CN116033032A - Mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method - Google Patents

Abstract

The application discloses a mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, which is used for obtaining the minimum maximum user time delay by jointly optimizing user unloading decisions, computing decisions, unmanned aerial vehicle deployment positions and computing resource allocation under the condition of considering CPU temperature constraint. According to the method, the optimization problem is further decomposed into three sub-problems, iterative processing is sequentially carried out, and in a specific embodiment, simulation experiments show that compared with other algorithms, the method provided by the application has better performance.

Description

Mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method

Technical Field

The application relates to the field of unmanned aerial vehicle and edge computing combined with the Internet of things, in particular to a mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method.

Background

Because Unmanned Aerial Vehicles (UAVs) have the characteristics of high mobility, low cost and the like, the unmanned aerial vehicles are used as airborne MEC (mobile edge computing) servers in the prior art, so that the service coverage range of the unmanned aerial vehicles in a resource shortage area is effectively enlarged. Secondly, the unmanned aerial vehicle can be used as an air relay to assist a user in achieving task unloading. The unmanned aerial vehicle network can provide various services for the Internet of things equipment by flexibly adjusting the deployment position of the unmanned aerial vehicle, such as calculation unloading data acquisition and content caching.

However, the computing power, size and weight of the on-board MEC server are relatively less, smaller, lighter, as compared to the MEC server integrated on the base station, subject to hardware costs, user experience design and deployment environment. In order to provide better computing services and to ensure greater maneuverability for the drone, the computing power, size and weight of the onboard MEC server need to be taken into account comprehensively. For example, the Dajiang manifield 2 platform uses Intel Kuui processor i7-8550U with a CPU dominant frequency of 1.8GHz. Meanwhile, the length (width) and thickness of DJI Manifold 2 are only 11 cm and 2.6 cm, respectively, and their weight is less than 200 g. On the other hand, with the development of semiconductor technology used in chips and the exponential increase of their performance, more and more embedded real-time systems are expected to be implemented on these power density computing platforms, which further brings new challenges to heat dissipation and temperature control of chips.

The existing literature on the unmanned aerial vehicle supporting the MEC network mainly focuses on energy consumption of the unmanned aerial vehicle in calculation, hovering or flight, and does not disclose how to solve the problem of excessive CPU temperature in the previous situation. The existing conventional DVFS scheduling relieves or solves the hardware limit (namely CPU temperature) of the on-board MEC server, and improves the computing resource allocation efficiency. For example CN201811143682.6, a computational offload scheduling method based on deep reinforcement learning; CN202111095551.7 is a distributed computing offloading method based on computational network coordination in a random network; according to the prior method, a composite scene set of residence time and waiting time delay is respectively established according to random movement and burst calculation requirements of a user; adopting posterior searching right actions to compensate game strategies, and establishing a game-based random planning model of the equipment end and the MEC server; and (3) constructing a scene tree, converting the multi-stage random regularization problem of the equipment end and the MEC server into an DEP problem, and mainly solving and obtaining an optimal task strategy unloaded by the MEC server and an optimal quotation strategy of the MEC server on the equipment end.

Under the condition of considering load balancing, the existing method does not consider the problem of communication stability between the unmanned aerial vehicle and the unmanned aerial vehicle. Second, CPU temperature control is not considered in minimizing user latency.

In the existing method, the calculation capacity and the energy supply of a calculation server at the mobile edge cannot be effectively optimized, the equipment supply at the mobile edge is limited, the unmanned aerial vehicle load cannot be optimally adjusted according to the characteristic of uneven distribution of users of the Internet of things, so that the unmanned aerial vehicle load is unbalanced, and the problem that the temperature of a part of overload CPU is too high cannot be solved.

Disclosure of Invention

The application provides a mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method aiming at the technical problems, which can effectively improve the utilization efficiency of communication and computing resources in an MEC network supported by an unmanned aerial vehicle, and simultaneously considers a multi-unmanned aerial vehicle collaborative computing system, in the system, an unmanned aerial vehicle hovers above a user to provide services for a resource shortage area, for example, the defect of fixed MEC can be well overcome by using unmanned aerial vehicle auxiliary communication under rural areas, temporary emergency rescue and other conditions. However, the unmanned aerial vehicle is used as a flight server, has limited computing capacity and energy supply, is applied to application with large computing capacity, or can not continuously process data normally due to serious degradation of CPU operation computing performance caused by the fact that the CPU temperature exceeds the normal operation temperature value when the unmanned aerial vehicle is in a scene of unbalanced load caused by uneven user distribution.

The application provides a mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, which comprises the following steps:

step S10: optimal deployment position Q= { Q of unmanned aerial vehicle by using BCD method _m ' optimal task offloading scheduling strategy

Optimal task scheduling strategy->

Computing resource allocation policy->

Decoupling is carried out to obtain a user unloading scheduling strategy and a computing scheduling strategy P ₁ Unmanned aerial vehicle deployment position P ₂ Resource allocation policy P ₃ The optimization objective is to minimize user processing time;

step S20: offloading scheduling policy and computing scheduling policy P for users ₁ : calculating resource allocation { F } by giving deployment position { Q }, and obtaining optimal user unloading strategy and calculation strategy { A } by adopting genetic algorithm ^* ，C ^* }；

Step S30: for unmanned aerial vehicle deployment position P ₂ : based on the initial computing resource allocation { F }, an optimal user offload policy { A } ^* Optimal task scheduling policy { C } ^* Converting the non-convex objective function and constraint into convex objective and constraint by using a first-order Taylor expansion, further converting the non-convex problem into a convex optimization problem, optimizing the deployment position { Q } of the unmanned aerial vehicle until the unmanned aerial vehicle converges to a tolerable precision, and obtaining the optimal { Q } ^* }；

Step S40: for resource allocation policy P ₃ : under given conditions { A ^* ，C ^* ,Q ^* Under }, P ₃ Is a convex problem based on CVXIterative obtaining of optimal computing resource allocation strategy { F ^* }；

Step S50: p obtained in steps S20-S40 after sequential iterative optimization ₁ 、P ₂ 、P ₃ And updating the corresponding variables to obtain an optimal deployment scheme and a resource allocation scheme of the unmanned aerial vehicle.

