Unmanned aerial vehicle cluster computing resource collaborative optimization method for edge intelligence

Abstract

The invention discloses an unmanned aerial vehicle cluster computing resource collaborative optimization method oriented to edge intelligence, and relates to the technical field of collaboration.The method comprises the steps that firstly, an unloading strategy of each piece of Internet of Things equipment is analyzed, the unloading transmission rate is calculated, then time delay and power consumption generated by task unloading are calculated, a reward function is constructed, unmanned aerial vehicles are selected according to the reward function, and the unmanned aerial vehicles are selected according to the reward function; constructing a replicator dynamic equation according to the unloading cost of each piece of Internet of Things equipment, carrying out the replacement of an unloading strategy, carrying out the iteration of each individual in the population, calculating the fitness of each individual after a preset iteration number threshold value, selecting the individual with the highest fitness, and taking the selected individual as the optimal position deployment decision of each unmanned aerial vehicle. The problems of low resource utilization efficiency and system performance reduction caused by dynamic scene change and load imbalance in the unmanned aerial vehicle cluster are solved, effective utilization of resources and system performance are guaranteed, and system cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of collaborative technologies, and more particularly to a method for collaborative optimization of computing resources for an unmanned aerial vehicle (UAV) cluster for edge intelligence. Background Art

[0002] With the rapid development of Internet of Things (IoT) technology and the wide deployment of 5G networks, a large number of computationally intensive and delay-sensitive tasks have emerged. However, due to the limited computing power and battery capacity of devices in the IoT, these tasks' requirements for high computing and low latency cannot be met. To address these issues, mobile edge computing is an effective way to meet these requirements. However, mobile edge computing servers are restricted by their static locations and cannot be deployed anywhere at any time, and the possibility of infrastructure being damaged by natural disasters is very high. Therefore, considering the flexibility of UAVs, multiple UAVs are introduced in mobile edge computing to provide flexible computing services for industrial IoT devices.

[0003] Traditional methods for collaborative optimization of computing resources for UAV clusters for edge intelligence mainly focus on minimizing system costs by optimizing the task offloading decisions and resource allocation of UAVs, or optimizing the deployment and flight trajectories of UAVs to further reduce system costs. Obviously, such collaborative computing methods within UAV clusters have at least the following deficiencies: 1. When traditional methods for collaborative optimization of computing resources for UAV clusters for edge intelligence reduce system costs, they optimize the offloading decisions of UAVs, resources, or UAV routes to reduce system costs, without considering the deployment of IoT devices and cannot determine whether the system cost is the lowest cost.

[0004] 2. The dynamically changing offloading requests of IoT devices result in significant unevenness in the load conditions of different UAVs. This load difference leads to a large difference in resource utilization rates among UAVs, thus limiting the further improvement of the overall system performance. Traditional collaborative computing methods within UAV clusters lack a solution to this problem and cannot ensure the effective utilization of resources and the performance of the system. Summary of the Invention

[0005] Aiming at the above-mentioned existing technical deficiencies, the purpose of the present invention is to provide a method for collaborative optimization of computing resources for an unmanned aerial vehicle (UAV) cluster for edge intelligence.

[0006] To solve the above technical problems, the present invention adopts the following technical solutions: A method for collaborative optimization of computing resources for an unmanned aerial vehicle (UAV) cluster for edge intelligence, comprising the following steps: Step 1, Task offloading: Analyze the offloading strategies of each IoT device, and each IoT device performs task offloading according to the offloading strategy, and calculate the task offloading transmission rate of each IoT device.

[0007] Step 2. Formulate collaborative computing decisions: After each IoT device offloads its tasks, each drone obtains the information of each other drone, and analyzes the collaborative computing decisions of the drones based on the information of each other drone.

[0008] Step 3. Update the offloading strategy: Calculate the offloading costs of each IoT device, construct a replicator dynamic equation for each IoT device according to the divided population, and determine whether the offloading strategy of each IoT device needs to be updated based on the replicator dynamic equation of each IoT device.

[0009] Step 4. Adjust the drone position deployment: Initialize the population, and call each individual obtained by the initialization as each parent individual. Mutate and cross each parent individual to obtain several offspring individuals, calculate the fitness of each parent individual and each offspring individual, and select the individual with the highest fitness as the optimal drone position deployment decision.

[0010] Preferably, the process of analyzing the offloading strategy of each IoT device is as follows: S11. Establish a three-dimensional coordinate system with the base station as the origin.

[0011] S12. Obtain the coordinates of each IoT device, the coordinates of each drone, and the service radius of each drone, and denote them as (x i , y i , 0), (X j , Y j , H j ), and R j , where i represents the number of each IoT device, i = 1, 2, 3,..., M, M represents the total number of IoT devices, j represents the number of each drone, j = 1, 2, 3,..., N, N represents the total number of drones, and i, j, M, and N are all positive integers.

[0012] S13. According to the analysis formula: Obtain the horizontal distance d ij from the i-th IoT device to the j-th drone. Taking each drone as the center, according to the analysis formula: Obtain the coverage radius d' j of the j-th drone. Compare the horizontal distance between each IoT device and each drone with the coverage radius of each drone. If the horizontal distance from a certain IoT device to a certain drone is less than the coverage radius of the drone, it means that the IoT device can choose to offload the task to the drone. If the horizontal distance from a certain IoT device to each drone is greater than the coverage radius of each drone, it means that the device can only offload the task to the base station.

[0013] Preferably, the task offloading transmission rate of each Internet of Things device is calculated as follows: If each Internet of Things device offloads tasks to the base station, obtain the transmission power of each Internet of Things device, the number of Internet of Things devices connected to the base station, and the bandwidth of the base station from the database, use a channel sounder to detect the signal power received by the base station and the signal power sent by each Internet of Things device, and according to the analysis formula: Obtain the channel gain h between the i-th Internet of Things device and the base station i0 , where A i represents the signal power sent by the i-th Internet of Things device, and A i ' represents the signal power received by the base station when the i-th Internet of Things device sends a signal: Then the task offloading transmission rate of each Internet of Things device is as follows:

[0014]

[0015] where B 0 represents the bandwidth of the base station, K 0 represents the number of Internet of Things devices connected to the base station, η i represents the transmission power of the i-th Internet of Things device, R i0 represents the task offloading transmission rate of the i-th Internet of Things device, and σ 2 represents the power of additive white Gaussian noise.

[0016] If each Internet of Things device offloads tasks to a certain drone, obtain the transmission power of each Internet of Things device from the database, and each drone obtains the number of connected Internet of Things devices and the bandwidth from the management module of the mobile edge computing system it carries. Then the task offloading transmission rate of each Internet of Things device is as follows:

[0017]

[0018] where B j represents the bandwidth of the j-th drone, K j represents the number of Internet of Things devices connected to the j-th drone, η i represents the transmission power of the i-th Internet of Things device, β represents the reference channel gain, and d ij represents the horizontal distance from the i-th Internet of Things device to the j-th drone, R ij represents the transmission rate at which the i-th Internet of Things device offloads tasks to the j-th drone, j represents the number of each drone, j = 1, 2, 3,..., N, N represents the total number of drones, and both j and N are positive integers.

