherits the parent gene value at the next position, and the next offspring gene value returns to inherit the parent gene value at this position. In exchange mutation, two different gene positions in the sequence are randomly selected for exchange. If the exchanged chromosomes are the same, the combination result is unchanged. If the exchanged chromosomes are different, a new combination result is generated.

[0029] Further, the second half of the real type task order sequence uses simulated binary crossover. From two parent gene values and , the formula for calculating the offspring gene values and is:

[0030]

[0031]

[0032]

[0033] wherein, is an intermediate calculation variable; is a random number in the interval [0, 1], the random number is used to control the randomness of the crossover operation. is a distribution function, the distribution function controls the distribution characteristics of the crossover operation, the greater the value, the closer the generated offspring to the parent, the smaller the value, the more diversity of the generated offspring.

[0034] Further, the latter half of the real type task order sequence adopts a polynomial mutation, and the parent gene value The formula for calculating the offspring gene value e is:

[0035]

[0036]

[0037] wherein, is a mutation step length ratio factor, is an intermediate calculation variable; is a random number in the interval [0, 1], the random number is used to control the randomness of the mutation operation. is a distribution index, the distribution index controls the distribution characteristics of the mutation operation, the greater the value, the closer the generated offspring to the parent, the smaller the value, the more diversity of the generated offspring; is an upper limit of the gene value, take 1; is a lower limit of the gene, take 0.

[0038] Further, in step S4, a non-dominated sorting genetic algorithm model is used for environmental selection, all individuals in the merged population are stratified according to the Pareto dominance relationship, and for individuals in the same non-dominated layer The crowded distance on the objective function space is calculated to reflect the state of crowding, and the formula is:

[0039]

[0040] wherein, is the current individual; is the number of objective functions; is the individual The next neighbor individual In the The values ​​of each objective function; For individuals The previous neighbor individual In the The values ​​of each objective function; In the objective function Above, the maximum value of all individuals in the current layer; In the objective function The minimum value among all individuals in the current layer.

[0041] Furthermore, the obtained crowding distance is combined with the dual optimization objective function to rank all individuals in the merged race and select the top... Individuals form a new generation of population; at the end of each generation, the Pareto optimal solutions in the current population are compared with the solutions pre-stored in the archive, and the archive is updated using a fast non-dominated ordering and crowding comparison method to ensure that only the latest and evenly distributed Pareto optimal solution set is saved.

[0042] Furthermore, in step S6, The drone swarm formed by the above Pareto optimal solution. The three tasks—location, detection, and firefighting—form a cooperative game defined as follows: , Among them, the alliance For the set of all tasks, It is the characteristic function of the coalition, used to quantify the coalition. The value of collaborative task execution; the characteristic function The calculation formula is

[0043]

[0044] in, These are the weighting coefficients. This is used to adjust the weight of task performance and energy consumption in value assessment; The timeliness value of the task is a decay function; It is the first in the set of tasks. Sub-tasks; The probability of the alliance successfully completing the sub-task is a function of the matching degree between the capabilities of the drones within the alliance and the mission requirements. The total energy consumption expected to be used to execute the subtask.

[0045] The above-mentioned greedy strategy-based alliance formation algorithm first requires creating an empty task alliance structure. , Wherein each alliance initial value is empty, and the unmanned aerial vehicle is one-to-one corresponding to the task alliance according to the positioning ability, the detection ability and the fire extinguishing ability;

[0046] For each non-full task alliance and each unassigned unmanned aerial vehicle, the maximum contribution of the unmanned aerial vehicle after joining the alliance is calculated;

[0047] Among all the completed combinations of unmanned aerial vehicles and alliances, the combination with the maximum contribution of the unmanned aerial vehicle to the alliance is selected, and then the unmanned aerial vehicle and the alliance are formally formed, and the assigned unmanned aerial vehicle is removed from the total unmanned aerial vehicle group;

[0048] The above steps are repeated until all unmanned aerial vehicles are assigned, so that the unmanned aerial vehicle group with the highest efficiency for completing each task is obtained, that is, the optimal solution of the unmanned aerial vehicle task assignment.

[0049] A computer device comprises a memory, a processor and a computer program running on the memory, and the processor implements the steps of the unmanned aerial vehicle group forest fire low-altitude cooperative task assignment method when executing the above program.