Preferably, step S20 comprises the steps of:

adopting GA method to meet constraint s.t.:

/>

obtaining the optimal solution { A } ^* ，C ^* -wherein constraint B1 represents a latency constraint at which a task needs to be constrained; constraint B2 represents the unmanned aerial vehicle acting as a MEC server, whose computational resources are also limited, and therefore, the resource constraints allocated to the users; constraint B3 represents the temperature constraint of the unmanned aerial vehicle and the CPU chip of the user, constraint B4 and constraint B5 represent the energy consumption constraint of the user and the unmanned aerial vehicle respectively; constraint B6 represents the distance constraint between the user and the unmanned aerial vehicle, which is required to be met when the user is within the maximum communication distance range of the unmanned aerial vehicle; constraint B7 represents a constraint between the offloading decision and the computation decision, constraint B8 and constraint B9 represent that the user task selecting offloading may be computed on the offloading drone and may be forwarded to other drone computations as well; constraint B10 and constraint B12 respectively represent unloading decision constraint and calculating the value range of the decision constraint;

Preferably, the GA process comprises the steps of:

step S21: initializing unloading or calculation of each body task in the population by adopting a binary coding mode;

step S22: calculating fitness of each body of the initialized population;

step S23: selecting individuals by adopting a roulette method;

step S24: crossing and mutating with certain probability to generate new individuals;

step S25: repeating the steps S23 and S24 for a plurality of times to obtain a plurality of new individual sets to form a new population, and repeating the step S22 to calculate the fitness value of each individual in the obtained population;

step S26: judging whether each body in each group obtained in the step S25 meets constraint s.t or not, and outputting the highest fitness and the corresponding individual if the constraint s.t is met; if not, the process returns to steps S22 to S25.

Preferably, the step S22 includes the steps of:

step S221: judging whether each individual in the initial population meets constraint s.t.;

step S222: if the individual n meets the constraint, its fitness is calculated according to the following formula:

fit _n ＝T-max(T _n ) (0.3)

wherein T is a constant for ensuring fitness is a positive value;

the fitness of an individual is set to 0 if it does not meet the constraint.

Preferably, step S23 is specifically:

Calculating the fitness fit of each individual _i The probability of each individual being selected is calculated from the ratio of the fitness of each individual to the fitness of all individuals as:

and calculating the cumulative probability of each individual according to the following formula:

Pr _s ＝{pr ₁ ,pr ₁ +pr ₂ ,pr ₁ +pr ₂ +pr ₃ ,...,pr ₁ +pr ₂ +...+pr _I }

wherein Pr is _s The cumulative probability of different intervals may be represented.

Preferably, step S24 comprises the steps of:

step S241: crossover operation: a process of crossing two parents according to a preset probability to generate new offspring, wherein the crossing probability is Pc, a random number between 0 and 1 exists for each individual, and when the random number is smaller than Pc, genes of two adjacent individuals are crossed to obtain two new individuals;

step S242: mutation operation: individual genes are mutated with a smaller probability of mutation.

Preferably, step S3 comprises the steps of:

step S31: initializing parameters and setting iteration times i: after the initial computing resource allocation { F }, optimal user offloading policy { A } ^* { C } optimal task calculation strategy ^* -and initial position of unmanned aerial vehicle { Q } ₀ Deployment of position P with drone ₂ The problem is the largest in realizing the user processing time and the smallest in realizing the problem, and the constraint condition s.t. shown below is satisfied as an iteration stop condition;

Step S32: calculating the communication speed and distance between a user and the unmanned aerial vehicle according to the initial value, and solving the unmanned aerial vehicle deployment position problem P ₂ Obtaining deployment position { Q ] of unmanned aerial vehicle ^* Communication rate between user and drone

Communication rate between unmanned aerial vehicles->

Then update:>

stopping until the constraint s.t. requirement as shown below is satisfied;

X _min ≤X _m ≤X _max (E7)

Y _min ≤Y _m ≤Y _max (E8)

constraint E1 represents the maximum latency requirement of the user, where { D _n The task size of the user is represented by { θ } _n Represents the computational resource occupancy of the user, constraint E2 represents the total energy consumption constraint of the drone, where

Power representing hovering flight of unmanned aerial vehicle, +.>

Represents unmanned aerial vehicle flight time, { beta } _m Represents a calculated power constant, constraint E3 represents a user erase transfer rate constraint, wherein +.>

Represents the gradient of the user upload rate, { B } represents the bandwidth size, constraint E4 represents the transfer rate constraint between unmanned aerial vehicles, where +.>

Representing a gradient in transmission rate between the unmanned aerial vehicles; constraint E5, E6 respectively represent the distance requirements between the user and the unmanned aerial vehicle and between the unmanned aerial vehicles; constraints E7, E8 then represent the range requirements of the deployment location between the drones.

Preferably, the step S4 specifically includes the following steps:

Based on the obtained optimal user offloading policy { A } ^* { C } optimal task calculation strategy ^* -and optimal deployment position Q of the drone ^* Under the condition of meeting the constraints of CPU temperature, energy consumption and computing resources, solving the problem of the resource allocation strategy P3 through CVX so as to minimize the maximum user processing time and enable the optimization result to meet the constraint condition st.

Wherein constraint F1 represents a task completion time constraint; constraint F2 represents the total CPU resource limit of the unmanned aerial vehicle; constraint F3 represents the total energy consumption constraint limit of the unmanned aerial vehicle, and constraint F4 and constraint F5 represent the CPU temperature limits of the user and the unmanned aerial vehicle respectively.

The beneficial effects that this application can produce include:

1) The application provides a mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, and provides an edge computing cooperation network based on multiple unmanned aerial vehicles. In the unmanned aerial vehicle auxiliary mobile edge computing system, a mobile edge computing network based on multiple unmanned aerial vehicles is defined, the multiple unmanned aerial vehicle systems can complete tasks more effectively and economically in a cooperative mode, total time delay multi-level reduction is achieved, in addition, computing efficiency and resource utilization rate can be remarkably improved, and network computing load is balanced. Compared with a single unmanned aerial vehicle system, the method can better cope with complex and changeable actual scenes. Secondly, through the optimization of unmanned aerial vehicle deployment position, can guarantee the coverage volume, reduce transmission cost simultaneously. .

2) According to the mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, the unloading process and the computing process are considered to be separated, and a multi-layer scheduling mechanism is realized. The separation of transmission and calculation is realized, the association of the unloading scheduling decision and the calculation scheduling decision is established, and the process that the user task is forwarded is further described. Through task forwarding scheduling, system resource redistribution is realized, energy consumption and time delay of users are reduced, and obvious effects on flow equalization and calculation equalization are also achieved.

3) According to the mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, the problem of total time delay fairness of task processing under temperature control is considered. As CPU utilization and memory utilization increase, energy consumption and CPU temperature increase, and the heat dissipation performance of the user equipment and the MEC server itself is poor. Thus, herein, temperature constraints at the device side and the server side are considered. The control of the CPU temperature is realized, the excessive loss of the CPU is reduced, and the service life of the equipment is prolonged.

4) According to the mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, a double-loop iterative algorithm is obtained by utilizing a continuous convex approximation (SCA) technology and a block coordinate reduction (BCD) method to solve a formulated MINLP problem. Furthermore, due to branch delimitation (B &B) Algorithm solution P ₁ Is computationally complex, especially for a large number of users and drones, therefore the GA algorithm is used to reduce solving P ₁ Is not limited by the complexity of (a).