[0019] Preferably, the collaborative task offloading decision is formulated as follows: S21. Define a matrix χ to represent the offloading decision of each Internet of Things device. When the i-th Internet of Things device offloads tasks to the j-th drone, χ = χij = 1. When the \(i\)-th Internet of Things device offloads a task to the base station, \(\chi=\chi\) i0 = 1.

[0020] S22. Define a matrix to represent the collaborative task offloading decision among drones. When the \(j\)-th drone offloads the \(i\)-th Internet of Things device task to the \(k\)-th drone, When the \(j\)-th drone has no Internet of Things device task, where \(k\) represents the numbers of other drones, \(k = 1, 2, 3,\cdots, N'\), and both \(k\) and \(N'\) are positive integers.

[0021] S23. Obtain the computing resources of each drone from the database. When each drone offloads its own Internet of Things device task to other drones, each drone exchanges data with other drones to obtain the computing resources and task queue information of other drones. At this time, construct a state space, denoted as \(s\) j , and the state space includes the computing resources and task queue information of each drone. Then the state space can be expressed as:

[0022] \(s\) j = \{F 1 , F 2 , F 3 ,\cdots, F j , T 1 , T 2 ,\cdots, T j \},

[0023] where \(T\) j represents the task queue information of the \(j\)-th drone, and \(F\) j represents the computing resources of the \(j\)-th drone. Obtain the number of cycles required for each historical Internet of Things device task from the database, and calculate the average value of the number of cycles required for each historical Internet of Things device task, and use it as the number of cycles required for each Internet of Things device task. According to the analysis formula: obtain the processing delay \(T\) ij ' and power consumption \(E\) i ' j of the \(j\)-th drone to process the \(i\)-th Internet of Things device task. In the formula, \(C\) i represents the number of cycles required for the \(i\)-th Internet of Things device task, \(K\) j represents the number of Internet of Things devices connected to the \(j\)-th drone, \(F\) j represents the computing resources of the \(j\)-th drone, and \(\kappa\) represents the capacitance of the effective switch.

[0024] S24. Construct the action space: Each drone executes the collaborative task offloading decision \(a\) j according to the observed state space:

[0025] a j ={u k , Q}, j≠k,

[0026] where u k represents that the j-th drone selects the k-th other drone as the target for collaborative task offloading, Q represents the number of offloading tasks from the j-th drone to the k-th other drone. The transmission rate between each drone is measured using a network speed measurement tool, the transmission power of each drone is obtained from the database, and the coordinates (X j , Y j , H j ) of each drone are obtained. Each drone obtains the data quantity of the tasks of the Internet of Things devices offloaded to itself from the mobile edge computing system carried. According to the analysis formula: the transmission delay T jk between the j-th drone and the k-th drone is obtained, where v jk represents the transmission rate between the j-th drone and the k-th drone. According to the analysis formula: the transmission delay T j ″ k and the transmission power consumption E j ″ k are obtained when the j-th drone offloads the task of the i-th Internet of Things device to the k-th drone. Here, D i represents the data quantity of the task of the i-th Internet of Things device, R jk represents the transmission rate between the j-th drone and the k-th drone, and p j represents the transmission power of the j-th drone.

[0027] S25. Construct a reward function: Calculate the reward when the j-th drone takes action a j in state s j :

[0028]

[0029] where r j represents the reward value after the j-th drone takes the collaborative computing action, w 1 , w 2 represent the weight coefficients of delay and power consumption respectively. Compare the calculated reward values. When the maximum reward value is selected, the system cost obtained after the drone adopts the collaborative offloading strategy is the lowest.

[0030] Preferably, the updated offloading strategy will be as follows: S31. Calculate the offloading delay of each Internet of Things device, denoted as T i and the power consumption, denoted as E i , according to the analysis formula: Ui = w 1 T i + w 2 E i Obtain the system cost U generated when the i-th Internet of Things device offloads a task i .

[0031] S32. Divide each Internet of Things device into various populations according to the three-dimensional coordinates of each Internet of Things device, the three-dimensional coordinates of each drone, and the coverage radius of each drone, and count the number of Internet of Things devices in each population.

[0032] S33. Construct a replicator dynamics equation for each Internet of Things device in each population:

[0033]

[0034] In the formula, δ represents the control coefficient, U a f represents the system cost generated when the a-th Internet of Things device in the f-th population offloads a task, C f represents the number of Internet of Things devices in the f-th population, represents the cost index of the a-th Internet of Things device in the f-th population, a represents the number of each Internet of Things device, a = 1, 2, 3,..., b, b represents the total number of Internet of Things devices, f represents the number of each population, f = 1, 2, 3,..., c, c represents the number of populations, and a, b, f, and c are all positive integers.

[0035] When the cost index of a certain Internet of Things device in a certain population is less than zero, it means that the offloading cost of this Internet of Things device is less than the average offloading cost of this population. At this time, this Internet of Things device does not need to change the offloading strategy. When the cost index of a certain Internet of Things device in a certain population is greater than zero, it means that the offloading cost of this Internet of Things device is higher than the average offloading cost of this population. At this time, this Internet of Things device needs to change the offloading strategy.

[0036] Preferably, the specific process of calculating the offloading delay and power consumption of each Internet of Things device is as follows: S41. Obtain the number of data in each Internet of Things device task from the database and obtain the transmission power of each Internet of Things device. According to the analysis formula: Obtain the transmission delay T i0 and transmission power consumption E i0 of the i-th Internet of Things device when offloading the task to the base station. In the formula, R i0 represents the task offloading transmission rate of the i-th Internet of Things device, and η i represents the transmission power of the i-th Internet of Things device.

[0037] S42. Obtain the computing resources of the base station and the number of Internet of Things devices connected to the base station from the database, and obtain the number of cycles required for each Internet of Things device task. According to the analysis formula: Obtain the processing delay T required for the task on the i-th Internet of Things device at the base station i ′ 0 , where C i represents the number of cycles required for the task of the i-th Internet of Things device, K 0 represents the number of Internet of Things devices connected to the base station, and F 0 represents the computing resources of the base station.

[0038] S43. Obtain the data volume of the tasks on each Internet of Things device from the database. According to the analysis formula: Obtain the transmission delay T ij and transmission power consumption E ij when the i-th Internet of Things device offloads the task to the j-th drone, where R ij represents the transmission rate when the i-th Internet of Things device offloads the task to the j-th drone, and D i represents the data volume of the task of the i-th Internet of Things device.

[0039] S44. Obtain the transmission power of each Internet of Things device from the database. According to the analysis formula: Obtain the processing delay T i ′ k and power consumption E i ′ k .

