Mobile Edge Computing Network Task Scheduling and UAV Resource Deployment Method

Abstract

This application discloses a method for task scheduling in a mobile edge computing network and UAV resource deployment. Considering the CPU temperature constraint, by jointly optimizing user offloading decisions, computing decisions, UAV deployment locations, and computing resource allocation, the minimum maximum user delay is obtained. The method provided in this application further decomposes the optimization problem into three sub-problems and processes them iteratively in sequence. In specific embodiments, through simulation experiments, the obtained results show that the method proposed in this application has better performance compared with other algorithms.

Description

Technical Field

[0001] The present application relates to the field of drones, edge computing combined with the Internet of Things, and particularly to a method for mobile edge computing network task scheduling and drone resource deployment. Background Art

[0002] Due to the characteristics of high mobility and low cost of unmanned aerial vehicles (UAVs), in the prior art, UAVs are used as airborne MEC (Mobile Edge Computing) servers, effectively expanding their service coverage in resource-scarce areas. Secondly, UAVs can also be used as aerial relays to assist users in task offloading. The UAV network can provide various services for Internet of Things devices by flexibly adjusting the deployment positions of UAVs, such as computing offloading, data collection, and content caching.

[0003] However, compared with the MEC servers integrated on base stations, affected by hardware costs, user experience design, and deployment environment, the computing power, size, and weight of airborne MEC servers are relatively less, smaller, and lighter. In order to provide better computing services for UAVs and ensure greater mobility, it is necessary to comprehensively consider the computing power, size, and weight of airborne MEC servers. For example, the DJI Manifold 2 platform uses an Intel Core i7-8550U processor with a CPU main frequency of 1.8 GHz. At the same time, the length (width) and thickness of the DJI Manifold 2 are only 11 cm and 2.6 cm respectively, and its weight is less than 200 grams. On the other hand, with the development of semiconductor technology used in chips and the exponential improvement 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 chip heat dissipation and temperature control.

[0004] Existing literature on UAV-supported MEC networks mainly focuses on the energy consumption of UAVs during computing, hovering, or flying, and does not disclose how to solve the problem of excessive CPU temperature in the foregoing situations. Existing commonly used DVFS scheduling alleviates or solves the hardware limitations (i.e., CPU temperature) of airborne MEC servers and improves the efficiency of computing resource allocation. For example, CN201811143682.6, a computing offloading scheduling method based on deep reinforcement learning; CN202111095551.7, a distributed computing offloading method based on computing-network collaboration in a random network; existing methods respectively establish a composite scenario set of residence time and waiting delay according to the random movement and burst computing requirements of users; adopt posterior recourse actions to compensate for game strategies, and establish a game-based stochastic programming model for both the device side and the MEC server; by constructing a scenario tree, transform the multi-stage stochastic programming problem of both the device side and the MEC server into a DEP problem, and mainly solve to obtain the optimal task strategy for MEC server offloading and the optimal quoting strategy of the MEC server for the device side.

[0005] In the case of considering load balancing, the above existing method does not consider the communication stability problem between drones. Secondly, CPU temperature control is not considered in minimizing the maximum user delay.

[0006] Moreover, in the above existing method, the computing power and energy supply of the computing server at the mobile edge cannot be effectively optimized, and the device supply at the mobile edge is limited. It is impossible to optimize and adjust the drone load according to the uneven distribution of IoT users, resulting in unbalanced drone loads and being unable to solve the problem of over-high CPU temperature caused by partial overload. Summary of the Invention

[0007] The present application provides a method for mobile edge computing network task scheduling and drone resource deployment to effectively improve the utilization efficiency of communication and computing resources in the MEC network supported by drones. At the same time, the method provided by the present application considers a multi-drone collaborative computing system. In this system, drones hovering above users can provide services for resource-scarce areas. For example, in rural areas, using drones to assist communication in cases such as temporary emergency rescue can well make up for the deficiencies of fixed MEC. However, as a flying server, the computing power and energy supply of drones are limited. When applied to applications with large amounts of computing, or in scenarios where the drone load is unbalanced due to uneven user distribution, after the CPU temperature exceeds the normal operating temperature value, the computing performance of the CPU drops severely, and it is unable to continue to process data normally.