[0050] Advantages:

[0051] The present application is a kind of unmanned aerial vehicle group forest fire ground-air cooperative task allocation method, through double target optimization, the performance of unmanned aerial vehicle is improved, then the crowding degree obtained by using non-dominated sorting genetic algorithm is combined with double target optimization function to obtain pareto optimal solution, so that the combination work efficiency of unmanned aerial vehicle can be maximized. Again through the alliance algorithm of greedy strategy, each unmanned aerial vehicle can better complete each task in the field of expertise, and improve the completion quality of each task. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 It is a flowchart of the present application;

[0053] Figure 2 It is a flowchart of the present application non-dominated sorting genetic algorithm combined with crowding degree and double optimization target function to find pareto optimal solution set;

[0054] Figure 3 It is a flowchart of the present application alliance algorithm based on greedy strategy, which calculates the maximum contribution of each unmanned aerial vehicle to each task;

[0055] Figure 4 It is a schematic diagram of the unmanned aerial vehicle used in the present application;

[0056] Figure 5 It is the use scenario of the present application. DETAILED DESCRIPTION

[0057] In order to make the technical purpose, innovative combination and actual effect of the present application clear and obvious, the following will be combined with Figures 1 to 5The specific embodiments are described and illustrated.

[0058] Embodiment 1:

[0059] A UAV swarm forest fire air-ground cooperative task allocation method combines Figure 1 As shown in the figure, the method comprises the following steps:

[0060] S1, obtaining the digital elevation model of the fire area, the fire point thermal map, the obstacle distribution and the airspace control information; defining a task allocation set to be executed, the task types including positioning tasks, detection tasks and fire extinguishing tasks; defining a UAV cluster, and each UAV in the UAV cluster has different load capacity and state attributes, such as Figure 4 and Figure 5 as shown in the figure;

[0061] S2, maximizing the endurance capability and the wind resistance capability of the UAV as the double optimization objective of the non-dominated sorting genetic algorithm, establishing an air-ground cooperative task allocation model, while satisfying multiple constraints such as task timing, UAV capability, airspace safety and task uniqueness, and forming a functionally mixed UAV swarm;

[0062] S3, calculating the double optimization objective function value of the non-dominated sorting genetic algorithm of each UAV in the UAV swarm, forming a UAV objective function set, and composing a plurality of UAV swarms with optimized performance;

[0063] S4, performing chromosome numerical encoding on the UAVs in the UAV swarm with optimized performance, and iteratively optimizing the UAVs by using the non-dominated sorting genetic algorithm based on simulated binary crossover and polynomial mutation, combining the crowding degree and the double optimization objective function value, and finding the Pareto optimal solution set of the UAV swarm combination mode;

[0064] S5, the different UAV swarms in the formed Pareto optimal solution set contain multiple UAVs with different efficiency values for positioning tasks, detection tasks and fire extinguishing tasks, and each UAV is allocated to each task;

[0065] S6, using a coalition formation algorithm based on a greedy strategy, matching the multiple UAV swarms obtained by calculating the Pareto optimal solution with the three tasks, establishing cooperative alliances for the positioning tasks, the detection tasks and the fire extinguishing tasks, respectively, calculating the maximum contribution of each UAV to each task, and thus forming the optimal solution for the UAV task allocation.

[0066] The air-ground cooperative task allocation model in step S2 is defined as follows:

[0067] The formula of the objective function for maximizing the endurance capability is as follows:

[0068]

[0069] where, is the endurance capability; is the flight speed; is the fuel consumption rate per unit time; is the lift; is the drag; is the initial weight; is the final weight

[0070] The objective function for maximizing the wind resistance capability is:

[0071]

[0072]

[0073] where, is the wind resistance capability; denotes the total number of wind force levels; denotes the total number of tests under the th wind force condition of the th level of wind force; denotes the th test; denotes the th measurement under the th wind force condition; is the prediction model, which predicts the aerodynamic force according to the environmental conditions at the time, with respect to the input is the feature vector, which represents the environmental and UAV state parameters related to the aerodynamic force at the th level of wind force and the th measurement; denotes the wind speed; denotes the wind direction; denotes the UAV airspeed; denotes the angle of attack; denotes the side slip angle; is the potential linear coefficient specific to each wind force condition;

[0074] The above constraints include:

[0075] Task uniqueness: ensure that each task is executed by at most one UAV;

[0076] UAV capability: ensure that the resource requirements of each task are within the UAV's bearing capacity, while not exceeding the UAV's load capacity;

[0077] airspace safety: at any time , the three-dimensional spatial distance , between any two UAVs Must be greater than the minimum safe distance ;

[0078] Task order: For task pairs that have execution order requirements. The task must be ensured The completion time was earlier than the task The start time.

