Method for optimizing overall performance of unmanned aerial vehicle assisted mobile edge computing network

Abstract

The invention provides a method for optimizing the overall performance of an unmanned aerial vehicle assisted mobile edge computing network, and the method comprises the steps: firstly providing an unmanned aerial vehicle assisted mobile edge computing network joint optimization model which comprises the steps: designing a multi-dimensional evaluation model for the system capacity, the total energy consumption of an unmanned aerial vehicle, the total time delay of the system and the like; and constraint conditions such as unmanned aerial vehicle computing resources, user task delay, unmanned aerial vehicle and user matching relationship and the like are determined. Secondly, providing a multi-target fusion evaluation method for rapidly evaluating a target function and a self-adaptive constraint processing method for processing constraint conditions; and finally, providing a constrained multi-target particle swarm algorithm to optimize the overall performance of the unmanned aerial vehicle assisted mobile edge computing network.

Description

Technical Field

[0001] The present invention belongs to the cross - field of computational intelligence and UAV optimization, and particularly relates to a method for optimizing the overall performance of a UAV - assisted mobile edge computing network. Background Art

[0002] With the continuous increase in the application requirements of B5G and 6G, future networks will face extremely severe challenges in terms of frequency resources and air - interface technologies. Although the intensive deployment of ground base stations can alleviate the tight demand for communication resources, in the case of large - scale user aggregation or sudden disasters, ground base stations often become overloaded or even paralyzed. If temporary ground base stations are deployed, not only are there serious time - consuming and costly problems, but also the requirements for overloaded resource demand and high dynamics cannot be met. UAV - assisted base stations, due to their excellent flexibility and mobility, enable UAV - assisted mobile edge computing networks to offload part of the user's computing tasks to UAVs, thereby effectively reducing latency and saving the computing resources of user equipment. Therefore, UAV - assisted edge computing networks are very conducive to assisting ground base stations in achieving user offloading for sudden hotspots, universal coverage, and rapid service recovery. However, due to problems such as the resource limitations of UAVs and communication interference, how to reasonably optimize the UAV system to comprehensively improve the overall performance of the system is a very complex constrained multi - objective optimization problem. Designing a dedicated constrained multi - objective algorithm for the optimization research of UAV - assisted mobile edge computing networks has very important theoretical significance and practical significance. Summary of the Invention

[0003] Object of the Invention: The technical problem to be solved by the present invention is to provide a method for optimizing the overall performance of a UAV - assisted mobile edge computing network in view of the deficiencies of the prior art, including the following steps:

[0004] Step 1: Establish a perception model of a UAV - coordinated mobile edge computing network;

[0005] Step 2: Establish a joint optimization model of a UAV - assisted mobile edge computing network;

[0006] Step 3: Establish a constrained multi - objective particle swarm algorithm;

[0007] Step 4: Use the constrained multi - objective particle swarm algorithm to optimize the joint optimization model of the UAV - assisted mobile edge computing network.

[0008] Step 1 includes: In a 3D Cartesian coordinate system, d is the Euclidean distance from the UAV to the user, h is the flight altitude of the UAV, R is the coverage radius of the UAV, θ is the pitch angle between the UAV and the user, and r is the Euclidean distance from the projection of the UAV on the ground to the user.

[0009] Step 2 includes:

[0010] Step 2-1: Establish the system capacity function;

[0011] Step 2-2: Establish the system delay function;

[0012] Step 2-3: Establish the energy consumption function of the UAV system;

[0013] Step 2-4: Establish a joint optimization model for the mobile edge computing network assisted by UAVs.