[0008] The present application provides a method for mobile edge computing network task scheduling and drone resource deployment, including the following steps:

[0009] Step S10: Apply the BCD method to decouple the optimal deployment position Q = {q m}, the optimal task offloading scheduling strategy optimal task scheduling strategy computing resource allocation strategy to obtain the user offloading scheduling strategy and computing scheduling strategy P1, the drone deployment position P2, and the resource allocation strategy P3. The optimization goal is to minimize the user processing time;

[0010] Step S20: For the user offloading scheduling strategy and computing scheduling strategy P1: Given the deployment position {Q} and the computing resource allocation {F}, use the genetic algorithm to obtain the optimal user offloading strategy and computing strategy {A * , C *};

[0011] Step S30: For the drone deployment position P2: According to the initial computing resource allocation {F} and the optimal user offloading strategy {A *}, the optimal task scheduling strategy {C *}, using the first-order Taylor expansion to transform the non-convex objective function and constraints into convex objectives and constraints, and then converting the non-convex problem into a convex optimization problem to optimize the UAV deployment location {Q}, until it converges to a tolerable accuracy to obtain the optimal {Q *};

[0012] Step S40: For the resource allocation strategy P3: Under the given conditions {A * , C * , Q *}, P3 is a convex problem, and the optimal computing resource allocation strategy {F *} is obtained based on CVX iteration;

[0013] Step S50: Sequentially and iteratively optimize P1, P2, and P3 obtained in Steps S20 to S40, and update the corresponding variables to obtain the optimal UAV deployment plan and resource allocation plan.

[0014] Preferably, Step S20 includes the following steps:

[0015] Adopt the GA method to satisfy the constraints s.t.:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029] Obtain the optimal solution {A * , C *}, where the constraint B1 represents the delay constraint within which the task needs to be completed; the constraint B2 means that as an MEC server, the computing resources of the UAV are also limited, so there are resource constraints allocated to users; the constraint B3 represents the CPU chip temperature constraints of the UAV and the user, and the constraints B4 and B5 respectively represent the energy consumption resource constraints of the user and the UAV; the constraint B6 means that the user needs to be within the maximum communication distance range of the UAV, and the distance constraint between the user and the UAV needs to be satisfied; the constraint B7 represents the constraint between the offloading decision and the computing decision, and the constraints B8 and B9 mean that the user tasks selected for offloading may be computed on the offloading UAV or may be forwarded to other UAVs for computing; the constraints B10 and B12 respectively represent the value ranges of the offloading decision constraint and the computing decision constraint.

[0030] Preferably, the GA method includes the following steps:

[0031] Step S21: Initialize the offloading or computing of each individual task in the population using binary coding mode;

[0032] Step S22: Calculate the fitness of each individual in the initialized population;

[0033] Step S23: Select individuals using the roulette wheel method;

[0034] Step S24: Perform crossover and mutation with a certain probability to generate new individuals;

[0035] Step S25: Repeating steps S23 and S24 multiple times, the multiple sets of new individuals obtained form a new population. Repeat step S22 to calculate the fitness value of each individual in the obtained population;

[0036] Step S26: Determine whether each individual in the populations obtained in step S25 satisfies the constraint s.t. If satisfied, output the highest fitness and its corresponding individual; if not satisfied, return to steps S22 - S25.

[0037] Preferably, step S22 includes the following steps:

[0038] Step S221: Determine whether each individual in the initial population satisfies the constraint s.t.;

[0039] Step S222: If individual n meets the constraint, then calculate its fitness according to the following formula:

[0040] fit n = T - max(T n ) (0.3)

[0041] where T is a constant used to ensure that the fitness is a positive value;

[0042] If an individual does not meet the constraints, its fitness is set to 0.

[0043] Preferably, step S23 is specifically:

[0044] Calculate the fitness fit of each individual i , and calculate the probability of each individual being selected by the ratio of the fitness of each individual to the fitness of all individuals as:

[0045]

[0046] And calculate the cumulative probability of each individual according to the following formula:

[0047] Pr s ={pr1, pr1 + pr2, pr1 + pr2 + pr3,..., pr1 + pr2 +... + pr I}

[0048] where Pr s can represent the cumulative probability of different intervals.

[0049] Preferably, step S24 includes the following steps:

[0050] Step S241: Crossover operation: The process of crossing two parents according to a preset probability to generate new offspring. Assuming the crossover probability is Pc, for each individual, there is a random number between 0 and 1. When the random number is less than Pc, the genes of two adjacent individuals are crossed to obtain two new individuals;

[0051] Step S242: Mutation operation: Mutate the individual genes with a relatively small mutation probability.

[0052] Preferably, step S3 includes the following steps:

[0053] Step S31: Initialize parameters and set the number of iterations i: After initialization, calculate the resource allocation {F}, the optimal user offloading strategy {A *}, the optimal task computing strategy {C *}, and the initial position {Q0} of the UAV. To achieve the minimum maximum user processing time for the UAV deployment position problem P2 and satisfy the following constraint conditions s.t. as the iteration stop condition;

[0054] Step S32: Calculate the communication rate and distance between the user and the UAV according to the initial values. By solving the UAV deployment position problem P2, obtain the deployment position {Q *} of the UAV, the communication rate between the user and the UAV The communication rate between UAVs Then update: Stop when the following constraints s.t. requirements are met;

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062] X min ≤X m ≤X max (E7)