[0079] In step S4, the genetic algorithm uses a bi-objective function as an indicator and employs a segmented chromosome numerical encoding method. The first half of the chromosome is the task allocation sequence, such as... Figure 2 As shown. Each gene position represents a task, and the gene value is an integer representing the drone number assigned to perform that task. The latter half is the task assignment sequence for each drone, where each gene position represents a task and the gene value is a real number. When executing tasks, the tasks are sorted in ascending order based on the real values, and the sorted task order is the global execution order of all tasks. A portion of the task assignment sequence is subjected to uniform crossover, and the other portion is subjected to exchange mutation. A portion of the task order sequence is subjected to sequential crossover, and the other portion is subjected to inversion mutation. Individual selection is performed using non-dominated sorting and crowding calculation in the non-dominated sorting genetic algorithm framework, and an external optimal set is maintained to store the Pareto optimal solution obtained each time.

[0080] The first part uses integer encoding to encode the drone, and the total number of gene tasks is [number missing]. Each locus Corresponding task The drones to be assigned to perform the missions are numbered as follows: If no task is assigned, it is assigned a value of 0; the latter part uses a real number encoding method, with each locus... Corresponding task The gene value for each task is in The tasks are arranged in ascending order based on the magnitude of these uniformly distributed real random numbers within the interval. This determines the global finite execution order of all tasks, thereby parsing the individual task sequence of each UAV.

[0081] In this process, the first half of the integer-type task allocation sequence is handled using uniform crossover and interchange mutation. In uniform crossover, each chromosome position is randomly filled with the number 0 or 1. When the position is 1, the offspring gene value inherits the parent gene value at that position. When the position is 0, the offspring gene value inherits the parent gene value at the next position, and the next offspring gene value returns the parent gene value inherited at that position. In interchange mutation, two different gene positions in the sequence are randomly selected and interchanged. If the interchanged chromosomes are the same, the combination result remains unchanged. If the interchanged chromosomes are different, a new combination result is generated.

[0082] wherein the latter half of the real type task order sequence adopts simulated binary crossover, and the child gene values and are calculated by the formula:

[0083]

[0084]

[0085]

[0086] wherein, is an expansion factor, and is an intermediate calculation variable; is a random number in the interval [0, 1], and the random number is used to control the randomness of the crossover operation; is a distribution function, and the distribution function controls the distribution characteristics of the crossover operation, the greater the value, the closer the generated child is to the parent, the smaller the value, the more diverse the generated child is.

[0087] wherein the latter half of the real type task order sequence adopts polynomial mutation, and the child gene value e is calculated by the formula:

[0088]

[0089]

[0090] wherein, is a mutation step size ratio factor, and is an intermediate calculation variable; is a random number in the interval [0, 1], and the random number is used to control the randomness of the mutation operation; is a distribution index, and the distribution index controls the distribution characteristics of the mutation operation, the greater the value, the closer the generated child is to the parent, the smaller the value, the more diverse the generated child is; is an upper limit of the gene value, taking 1; is a lower limit of the gene, taking 0.

[0091] wherein the non-dominated sorting genetic algorithm model is used for environmental selection, and all individuals in the merged population are stratified according to the Pareto dominance relationship. For individuals in the same non-dominated layer, ​​​The crowding distance in the objective function space is calculated to reflect the crowding level. The formula is as follows:

[0092]

[0093] in, For the current individual; The number of objective functions; For individuals The next neighbor individual In the The values ​​of each objective function; For individuals The previous neighbor individual In the The values ​​of each objective function; In the objective function Above, the maximum value of all individuals in the current layer; In the objective function The minimum value among all individuals in the current layer.

[0094] The obtained crowding distance is combined with the dual optimization objective function to rank all individuals in the merged race, and the top [individuals] are selected. Individuals form a new generation of population. At the end of each generation, the Pareto optimal solutions in the current population are compared with the solutions pre-stored in the archive. The archive is updated using a fast non-dominated ordering and crowding comparison method to ensure that only the latest and evenly distributed Pareto optimal solution set is saved.