[0014] Step 2-1 includes: Each user can only connect to one UAV, and one UAV can simultaneously process the offloading data from more than two users. Let \(L_{ij}=1\) indicate that the \(i\)-th user and the \(j\)-th UAV have established a connection, and \(L_{ij}=0\) indicate that the \(i\)-th user and the \(j\)-th UAV have not established a connection; ij Let \(L_{ij}=1\) indicate that the \(i\)-th user and the \(j\)-th UAV have established a connection, ij and \(L_{ij}=0\) indicate that the \(i\)-th user and the \(j\)-th UAV have not established a connection;

[0015] In the established air-ground communication model, the line-of-sight transmission probability is:

[0016]

[0017]

[0018] where \(a\) and \(b\) are path loss parameters determined by the environment; \(P_{los}\) is the line-of-sight transmission probability, and exp is the exponential function; LoS \(P_{los}\) is the line-of-sight transmission probability, and exp is the exponential function;

[0019] The path loss in air-ground communication is divided into line-of-sight propagation loss \(PL_{los}\) LoS and non-line-of-sight propagation loss \(PL_{nlos}\), NLoS as shown in formulas (3) and (4) respectively:

[0020]

[0021]

[0022] where \(f\) c is the carrier frequency, \(c\) is the speed of light, \(\eta_{los}\) LoS and \(\eta_{nlos}\) NLoS are the additional path losses added in free-space propagation for line-of-sight transmission and non-line-of-sight transmission respectively;

[0023] The average path loss \(\overline{PL}_{ij}\) between user \(i\) and UAV \(j\) is established as: as:

[0024]

[0025] \(P_{nlos}=1 - P_{los}\) LoS \(P_{nlos}=1 - P_{los}\) NLoS (6)

[0026] P NLoS is the non-line-of-sight transmission probability;

[0027] Model the signal-to-noise ratio SNR when user i receives the signal from drone j as: i,j Modeled as:

[0028]

[0029] where indicates that user i has established communication with drone j and the transmission power of drone j is the maximum, and σ 2 is the noise power, represents the received power of user i from drones other than drone j, and M is the number of users;

[0030] The channel capacity C between user i and drone j ij and the total system channel capacity C total are respectively:

[0031] C ij = Blog(1 + SNR i,j ) (8)

[0032]

[0033] where B is the channel bandwidth and N is the number of drones; T snr is the set first signal-to-noise ratio threshold, that is, user i only communicates with drone j whose signal-to-noise ratio is greater than SNR th ; when SNR i,j ≥ SNR th , T snr = 1, otherwise, T snr = 0; at the same time, another communication distance threshold T dis is set, that is, user i only communicates with drone j whose distance is less than d th ; when d i,j ≥ SNR th , T dis = 1; otherwise, T dis = 0.

[0034] Step 2-2 includes: The drone system delay includes local delay, offloading delay, and drone computing delay;

[0035] Local delay: The time required for local computing of part of the data of user i is

[0036]

[0037] where α is the offloading coefficient, indicating the ratio of user data offloaded to the drone, is the total data volume of user i, C unit is the number of CPU cycles required for unit data, and f0 is the local computing frequency;

[0038] Offloading delay: After user i establishes communication with drone j, the transmission time when user i offloads part of the data to drone j is:

[0039]

[0040] Drone computing delay: The time required for drone j to compute the offloaded data of user i is:

[0041]

[0042] In the formula, C unit is the number of CPU cycles required for unit data, f ij is the computing frequency allocated by drone j to user i;

[0043] The total system delay T total is:

[0044]

[0045] Step 2-3 includes: When the user offloads data, the energy consumption generated by the drone includes: the energy consumption for local computing of user data, the communication energy consumption generated when the drone receives the offloaded data from the user, and the energy consumption generated when the drone processes the offloaded data of the user;

[0046] The energy consumption for local computing of user i's data is:

[0047]

[0048] where k is the energy consumption factor;

[0049] The communication energy consumption for drone j to receive the offloaded data of user i is:

[0050]

[0051] where is the offloading power;

[0052] The energy consumption generated when drone j computes the offloaded data of user i is:

[0053]

[0054] The total system energy consumption E totalis as follows:

[0055]

[0056] Step 2-4 includes: The jointly optimized model of the UAV-assisted mobile edge computing network is:

[0057]

[0058] where C1 represents the variables to be optimized, including the position (x, y, h) of the UAV, the computing resources and offloading coefficients allocated by the UAV to users; C2 represents that the tasks need to be completed within the total delay T, and C3 ensures that the total frequency allocated by the UAV to users is within the total frequency range of the UAV; C4 ensures that each user can only connect to one UAV.