[0063] Y min ≤Y m ≤Y max (E8)

[0064] Constraint E1 represents the maximum delay requirement of the user, where {D n} represents the size of the user's task volume, {θ n} represents the amount of computing resources occupied by the user. Constraint E2 represents the total energy consumption constraint of the UAV, where represents the hovering power of the UAV, represents the flight time of the UAV, {β m} represents the computing power constant. Constraint E3 represents the user's transmission rate constraint, where represents the gradient of the user's upload rate, {B} represents the bandwidth size. Constraint E4 represents the transmission rate constraint between UAVs, where represents the gradient of the transmission rate between UAVs; Constraints E5 and E6 respectively represent the distance requirements between the user and the UAV and between UAVs; Constraints E7 and E8 represent the range requirements for the deployment positions of UAVs.

[0065] Preferably, step S4 specifically includes the following steps:

[0066] Based on the obtained optimal user offloading strategy {A *}, optimal task computing strategy {C *}, and the optimal deployment position Q of the UAV *, under the constraints of CPU temperature, energy consumption, and computing resources, the resource allocation strategy P3 problem is solved by CVX to minimize the maximum user processing time and ensure that the optimization result meets the constraint conditions st.

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073] Among them, constraint F1 represents the task completion time constraint; constraint F2 represents the total CPU resource limit of the UAV; constraint F3 represents the total energy consumption constraint limit of the UAV, and constraints F4 and F5 respectively represent the CPU temperature limits of the user and the UAV.

[0074] The beneficial effects that this application can produce include:

[0075] 1) The mobile edge computing network task scheduling and UAV resource deployment method provided by this application proposes an edge computing cooperation network based on multiple UAVs. In the UAV-assisted mobile edge computing system, a mobile edge computing network based on multiple UAVs is defined. The multi-UAV system can cooperate to complete tasks more effectively and economically, achieve a multi-level reduction in the total delay, and in addition, can significantly improve the computing efficiency and resource utilization rate, and balance the network computing load. Compared with the single-UAV system, it can better handle complex and changeable actual scenarios. Secondly, by optimizing the UAV deployment location, the coverage can be guaranteed while reducing the transmission cost..

[0076] 2) The mobile edge computing network task scheduling and UAV resource deployment method provided by this application considers separating the offloading process and the computing process to implement a multi-level scheduling mechanism. Realize the separation of transmission and computing, establish the association between the offloading scheduling decision and the computing scheduling decision, and further describe the process of user tasks being forwarded. Through task forwarding scheduling, the system resources are reallocated, the energy consumption and delay of users are reduced, and obvious effects are also achieved in traffic balance and computing balance.

[0077] 3) The mobile edge computing network task scheduling and UAV resource deployment method provided by this application considers the fairness problem of the total task processing delay under temperature control. As the CPU utilization rate and memory utilization rate increase, the energy consumption and CPU temperature will rise, while the heat dissipation performance of user equipment and MEC servers themselves is poor. Therefore, in this article, the temperature constraints of the device side and the server side are considered. The control of the CPU temperature is achieved, reducing the excessive loss of the CPU and extending the service life of the device.

[0078] 4) The mobile edge computing network task scheduling and UAV resource deployment method provided by this application uses the Successive Convex Approximation (SCA) technique and the Block Coordinate Descent (BCD) method to obtain a double-loop iterative algorithm to solve the formulated MINLP problem. In addition, since the computational complexity of the Branch and Bound (B&B) algorithm for solving P1 is high, especially for a large number of users and UAVs, the GA algorithm is used to reduce the complexity of solving P1.

[0079] 5) The mobile edge computing network task scheduling and UAV resource deployment method provided by this application evaluates the performance of this method from two perspectives: the maximum user task processing delay and the UAV load balancing. It can be seen that the performance of the OFCS scheme proposed in this method is significantly better than the comparison scheme OCS. Secondly, because the influence of CPU temperature is considered, it will increase the time for user task processing, but it can meet the service quality requirements of user tasks while ensuring the system performance. Description of the Drawings

[0080] Figure 1 It is a schematic diagram of the MEC system structure for multi-UAV collaborative computing in a specific embodiment of this application;

[0081] Figure 2 It is a line graph comparing the maximum user delays under different numbers of users and different CPU frequencies in an embodiment of this application;

[0082] Figure 3 It is a line graph comparing the maximum user delays under different temperature constraints and different CPU frequencies in an embodiment of this application;

[0083] Figure 4 It is a line graph comparing the maximum user delays under different temperature thresholds and the number of users in an embodiment of this application;

[0084] Figure 5 It is a line graph comparing the maximum user delays under different bandwidths and scheduling mechanisms in an embodiment of this application;

[0085] Figure 6 It is a line graph comparing the maximum user delays under different numbers of UAVs and scheduling mechanisms in an embodiment of this application;

[0086] Figure 7 This is a comparative line chart of the communication delay of users with different user distribution positions and the number of users in the embodiments of this application;

[0087] Figure 8 This is a comparative line chart of the influence of different resource allocation methods and CPU frequencies on task delay in the embodiments of this application;

[0088] Figure 9 This is a bar chart and a line chart of the drone load under different scheduling schemes in the embodiments of this application; Figure 10 This is a schematic flow chart of the task scheduling of the mobile edge computing network and the drone resource deployment method provided by this application.