[0040] S45. Calculate the offloading delay and power consumption of each Internet of Things device:

[0041]

[0042] where T i represents the offloading delay of the i-th Internet of Things device, E i represents the offloading energy consumption of the i-th Internet of Things device, χ i0 represents the return value when the i-th Internet of Things device offloads the task to the base station, represents the return value when the j-th drone offloads the task of the i-th Internet of Things device to the k-th drone.

[0043] Preferably, the population division is performed as follows: Regarding the deployable decisions of the drone cluster as individuals, multiple individuals form a population, and the individual with the highest fitness in this population is selected as the current optimal deployment decision.

[0044] Preferably, the initialization of the population is carried out as follows: each individual in the population is a feasible solution to the deployment coordinates of each UAV in the population. This feasible solution consists of N genes, and each gene represents the deployment coordinate l of each UAV. j =(X j ,Y j ,H j ). If there are E individuals in the population, then each individual can be encoded as Ψ n ={l n1 ,l n2 ,...,l nN}. Then the initialization process of the population is a process of randomly generating E individuals, which can be expressed as:

[0045] Ψ nj =Ψ n ′+random(0,1)(Ψ n ″-Ψ n ′),

[0046] In the formula, Ψ nj represents the j-th gene of the n-th individual in the population, Ψ n ′ and Ψ n ″ respectively represent the lower limit of the deployable coordinates and the upper limit of the deployable coordinates, n represents the number of each individual, n = 1, 2, 3,..., E, E represents the total number of individuals, and both n and E are positive integers.

[0047] Preferably, according to the collaborative computing method in a UAV cluster described in claim 1, the mutation and crossover of each parent individual are carried out as follows: S51. Each individual obtained by initialization is called each parent individual, and the differential algorithm that can be selected by the mutation strategy is used to mutate the parent individuals to obtain each mutant individual. Three mutation strategies are provided in this algorithm: The formula of the first mutation strategy: In the formula represents the n-th 1 parent individual obtained in the g-th iteration, in the formula represents the n-th 2 parent individual obtained in the g-th iteration, in the formula represents the n-th 3 parent individual obtained in the g-th iteration, V r 1 (g + 1) represents the r-th mutant individual obtained in the (g + 1)-th iteration, r represents the number of each mutant individual, r = 1, 2, 3,..., I, I represents the total number of mutant individuals, and both r and I are positive integers, n 1 , n 2 and n 3 represent the numbers of each individual, n 1 , n2 , n 3 ∈ {1, 2, 3, ..., E} and n 1 ≠ n 2 ≠ n 3 , Q represents the mutation factor, g represents the number of iterations, g = 1, 2, 3, ..., G, and G represents the maximum number of iterations.

[0048] The formula for the second mutation strategy: where Ψ′(g) represents the individual vector with the best fitness in the g-th iteration, and V r 2 (g + 1) represents the r-th mutated individual obtained in the (g + 1)-th iteration.

[0049] The formula for the third mutation strategy: where Ψ n (g) represents the r-th parental individual obtained in the g-th iteration, and V r 3 (g + 1) represents the r-th mutated individual obtained in the (g + 1)-th iteration.

[0050] S52. When mutating each parental individual, it is necessary to select among the three mutation strategies. The selection function of the mutation strategy is expressed as:

[0051]

[0052] where f represents a random number, z represents the number of consecutive times the optimal fitness value of the neighboring generation is the same, ε represents the upper limit of the number of consecutive times the optimal fitness value of the neighboring generation is the same, and V r (g + 1) represents the formula of the selected mutation strategy.

[0053] S53. Cross the genes in each mutated individual with the genes in each parental individual to obtain the genes of each offspring individual:

[0054]

[0055] where T represents the preset crossover probability, C mj (g + 1) represents the j-th gene in the m-th offspring individual at the (g + 1)-th iteration, V rj (g + 1) represents the j-th gene in the r-th mutated individual at the (g + 1)-th iteration, and Ψ nj (g) represents the j-th gene in the n-th parental individual at the g-th iteration, j′ represents the number of a certain gene in a randomly selected mutated individual, m represents the number of each offspring individual, m = 1, 2, 3, ..., q, q represents the total number of offspring individuals, and both m and q are positive integers.

[0056] Preferably, calculate the fitness of each parent individual and each offspring individual, and select the individual with the highest fitness as the optimal UAV position deployment decision. The specific process is as follows: S61. Calculate the fitness of each offspring individual and each parent individual:

[0057]

[0058] In the formula, f(Ψ n ) represents the fitness of the nth parent individual when each UAV is deployed according to the coordinates in Ψ n , f(C m ) represents the fitness of the mth offspring individual when each UAV is deployed according to the coordinates in C m , represents the system cost generated when the ith IoT device offloads tasks when each UAV is deployed according to the coordinates in Ψ n , Ψ n represents the nth parent individual, represents the system cost generated when the ith IoT device offloads tasks when each UAV is deployed according to the coordinates in C m , C m represents the mth offspring individual, m represents the number of each offspring individual, m = 1, 2, 3,..., q, q represents the total number of offspring individuals, and both m and q are positive integers.

[0059] S62. Compare the fitness of each offspring individual with that of its parent. If the fitness of a certain offspring individual is higher than that of its parent individual, then use this offspring individual as the parent individual for the next round of iteration. If the fitness of a certain offspring individual is lower than that of its parent individual, then use this parent individual as the parent individual for the next round of iteration. Use this method to obtain each parent individual for each iteration. After a preset number of iterations, compare the fitness of each parent individual and each offspring individual, select the individual with the highest fitness, and deploy each UAV according to the coordinates in this individual.

[0060] The beneficial effects of the present invention are as follows: 1. The present invention provides a collaborative optimization method for computing resources of an unmanned aerial vehicle (UAV) cluster for edge intelligence. First, the offloading strategies of each Internet of Things (IoT) device are analyzed and the offloading transmission rate is calculated. Then, the delay and power consumption generated by task offloading are calculated, and a reward function is constructed. UAVs are selected according to the reward function. After that, a replicator dynamic equation is constructed based on the offloading cost of each IoT device, and the offloading strategy is replaced. Finally, each individual in the population is iterated. After passing the preset iteration number threshold, the fitness of each individual is calculated, and the individual with the highest fitness is selected and used as the optimal position deployment decision for each UAV, solving the problems of low resource utilization efficiency and system performance degradation caused by dynamic changes in the scenario and uneven load within the UAV cluster, ensuring the effective utilization of resources and the performance of the system, and reducing the system cost.

[0061] 2. When each IoT device offloads tasks, the present invention solves the problems of low resource utilization efficiency and system performance degradation caused by dynamic changes in the scenario and uneven load within the UAV cluster by jointly optimizing the deployment decisions of each UAV, the collaborative task offloading strategy among UAVs, and the task offloading decisions of each IoT device, ensuring the effective utilization of resources and the performance of the system.