[0095] In step S3, combined with Figure 3 As shown, The drone swarm formed by the above Pareto optimal solution. The three tasks—location, detection, and firefighting—form a cooperative game defined as follows: , Among them, the alliance For the set of all tasks, That is, the characteristic function of the coalition, used to quantify the coalition. The value of collaborative task execution; the characteristic function The calculation formula is

[0096]

[0097] in, These are the weighting coefficients. This is used to adjust the weight of task performance and energy consumption in value assessment; The timeliness value of the task is a decay function; It is the first in the set of tasks. Sub-tasks; The probability of successfully completing the sub-task by the alliance, i.e., the matching degree of the capabilities of the unmanned aerial vehicles in the alliance and the task requirements; The total energy consumption expected to be consumed for executing the sub-task.

[0098] The above-mentioned alliance formation algorithm based on the greedy strategy first needs to create an empty task alliance structure, wherein each alliance initially has a null value, and the unmanned aerial vehicles are one-to-one corresponding to the task alliance according to the positioning capability, the detection capability and the fire extinguishing capability;

[0099] For each non-full task alliance and each unassigned unmanned aerial vehicle, the maximum contribution brought by the unmanned aerial vehicle after joining the alliance is calculated;

[0100] Among all the completed combinations of the unmanned aerial vehicles and the alliances, the combination in which the unmanned aerial vehicle contributes the most to the alliance is selected, the unmanned aerial vehicle and the alliance are formally formed, and the assigned unmanned aerial vehicle is removed from the total unmanned aerial vehicle group;

[0101] The above steps are repeated until all the unmanned aerial vehicles are assigned, so as to obtain the unmanned aerial vehicle group with the highest efficiency for completing each task, which is the optimal solution of the task assignment of the unmanned aerial vehicle group.

[0102] Embodiment 2

[0103] The embodiment discloses a computer device, which comprises a memory, a processor and a computer program running on the memory, and the processor implements the steps of the unmanned aerial vehicle group forest fire air-ground cooperative task assignment method when executing the above-mentioned program.

[0104] Embodiment 3

[0105] The embodiment discloses a computer scale storage medium for storing a computer program, which is executed by a processor to implement the steps of the unmanned aerial vehicle group forest fire low-altitude cooperative task assignment method.

[0106] Specifically,

[0107] The method comprises five steps, the first step is to obtain a fire digital elevation model, a fire point thermal map, an obstacle distribution and airspace control information; a task assignment set to be executed is defined, and the task types include a positioning task, a detection task and a fire extinguishing task; and a unmanned aerial vehicle cluster is defined, and the unmanned aerial vehicles in the unmanned aerial vehicle cluster have different load capabilities and state attributes.

[0108] The second step is to maximize the endurance capability and the wind resistance capability as the dual optimization objectives of the non-dominated sorting genetic algorithm, to improve the performance of the unmanned aerial vehicles, so that the unmanned aerial vehicles form a functional mixed unmanned aerial vehicle group, as shown in Figure 5 .

[0109] Thirdly, the double-objective function value of each UAV is calculated to form a UAV objective function set, which constitutes the performance-optimized UAV group.

[0110] Fourthly, the UAVs in the performance-optimized UAV group are real-number coded, and a non-dominated sorting genetic algorithm based on improved simulated binary crossover and polynomial mutation is used for iterative optimization, and the Pareto optimal solution set is found by combining the crowding degree and the double-optimized objective function.

[0111] Fifthly, the different UAV groups in the formed Pareto optimal solution set contain multiple UAVs with different efficiency values for the positioning task, the detection task and the fire-extinguishing task, and each UAV is assigned to each task.

[0112] Sixthly, a coalition formation algorithm based on the greedy strategy is used to match the multiple UAV groups obtained by calculating the Pareto optimal solution with the three tasks, and cooperative coalitions are established for the positioning task, the detection task and the fire-extinguishing task, respectively, the maximum contribution of each UAV to each task is calculated, and thus the optimal solution for the UAV task assignment is formed.