[0089] Symbol definition:

[0090] 1. The deployment position variable Q of the drone = {q m};

[0091] 2. The offloading scheduling decision variable

[0092] 3. The computing scheduling decision variable

[0093] 4. The computing resource allocation strategy

[0094] 5. The individual fitness parameter fit n ;

[0095] 6. The user location information w n ;

[0096] 7. The drone location information q m ;

[0097] 8. The user upload rate r n,m ;

[0098] 9. The transmission rate between drones

[0099] 10. The user task computing volume D n ;

[0100] 11. The maximum CPU resource F of the drone m ;

[0101] 12. The maximum energy consumption of the drone Detailed implementation manners

[0102] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0103] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0104] Technical means not detailed in this application and not used to solve the technical problems of this application are set according to common general knowledge in the art, and various common general knowledge setting methods can be implemented.

[0105] This application proposes a thermal perception task scheduling and resource allocation strategy to minimize the maximum user processing task time; since the problem described is a complex mixed-integer non-linear programming problem (MINLP) with strongly coupled variables, in order to further decouple these variables, it is transformed into three sub-problems that are easier to handle:

[0106] 1) P1, user offloading scheduling strategy and computing scheduling strategy;

[0107] 2) P2, the deployment location of the unmanned aerial vehicle;

[0108] 3) P3, resource allocation strategy, and process iteratively in sequence.

[0109] Driven by the aforementioned pioneering work, in a multi-unmanned aerial vehicle collaborative computing system, in order to balance the load balance between unmanned aerial vehicles, each unmanned aerial vehicle can be regarded as an additional assistant to other unmanned aerial vehicles. The device offloads its tasks to the unmanned aerial vehicle and calculates or forwards them to other unmanned aerial vehicles for collaborative computing.

[0110] I. System Model and Problem Formulation

[0111] Refer to Figure 1 , the multi-unmanned aerial vehicle collaborative system used in this application, which is an edge collaboration scenario composed of "multi-unmanned aerial vehicles, multi-users", including a group of unmanned aerial vehicles represented by M = {1, 2, 3,... M}, and a group of random users N = {1, 2, 3,..., N}.

[0112] Specifically, a computing processor with limited computing power is deployed on each UAV. Suppose each user has a computing-sensitive task, and the user task is represented by a triple, where D n represents the size of the nth user task, in MB. θ n represents the CPU cycles required for the user task to be completed, represents the upper limit of the delay tolerable by the user, and the tolerances of different users are different. The UAVs are deployed above the users to assist the users in task processing, and the height is a fixed constant H.

[0113] Meanwhile, considering the three-dimensional coordinate system (3D), the position of the user is w n ={x n , y n , 0}. In addition, the 3D position of the mth UAV is defined as q m ={x m , y m , H}, so the distance between the nth user and the mth UAV is

[0114] Suppose there are m and j representing any two UAVs, then the distance between the mth UAV and the jth UAV is expressed as where q j is the 3D position of the jth UAV.

[0115] Define the binary variable to control the task offloading decision, represents the control task computing decision.

[0116] Specifically, means that user n offloads the task to UAV m, and define then it means that the task of the nth user is offloaded to the a n th server.

[0117] Define If c n =m, it means that the task of the nth user is computed on the mth UAV.

[0118] If a n =c n , it means that the UAV for task offloading and the UAV for computing are the same, and this task is not forwarded after offloading. Conversely, the nth task is forwarded from the a n th UAV to the c n th UAV.

[0119] For a certain user task, it may be computed locally or offloaded to a UAV for computing. Therefore, it needs to satisfy For the tasks offloaded to the UAV for computing, the prerequisite is that offloading is selected, and finally, the computing can only be performed on one UAV. Therefore, the following conditions need to be satisfied

[0120] In addition, for the tasks computed on the UAV, they may be computed directly after offloading or after being forwarded. It is only considered that the user's tasks are forwarded once.

[0121] Communication mode

[0122] This paper considers the A2G link and the A2A link, and assumes that in the wireless channels of the above links, the line-of-sight channel based on the free space transmission loss model is considered. Assume that the channel gain between the nth user and the mth UAV is Then is defined as follows:

[0123]

[0124] where β0 is the reference value of the channel gain at 1m.