[0062] 3. When each IoT device offloads tasks, the present invention divides each UAV into a population, initializes the population as individuals, mutates and crosses each individual. After passing the preset number of iterations, the fitness of each individual is calculated according to the offloading cost of each individual, and the individual with the highest fitness is selected and used as the optimal position deployment decision for each UAV, reducing the system cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0064] Figure 1 It is a schematic diagram of the system structure connection of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0066] Please refer to Figure 1 As shown, the present invention provides a collaborative computing method within a drone swarm, including the following steps: Step 1, task offloading: Analyze the offloading strategies of each Internet of Things device. Each Internet of Things device performs task offloading according to the offloading strategy, and calculates the task offloading transmission rate of each Internet of Things device.

[0067] In a specific embodiment, the process of analyzing the offloading strategies of each Internet of Things device is as follows: S11. Establish a three-dimensional coordinate system with the base station as the origin.

[0068] S12. Obtain the coordinates of each Internet of Things device, the coordinates of each drone, and the service radius of each drone, which are respectively denoted as (x i , y i , 0), (X j , Y j , H j ), and R j , where i represents the number of each Internet of Things device, i = 1, 2, 3,..., M, M represents the total number of Internet of Things devices, j represents the number of each drone, j = 1, 2, 3,..., N, N represents the total number of drones, and i, j, M, and N are all positive integers.

[0069] S13. According to the analysis formula: Obtain the horizontal distance d ij from the i-th Internet of Things device to the j-th drone. Taking each drone as the center, according to the analysis formula: Obtain the coverage radius d j ′ of the j-th drone. Compare the horizontal distance between each Internet of Things device and each drone with the coverage radius of each drone. If the horizontal distance from a certain Internet of Things device to a certain drone is less than the coverage radius of the drone, it means that the Internet of Things device can choose to offload the task to the drone. If the horizontal distance from a certain Internet of Things device to each drone is greater than the coverage radius of each drone, it means that the device can only offload the task to the base station.

[0070] It should be noted that when the horizontal distance from an Internet of Things device to a certain drone is greater than the coverage radius of the drone, the Internet of Things device is not within the service range of the drone, and at this time, the Internet of Things device cannot offload the task to the drone.

[0071] In another specific embodiment, the task offloading transmission rate of each Internet of Things device is calculated as follows: If each Internet of Things device offloads tasks to the base station, obtain the transmission power of each Internet of Things device, the number of Internet of Things devices connected to the base station, and the bandwidth of the base station from the database. Use a channel sounder to detect the signal power received by the base station and the signal power sent by each Internet of Things device. According to the analysis formula: Obtain the channel gain h between the i-th Internet of Things device and the base station i0 , where A i represents the signal power sent by the i-th Internet of Things device, and A i ' represents the signal power received by the base station when the i-th Internet of Things device sends a signal: Then the task offloading transmission rate of each Internet of Things device is as follows:

[0072]

[0073] where B 0 represents the bandwidth of the base station, K 0 represents the number of Internet of Things devices connected to the base station, η i represents the transmission power of the i-th Internet of Things device, R i0 represents the task offloading transmission rate of the i-th Internet of Things device, and σ 2 represents the power of additive white Gaussian noise.

[0074] It should be noted that the manufacturers of each Internet of Things device set the transmission power of each Internet of Things device during production.

[0075] It should also be noted that after each Internet of Things device is connected to the base station, it actively reports its own connection information to the base station. After the server of the base station receives these reported information, it makes statistics and records the number of Internet of Things devices connected to itself. The nameplate information on the base station device contains bandwidth information.

[0076] It should also be noted that the staff uses network detection software to collect relevant parameters of network transmission and uses power spectrum estimation technology to deduce the power of additive white Gaussian noise.

[0077] If each Internet of Things device offloads tasks to a certain drone, obtain the transmission power of each Internet of Things device from the database. Each drone obtains the number of connected Internet of Things devices and the bandwidth from the management module of the mobile edge computing system it carries. Then the task offloading transmission rate of each Internet of Things device is as follows:

[0078]

[0079] where B j represents the bandwidth of the j-th drone, K jrepresents the number of Internet of Things devices connected to the j-th drone, η i represents the transmission power of the i-th Internet of Things device, β represents the reference channel gain, d ij represents the horizontal distance from the i-th Internet of Things device to the j-th drone, R ij represents the transmission rate at which the i-th Internet of Things device offloads tasks to the j-th drone, j represents the number of each drone, j = 1, 2, 3,..., N, N represents the total number of drones, and both j and N are positive integers.

[0080] It should be noted that the manufacturer sets the bandwidth of each drone during the production of each drone.

[0081] It should also be noted that there is a list of devices connected to the drone in the mobile edge computing system, and the number of connected Internet of Things devices is obtained by querying the length of this list.

[0082] It should also be noted that when the horizontal distance from the Internet of Things device to the drone is 1 meter, the staff calculates the channel gain between the Internet of Things device and the drone and uses it as the reference channel gain.

[0083] Step 2: Make a collaborative computing decision: After each Internet of Things device offloads tasks, each drone obtains the information of each other drone and analyzes the collaborative computing decision of the drone based on the information of each other drone.

[0084] It should be noted that the information of each other drone includes the computing resources and bandwidth of each other drone.

[0085] In a specific embodiment, the process of making the collaborative task offloading decision is as follows: S21. Define a matrix χ to represent the offloading decision of each Internet of Things device. When the i-th Internet of Things device offloads tasks to the j-th drone, χ = χ ij = 1. When the i-th Internet of Things device offloads tasks to the base station, χ = χ i0 = 1.

[0086] S22. Define a matrix to represent the collaborative task offloading decision between drones. When the j-th drone offloads the task of the i-th Internet of Things device to the k-th drone, When there is no task of the Internet of Things device on the j-th drone, k represents the number of each other drone, k = 1, 2, 3,..., N′, and both k and N′ are positive integers.

[0087] S23. Obtain the computing resources of each UAV from the database. When each UAV offloads its Internet of Things device tasks to other UAVs, each UAV exchanges data with other UAVs to obtain the computing resources and task queue information of other UAVs. At this time, construct a state space, denoted as s j , and the state space includes the computing resources and task queue information of each UAV. Then the state space can be expressed as:

[0088] s j = {F 1 , F 2 , F 3 ,..., F j , T 1 , T 2 ,..., T j},

[0089] In the formula, T j represents the task queue information of the j-th UAV, and F j represents the computing resources of the j-th UAV. Obtain the number of cycles required for each historical Internet of Things device task from the database, and calculate the average value of the number of cycles required for each historical Internet of Things device task, and use it as the number of cycles required for each Internet of Things device task. According to the analysis formula: Obtain the processing delay T ij ' and power consumption E i ' j of the j-th UAV for processing the i-th Internet of Things device task. In the formula, C i represents the number of cycles required for the i-th Internet of Things device task, K j represents the number of Internet of Things devices connected to the j-th UAV, F j represents the computing resources of the j-th UAV, and κ represents the capacitance of the effective switch.