5) According to the mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method, the performance of the method is evaluated from the two angles of maximum user task processing time delay and unmanned aerial vehicle load balancing, and the performance of the OFCS scheme provided in the method is obviously superior to that of the OCS scheme. Secondly, because the influence of the CPU temperature is considered, the processing time of the user task is increased, but the service quality requirement of the user task can be met while the system performance is ensured.

Drawings

Fig. 1 is a schematic structural diagram of an MEC system cooperatively calculated by multiple unmanned aerial vehicles in an embodiment of the present application;

FIG. 2 is a graph illustrating a comparison of the maximum user delay at different CPU frequencies with different user numbers in an embodiment of the present application;

FIG. 3 is a graph of maximum user latency versus line for different temperature constraints and different CPU frequencies in an embodiment of the present application;

FIG. 4 is a graph of comparison between different temperature thresholds and maximum user delay in the number of users in an embodiment of the present application;

FIG. 5 is a graph of a comparison of maximum user delay under different bandwidths and scheduling mechanisms in an embodiment of the present application;

FIG. 6 is a comparison line diagram of maximum user delay under different numbers of unmanned aerial vehicles and scheduling mechanisms in an embodiment of the present application;

FIG. 7 is a comparison line diagram of user communication delay versus user distribution positions and user numbers of different users in an embodiment of the present application;

FIG. 8 is a graph of a comparison of the impact of different resource allocation schemes and CPU frequency on task delay in an embodiment of the present application;

fig. 9 is a bar graph and a line graph of unmanned aerial vehicle load under different scheduling schemes in the embodiment of the present application, and fig. 10 is a flowchart of a method for scheduling a task of a mobile edge computing network and deploying unmanned aerial vehicle resources.

Symbol definition:

1. unmanned aerial vehicle's deployment position variable Q= { Q _m }；

2. Unloading scheduling decision variables

3. Calculating scheduling decision variables

4. Computing resource allocation policies

5. Individual fitness parameter fit _n ；

6. User location information w _n ；

7. Unmanned plane position information q _m ；

8. User upload rate r _n,m ；

9. Transfer rate between unmanned aerial vehicles

10. User character calculation amount D _n ；

11. Unmanned aerial vehicle maximum CPU resource F _m ；

12. Maximum energy consumption of unmanned aerial vehicle

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the invention.

Technical means which are not described in detail in the application and are not used for solving the technical problems of the application are all arranged according to common general knowledge in the field, and various common general knowledge arrangement modes can be realized.

The application provides a thermal perception task scheduling and resource allocation strategy, so that the maximum user processing task time is minimized; since the stated problem is a complex mixed integer nonlinear programming problem (MINLP) with strongly coupled variables, to further decouple these variables, it is converted into three more easily handled sub-problems:

1) P1, a user uninstalls a scheduling strategy and calculates the scheduling strategy;

2) P2, unmanned aerial vehicle deployment position;

3) And P3, a resource allocation strategy and iterative processing are carried out in sequence.

Under the push of the aforementioned pioneering work, in the multi-unmanned aerial vehicle collaborative computing system, in order to balance the load balance between the unmanned aerial vehicle and the unmanned aerial vehicle, each unmanned aerial vehicle can be regarded as additional assistance of other unmanned aerial vehicles, and the device offloads the tasks thereof to the unmanned aerial vehicle and calculates or forwards the tasks to other unmanned aerial vehicles to perform collaborative computing at the current unmanned aerial vehicle.

1. System model and problem formulation

Referring to fig. 1, a multi-unmanned cooperative system used in the present application is composed of a side cooperative scenario consisting of "multi-unmanned, multi-user", including a set of unmanned aerial vehicles denoted by m= {1,2,3,..m } and a set of random users n= {1,2,3,..n.

Specifically, each drone has a computing processor disposed thereon that has limited computing power. Assuming that each user has a computation-sensitive task, the user tasks are represented by triples, where D _n The size of the nth user task is represented in MB. θ _n The CPU period required by the task of the user to be completed is represented, the upper limit of the tolerable delay of the user is represented, and the tolerance of different users is different. The unmanned aerial vehicle is deployed above the user to assist the user in task processing, and the height is a fixed constant H.

Considering a three-dimensional coordinate system (3D) at the same time, the user's position is w _n ＝{x _n ,y _n ,0}. In addition, the 3D position where the mth unmanned aerial vehicle is located is defined as q _m ＝{x _m ,y _m H, so the distance between the nth user and the mth drone is

Assuming that m and j represent any two unmanned aerial vehicles, the distance between the mth unmanned aerial vehicle and the jth unmanned aerial vehicle is represented as

Wherein q is _j Is the 3D position of the j-th unmanned plane.

Definition of binary variables

Control task offloading decision, ++>

Representing control task calculation decisions.

In particular the number of the elements,

representing that user n offloads tasks to unmanned plane m and defining +.>

Then the task representing the nth user is offloaded to the a-th _n And a server.

Definition of the definition

If c _n =m represents that the task of the nth user is calculated on the mth drone.

If a is _n ＝c _n And the unmanned aerial vehicle for task unloading is the same as the unmanned aerial vehicle for calculation, and the task is not forwarded after being unloaded. Conversely, the nth task is from the a _n The unmanned aerial vehicle forwards to the c _n And (5) the unmanned plane.

For a certain user task, it may be either locally calculated or offloaded to the drone calculation, so it needs to meet

The precondition for offloading the tasks calculated to the drones is that offloading is selected and eventually only one drone can be used for calculation, thus requiring +.>

Furthermore, the tasks calculated on the unmanned aerial vehicle may be calculated directly after unloading or may be calculated after being forwarded, considering only the tasks of the user to be forwarded once.

Communication mode

The A2G link and the A2A link are considered herein, and it is assumed that, among the radio channels of the above links, a visible distance channel based on a free space transmission loss model is considered. Assume that the channel gain between the nth user and the mth unmanned aerial vehicle is

Then

Is defined as follows:

wherein beta is ₀ Is a reference value for the channel gain at 1 m.

In order for the user's offloaded task to be successfully received by the drone, the user must be within the coverage of the drone that he offloads, i.e. not exceed the maximum distance D covered by the drone _max Which may also be referred to as the maximum communication distance. The concrete representation is as follows:

secondly, to better utilize the resources within the system, collaboration can be made between e M for any M. That is, when the load of the mth unmanned aerial vehicle is greater than the load of the jth unmanned aerial vehicle, in order to balance the load therebetween, the mth unmanned aerial vehicle can forward the user task offloaded thereto to the jth unmanned aerial vehicle for calculation. The channel gain is calculated by the following formula:

in order to avoid collision between unmanned aerial vehicles, the distance between any two unmanned aerial vehicles should not be smaller than the minimum safety distance D _min I.e.