[0059] Step 3 includes:

[0060] Step 3-1, initialize the population: Randomly generate the initial population, which altogether includes N P particles, and each particle X i =[x i1 , y i1 , h i1 , …, x iN , y iN , h iN , f i1 , …, f iM , α i1 , …, α iM ; where x iN , y iN , h iN are the three-dimensional coordinates of the UAV, f iM is the computing frequency allocated by the UAV to the i-th user, and α iM is the offloading coefficient of the i-th user;

[0061] The update method of the particle swarm is:

[0062]

[0063] X i (t)=X i (t - 1)+V i (t) (20)

[0064] where i = 1, 2, …, N p , N p is the population size, t is the generation of evolution, ω is the particle velocity weight, c1 and c2 are learning factors used to balance global exploration and local exploitation, and r1 and r2 are random numbers between 0 and 1; V i(t) is the movement speed of particle i at the current evolutionary generation t; X i (t) is the position of particle i at the current evolutionary generation t; P i (t - 1) is the optimal position of the i-th particle in the (t - 1)-th generation, and G(t - 1) is the optimal position of the particle population in the (t - 1)-th generation;

[0065] Step 3-3, multi-objective fusion evaluation method for particles: The calculation formula for the fitness Fit(X) of each particle X is:

[0066]

[0067] where, G max is the maximum number of evolutionary generations, X is an individual in the population, γ(X) is the angle between individual X and its nearest individual Y in the Euclidean space, and d(X) is the norm of the normalized objective vector of X:

[0068]

[0069]

[0070] where, O is the number of objectives, is the normalized objective vector; are respectively the norm of the normalized objective vector of individual X and the norm of the normalized objective vector of individual Y;

[0071] Step 3-4, perform constraint processing on the particles;

[0072] Step 3-5, determine whether the maximum number of evolutionary generations G max is reached. If so, output the particle with the optimal fitness value in the particle population; otherwise, go to Step 3-2.

[0073] In Step 3-3, the calculation formula of

[0074]

[0075] In Step 3-4, the constraint processing method for particles is:

[0076]

[0077] where, G(X1) and G(X2) are respectively the constraint violation degrees of particle X1 and particle X2, and ε is a very small positive constant.

[0078] Beneficial effects: By establishing a performance model for comprehensively evaluating the performance of a drone-assisted mobile edge computing network, the comprehensive performance of the mobile edge computing network is fully reflected; on this basis, a fast and efficient constrained multi-objective particle swarm optimization algorithm is proposed to solve the above-mentioned complex and time-consuming multi-objective and multi-constraint optimization problem; ultimately, the total capacity of the mobile edge computing network is greatly enhanced, while the total energy consumption and total delay are significantly reduced, achieving the goal of comprehensively improving the performance of the drone-assisted mobile edge computing network. The invention can provide solid theoretical guidance and technical support for the optimal deployment of drones in practical scenarios such as large-scale user aggregation and emergency communication after disasters. Brief Description of the Drawings

[0079] The following further specifically describes the present invention in conjunction with the drawings and specific embodiments, and the above and / or other advantages of the present invention will become clearer.

[0080] Figure 1 It is a schematic diagram of a drone communication system.

[0081] Figure 2 It is a schematic structural diagram of a constrained multi-objective particle swarm optimization algorithm for optimizing a drone-assisted mobile edge computing network disclosed in an embodiment of the present invention. Specific Embodiments

[0082] The present invention provides a method for optimizing the overall performance of a drone-assisted mobile edge computing network. First, a joint optimization model for a multi-drone-assisted mobile edge computing network is proposed, specifically including the establishment of multi-dimensional evaluation models such as system capacity, total drone energy consumption, and total system delay, as well as the construction of constraint conditions such as drone computing resources, user task delay, and the matching relationship between drones and users; a multi-objective fusion evaluation method for quickly evaluating the performance of the objective function and an adaptive constraint processing technology for processing constraint conditions are proposed; finally, a constrained multi-objective particle swarm optimization algorithm is proposed to optimize the performance of the drone-assisted mobile edge computing network. The method includes the following steps:

[0083] Step 1: Establish a perception model for a drone-assisted mobile edge computing network, as Figure 1 shown. In a 3D Cartesian coordinate system, d is the Euclidean distance from the drone to the user, h is the flight altitude of the drone, R is the coverage radius of the drone, θ is the pitch angle between the drone and the user, and r is the Euclidean distance from the projection of the drone on the ground to the user.