[0090] S24. Construct an action space: Each UAV executes a collaborative task offloading decision a j according to the observed state space:

[0091] a j = {u k , Q}, j ≠ k,

[0092] In the formula, u k represents that the j-th UAV selects the k-th other UAV as the target for collaborative task offloading, and Q represents the number of offloading tasks from the j-th UAV to the k-th other UAV. Use a network speed measurement tool to measure the transmission rate between each UAV, obtain the transmission power of each UAV from the database, and obtain the coordinates (X j , Y j , H j), each drone obtains the amount of data of the tasks of the Internet of Things devices unloaded to itself from the mobile edge computing system carried, and according to the analysis formula: The transmission delay T between the j-th drone and the k-th drone is obtained jk , where v jk represents the transmission rate between the j-th drone and the k-th drone. According to the analysis formula: The transmission delay T generated when the j-th drone offloads the task of the i-th Internet of Things device to the k-th drone is obtained j ″ k and the transmission power consumption E j ″ k , where D i represents the amount of data of the task of the i-th Internet of Things device, R jk represents the transmission rate between the j-th drone and the k-th drone, and p j represents the transmission power of the j-th drone.

[0093] S25. Construct a reward function: Calculate the reward when the j-th drone takes action a j in state s j :

[0094]

[0095] where r j represents the reward value after the j-th drone takes the collaborative computing action, and w 1 , w 2 represent the weight coefficient of delay and the weight coefficient of power consumption respectively. Compare the calculated reward values. When the maximum reward value is selected, the system cost obtained after the drone adopts the collaborative offloading strategy is the lowest.

[0096] It should be noted that this algorithm adopts an architecture design of centralized training - decentralized execution. In the training stage, each drone independently maintains a network for decision-making actions. At the same time, all drones share a global network to uniformly evaluate the state value using global state information. When making a collaborative task offloading decision, the drone needs to select a drone as the offloading target among other drones and at the same time determine the amount of tasks to be offloaded.

[0097] It also should be noted that the staff obtains the influence degree of delay and power consumption on the system cost through multiple experiments, and allocates the weight coefficient of delay and the weight coefficient of power consumption according to the influence degree. The higher the influence degree, the greater the allocated weight coefficient, and the sum of the weight coefficient of delay and the weight coefficient of power consumption is 1.

[0098] Step 3. Update the offloading strategy: Calculate the offloading cost of each IoT device, and construct a replicator dynamic equation for each IoT device according to the divided population. Determine whether the offloading strategy of each IoT device needs to be updated according to the replicator dynamic equation of each IoT device

[0099] In a specific embodiment, the updating of the offloading strategy is as follows: S31. Calculate the offloading delay of each IoT device, denoted as T i and the power consumption, denoted as E i , and according to the analysis formula: U i = w 1 T i + w 2 E i obtain the system cost U generated when the i-th IoT device offloads the task i .

[0100] S32. Divide each IoT device into various populations according to the three-dimensional coordinates of each IoT device, the three-dimensional coordinates of each drone, and the coverage radius of each drone, and count the number of IoT devices in each population.

[0101] S33. Construct a replicator dynamic equation for each IoT device in each population:

[0102]

[0103] In the formula, δ represents the control coefficient, U a f represents the system cost generated when the a-th IoT device in the f-th population offloads the task, C f represents the number of IoT devices in the f-th population, represents the cost index of the a-th IoT device in the f-th population, a represents the number of each IoT device, a = 1, 2, 3,..., b, b represents the total number of IoT devices, f represents the number of each population, f = 1, 2, 3,..., c, c represents the number of populations, and a, b, f, and c are all positive integers.

[0104] When the cost index of a certain IoT device in a certain population is less than zero, it means that the offloading cost of this IoT device is less than the average offloading cost of this population. At this time, this IoT device does not need to change the offloading strategy. When the cost index of a certain IoT device in a certain population is greater than zero, it means that the offloading cost of this IoT device is higher than the average offloading cost of this population. At this time, this IoT device needs to change the offloading strategy.

[0105] In the above, the calculation of the offloading delay and power consumption of each IoT device is as follows: S41. Obtain the data quantity in the task of each IoT device from the database and the transmission power of each IoT device. According to the analysis formula: Obtain the transmission delay \(T\) for the \(i\)-th Internet of Things device to offload the task to the base station i0 and the transmission power consumption \(E\) i0 , where \(R\) i0 represents the task offloading transmission rate of the \(i\)-th Internet of Things device, and \(\eta\) i represents the transmission power of the \(i\)-th Internet of Things device.

[0106] S42. Obtain the computing resources of the base station and the number of Internet of Things devices connected to the base station from the database, and obtain the number of cycles required for the tasks of each Internet of Things device. According to the analysis formula: Obtain the processing delay \(T'\) required for the task on the \(i\)-th Internet of Things device at the base station i ', 0 , where \(C\) i represents the number of cycles required for the task of the \(i\)-th Internet of Things device, \(K\) 0 represents the number of Internet of Things devices connected to the base station, and \(F\) 0 represents the computing resources of the base station.

[0107] S43. Obtain the data volume of the tasks on each Internet of Things device from the database. According to the analysis formula: Obtain the transmission delay \(T\) for the \(i\)-th Internet of Things device to offload the task to the \(j\)-th drone ij and the transmission power consumption \(E\) ij , where \(R\) ij represents the transmission rate for the \(i\)-th Internet of Things device to offload the task to the \(j\)-th drone, and \(D\) i represents the data volume of the task of the \(i\)-th Internet of Things device.

[0108] S44. Obtain the transmission power of each Internet of Things device from the database. According to the analysis formula: Obtain the processing delay \(T'\) of the task of the \(i\)-th Internet of Things device on the \(k\)-th other drone i ' k and the power consumption \(E\) i ' k .

[0109] S45. Calculate the offloading delay and power consumption of each Internet of Things device:

[0110]

[0111] where \(T\) i represents the offloading delay of the \(i\)-th Internet of Things device, \(E\) i represents the offloading energy consumption of the \(i\)-th Internet of Things device, \(\chi\) i0 represents the return value for the \(i\)-th Internet of Things device to offload the task to the base station, represents the return value for the \(j\)-th drone to offload the task of the \(i\)-th Internet of Things device to the \(k\)-th drone.

[0112] Among the above, the population division is carried out as follows: The deployable decision of the UAV cluster is taken as an individual, and multiple individuals form a population. The individual with the highest fitness in this population is selected as the current optimal deployment decision.

[0113] Step 4: Adjust the UAV position deployment: Initialize the population, and call each individual obtained by initialization as each parent individual. Mutate and cross each parent individual to obtain several offspring individuals. Calculate the fitness of each parent individual and each offspring individual, and select the individual with the highest fitness as the optimal UAV position deployment decision.