[0125] In order for the tasks offloaded by the user to be successfully received by the UAV, the user must be within the coverage area of the UAV to which it offloads, that is, not exceeding the maximum distance D covered by the UAV max , which can also be called the maximum communication distance. It is specifically expressed as follows:

[0126]

[0127] Secondly, in order to better utilize the resources within the system, any m, j ∈ M can cooperate. That is, when the load of the mth UAV is greater than the load of the jth UAV, in order to balance the load between the two, the mth UAV can forward the user tasks offloaded to it to the jth UAV for computing. The channel gain is calculated by the following formula:

[0128]

[0129] In order to avoid collisions between UAVs, the distance between any two UAVs should not be less than the minimum safety distance D min , that is It is specifically expressed as follows:

[0130]

[0131] Since the considered multi-UAV cooperative computing system is based on FDMA (Frequency Division Multiple Access), the UAVs and devices will share the common bandwidth B during the offloading and forwarding processes. Therefore, the transmission rates of offloading and forwarding from the nth user to the mth UAV and from the mth UAV to the jth UAV are respectively expressed as the following relationships:

[0132]

[0133]

[0134] Among them, P n and P m They represent the transmission power of the nth user and the mth UAV respectively, and N0 is the noise power spectral density of the UAV and the base station.

[0135] B. Delay correlation

[0136] (1) Local computing

[0137] User tasks can be computed locally or offloaded to the drone. , it means that a user's task is calculated locally, and the calculation time of the task executed locally is as follows:

[0138]

[0139] Among them, D n represents the task size of user n, θ n Indicates the CPU cycles required for the local user to calculate 1 bit of data, f n represents the computing power of the local user processor. In this paper, all users are considered to be homogeneous, so it is assumed that f n is the same constant, and these users are only powered by their own limited battery capacity. However, in reality, different users have different latency requirements. Users who are more sensitive to latency requirements may use their own computing power as much as possible to meet latency-sensitive requirements, which will lead to higher CPU energy consumption. At the same time, the limited battery capacity may not be able to support their latency requirements. In this regard, users need to use drone servers with greater computing power for auxiliary calculations, which can not only meet their own latency requirements, but also save their own energy consumption.

[0140] (2) Offloading calculation

[0141] when When the user offloads the task to the drone, the task may be calculated on the unloaded drone or forwarded by the current drone to other drones for execution. This involves two situations: on the one hand, the task is directly calculated after being offloaded, and on the other hand, the user unloads and forwards it for calculation.

[0142] First is the unloading process. The transmission time from the user to the drone can be calculated as follows:

[0143]

[0144] Next is the calculation process, If the task of the nth user is executed on the jth drone, the calculation time of the task is as follows:

[0145]

[0146] f n m represents the computing resources allocated by the drone to the user.

[0147] Then the forwarding delay is defined as follows:

[0148]

[0149] In summary, for the task processing total delay of any user n, it includes the time of four processes, namely local computing time, offloading time, computing time, and forwarding time. Specifically, when the drone for user offloading and the drone for task computing are the same, then the total delay includes local computing delay, communication delay, and drone computing delay, which is expressed by the following formula:

[0150]

[0151] Expanding the above formula, we get the following formula:

[0152]

[0153] Otherwise, the total delay includes four parts, namely local computing delay, drone computing delay, and communication delay, where the communication delay includes the communication between drones and other drones as well as between drones and users. It is expressed by the following formula:

[0154]

[0155] Expanding the above formula is as follows:

[0156]

[0157] Specifically, in the above formula, Φ n represents the task forwarding matrix of user n, indicating that the user task is forwarded from the offloading drone to the computing drone. The rows and columns of the matrix represent the offloading drone serial number and the computing drone serial number respectively. In the above formula, m represents the offloading drone and j represents the computing drone.

[0158] C. Energy Consumption Analysis

[0159] In this paper, different computing modes are considered for tasks that need to be calculated, including local computing and offloading computing. Among them, offloading computing includes direct computing after offloading and computing after offloading and forwarding. Among the different computing modes considered, whether the user's task chooses local computing or offloads it to the drone for computing, there is energy consumption. First, if the user task is executed locally, the energy consumption during its computing process can be expressed as follows:

[0160]

[0161] In the above formula, β n represents the effective capacitance coefficient of the nth user, which mainly depends on the chip structure of the processor (CPU).