[0084] (Step 2: According to Step 1, establish a joint optimization model for a drone-assisted mobile edge computing network)

[0085] Step 2: Establish a joint optimization model for the UAV-assisted mobile edge computing network, including a multi-dimensional evaluation model for objectives such as system capacity, total UAV energy consumption, and total system latency, as well as the construction of constraint conditions such as UAV computing resources, user task latency, and the matching relationship between UAVs and users, providing a theoretical guarantee for optimizing UAV performance.

[0086] Step 2.1: Establish the system capacity function. Each user can only be connected to one UAV, while one UAV can simultaneously process the offloaded data from multiple users. Let L ij = 1 indicate that the i-th user and the j-th UAV are connected, and L ij = 0 indicates that they are not yet connected. Generally, the signal transmission between UAVs and users is divided into two categories: line-of-sight transmission and non-line-of-sight transmission. In line-of-sight transmission, the wireless signal propagates in a straight line between the user and the UAV. In contrast, in non-line-of-sight transmission, the wireless signal will be reflected or scattered due to obstacle occlusion. Therefore, the probability of line-of-sight transmission in the established air-ground communication model is:

[0087]

[0088]

[0089] In the formula, θ is the elevation angle of line-of-sight transmission, and a and b are path loss parameters determined by the environment, taking values of 9.61 and 0.16 respectively. Therefore, the path loss in air-ground communication can also be divided into line-of-sight propagation loss and non-line-of-sight propagation loss, as shown in equations (3) and (4) respectively.

[0090]

[0091]

[0092] In the formula, d is the distance between the user and the UAV, f c is the carrier frequency, c is the speed of light, η LoS and η NLoS are the additional path losses in free space propagation for line-of-sight transmission and non-line-of-sight transmission respectively. Therefore, the average path loss between user i and UAV j is established as:

[0093]

[0094] P LoS = 1 - P NLoS (6)

[0095] Considering that the user will be interfered by other UAVs when offloading data, the signal-to-noise ratio when user i receives the signal from UAV j is modeled as:

[0096]

[0097] In the formula, indicates that user i has established communication with drone j, and the transmission power of drone j is the maximum, and σ 2 is the noise power, represents the received power of user i from drones other than drone j, that is, it is considered as the received interference power. Therefore, the channel capacity between user i and drone j and the total system channel capacity are established as follows:

[0098] C ij = Blog(1 + SNR i,j ) (8)

[0099]

[0100] In the formula, B is the channel bandwidth, N is the number of drones, and M is the number of users. To ensure the communication quality between users and drones, T snr is the first set signal-to-noise ratio threshold, that is, user i only communicates with drone j whose signal-to-noise ratio is greater than SNR th . When SNR i,j ≥SNR th , T snr = 1; otherwise, T snr = 0. At the same time, another communication distance threshold T dis is set, that is, user i only communicates with drone j whose distance is less than d th . When d i,j ≥SNR th , T dis = 1; otherwise, T dis = 0.

[0101] Step 2.2: Establish the system delay function. Since the local computing power and battery capacity of users are insufficient, the ground equipment will retain some data for local computing and offload other data to the drone edge computing server with powerful computing power for processing. Therefore, the established drone system delay mainly includes local delay, offloading delay, and drone computing delay.

[0102] Local delay: The time required for local computing of part of the data of user i is:

[0103]

[0104] In the formula, α is the offloading coefficient, indicating the ratio of user data offloaded to the drone, is the total data volume of user i, C unit is the number of CPU cycles required for unit data, and f0 is the local computing frequency.

[0105] Offloading delay: After user i establishes communication with drone j, the transmission time when user i offloads part of the data to drone j is:

[0106]

[0107] where α is the offloading coefficient, is the total data volume of user i, and C ij is the channel capacity between user i and drone j.