The concrete representation is as follows:

since the multi-drone collaborative computing system under consideration is based on FDMA (frequency division multiple access), the drones and devices will share the common bandwidth B during offloading and forwarding. Thus, the transmission rates of the unloading and forwarding of the nth user to the mth unmanned aerial vehicle and the mth unmanned aerial vehicle to the jth unmanned aerial vehicle are expressed as the following relationships, respectively:

Wherein P is _n And P _m Respectively representing the transmission power of the nth user and the mth unmanned aerial vehicle, N ₀ Noise power spectral density for drones and base stations.

B. Delay correlation

(1) Local computing

The user tasks may be calculated locally or offloaded to the drone. Is known to be

When a task representing a certain user is calculated locally, the calculation time for the task to be executed locally is as follows:

wherein D is _n Representing the task size, θ, of user n _n Representing the CPU cycles, f, required by the local user to compute 1bit of data _n Representing the computing power of the local user processor, it is considered herein that all users are isomorphic, and therefore f for all users is assumed _n Is a constant and these users are powered by their own limited capacity battery only. In practice, however, different users have different latency requirements. Users who are sensitive to the delay requirement may use as much computing power as possible in order to meet the delay requirement, which may result in higher CPU power consumption, while limited battery capacity may not necessarily support the requirement on delay. For this purpose, the user needs to haveThe unmanned aerial vehicle server with larger calculation capability performs auxiliary calculation, so that the demand of the unmanned aerial vehicle server for time delay is met, and the energy consumption of the unmanned aerial vehicle server can be saved.

(2) Offloading computing

When (when)

When the user offloads the task to the unmanned aerial vehicle, the task may be calculated on the offloaded unmanned aerial vehicle, and may be forwarded to other unmanned aerial vehicles by the current unmanned aerial vehicle for execution. This involves two cases, on the one hand, the direct calculation after task offloading and on the other hand, the calculation after user offloading and being forwarded.

Firstly, the unloading process, the transmission time from the user to the unmanned aerial vehicle can be calculated by the following steps:

the second is the calculation process of the method,

the task representing the nth user is executed on the jth unmanned plane, and the calculation time of the task is as follows:

f _n ^m representing the computing resources allocated to the user by the drone.

The forwarding delay is defined as follows:

in summary, the total time delay for processing the task of any user n includes four processes, namely, a local calculation time, an unloading time, a calculation time and a forwarding time. In particular, whenWhen the unmanned aerial vehicle unloaded by the user and the unmanned aerial vehicle calculated by the task are the same,

the total delay includes a local computation delay, a communication delay, and a drone computation delay, expressed by the following formulas:

/>

expanding the formula to obtain the following formula:

otherwise, the total time delay comprises four parts, namely local calculation time delay, unmanned aerial vehicle calculation time delay and communication time delay, wherein the communication time delay comprises communication between the unmanned aerial vehicle and other unmanned aerial vehicles and communication between the unmanned aerial vehicle and a user. Is expressed by the following formula:

The above formula is developed as follows:

specifically, in the above formula Φ _n The task forwarding matrix of the user n is represented, and the task of the user is forwarded from the unloaded unmanned aerial vehicle to the calculated unmanned aerial vehicle, wherein the row and the column of the matrix respectively represent the sequence number of the unloaded unmanned aerial vehicle and the sequence number of the calculated unmanned aerial vehicle, m represents the unloaded unmanned aerial vehicle in the above formula, and j represents the calculated unmanned aerial vehicle.

C. Energy consumption analysis

Herein, different modes of computation are considered for tasks requiring computation, including local computation and offloading computation. The unloading calculation comprises direct calculation after unloading and forwarding. In the different calculation modes considered, there is energy consumption whether the user's task is to select a local calculation or to offload to the drone for calculation. First, the user task is executed locally, and then the energy consumption of the calculation process can be expressed as follows:

beta in the above formula _n The effective capacitance coefficient representing the nth user is primarily dependent on the chip architecture of the processor (CPU).

Second, for mobile devices with computationally intensive and time-delay sensitive tasks, their own power may not be able to complete within the required deadline, and their battery's low power capacity may also be difficult to support the power consumption by moving and performing the task. At this time, the user can unload the task to the unmanned aerial vehicle for calculation, can effectively alleviate the dilemma that the self ability of the user is insufficient for meeting the demand, but in the process of unloading the task to the unmanned aerial vehicle, the user can also have unloading energy consumption, and the unloading required by the user n to unload the task to the unmanned aerial vehicle m is represented as follows:

If the user task selects to offload to the drone calculation, then the drone needs to allocate computing resources to the user, so the task of user n calculates on drone m that the energy that needs to be consumed is

The method can obtain:

wherein beta is _m Is the effective capacitance coefficient depending on the unmanned aerial vehicle processor chip structure. This isIn addition, if there is a task to forward from the mth unmanned aerial vehicle to the jth unmanned aerial vehicle, there is energy consumption in unmanned aerial vehicle m in the process

Namely>

Finally, both hover and movement of the drone require propulsion energy consumption maintenance. Typically, the propulsion energy consumption of a drone is determined by its acceleration and speed. In the scene, only when a user has a calculation requirement, the unmanned aerial vehicle needs to hover to the upper air of the user for a period of time to provide calculation service for the user, and when the speed of the unmanned aerial vehicle is 0, the propulsion energy consumption of the unmanned aerial vehicle is a constant. Therefore, in this context, the hover energy consumption of the drone is mainly considered. The hovering time and energy consumption are respectively as follows:

in order to ensure that all users needing to be offloaded in the range are offloaded and all tasks that the users have offloaded are calculated, the unmanned aerial vehicle hovers for the maximum time for completing the user tasks. In addition, in the case of the optical fiber,

is the received power of the unmanned aerial vehicle when hovering.

In summary, the total energy consumption required by a certain unmanned aerial vehicle to complete a computing task of a user within a certain range includes computing energy consumption, forwarding energy consumption and hovering energy consumption. The method is specifically as follows:

D. unmanned aerial vehicle's power and heat energy model

Whether the user task is computed locally or is chosen to be offloaded to the MEC server for computation, it may be an order of magnitude larger task due to the millisecond latency requirements that it may have. In order to meet the time delay requirement, the task needs to be processed by the computing resources as large as possible, and the heat dissipation performance of the user equipment and the MEC server is poor, which tends to result in higher CPU chip temperature and even damages the CPU chip performance. For periodic task τ _i There is an associated period p _i And an execution time e _i . At this time, task τ _i Processor utilization of (2) is

For periodic tasks, the dynamic power consumption specific to the task increases with the processing time of the task or the utilization of the processor, while the wasted power increases with increasing temperature. Thus, for a given task set τ _i The total power consumption is expressed as follows:

P _tot ＝P _dyn +P _leak

in the formula, P _dyn Representing dynamic power consumption, P _leak Indicating wasted power. P (P) _dyn Is expressed by the following formula:

In the above formula, V, f, e _i ，p _i ，

Respectively representing the operating voltage, the calculation frequency, the calculation time, the period and the task switching activity factor.