[0162] Secondly, for mobile devices with computationally intensive and latency-sensitive tasks, their own computing power may not be able to complete within the required deadline, and the small energy capacity of their batteries also makes it difficult to support the energy consumption generated by their movement and task execution. At this time, the user can offload the task to the drone for computing, which can effectively relieve the dilemma that the user's own ability is insufficient to meet the demand. However, during the process of offloading the task to the drone, the user will also have offloading energy consumption. The offloading required for user n to offload the task to drone m is expressed as follows:

[0163]

[0164] If the user task chooses to offload to the drone for computing, then the drone needs to allocate computing resources to the user. Therefore, the energy consumption required for the task of user n to be computed on drone m is It can be obtained that:

[0165]

[0166] Among them, β m is the effective capacitance coefficient depending on the chip structure of the drone processor. In addition, if there is a task forwarded from the mth drone to the jth drone, there will be energy consumption on drone m during this process That is:

[0167]

[0168] Finally, both the hovering and movement of the drone need to maintain propulsion energy consumption. Usually, the propulsion energy consumption of the drone is determined by its acceleration and speed. In the scenario, only when the user has a computing demand, the drone needs to hover above the user's head for a period of time to provide computing services for the user. At this time, its speed is 0, so the propulsion energy consumption of the drone is a constant. Therefore, in this paper, the hovering energy consumption of the drone is mainly considered. Its hovering time and energy consumption are shown as follows:

[0169]

[0170]

[0171] To ensure that all users who need to uninstall within the scope are uninstalled and all tasks that users have uninstalled are calculated, the hovering time of the UAV is the maximum user task completion time. In addition, is the received power when the UAV hovers.

[0172] In summary, the total energy consumption required for a certain UAV to complete the computing tasks of users within a certain range includes computing energy consumption, forwarding energy consumption, and hovering energy consumption. Specifically, it is as follows:

[0173]

[0174] D. Power and Thermal Energy Model of UAV

[0175] Whether the user task is calculated locally or selected to be offloaded to the MEC server for calculation, due to its possible millisecond-level delay requirements and at the same time it may be a task of a relatively large order of magnitude. And at this time, in order to meet the delay requirements, it is necessary to process the task with as much computing resource as possible, while the heat dissipation performance of the user device and the MEC server itself is not good, which will inevitably lead to a relatively high CPU chip temperature and even damage the performance of the CPU chip. For the periodic task τ i , there is an associated period p i and an execution time e i . At this time, the processor utilization rate of the task τ i is For periodic tasks, the task-specific dynamic power consumption increases with the processing time of the task or the utilization rate of the processor, while the wasted power increases with the increase in temperature. Therefore, for a given task set τ i , its total power consumption is expressed as follows:

[0176] P tot = P dyn + P leak

[0177] In the formula, P dyn represents the dynamic power consumption, and P leak represents the wasted power. P dyn is expressed by the following formula:

[0178]

[0179] In the above formula, V, f, e i , p i , represent the working voltage, computing frequency, computing time, period, and task switching activity factor respectively.

[0180] The temperature of the processor chip will finally reach the steady-state temperature determined by the total power consumption, and it is measured that this temperature range is 20°C to 60°C. The temperature of the processor chip can be specifically expressed as follows:

[0181] Γ chip = P tot R th + Γ a

[0182] Γ chip = P tot R th + Γ a

[0183] In the above formula, Γ a represents the temperature of the surrounding environment. R th represents the thermal resistance of the chip. Assuming that the chip temperature threshold is represented by , then the above formula is transformed to obtain the total power consumption as follows:

[0184]

[0185] In the above formula, the left side of the inequality is the total power consumption, and the terms on the right side can be called the available thermal budget.

[0186] Combined with the scenario considered in this article, this article uses the ratio of the actually allocated computing resources to the total computing resources as the CPU utilization rate. When the nth user task is calculated on the mth drone, the total power consumption of the CPU chip of the drone is expressed as follows:

[0187]

[0188]

[0189] Similarly, when the user task is calculated locally, the total power of its CPU chip is as follows:

[0190]

[0191] In the above formula, since the total resources F n owned by the user do not need to be allocated again, that is, f n = F n , and at this time the CPU utilization rate of the user is 1.

[0192] E. Problem description

[0193] The deployment location variable of the drone is expressed as Q = {q m}, the offloading scheduling variable is expressed as The computing scheduling variable is expressed as And the computing resources are expressed as In this paper, the above variables are jointly optimized to minimize the maximum user processing time, and the formula is:

[0194]

[0195]

[0196]

[0197]

[0198]

[0199]

[0200]

[0201]

[0202]

[0203]

[0204]

[0205]

[0206]

[0207]

[0208]

[0209]