[0108] Drone computing delay: The time required for drone j to compute the offloaded data of user i is:

[0109]

[0110] where α is the offloading coefficient, is the total user data volume, and C unit is the number of CPU cycles required for unit data, and f ij is the computing frequency allocated by drone j to user i.

[0111] In summary, the total system delay established is:

[0112]

[0113] where M is the number of users.

[0114] Step 2.3: Establish the energy consumption function of the drone system. When executing user offloaded data, the energy consumption generated by the drone mainly includes three aspects: the energy consumption for local computing of user data, the communication energy consumption generated by the drone receiving user offloaded data, and the energy consumption generated by the drone processing user offloaded data.

[0115] The energy consumption for local computing of user i's data is:

[0116]

[0117] where k is the energy consumption factor, f0 is the local computing frequency, is the local computing delay of user i, α is the offloading coefficient, is the total user data volume, and C unit is the number of CPU cycles required for unit data.

[0118] The communication energy consumption of drone j receiving user i's offloaded data is:

[0119]

[0120] where is the offloading delay, is the offloading power.

[0121] The energy consumption generated by the UAV j for calculating the offloading data of user i is:

[0122]

[0123] Where k is the energy consumption factor, is the UAV calculation delay, and f ij is the computing frequency allocated by UAV j to user i.

[0124] In summary, the total energy consumption of the established system is:

[0125]

[0126] Where M is the number of users.

[0127] Step 2.4: Establish a joint optimization model for the UAV-assisted mobile edge computing network:

[0128]

[0129] Where C1 represents the variables to be optimized, including the position (x, y, h) of the UAV, the computing resources (f) allocated by the UAV to users, and the offloading coefficient (α); C2 indicates that the task needs to be completed within the total delay T, and C3 ensures that the total frequency allocated by the UAV to users is within the total frequency range of the UAV; C4 ensures that each user can only connect to one UAV.

[0130] (Step 3 According to Step 2, a constrained multi-objective particle swarm optimization algorithm is proposed to optimize the UAV-assisted mobile edge computing network model.)

[0131] Step 3: Propose a constrained multi-objective particle swarm optimization algorithm to optimize the UAV-assisted mobile edge computing network model. It mainly includes: the update method for the particle swarm, the multi-objective fusion evaluation method for particles, and the constraint processing method for particles.

[0132] Step 3.1: Initialize the population. Randomly generate the initial population, which includes a total of N P particles, and each particle X i =[x i1 ,y i1 ,h i1 ,…,x iN ,y iN ,h iN ,f i1 ,…,f iM ,α i1 ,…,α iM .

[0133] Step 3.2: Particle swarm update method. To continuously generate new particles and thus promote the continuous evolution of the particle population towards the optimal direction, a particle swarm update method needs to be designed, specifically as follows:

[0134] V i (t) = ωV i (t - 1) + c1r1(P i (t - 1) - X i (t - 1))

[0135] + c2r2(G(t - 1) - X i (t - 1)) (19)

[0136] X i (t) = X i (t - 1) + V i (t) (20)

[0137] In the formula, i = 1, 2, …, N p , N P is the population size, t is the evolutionary generation, ω is the particle velocity weight, c1 and c2 are learning factors used to balance global exploration and local exploitation, and r1 and r2 are random numbers between 0 and 1 to ensure the randomness of the search. V i (t) is the movement velocity of particle i at the current evolutionary generation t. X i (t) is the position of particle i at the current evolutionary generation t.

[0138] Step 3.3: Multi-objective fusion evaluation method for particles. To quickly evaluate the advantages and disadvantages of different particles, the fitness value of the particles needs to be calculated. The fitness calculation method for each particle X is as follows:

[0139]

[0140] In the formula, t is the evolutionary generation, G max is the maximum evolutionary generation, X is the population individual, γ(X) is the angle between individual X and its nearest individual Y in the Euclidean space, and d(X) is the normalized objective vector norm of X:

[0141]

[0142] In the formula, X and Y represent individuals in the evolutionary population.