The temperature of the processor chip will eventually reach a steady state temperature determined by the total power consumption and this temperature range is 20-60 c as measured. The processor chip temperature can be expressed as follows:

Γ _chip ＝P _tot R _th +Γ _a

Γ _chip ＝P _tot R _th +Γ _a

in the above formula Γ _a Indicating the temperature of the surrounding environment. R is R _th Representing the thermal resistance of the chip. Assume for chip temperature threshold

The overall power consumption obtained by deforming the above formula is shown as follows: />

In the above equation, the left of the inequality is the total power consumption and the right term may be referred to as the available thermal budget.

In connection with the scenario considered herein, the ratio of the actually allocated computing resources to the total computing resources is used herein as the utilization of the CPU, and when the nth user task is computed at the mth drone, the total power consumption of the CPU chip of the drone is expressed as follows:

similarly, when the user task is calculated locally, the total power of the CPU chip is as follows:

in the above formula, because of the total resource F owned by the user _n No further allocation is required, i.e. f _n ＝F _n At this time, the CPU utilization of the user is 1.

E. Description of the problem

Representing the deployment position variable of the unmanned aerial vehicle as Q= { Q _m ' express the unload schedule variable as

Representing the computational schedule variable as +.>

The computing resources are denoted +.>

In this context, the variables are jointly optimized to minimize the maximum user processing time, given by:

/>

in the above problem, the objective function is to minimize the total delay, optimize the deployment variable q= { Q _m Unloading schedule variables

And calculate schedule variable +.>

Computing resources->

Secondly, constraint A1 indicates that tasks need to be completed within the time delay constraint range, and different QoS (Quality of Service) requirements of different users are considered, so that the maximum time delay tolerable by each user is different; constraint A2 represents that UAV acts as MEC server, its computational resources are also limited, and therefore the resources allocated to the user cannot exceed the resources they have themselves; constraint A3 and constraint A4 represent the UAV and the CPU chip temperature constraint of the user respectively, and constraint A5 and constraint A6 represent the energy consumption constraint of the user and the UAV respectively; constraint A7 indicates that in order to avoid collisions between UAVs, a minimum safe communication distance between different UAVs needs to be maintained; constraint A8 indicates that the user needs to be within the maximum communication distance range of the UAV in order to offload tasks to the UAV; constraint A9 indicates that the user task selected for offloading may be calculated on the offloaded UAV and may be forwarded to other UAV calculations; constraint a10 and constraint a11 represent, respectively, that a task can only be offloaded to and a task can only be calculated on one UAV; constraint a12 indicates that a task can only be offloaded once and forwarded once. Constraint A13 and constraint A14 represent the variable +. >

And->

The value range of (2) can only be 0 or 1; constraint a15 indicates that the deployment range of UAV locations is to cover the user's distribution area.

2. The algorithm proposes

From observations of P, due to the presence in the objective function and constraints

And q _m Coupling (I)>

And->

Is provided with a coupling of (a) and (b),

and q _m It can be seen that this is a very troublesome MINLP problem. Furthermore, the integer variable { A, C } further increases the difficulty in solving this problem. Thus, it is not possible to directly get a solution of the current form of P. Inspired by iterative design, the variables are decoupled by applying a BCD method, and P is divided into the following sub-problems:

1)P ₁ the user uninstalls the scheduling policy and calculates the scheduling policy;

2)P ₂ a deployment location for the unmanned aerial vehicle;

3)P ₃ a resource allocation policy;

firstly, given a deployment position Q, calculating resource allocation { F }, and obtaining an optimal user unloading strategy and a calculation strategy { A } by adopting a genetic algorithm ^* ，C ^* And update { T } _n }。

Second, according to the initial computing resource allocation { F }, the optimal user offload policy { A } ^* { C } optimal task calculation strategy ^* Converting the non-convex objective function and constraint into a convex objective and constraint by using a first-order Taylor expansion, further converting the non-convex problem into a convex optimization problem, optimizing the deployment position Q of the unmanned aerial vehicle until the unmanned aerial vehicle converges to a tolerable precision, and obtaining the optimal Q ^* At the same time update { T } _n }；

Third, under given conditions { A ^* ，C ^* ,Q ^* Under }, P ₃ As a convex problem, the optimal computing resource allocation strategy { F ] is obtained based on CVX iteration ^* }；

Finally, optimizing P by sequential iteration ₁ 、P ₂ 、P ₃ And updates the related variables, the process of which is called BCD method;

the joint optimization is carried out on the aspects of task unloading scheduling, task calculation scheduling, unmanned aerial vehicle deployment position, calculation resource allocation and the like through an iteration method based on a BCD method, and the calculation complexity of the algorithm is given.

A. Task unloading scheduling and task computing scheduling

As described above, the offloaded user task will be scheduled to the drone for calculation according to two scenarios:

1) Forwarding, the task unloaded to a certain unmanned aerial vehicle can be forwarded to another unmanned aerial vehicle for calculation;

2) And (5) not forwarding, and immediately calculating on the unloaded unmanned aerial vehicle.

Since the unloading decision and the calculation decision are strongly coupled and have larger mutual influence, the P under the given condition { Q, F } is solved ₁ The user task unloading scheduling and task computing scheduling are jointly optimized, so that the maximum user processing time is minimum,

due to the coupling of the binary variable, it is strongly non-convex. To deal with this integer-dependent sub-problem, the GA method is used to obtain a suboptimal solution { A } ^* C }. The GA solving process is shown in FIG. 2:

(1) Initializing a population.

A population includes a plurality of individuals, each individual having a chromosome with a plurality of genes thereon.

The variables solved are binary variables, so that in binary coding mode, one variable in the problem corresponds to one gene in the chromosome. Assuming that N users and M drones are considered, the length of one chromosome is 2×n×m. As shown in table 1 below, n=3 _, One chromosome at m=2, each two genes representing the offloading or computation of one user task, such as gene 1 and gene 2At 0, the user 1 task is not offloaded to either drone 1 or drone 2. Gene 9 and gene 10 are 0,1, respectively, then the task representing user 2 is calculated at drone 2. Thus in the following chromosomes, the blue part represents the offloaded gene representation of all user tasks and the orange part represents the calculated gene representation of all user tasks.

Table 1: chromosomal sequence

(2) The fitness of each individual is calculated.

For an individual meeting all the constraints, the smaller the delay is, the larger the fitness value is, and therefore, the opposite number of objective functions is used as the fitness function. Second, to ensure that fitness is positive, the fitness function may be expressed as fit _n ＝T-max(T _n ) Where T is a constant. For individuals not meeting the constraint condition, the fitness value is directly assigned to 0.