Claims

1. A method for task allocation of forest fire ground-air cooperation of UAV swarm, characterized in that, It comprises the following steps: S1, obtaining a digital elevation model of a fire area, a fire point thermal map, an obstacle distribution and airspace control information; defining a task allocation set to be executed, the task types including positioning tasks, detection tasks and fire extinguishing tasks; and defining a UAV cluster, each UAV in the UAV cluster having different load capacity and state attributes; S2, maximizing the endurance capacity and the wind resistance of the UAVs as double optimization objectives of a non-dominated sorting genetic algorithm, establishing a ground-air collaborative task allocation model, and simultaneously satisfying multiple constraints of task timing, UAV capacity, airspace safety and task uniqueness to form a hybrid functional UAV group; S3, calculating the double optimization objective function value of the non-dominated sorting genetic algorithm of each UAV in the UAV group to form a UAV objective function set and constitute a plurality of performance-optimized UAV groups; S4, performing chromosome numerical coding on the UAVs in the performance-optimized UAV group, iteratively optimizing the UAVs by using a non-dominated sorting genetic algorithm based on simulated binary crossover and polynomial mutation, combining crowding degree and double optimization objective function value, and finding a Pareto optimal solution set of the UAV group combination mode; S5, forming a plurality of UAVs with different efficiency values for positioning tasks, detection tasks and fire extinguishing tasks in different UAV groups in the Pareto optimal solution set, and allocating each UAV to each task; S6, using a coalition formation algorithm based on a greedy strategy to match the plurality of UAV groups obtained by calculating the Pareto optimal solution with the three tasks, establishing cooperative alliances for the positioning tasks, detection tasks and fire extinguishing tasks, respectively, calculating the maximum contribution of each UAV to each task, and thus forming an optimal solution for UAV task allocation. 2.The method of claim 1, wherein, The ground-air collaborative task allocation model in step S2 is defined as follows: The formula of the objective function for maximizing the endurance capacity is: wherein, is the endurance capability; is the flight speed; is the fuel consumption rate per unit time; is the lift; is the drag; is the initial weight; is the final weight; The formula of the objective function for maximizing the wind resistance is: wherein, is the wind resistance capacity; denotes the total number of wind force levels; denotes the total number of tests performed under the wind force condition; denotes the test; is the true aerodynamic force measured at the wind force condition at the measurement; is the prediction model that predicts the aerodynamic force according to the environmental conditions at the time, with respect to the values of the inputs ; is the feature vector that represents the environmental and drone state parameters related to the aerodynamic force at the wind force level, at the measurement; denotes the wind speed; denotes the wind direction; denotes the drone airspeed; denotes the angle of attack; denotes the sideslip angle; is the latent linear coefficient specific to each wind force condition; The constraint conditions include: Task uniqueness: ensuring that each task is executed by at most one UAV; UAV capacity: ensuring that the resource demand of each task is within the bearing capacity of the UAV, and at the same time, not exceeding the load capacity of the UAV; Airspace security: at any time Any two drones Three-dimensional spatial distance between Must be greater than the minimum safe distance ; Task timing: for task pairs that have execution order requirements , the completion time of task must be ensured to be earlier than the start time of task .

3. The method of claim 1, wherein, In step S4, the genetic algorithm uses a segmented chromosome numerical coding mode with double objective functions as indexes, the first half of the chromosome is a task allocation sequence, each gene position represents a task, and the gene value is an integer representing the UAV number allocated to execute the task; the second half is a task sequence sequence of each UAV, each gene position represents a task, and the gene value is a real number; when executing a task, the real number value is sorted in ascending order, and the sorted task sequence is the global execution sequence of all tasks; one part of the task allocation sequence is uniformly crossed, and the other part is mutually varied; one part of the task sequence sequence is sequentially crossed, and the other part is inverted; the non-dominated sorting and crowding degree calculation in the non-dominated sorting genetic algorithm framework are used for individual selection, and an external optimal set is reserved to save the Pareto optimal solution obtained each time.

4. The method of claim 3, wherein, In step S4, the first half part adopts the method of integer encoding to encode the UAV, and the total number of genes of the task is Each gene locus corresponds to a task , and the UAV number to be assigned to execute the task is , and if the task is not assigned, it is assigned a value of 0; the second half part adopts the method of real number encoding, each gene locus corresponds to a task , and the gene value of each task is a real random number uniformly distributed in the interval , when executing the task, the tasks are arranged in ascending order according to the size of the real random number, and the global limited execution order of all tasks is determined, so as to analyze the individual task order sequence of each UAV.