[0210] In the above problem, the objective function is to minimize the maximum total delay, and the optimized deployment variable Q = {q m}, the offloading scheduling variable and the computing scheduling variable as well as the computing resources Secondly, the constraint A1 indicates that the task needs to be completed within its delay constraint range. Considering that different users have different QoS (Quality of Service) requirements, the maximum tolerable delay for each user is different; the constraint A2 indicates that as an MEC server, the computing resources of the UAV are also limited. Therefore, the resources allocated to users cannot exceed the resources it itself has; the constraints A3 and A4 represent the CPU chip temperature constraints of the UAV and the user respectively, and the constraints A5 and A6 represent the energy consumption resource constraints of the user and the UAV respectively; the constraint A7 indicates that in order to avoid collisions between UAVs, a minimum safe communication distance needs to be maintained between different UAVs; the constraint A8 indicates that the user needs to be within the maximum communication distance range of the UAV in order to offload the task to the UAV; the constraint A9 indicates that the selected offloaded user task may be calculated on the offloaded UAV or may be forwarded to other UAVs for calculation; the constraints A10 and A11 respectively indicate that a task can only be offloaded to one UAV and a task can only be calculated on one UAV; the constraint A12 indicates that a task can only be offloaded once and can only be forwarded once. The constraints A13 and A14 indicate that the value ranges of the variables and can only be 0 or 1; the constraint A15 indicates that the deployment range of the UAV position should be able to cover the distribution area of the users.

[0211] II. Algorithm Proposal

[0212] Based on the observation of P, due to the coupling of and q m in the objective function and constraints, and coupling, and q m coupling, etc., it can be seen that this is a very difficult MINLP problem. In addition, the integer variables {A, C} further increase the difficulty of solving this problem. Therefore, it is impossible to directly obtain the solution of P in the current form. Inspired by the iterative design, the BCD method is applied to decouple these variables, and P is divided into the following sub-problems:

[0213] 1) P1, the user offloading scheduling strategy and the computing scheduling strategy;

[0214] 2) P2, the UAV deployment location;

[0215] 3) P3, the resource allocation strategy;

[0216] First, given the deployment location Q, calculate the resource allocation {F}, and use the genetic algorithm to obtain the optimal user offloading strategy and computing strategy {A * , C *}, and update {T n}.

[0217] Second, according to the initial computing resource allocation {F}, the optimal user offloading strategy {A *}, and the optimal task computing strategy {C *}, use the first-order Taylor expansion to transform the non-convex objective function and constraints into convex objectives and constraints, and then convert the non-convex problem into a convex optimization problem. Optimize the UAV deployment position Q until it converges to a tolerable accuracy to obtain the optimal Q * , and at the same time update {T n} again;

[0218] Third, under the given conditions {A * , C * , Q *}, P3 is a convex problem. Based on CVX iteration, obtain the optimal computing resource allocation strategy {F *};

[0219] Finally, iteratively optimize P1, P2, and P3 in sequence and update the relevant variables. This process is called the BCD method;

[0220] Through the iterative method based on the BCD method, jointly optimize task offloading scheduling, task computing scheduling, UAV deployment position, computing resource allocation, etc., and give the computational complexity of this algorithm.

[0221] A. Task offloading scheduling, task computing scheduling

[0222] As described above, the offloaded user tasks will be scheduled to the UAV for computing according to the following two cases:

[0223] 1) Forwarding. The tasks offloaded to a certain UAV can be forwarded to another UAV for computing;

[0224] 2) Non-forwarding. Compute immediately on the offloaded UAV.

[0225] Since the offloading decision and the computing decision are strongly coupled and have a large mutual influence, by solving P1 under the given conditions {Q, F}, jointly optimize the user task offloading scheduling and task computing scheduling to minimize the maximum user processing time.

[0226]

[0227]

[0228]

[0229]

[0230]

[0231]

[0232]

[0233]

[0234]

[0235]

[0236]

[0237]

[0238]

[0239] Due to the coupling of binary variables, it is strongly non-convex. To handle this integer-related subproblem, the GA method is adopted to obtain a suboptimal solution {A * , C*}. The GA solution process is as Figure 2 shown below:

[0240] (1) Initialize the population.

[0241] A population consists of multiple individuals, each individual has a chromosome, and there are multiple genes on the chromosome.

[0242] The variables to be solved are binary variables, so the binary coding mode is adopted. Then one variable in the problem corresponds to one gene in the chromosome. Assuming that N users and M UAVs are considered, the length of one chromosome is 2×N×M. As shown in Table 1 below, it shows a chromosome when N = 3 , M = 2. Every two genes represent the offloading or computing of a user's task. For example, if both gene 1 and gene 2 are 0, then the task of user 1 is neither offloaded to UAV 1 nor to UAV 2. If gene 9 and gene 10 are 0 and 1 respectively, it means that the task of user 2 is computed on UAV 2. Therefore, in the following chromosome, the blue part represents the offloading gene representation of all user tasks, and the orange part represents the computing gene representation of all user tasks.

[0243] Table 1: Chromosome sequence

[0244]

[0245] (2) Calculate the fitness of each individual.

[0246] For individuals that meet all the constraints, when evaluating their quality, the smaller the time delay, the larger the fitness value. Therefore, the negative of the objective function is used as the fitness function. Secondly, to ensure that the fitness is positive, the fitness function can be expressed as fit n = T - max(T n ), where T is a constant. For individuals that do not meet the constraints, their fitness values are directly assigned as 0.