(3) Selection of

The roulette wheel selection method (also known as the proportional selection method) is used herein for selection. First, calculate the fitness fit of each individual _n The probability that each individual is selected is calculated from the ratio of the fitness of each individual to the fitness of all individuals as follows:

and calculates the cumulative probability for each individual, as calculated below:

Pr _s ＝{pr ₁ ,pr ₁ +pr ₂ ,pr ₁ +pr ₂ +pr ₃ ,...,pr ₁ +pr ₂ +...+pr _I }

further, different sections are formed, and the section size indicates the fitness value size, so the larger the section span is, the larger the probability of being selected is.

(4) Crossover

The crossover operation is the process of crossing two parents with a certain probability to produce new children. Assuming that the crossover probability is Pc, there is a random number between 0 and 1 for each individual, and when the random number is smaller than Pc, genes of adjacent two individuals are crossed, thereby forming two new individuals.

(5) Variation of

Mutation refers to mutation of individual genes, such as from "0" to "1" or from "1" to "0", with a small probability of mutation, thereby expanding the diversity of the population.

And (3) forming a new population through the steps (3), (4) and (5), repeating the step (2) to calculate the fitness value of each individual in the population, and outputting the highest fitness and the corresponding individual if the fitness value meets the exit condition. If not, continuing to repeat the steps (2), (3), (4) and (5).

The algorithm pseudocode described above is shown in table 2 below:

table 2: genetic algorithm solution unloading decision and calculation decision sub-problem

B. Unmanned aerial vehicle position deployment optimization sub-problem

By solving the P1 sub-problem, the optimal offloading scheduling decision is known

Calculate scheduling decision->

Variable->

Solving the sub-problem of the deployment position of the unmanned aerial vehicle. :

X _min ≤X _m ≤X _max (C5)

Y _min ≤Y _m ≤Y _max (C7)

specific treatments for the non-convex portion in the above problems are as follows:

(1) As the velocity formula pertains to the drone position q _m Non-convex

The communication rate expansion involved in the communication process of the nth user and the mth unmanned aerial vehicle is as follows:

approximating the non-convex function to a convex function using SCA (Successive Convex Approximation) introduces a relaxation variable for ease of solving the problem

The objective function is processed on the rate non-convex part in the constraint, while the following constraint is newly added:

first, assume that

The above formula is expressed as follows: />

At this time, the variable q is referred to in the left side of the above formula _m Is not convex, but is related to ||q _m -w _n The term is convex, and the lower bound of the rate is obtained by a first-order taylor expansion. Assume that

Q at the ith iteration _m The lower bound for the rate is then found by:

wherein, the liquid crystal display device comprises a liquid crystal display device,

representing the rate at the ith iteration:

r represents _n,m Is a first derivative of (a).

Similarly, the same processing is performed on the communication rate non-convex part between any two unmanned aerial vehicles shown in the following formula, and a relaxation variable is introduced first

The newly added constraints are as follows:

second, assume that

The above formula can be expressed as follows:

then, the first-order Taylor expansion is performed on the formula (4.3.20) to obtain

Is shown below:

wherein the method comprises the steps of

Is the rate at the ith iteration,

is->

Is a first derivative of (a). By the above processing, the non-convex objective function is converted as follows:

the original problem P2 is converted into the following problem P ₂ ：

/>

X _min ≤X _m ≤X _max (D7)

Y _min ≤Y _m ≤Y _max (D8)

Since constraint D5 and constraint D6 are not convex;

in problem P' ₂ In which constraint D5 and constraint D6 equations remain non-convex to the left, the constraint is relaxed using SCA techniques, first, for the constraint

The sides of the inequality sign are respectively squared to obtain the following formula:

for any given point

After a section of taylor expansion, the following inequality exists:

similarly, for any given point

And->

For constraint D6, the following inequality exists:

the original non-convex problem is converted into a convex optimization problem by converting the non-convex objective function in the original problem and converting the non-convex constraint, as follows:

X _min ≤X _m ≤X _max (E7)

Y _min ≤Y _m ≤Y _max (E8)

problem P ₂ "in which the constraint is convex and the objective function is convex, the problem is thus a convex optimization problem that can be solved using standard convex algorithms, such as the interior point method and Lagrangian dual method, and convex optimization tools, such as CVX. The CVX is used herein for solving, and the problem solving process is as follows:

(1) Initializing parameters.

And->

For the optimal solution of the problem P1,

CPU resources initially allocated to the user for CPU, +.>

Indicating the initial position of the drone.

Pseudo-code using the CVX solving Algorithm is shown in Table 3 below:

table 3: unmanned plane position deployment optimization algorithm

C. Resource allocation sub-problem

By solving the above two sub-problems, an optimized solution unloading matrix A has been obtained ^* Calculate matrix C ^* Optimized deployment location Q ^* These values are used as inputs to the resource allocation sub-problem, which is described below, to optimize the resource allocation:

wherein constraint F1 represents a task completion time constraint; constraint F2 represents the total CPU resource limit of the unmanned aerial vehicle; constraint F3 represents the total energy consumption constraint limit of the unmanned aerial vehicle, and constraint F4 and constraint F5 represent the CPU temperature limits of the user and the unmanned aerial vehicle respectively.

In problem P3, except for variables

Other than that, being constant or of known variable, theThe question is about->

Single variable optimization problem of (2).

In constraint F1, the first term, the second term, and the 4 th term on the left of the inequality sign are constant terms in the present problem, as follows:

order the

The above conversion is as follows:

the above formula is converted to obtain the following:

in the above formula, the right side of the inequality is a constant, and the constraint is related to the variable

Is linear and therefore the constraint is a convex constraint.

The same holds that the objective function and constraints F2-F5 relate to variables

Both affine, the objective function and constraints of the problem are convex, so the resource allocation problem is a convex optimization problem that can be solved directly using the CVX tool.

D. Overall algorithm and complexity analysis

By P pair ₁ ,P ₂ ,P ₃ Iterative optimization is carried out until the outer ring is formedAnd the tolerable precision is achieved, and the suboptimal solution of the original algorithm can be obtained. The algorithm pseudocode is shown in table 4 below:

table 4: user minimum maximum task total time delay problem algorithm pseudo code

The complexity analysis is as follows: for unloading decision and computational decision sub-problems, GA solution was used, reference (Fengxin Guo, heli Zhang, hong Ji, xi Li, victor C.M. Leung. An Efficient Computation Offloading Management Scheme in the Densely Deployed Small Cell Networks With Mobile Edge Computing [ J) ]GA time complexity analysis was performed by IEEE/ACM Transactions on Networking,2018-10-12,26 (6): 2651-2664), considering N users and M unmanned aerial vehicles, the program operation termination condition was genetic algebra I. First, the complexity is O (K (2N)) for the initial population of K individuals. The time complexity in calculating the fitness is O (I (KNM) +ik), abbreviated O (I (KNM)). The time complexity of the select operation is O (IK), the time complexity of the cross operation is

The time complexity of the mutation operation is O (IK), and in conclusion, the time complexity of the GA algorithm is

Also because K < NM, therefore +.>

The unmanned aerial vehicle deployment sub-problem uses Successive Convex Approximation (SCA) to transform the problem into a convex problem, and finally uses CVX for solution. The complexity of its algorithm is O (l ₁ (M(N+M)) ³ ). The complexity of the overall algorithm is therefore O (g (I (KNM) +l) ₁ (M(N+M)) ³ ) G is the outer layer cycle number, l1 is the inner layer cycle number).