[0143]

[0144] In the formula, O is the number of objectives, is the normalized objective vector:

[0145]

[0146] Step 3.4: Particle constraint handling method. To effectively handle the constraint condition problem and thus ensure the feasibility of particles, a constraint handling technique is proposed to enable a reasonable comparison between feasible particles and infeasible particles, thereby ensuring the evolution of the particle population towards the optimal feasible direction:

[0147]

[0148] In the formula, G(X1) and G(X2) are the constraint violation degrees of particles X1 and X2.

[0149] Step 3.5: Determine whether the maximum number of evolutionary generations G has been reached max If so, output the particle with the optimal fitness value in the particle population; otherwise, go to Step 3.2.

[0150] The present invention provides a method for optimizing the overall performance of a drone-assisted mobile edge computing network. There are many methods and ways to specifically implement this technical solution. The above description is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. Each component not clearly defined in this embodiment can be implemented by existing technologies.

Claims

1. A method for optimizing the overall performance of a drone-assisted mobile edge computing network, characterized in that, It includes the following steps: Step 1, establish a mobile edge computing network sensing model for UAV cooperation; Step 2, establish a joint optimization model for the mobile edge computing network assisted by UAVs; Step 3, establish a constrained multi-objective particle swarm optimization algorithm; Step 4, use the constrained multi-objective particle swarm optimization algorithm to optimize the joint optimization model for the mobile edge computing network assisted by UAVs.

2. The method according to claim 1, characterized in that, Step 1 includes: In a 3D Cartesian coordinate system, d is the Euclidean distance from the UAV to the user, h is the flight altitude of the UAV, R is the coverage radius of the UAV, θ is the pitch angle between the UAV and the user, and r is the Euclidean distance from the projection of the UAV on the ground to the user.

3. The method according to claim 2, wherein Step 2 includes: Step 2-1, establish a system capacity function; Step 2-2, establish a system delay function; Step 2-3, establish a UAV system energy consumption function; Step 2-4, establish a joint optimization model for the mobile edge computing network assisted by UAVs.

4. The method according to claim 3, wherein Step 2-1 includes: Each user can only connect to one drone, and one drone can simultaneously process offloading data from more than two users, where $L_{ij} = 1$ indicates that the $i$-th user and the $j$-th drone have established a connection, and $L_{ij} = 0$ indicates that the $i$-th user and the $j$-th drone have not established a connection. ij $L_{ij} = 1$ indicates that the $i$-th user and the $j$-th drone have established a connection, where ij $L_{ij} = 0$ indicates that the $i$-th user and the $j$-th drone have not established a connection; The line-of-sight transmission probability in the established air-ground communication model is: where a and b are path loss parameters determined by the environment; P LoS is the probability of line-of-sight transmission, and exp is the exponential function; The path loss in air-to-ground communication is divided into line-of-sight propagation loss PL LoS and non-line-of-sight propagation loss PL NLoS , as shown in formula (3) and formula (4) respectively: where f c is the carrier frequency, c is the speed of light, η LoS and η NLoS are the additional path losses added by line-of-sight transmission in free-space propagation and the additional path losses added by non-line-of-sight transmission in free-space propagation, respectively; Establish the average path loss between user i and drone j which is P LoS = 1 - P NLoS (6) P NLoS is the non-line-of-sight transmission probability; Model the signal-to-noise ratio SNR when user i receives the signal of drone j as follows: i,j as: Among them, indicates that user i has established communication with drone j, and the transmission power of drone j is the maximum, and σ 2 is the noise power, indicates the received power of user i from drones other than drone j, and M is the number of users; Channel capacity C between user i and UAV j ij and the total system channel capacity C total are respectively as follows: C ij = Blog(1 + SNR i,j ) (8) where B is the channel bandwidth and N is the number of UAVs; T snr is the first set signal-to-noise ratio threshold, that is, user i only communicates with UAV j whose signal-to-noise ratio is greater than SNR th ; when SNR i,j ≥SNR th , T snr = 1; otherwise, T snr = 0; At the same time, another communication distance threshold T dis is set, that is, user i only communicates with UAV j whose distance is less than d th ; when d i,j ≥SNR th , T dis = 1; otherwise, T dis = 0.