[0114] In a specific embodiment, the initialization of the population is carried out as follows: Each individual in the population is a feasible solution to the deployment coordinates of each UAV in the population. This feasible solution consists of N genes, and each gene represents the deployment coordinates l j =(X j , Y j , H j ). If there are E individuals in the population, then each individual can be encoded as Ψ n ={l n1 , l n2 ,..., l nN}. Then the initialization process of the population is a process of randomly generating E individuals, which can be expressed as:

[0115] Ψ nj =Ψ n '+random(0,1)(Ψ n ''-Ψ n '),

[0116] In the formula, Ψ nj represents the jth gene of the nth individual in the population, Ψ n ', Ψ n '' represent the lower limit and upper limit of the deployable coordinates respectively, n represents the number of each individual, n = 1, 2, 3,..., E, E represents the total number of individuals, and both n and E are positive integers.

[0117] It should be noted that random(0,1) represents randomly generating a decimal number, and this decimal number is between 0 and 1.

[0118] It should also be noted that the lower limit and upper limit of the deployable coordinates are determined according to the coverage radius of each UAV and the positions of each Internet of Things device.

[0119] In another specific embodiment, the mutation and crossover of each parent individual are carried out as follows: S51. Each individual obtained by initialization is called each parent individual. The differential algorithm that can be selected by the mutation strategy is used to mutate the parent individuals to obtain each mutant individual. Three mutation strategies are provided in this algorithm: The formula of the first mutation strategy: In the formula represents the nth 1 parent individual obtained in the gth iteration. In the formula represents the nth 2 parent individual obtained in the gth iteration. In the formula represents the nth 3 parent individual obtained in the gth iteration. V r 1 (g + 1) represents the rth mutant individual obtained in the (g + 1)th iteration. r represents the number of each mutant individual, r = 1, 2, 3,..., I, I represents the total number of mutant individuals, both r and I are positive integers, n 1 , n 2 and n 3 represent the numbers of each individual. n 1 , n 2 , n 3 ∈{1, 2, 3,..., E} and n 1 ≠n 2 ≠n 3 , Q represents the mutation factor, g represents the number of iterations, g = 1, 2, 3,..., G, G represents the maximum number of iterations.

[0120] The formula of the second mutation strategy: In the formula, Ψ′(g) represents the individual vector with the best fitness in the gth iteration. V r 2 (g + 1) represents the rth mutant individual obtained in the (g + 1)th iteration.

[0121] The formula of the third mutation strategy: In the formula, Ψ n (g) represents the rth parent individual obtained in the gth iteration. V r 3 (g + 1) represents the rth mutant individual obtained in the (g + 1)th iteration.

[0122] S52. When mutating each parent individual, it is necessary to select among the three mutation strategies. The selection function of the mutation strategy is expressed as:

[0123]

[0124] Where f represents a random number, z represents the number of consecutive identical times of the optimal fitness value of the neighboring generation, ε represents the upper limit of the number of consecutive identical times of the optimal fitness value of the neighboring generation, and V r (g + 1) represents the formula of the selected mutation strategy.

[0125] It should be noted that f is a decimal between 0 and 1.

[0126] It should also be noted that the staff determines the number of consecutive identical times of the optimal fitness value of the neighboring generation and the upper limit of the number of consecutive identical times of the optimal fitness value of the neighboring generation through experiments.

[0127] S53. Cross each gene in each mutant individual with each gene in each parental individual to obtain each gene of each offspring individual:

[0128]

[0129] Where T represents a preset crossover probability threshold, C mj (g + 1) represents the j-th gene in the m-th offspring individual at the (g + 1)-th iteration, V rj (g + 1) represents the j-th gene in the r-th mutant individual at the (g + 1)-th iteration, Ψ nj (g) represents the j-th gene in the n-th parental individual at the g-th iteration, j′ represents the number of a certain gene in a randomly selected mutant individual, m represents the number of each offspring individual, m = 1, 2, 3,..., q, q represents the total number of offspring individuals, and both m and q are positive integers.

[0130] It should be noted that when obtaining each mutant individual and parental individual from the database at each historical crossover, the crossover probability used is obtained, and the average value of the historical crossover probabilities is calculated and used as the preset crossover probability threshold.

[0131] In another specific embodiment, calculating the fitness of each parental individual and each offspring individual, and selecting the individual with the highest fitness as the optimal UAV position deployment decision, the specific process is: S61. Calculate the fitness of each offspring individual and each parental individual:

[0132]

[0133] Where f(Ψ n ) represents the fitness of the n-th parental individual when each UAV is deployed according to the coordinates in Ψ n , f(C m ) represents the fitness of the m-th offspring individual when each UAV is deployed according to the coordinates in C m , represents that each UAV is deployed according to Ψ nThe system cost generated when the i-th Internet of Things device offloads tasks during the deployment of each coordinate, Ψ n represents the n-th parental individual represents that each drone is deployed according to C m The system cost generated when the i-th Internet of Things device offloads tasks during the deployment of each coordinate in C m represents the m-th offspring individual, m represents the number of each offspring individual, m = 1, 2, 3,..., q, q represents the total number of offspring individuals, and both m and q are positive integers.

[0134] S62. Compare the fitness of each offspring individual with that of its parent. If the fitness of a certain offspring individual is higher than that of its parent individual, then use this offspring individual as the parental individual for the next round of iteration. If the fitness of a certain offspring individual is lower than that of its parent individual, then use this parent individual as the parental individual for the next round of iteration. Use this method to obtain the parental individuals for each iteration. After a preset number of iterations, compare the fitness of each parental individual and each offspring individual, select the individual with the highest fitness, and deploy each drone according to the coordinates in this individual.

[0135] In the embodiment of the present invention, first, analyze the offloading strategies of each Internet of Things device and calculate the offloading transmission rate, then calculate the delay and power consumption generated by task offloading, and construct a reward function. Select drones according to the reward function. Then, construct a replicator dynamics equation according to the offloading cost of each Internet of Things device and perform replacement of offloading strategies. Finally, iterate each individual in the population. After passing a preset number of iteration thresholds, calculate the fitness of each individual, select the individual with the highest fitness, and use it as the optimal position deployment decision for each drone, which solves the problems of low resource utilization efficiency and system performance degradation caused by dynamic changes in the scenario and unbalanced load within the drone cluster, ensures the effective utilization of resources and the performance of the system, and reduces the system cost.

[0136] The above content is only an example and explanation of the concept of the present invention. Those skilled in the art of this technology can make various modifications or supplements to the described specific embodiments or use similar methods for substitution, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. A method for collaborative optimization of computing resources of drone clusters for edge intelligence, characterized in that: The steps include: Step 1: Task offloading: Analyze the offloading strategy of each IoT device, each IoT device offloads tasks according to the offloading strategy, and calculate the task offloading transmission rate of each IoT device; Step 2: Make collaborative computing decisions: After each IoT device unloads the task, each drone obtains the information of other drones and analyzes the collaborative computing decisions of the drones based on the information of other drones; Step 3: Update the uninstallation strategy: Calculate the uninstallation cost of each IoT device, and construct a replicator dynamic equation for each IoT device based on the divided population. According to the replicator dynamic equation of each IoT device, determine whether the uninstallation strategy of each IoT device needs to be updated; Step 4: Adjust the position deployment of drones: Initialize the population, and call the individuals obtained by initialization the parent individuals. Perform mutation and crossover on each parent individual to obtain several offspring individuals. Calculate the fitness of each parent individual and each offspring individual, and select the individual with the highest fitness as the optimal drone position deployment decision.

2. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 1 is characterized in that: The specific process of analyzing the uninstallation strategy of each IoT device is as follows: S11, establishing a three-dimensional coordinate system with the base station as the origin; S12, obtain the coordinates of each IoT device, the coordinates of each drone, and the service radius of each drone, respectively recorded as (x i ,y i ,0)、(X j ,Y j ,H j ) and R j , where i represents the number of each IoT device, i=1,2,3,...,M, M represents the total number of IoT devices, j represents the number of each drone, j=1,2,3,...,N, N represents the total number of drones, i, j, M and N are all positive integers; S13, according to the analysis formula: Get the horizontal distance d from the i-th IoT device to the j-th drone ij , taking each drone as the center, according to the analysis formula: Get the coverage radius d of the jth drone j ′, according to the horizontal distance between each IoT device and each drone and the coverage radius of each drone, the horizontal distance between each IoT device and each drone is compared with the coverage radius of each drone. If the horizontal distance from an IoT device to a drone is smaller than the coverage radius of the drone, it means that the IoT device can choose to offload the task to the drone. If the horizontal distance from an IoT device to each drone is larger than the coverage radius of each drone, it means that the device can only offload the task to the base station.

3. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 2 is characterized in that: The task offloading transmission rate of each IoT device is described and calculated, and the specific process is as follows: If each IoT device offloads the task to the base station, the transmission power of each IoT device, the number of IoT devices connected to the base station, and the bandwidth of the base station are obtained from the database. A channel detector is used to detect the signal power received by the base station and the signal power sent by each IoT device. According to the analysis formula: Get the channel gain h between the i-th IoT device and the base station i0 , where A i represents the signal power sent by the i-th IoT device, A i ′ represents the signal power received by the base station when the i-th IoT device sends a signal: Then the task offloading transmission rate of each IoT device is as follows: Where B0 represents the bandwidth of the base station, K0 represents the number of IoT devices connected to the base station, and η i represents the transmission power of the ith IoT device, R i0 represents the task offloading transmission rate of the i-th IoT device, σ 2 represents the power of additive white Gaussian noise; If each IoT device offloads the task to a certain drone, the transmission power of each IoT device is obtained from the database, and each drone obtains the number and bandwidth of connected IoT devices from the management module of the mobile edge computing system. The task offloading transmission rate of each IoT device is as follows: Where B j represents the bandwidth of the jth UAV, K j represents the number of IoT devices connected to the jth drone, η i represents the transmission power of the ith IoT device, β represents the reference channel gain, and d ij represents the horizontal distance from the i-th IoT device to the j-th UAV, R ij represents the transmission rate of the i-th IoT device offloading the task to the j-th UAV, j represents the number of each UAV, j=1,2,3,...,N, N represents the total number of UAVs, and both j and N are positive integers.

4. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 2 is characterized in that: The specific process of formulating the collaborative computing strategy is as follows: S21. Define a matrix χ to represent the offloading decision of each IoT device. When the i-th IoT device offloads the task to the j-th drone, χ=χ ij =1, when the i-th IoT device offloads the task to the base station, χ = χ i0 =1; S22. Define a matrix To represent the collaborative task offloading decision between drones, when the jth drone offloads the i-th IoT device task to the k-th drone, When the jth drone does not have an IoT device task, k represents the serial number of each other UAV, k = 1, 2, 3, ..., N', k and N' are both positive integers; S23, obtain the computing resources of each drone from the database. When each drone offloads its own IoT device tasks to other drones, each drone exchanges data with other drones to obtain the computing resources and task queue information of other drones. At this time, a state space is constructed, denoted as s j , the state space includes the computing resources and task queue information of each drone, then the state space can be expressed as: s j ={F1,F2,F3,...,F j ,T1,T2,...,T j }, Where T j represents the task queue information of the jth UAV, F j Representing the computing resources of the jth drone, obtain the number of cycles required for each IoT device task in history from the database, and calculate the average number of cycles required for each IoT device task in history, and use it as the number of cycles required for each IoT device task, according to the analysis formula: Get the processing delay T of the jth drone processing the i-th IoT device task ij ′ and power consumption E i ' j , where C i represents the number of cycles required for the i-th IoT device task, K j represents the number of IoT devices connected to the jth drone, F j represents the computing resources of the jth UAV; S24, construct action space: each drone executes collaborative task offloading decision a according to the observed state space j : a j ={u k ,Q},j≠k, Where u k represents that the jth UAV selects the kth other UAV as the target of collaborative task offloading, Q represents the number of tasks offloaded by the jth UAV to the kth other UAV, the transmission rate between each UAV is measured using the network speed measurement tool, the transmission power of each UAV is obtained from the database, and the coordinates of each UAV (X j ,Y j ,H j ), each drone obtains the amount of data of IoT device tasks offloaded to itself from the mobile edge computing system it carries, according to the analysis formula: Get the transmission delay T between the jth UAV and the kth UAV jk , where v jk represents the transmission rate between the jth UAV and the kth UAV, according to the analytical formula: Get the transmission delay T generated when the j-th UAV offloads the task of the i-th IoT device to the k-th UAV j ″ k and transmission power consumption E j ″ k , where D i represents the amount of data for the i-th IoT device task, R jk represents the transmission rate between the jth UAV and the kth UAV, p j represents the transmission power of the jth UAV; S25. Construct reward function: Calculate the reward function when the jth drone is in state s j Take action a j Rewards: Where r j represents the reward value after the jth UAV takes collaborative computing action. w1 and w2 represent the weight coefficient of latency and the weight coefficient of power consumption respectively. The calculated reward values ​​are compared. When the largest reward value is selected, the system cost obtained by the UAV after adopting the collaborative unloading strategy is the lowest.

5. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 4 is characterized in that: The uninstallation policy will be updated as follows: S31. Calculate the unloading delay of each IoT device and record it as T i And the power consumption is recorded as E i , according to the analytical formula: U i =w1T i +w2E i Get the system cost U generated when the i-th IoT device offloads the task i ; S32, dividing the population according to the three-dimensional coordinates of each IoT device, the three-dimensional coordinates of each drone, and the coverage radius of each drone, and counting the number of IoT devices in each population; S33. Construct replicator dynamic equations for each IoT device in each group: Where δ represents the control coefficient, U a f represents the system cost incurred when the ath IoT device in the fth population offloads tasks, C f represents the number of IoT devices in the fth population, represents the cost index of the ath IoT device in the fth population, a represents the number of each IoT device, a=1,2,3,...,b, b represents the total number of IoT devices, f represents the number of each population, f=1,2,3,...,c, c represents the number of populations, and a, b, f and c are all positive integers; When the cost index of an IoT device in a certain population is less than zero, it means that the uninstallation cost of the IoT device is less than the average uninstallation cost of the population. At this time, the IoT device does not need to change the uninstallation strategy. When the cost index of an IoT device in a certain population is greater than zero, it means that the uninstallation cost of the IoT device is higher than the average uninstallation cost of the population. At this time, the IoT device needs to change the uninstallation strategy.

6. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 5 is characterized in that: The specific process of calculating the unloading delay and power consumption of each IoT device is as follows: S41. Obtain the amount of data in each IoT device task and the transmission power of each IoT device from the database according to the analysis formula: Get the transmission delay T of the i-th IoT device to offload the task to the base station i0 and transmission power consumption E i0 , where R i0 represents the task offloading transmission rate of the i-th IoT device, η i represents the transmission power of the i-th IoT device; S42. Obtain the computing resources of the base station and the number of IoT devices connected to the base station from the database, and obtain the number of cycles required for each IoT device task, according to the analysis formula: Get the processing delay T required for the task on the i-th IoT device on the base station i ′0, where C i represents the number of cycles required for the i-th IoT device task, K0 represents the number of IoT devices connected to the base station, and F0 represents the computing resources of the base station; S43. Obtain the data quantity of tasks on each IoT device from the database according to the analysis formula: Get the transmission delay T of the i-th IoT device offloading the task to the j-th drone ij and transmission power consumption E ij , where R ij represents the transmission rate of the i-th IoT device offloading tasks to the j-th UAV, D i Represents the number of data for the i-th IoT device task; S44. Obtain the transmission power of each IoT device from the database according to the analysis formula: Get the processing delay T of the i-th IoT device task on the k-th other drone i ' k and power consumption E i ' k ; S45. Calculate the unloading delay and power consumption of each IoT device: Where T i represents the offloading delay of the ith IoT device, E i represents the offloading energy consumption of the ith IoT device, χ i0 The return value representing the i-th IoT device offloading the task to the base station, The return value representing the j-th UAV offloading the task from the i-th IoT device to the k-th UAV.

7. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 5 is characterized in that And carry out population division, the specific process is as follows: The deployable decision of the drone cluster is regarded as an individual, and multiple individuals form a population. The individual with the highest fitness is selected from the population as the current optimal deployment decision.

8. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 2 is characterized in that: The specific process of initializing the population is as follows: Each individual in the population is a feasible solution θ for the deployment coordinates of each drone in the population. The feasible solution consists of N genes, each gene represents the deployment coordinates l of each drone. j =(X j ,Y j ,H j ), the population contains E individuals, then each individual can be encoded as Ψ n = {l n1 ,l n2 ,...,l nN }, then the population initialization process is the process of randomly generating E individuals, which can be expressed as: P nj =Ψ n ′+random(0,1)(Ψ n ″-Ψ n ′), Where Ψ nj represents the jth gene of the nth individual in the population, Ψ n ′、Ψ n ″ respectively represent the lower limit and the upper limit of the deployable coordinates, n represents the number of each individual, n=1,2,3,...,E, E represents the total number of each individual, and n and E are both positive integers.

9. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 1, characterized in that: The specific process of performing mutation and crossover on each parent individual is as follows: S51. Each individual obtained by initialization is called a parent individual. The parent individual is mutated using a differential algorithm with a selectable mutation strategy to obtain each mutant individual. Three mutation strategies are provided in the algorithm: The first mutation strategy formula: In the formula represents the n1th parent individual obtained in the gth iteration, where represents the n2-th parent individual obtained in the g-th iteration, where represents the n3th parent individual obtained in the gth iteration, V r 1 (g+1) represents the rth mutant individual obtained in the (g+1)th iteration, r represents the number of each mutant individual, r=1,2,3,...,I, I represents the total number of mutant individuals, r and I are both positive integers, n1, n2 and n3 represent the numbers of each individual, n1, n2, n3∈{1,2,3,...,E} and n1≠n2≠n3, Q represents the mutation factor, g represents the number of iterations, g=1,2,3,,...,G, G represents the maximum number of iterations; The second mutation strategy formula: Where Ψ′(g) represents the individual vector with the best fitness in the g-th iteration, V r 2 (g+1) represents the rth mutant individual obtained in the (g+1)th iteration; The third mutation strategy formula: Where Ψ n (g) represents the r-th parent individual obtained in the g-th iteration, V r 3 (g+1) represents the rth mutant individual obtained in the (g+1)th iteration; S52. When mutating each parent individual, three mutation strategies need to be selected. The selection function of the mutation strategy is expressed as: Where f represents a random number, z represents the number of times the optimal fitness value of the adjacent generation is the same in a row, ε represents the upper limit of the number of times the optimal fitness value of the adjacent generation is the same in a row, V r (g+1) represents the selected mutation strategy formula; S53, cross each gene in each variant individual with each gene of each parent individual to obtain each gene of each offspring individual: Where T represents the preset crossover probability, C mj (g+1) represents the jth gene in the mth offspring individual at the g+1th iteration, V rj (g+1) represents the jth gene in the rth mutant individual at the g+1th iteration, Ψ nj (g) represents the jth gene in the nth parent individual at the gth iteration, j′ represents the number of a gene in a randomly selected mutant individual, m represents the number of each offspring individual, m=1,2,3,...,q, q represents the total number of offspring individuals, and both m and q are positive integers.

10. The method for collaborative optimization of computing resources of drone clusters for edge intelligence according to claim 8, characterized in that: The fitness of each parent individual and each offspring individual is calculated, and the individual with the highest fitness is selected as the optimal drone location deployment decision. The specific process is: S61. Calculate the fitness of each offspring individual and each parent individual: Where f(Ψ n ) represents each drone according to Ψ n The fitness of the nth parent individual when deploying at each coordinate in, f(C m ) represents each drone according to C m The fitness of the mth offspring individual when deployed at each coordinate in, Represents each drone according to Ψ n The system cost incurred when the ith IoT device unloads tasks when deploying at each coordinate in , Ψ n represents the nth parent individual, Represents each drone according to C m The system cost generated when the ith IoT device offloads tasks when deploying at each coordinate in , C m represents the mth offspring individual, m represents the number of each offspring individual, m = 1, 2, 3, ..., q, q represents the total number of offspring individuals, and both m and q are positive integers; S62. Compare the fitness of each offspring individual with the fitness of its parent. If the fitness of a certain offspring individual is higher than that of its parent, then the offspring individual is used as the parent of the next iteration. If the fitness of a certain offspring individual is lower than that of its parent, then the parent is used as the parent of the next iteration. In this way, the parent of each iteration is obtained. After a preset number of iterations, the fitness of each parent and offspring individual is compared, and the individual with the highest fitness is selected, and each UAV is deployed according to the coordinates of the individual.

Images