Description the complexity of the proposed algorithm is within a tolerable range, and is lower than that of the solution mode directly adopting bnb.

Examples

The embodiment simulates according to the method provided by the application, and the simulation parameters are set:

the plurality of devices are randomly distributed in a 200m multiplied by 200m area, each device has task calculation requirements, and the unmanned aerial vehicle can be used as an MEC server and a relay to help a user calculate or forward tasks. The simulation parameters are summarized in table 5, unless otherwise indicated.

Table 5: simulation parameters

Simulation results and analysis:

the numerical results obtained by simulation verify the effectiveness and performance of the proposed multi-unmanned aerial vehicle collaborative computing (OFCS) scheme.

In order to evaluate the performance of the proposed overall algorithm for multi-unmanned collaborative computing OFCS, consider the following reference scheme, named:

1) The unloading scheduling is separated from the calculation scheduling, so that each unmanned aerial vehicle is endowed with double identities serving as an MEC server and a relay, the unmanned aerial vehicle can assist user task calculation, and when overload occurs, tasks can be forwarded to other unmanned aerial vehicles for calculation.

2) The offloading schedule and the computing schedule are not separated, and the unmanned aerial vehicle only serves as an MEC server, and the task of the user is offloaded to the computing position.

3) Without temperature constraints, consider the CPU temperature limit of the onboard MEC server.

4) Random allocation and optimization, i.e. taking into account the random initialization of the user and optimizing its offloading decisions.

Fig. 2 shows experiments performed for different numbers of users, which shows that as the total resources of the system increase, the total delay decreases. Because, in the case that the number of users remains stable and the allocation proportion is consistent in the experiment, increasing the total resources of the system will result in increasing the resources allocated to each user, and thus, the corresponding time delay will be reduced.

For the same computing resource, the time delay overall tends to rise as the number of users increases. Because, when the total computing resources are unchanged within the system, an increase in the number of users results in a decrease in the computing resources allocated to each user. Thus, the overall latency of the user processing tasks increases.

In addition, fig. 3 shows the effect of considering the temperature on the actual calculation, but in fig. 3, when n=12, the 3 rd point and the following points tend to be stable, because at the current point, the maximum computing resource that can be allocated under the temperature constraint is reached, and in the case of no temperature constraint, more computing resources continue to be allocated, so that the total task delay still tends to be reduced. The time delay of the first two points under the condition of temperature and the time delay of the first two points under the condition of no temperature coincide, because the total CPU resource is smaller at the moment, even if the threshold value of the unmanned aerial vehicle computing resource allocation is reached, the threshold value is smaller than or equal to the maximum computing resource which can be allocated under the temperature constraint at the moment.

In addition, fig. 4 further shows that as the temperature threshold increases, the time delay is in a decreasing trend, which means that the user can allocate more computing resources at a higher temperature threshold, and secondly, for the same temperature threshold, when the number of users is greater, the requirement on resources in the system is higher, and compared with the smaller number of users, the greater the number of users is, the earlier the number of users reaches the threshold of the computing resources which can be allocated under the temperature constraint, because the higher the number of users is, the higher the resource utilization is, and the faster the temperature threshold is reached.

As shown in fig. 5, the overall user delay tends to decrease as bandwidth resources increase. Because the total bandwidth resources are evenly distributed to each user, and the number of users is unchanged, the time delay is reduced along with the increase of the bandwidth resources. While the influence of the number of different unmanned aerial vehicles on the time delay is shown in fig. 6, it can be seen from the figure that the maximum user processing time delay in the two comparison schemes is gradually reduced along with the increase of the number of deployed unmanned aerial vehicles. Because the number of unmanned aerial vehicles in the system is increased, which means that the computing resources in the whole system are increased.

As shown in fig. 7, in general, the communication delay increases with the number of users. On the other hand, the relative positions of the unmanned plane and the users and the densely distributed degree of the users comprehensively determine the allocation of CPU resources, thereby influencing the calculation time delay. Secondly, the communication time delay and the maximum user time delay of the W3 with the most concentrated user distribution are smaller than those of the other two cases, and the main reason is that the unmanned aerial vehicle can be deployed close to the user distribution, and the users are concentrated, so that the whole communication distance is smaller. And in the case of W1, the user distribution is two-pole differentiated, and on the basis of ensuring the minimum safety distance, when the probability of the occurrence of the long user transmission distance is larger than that of W2 and W3, the more easily the load imbalance occurs.

Fig. 8 shows that as the total computing resources within the system increase, the overall latency of user task processing is progressively reduced in all three schemes. The time delay obtained for the initial calculation resources which are randomly allocated is slightly higher than the time delay after optimization, on one hand, the reason is that the total resources in the system are increased, and on the other hand, scheduling of unloading and calculation layers exists, so that the utilization rate of the system resources can be effectively improved, and the total time delay of task processing is smaller.

Fig. 9 illustrates that the load of the unmanned aerial vehicle is analyzed from the number of users calculated by the unmanned aerial vehicle and the duty ratio of the calculated task amount, and it can be seen that the load gap between the unmanned aerial vehicles is relatively small when there is a calculation schedule.

Although the present invention has been described with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements and changes may be made without departing from the spirit and principles of the present invention.

Claims (8)

1. A method for scheduling a mobile edge computing network task and deploying unmanned aerial vehicle resources is characterized by comprising the following steps:

Step S10: optimal deployment position Q= { Q of unmanned aerial vehicle by using BCD method _m ' optimal task offloading scheduling strategy

Optimal task scheduling strategy->

Computing resource allocation policy->

Decoupling is carried out to obtain a user unloading scheduling strategy and a computing scheduling strategy P ₁ Unmanned aerial vehicle deployment position P ₂ Resource allocation policy P ₃ The optimization objective is to minimize user processing time;

step S20: offloading scheduling policy and computing scheduling policy P for users ₁ : calculating resource allocation { F } by giving deployment position { Q }, and obtaining optimal user unloading strategy and calculation strategy { A } by adopting genetic algorithm ^* ，C ^* }；

Step S30: for unmanned aerial vehicle deployment position P ₂ : based on the initial computing resource allocation { F }, an optimal user offload policy { A } ^* Optimal task scheduling policy { C } ^* Converting the non-convex objective function and constraint into convex objective and constraint by using a first-order Taylor expansion, further converting the non-convex problem into a convex optimization problem, optimizing the deployment position { Q } of the unmanned aerial vehicle until the unmanned aerial vehicle converges to a tolerable precision, and obtaining the optimal { Q } ^* }；

Step S40: for resource allocation policy P ₃ : under given conditions { A ^* ，C ^* ,Q ^* Under }, P ₃ As a convex problem, the optimal computing resource allocation strategy { F ] is obtained based on CVX iteration ^* }；

Step S50: p obtained in steps S20-S40 after sequential iterative optimization ₁ 、P ₂ 、P ₃ And updating the corresponding variables to obtain an optimal deployment scheme and a resource allocation scheme of the unmanned aerial vehicle.