5. The method according to claim 4, characterized in that, Step 2-2 includes: The UAV system delay includes local delay, offloading delay, and UAV computing delay; Local latency: The time required for local computing of a part of the data of user i is Among them, α is the offloading coefficient, representing the ratio of user data offloaded to the drone, is the total data volume of user i, C unit is the number of CPU cycles required for unit data, and f0 is the local computing frequency; Offloading delay: After user i establishes communication with drone j, the transmission time that occurs when user i offloads some data to drone j is as follows: UAV computing delay: the time required for UAV j to compute the offloaded data of user i is as follows: where C unit is the number of CPU cycles required for unit data, and f ij is the computing frequency allocated to user i by drone j; Total system delay T total is as follows:

6. The method according to claim 5, characterized in that, Step 2-3 includes: When the user offloads data, the energy consumption generated by the UAV includes: the energy consumption for local computing of user data, the communication energy consumption generated by the UAV receiving user offloaded data, and the energy consumption generated by the UAV processing user offloaded data; The energy consumption for local computing of user i's data is as follows: Among them, k is the energy consumption factor; Communication energy consumption of the drone j for receiving data unloaded by the user i is as follows: Among them, is the unloading power; The energy consumption generated by the drone j for calculating the offloading data of user i is as follows: Total system energy consumption E total is as follows:

7. The method according to claim 6, characterized in that, Step 2-4 includes: The joint optimization model for the mobile edge computing network assisted by UAVs is: Among them, C1 represents the variables to be optimized, including the position (x, y, h) of the UAV, the computing resources allocated by the UAV to the user, and the offloading coefficient; C2 represents that the task needs to be completed within the total delay T, C3 ensures that the total frequency allocated by the UAV to the user is within the total frequency range of the UAV; C4 ensures that each user can only be connected to one UAV.

8. The method according to claim 7, characterized in that, Step 3 includes: Step 3-1, initialize the population: randomly generate the initial population, which includes a total of N P particles, and each particle X i = [x i1 , y i1 , h i1 , …, x iN , y iN , h iN , f i1 , …, f iM , α i1 , …, α iM ; where x iN , y iN , h iN are the three-dimensional coordinates of the UAV, f iM is the computing frequency assigned by the UAV to the i-th user, and α iM is the offloading coefficient of the i-th user; Step 3-2, the update method of the particle swarm is: X i X(t) = i X(t - 1)+V i (t) (20) where \(i = 1, 2, \ldots, N\) p , \(N\) p is the population size, \(t\) is the generation number of evolution, \(\omega\) is the particle velocity weight, \(c_1\) and \(c_2\) are learning factors used to balance global exploration and local exploitation, and \(r_1\) and \(r_2\) are random numbers between 0 and 1; \(V\) i (t) is the movement velocity of particle \(i\) at the current generation number of evolution \(t\); \(X\) i (t) is the position of particle \(i\) at the current generation number of evolution \(t\); \(P\) i (t - 1) is the optimal position of the \(i\)-th particle in the \((t - 1)\)-th generation, and \(G(t - 1)\) is the optimal position of the particle population in the \((t - 1)\)-th generation; Step 3-3, the multi-objective fusion evaluation method of the particle: The calculation formula for the fitness Fit(X) of each particle X is: Among them, G max is the maximum number of evolutionary generations, X is an individual in the population, γ(X) is the angle between individual X and its nearest individual Y in Euclidean space, and d(X) is the norm of the normalized objective vector of X: where O is the number of targets, is the normalized target vector; are the norms of the normalized target vector of individual X and the normalized target vector of individual Y, respectively; Step 3-4, perform constraint processing on the particles; Step 3-5, determine whether the maximum number of evolutionary generations G is reached max , if so, output the particle with the optimal fitness value in the particle population, otherwise go to Step 3-2.

9. The method according to claim 8, wherein, In step 3-3, The calculation formula is:

10. The method according to claim 9, wherein In Step 3-4, the constraint processing method of the particle is: Among them, G(X1) and G(X2) are the constraint violation degrees of particle X1 and particle X2 respectively, and ε is a positive constant.

Images