5. The method of claim 4, wherein, The first half of the integer type task allocation sequence is allocated by uniform crossover and exchange mutation; in the uniform crossover, each chromosome position is randomly filled with 0 or 1, when the position is 1, the offspring gene value inherits the parent gene value of the position; when the position is 0, the offspring gene value inherits the parent gene value of the next position, the next offspring gene value returns to inherit the parent gene value of the position; in the exchange mutation, two different gene positions in the sequence are randomly selected for exchange, and the combination result is unchanged when the exchanged chromosomes are the same, and a new combination result is generated when the exchanged chromosomes are different.

6. The method of claim 4, wherein, The latter half of the real type task order sequence adopts analog binary crossover, by two parent gene values and The formula for calculating child gene values and is: wherein, is an expansion factor, is an intermediate calculation variable; is a random number in the interval [0, 1], the random number is used to control the randomness of the crossover operation; is a distribution function, the distribution function controls the distribution characteristics of the crossover operation, the larger the value, the closer the generated offspring to the parent, the smaller the value, the more diversity of the generated offspring.

7. The method of claim 4, wherein, The latter half of the real type of task order sequence adopts polynomial mutation, and the parent gene value The formula for calculating the child gene value e is: wherein, is a mutation step size scaling factor, is an intermediate calculation variable; is a random number in the interval [0, 1], the random number is used to control the randomness of the mutation operation; is a distribution exponent, the distribution exponent controls the distribution characteristics of the mutation operation, the larger the value, the closer the generated offspring to the parent, the smaller the value, the more diversity of the generated offspring; is an upper bound of the gene value, takes 1; is a lower bound of the gene, takes 0.

8. The method of claim 1, wherein, In step S4, the non-dominated sorting genetic algorithm is used for environment selection, all individuals in the combined population are stratified according to the Pareto dominance relationship, and the individuals in the same non-dominated layer are selected The crowding distance on the objective function space is calculated to reflect the state of crowding, and the formula is as follows: wherein, is the current individual; is the number of objective functions; is the individual next neighbor individual of the value on the th objective function; is the individual previous neighbor individual of the value on the th objective function; is the maximum value on the objective function over all individuals in the current layer; is the minimum value on the objective function over all individuals in the current layer; The crowded distance is combined with the bi-objective function to rank all individuals in the merged population, and the top individuals form the new population; at the end of each generation, the current Pareto-optimal solutions in the population are compared with the pre-stored solutions in the archive, and the archive is updated by using the fast non-dominated sorting and crowded comparison method to ensure that only the newest and uniformly distributed Pareto-optimal solution set is saved.

9. The method of claim 1, wherein, In step S6, a UAV group formed for a Pareto optimal solution, , the positioning task, the detection task, and the fire extinguishing task form a coalition defined as , where the coalition is a set of all tasks, is a characteristic function on the coalition, used to quantify the value of the coalition in the cooperative execution of tasks; the calculation formula of the characteristic function is wherein, is a weight coefficient, is used to adjust the proportion of task performance and energy consumption in the value evaluation; is the time value of the task, which is a decay function; is the i-th subtask in the set of tasks; is the i-th subtask in the set of tasks; is the probability of successfully completing the subtask by the alliance, i.e., a function of the matching degree of the capabilities of the UAVs in the alliance and the task requirements; is the total energy consumption expected to be consumed in executing the subtask; The union formation algorithm based on the greedy strategy first creates an empty task union, wherein each union initially has no value, and the unmanned aerial vehicle is one-to-one corresponding to the task union according to three indexes of positioning ability, detection ability and fire extinguishing ability; For each non-full task union and each unassigned unmanned aerial vehicle, the maximum contribution brought by the unmanned aerial vehicle to the union after joining the union is calculated; In all completed combinations of unmanned aerial vehicles and unions, the combination in which the unmanned aerial vehicle contributes most to the union is selected, the unmanned aerial vehicle and the union are formally formed, and the assigned unmanned aerial vehicle is removed from the total unmanned aerial vehicle group; The above steps are repeated until all unmanned aerial vehicles are assigned, so that the unmanned aerial vehicle group with the highest efficiency for completing each task is obtained, which is the optimal solution of the unmanned aerial vehicle task allocation.

10. A computer device comprising a memory, a processor and a computer program running on the memory, characterized in that, The processor implements the steps of the unmanned aerial vehicle group forest fire low-altitude cooperative task allocation method according to any one of claims 1-9 when executing the above program.

Images