2. The mobile edge computing network task scheduling, unmanned aerial vehicle resource deployment method of claim 1, wherein step S20 comprises the steps of:

adopting GA method to meet constraint s.t.:

P ₁ ：

/>

_{obtaining the optimal solution} {A ^* ，C*}， _{Wherein the method comprises the steps of} Constraint B1 represents the latency constraint of a task; constraint B2 represents the drone acting as a MEC server, _{which is a kind of} Computing resources are also limited and, therefore, the resource constraints allocated to users; constraint B3 represents the temperature constraint of the unmanned aerial vehicle and the CPU chip of the user, constraint B4 and constraint B5 represent the energy consumption constraint of the user and the unmanned aerial vehicle respectively; constraint B6 represents the distance constraint between the user and the unmanned aerial vehicle, which is required to be met when the user is within the maximum communication distance range of the unmanned aerial vehicle; constraint B7 represents a constraint between the offloading decision and the computation decision, constraint B8 and constraint B9 represent that the user task selecting offloading may be computed on the offloading drone and may be forwarded to other drone computations as well; constraint B10 and constraint B12 represent unloading decision constraints and calculating the range of values of the decision constraints, respectively.

3. The mobile edge computing network task scheduling, unmanned aerial vehicle resource deployment method of claim 2, wherein the GA method comprises the steps of:

Step S21: initializing unloading or calculation of each body task in the population by adopting a binary coding mode;

step S22: calculating fitness of each body of the initialized population;

step S23: selecting individuals by adopting a roulette method;

step S24: crossing and mutating with certain probability to generate new individuals;

step S25: repeating the steps S23 and S24 for a plurality of times to obtain a plurality of new individual sets to form a new population, and repeating the step S22 to calculate the fitness value of each individual in the obtained population;

step S26: judging whether each body in each group obtained in the step S25 meets constraint s.t or not, and outputting the highest fitness and the corresponding individual if the constraint s.t is met; if not, the process returns to steps S22 to S25.

4. The mobile edge computing network task scheduling, unmanned aerial vehicle resource deployment method of claim 3, wherein step S22 comprises the steps of:

step S221: judging whether each individual in the initial population meets constraint s.t.;

step S222: if the individual n meets the constraint, its fitness is calculated according to the following formula:

fit _n ＝T-max(T _n ) (0.1)

wherein T is a constant for ensuring fitness is a positive value;

the fitness of an individual is set to 0 if it does not meet the constraint.

5. The mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method according to claim 3, wherein step S23 is specifically:

calculating the fitness fit of each individual _i The probability of each individual being selected is calculated from the ratio of the fitness of each individual to the fitness of all individuals as:

and calculating the cumulative probability of each individual according to the following formula:

Pr _s ＝{pr ₁ ,pr ₁ +pr ₂ ,pr ₁ +pr ₂ +pr ₃ ,...,pr ₁ +pr ₂ +...+pr _I }

wherein Pr is _s The cumulative probability of different intervals may be represented.

6. A mobile edge computing network task scheduling, unmanned aerial vehicle resource deployment method according to claim 3, wherein step S24 comprises the steps of:

step S241: crossover operation: a process of crossing two parents according to a preset probability to generate new offspring, wherein the crossing probability is Pc, a random number between 0 and 1 exists for each individual, and when the random number is smaller than Pc, genes of two adjacent individuals are crossed to obtain two new individuals;

step S242: mutation operation: individual genes are mutated with a smaller probability of mutation.

7. The mobile edge computing network task scheduling, unmanned aerial vehicle resource deployment method according to claim 1, wherein step S3 comprises the steps of:

Step S31: initializing parameters and setting iteration times i: after the initial computing resource allocation { F }, optimal user offloading policy { A } ^* { C } optimal task calculation strategy ^* -and initial position of unmanned aerial vehicle { Q } ₀ Deployment of position P with drone ₂ The problem is the largest in realizing the user processing time and the smallest in realizing the problem, and the constraint condition s.t. shown below is satisfied as an iteration stop condition;

step S32: calculating the communication speed and distance between a user and the unmanned aerial vehicle according to the initial value, and solving the unmanned aerial vehicle deployment position problem P ₂ Obtaining deployment position { Q ] of unmanned aerial vehicle ^* Communication rate between user and drone

Communication rate between unmanned aerial vehicles->

Then update:>

stopping until the constraint s.t. requirement as shown below is satisfied;

P ₂ :

X _min ≤X _m ≤X _max (E7)

Y _min ≤Y _m ≤Y _max (E8)

constraint E1 represents the maximum latency requirement of the user, where { D _n The task size of the user is represented by { θ } _n Represents the computational resource occupancy of the user, constraint E2 represents the total energy consumption constraint of the drone, where

Representing the power of the unmanned aerial vehicle hovering flight,

represents unmanned aerial vehicle flight time, { beta } _m Represents a calculated power constant, constraint E3 represents a user erase transfer rate constraint, wherein +.>

Represents the gradient of the user upload rate, { B } represents the bandwidth size, constraint E4 represents the transfer rate constraint between unmanned aerial vehicles, where +. >

Representing a gradient in transmission rate between the unmanned aerial vehicles; constraint E5, E6 respectively represent the distance requirements between the user and the unmanned aerial vehicle and between the unmanned aerial vehicles; constraints E7, E8 then represent the range requirements of the deployment location between the drones.

8. The mobile edge computing network task scheduling and unmanned aerial vehicle resource deployment method according to claim 1, wherein step S4 specifically comprises the following steps:

based on the obtained optimal user offloading policy { A } ^* { C } optimal task calculation strategy ^* -and optimal deployment position Q of the drone ^* Solving a resource allocation strategy P through CVX under the condition of meeting the constraints of CPU temperature, energy consumption and computational resources ₃ The problem is that the maximum user processing time is minimized and the optimization results meet constraints.

P ₃ :

Wherein constraint F1 represents a task completion time constraint; constraint F2 represents the total CPU resource limit of the unmanned aerial vehicle; constraint F3 represents the total energy consumption constraint limit of the unmanned aerial vehicle, and constraint F4 and constraint F5 represent the CPU temperature limits of the user and the unmanned aerial vehicle respectively.

Images