[0247] (3) Selection

[0248] In this paper, the roulette wheel selection method (also known as the proportional selection method) is adopted for selection. First, calculate the fitness fit n of each individual, and calculate the probability of each individual being selected through the ratio of the fitness of each individual to the fitness of all individuals as follows:

[0249]

[0250] And calculate the cumulative probability of each individual, which can be calculated as follows:

[0251] Pr s = {pr1, pr1 + pr2, pr1 + pr2 + pr3,..., pr1 + pr2 +... + pr I}

[0252] Furthermore, different intervals are formed, and the size of the interval represents the size of the fitness value. Therefore, the larger the interval span, the greater the probability of its being selected.

[0253] (4) Crossover

[0254] The crossover operation is a process of crossing two parent generations with a certain probability to generate new offspring. Assume the crossover probability is Pc. For each individual, there is a random number between 0 and 1. When the random number is less than Pc, the genes of two adjacent individuals are crossed to form two new individuals.

[0255] (5) Mutation

[0256] The mutation operation refers to mutating the genes of an individual with a very small mutation probability, such as from "0" to "1", or from "1" to "0", so as to expand the diversity of the population.

[0257] After steps (3), (4), and (5), a new population is formed. Repeat step (2) to calculate the fitness value of each individual in the population. If the exit condition is met, output the highest fitness and its corresponding individual. If not, continue to repeat steps (2), (3), (4), and (5).

[0258] The pseudo-code of the above algorithm is shown in Table 2 below:

[0259] Table 2: Solving the offloading decision and computing decision sub - problems by genetic algorithm

[0260]

[0261] B. UAV position deployment optimization sub - problem

[0262] By solving the P1 sub - problem, the optimal offloading scheduling decision is known Compute the scheduling decision and the initial values of the variables to solve the UAV deployment location sub - problem.:

[0263]

[0264]

[0265]

[0266]

[0267]

[0268] X min ≤X m ≤X max (C5)

[0269] Y min ≤Y m ≤Y max (C7)

[0270] For the non - convex part in the above - mentioned problem, the specific treatment is as follows:

[0271] (1) Since the rate formula is non - convex with respect to the UAV position q m non - convex

[0272] The communication rate involved in the communication process between the nth user and the mth UAV is expanded as follows:

[0273]

[0274] Use SCA (Successive Convex Approximation) to approximate the non - convex function into a convex function. To facilitate the solution of this problem, slack variables are introduced Process the non - convex part of the rate in the objective function and constraints, and at the same time add the following new constraints:

[0275]

[0276] First, assume The above formula is expressed as follows:

[0277]

[0278] At this time, regarding the variable q in the left side of the above formula m is non-convex, but regarding ||q m - w n || term is convex, and the lower bound of the rate is obtained through the first-order Taylor expansion. Assume is q at the i-th iteration m , then the lower bound of the rate is obtained by the following formula:

[0279]

[0280] where represents the rate at the i-th iteration:

[0281] represents the first derivative of r n,m .

[0282] Similarly, for the non-convex part of the communication rate between any two drones shown in the following formula, the same treatment is also carried out. First, introduce the slack variable The newly added constraints are as follows:

[0283]

[0284] Secondly, assume then the above formula can be expressed as follows:

[0285]

[0286] Then, the approximation of is obtained by performing the first-order Taylor expansion on formula (4.3.20), as shown below:

[0287]

[0288] where is the rate at the i-th iteration,

[0289] is 's first derivative. Through the above treatment, the non-convex objective function is converted to the following:

[0290]

[0291] The original problem P2 is converted to the following problem P2:

[0292]

[0293]

[0294]

[0295]

[0296]

[0297]

[0298]

[0299] X min ≤ X m ≤ X max (D7)

[0300] Y min ≤ Y m ≤ Y max (D8)

[0301] Since constraints D5 and D6 are non - convex;

[0302] In problem P′2, the left - hand sides of constraints D5 and D6 are still non - convex. Using the SCA technique to relax the constraints. First, for the constraint Square both sides of its inequality sign to obtain the following formula:

[0303]

[0304] For any given point After a first - order Taylor expansion, the following inequalities exist:

[0305]

[0306] Similarly, for any given points and For constraint D6, the following inequalities exist:

[0307]

[0308] By transforming the non - convex objective function and non - convex constraints in the original problem, the original non - convex problem is transformed into a convex optimization problem as follows:

[0309]

[0310]

[0311]

[0312]

[0313]

[0314]

[0315]

[0316] X min ≤ X m ≤ X max (E7)

[0317] Y min ≤ Y m ≤ Y max (E8)

[0318] In problem P2, the constraints are convex and the objective function is convex. Therefore, this problem is a convex optimization problem and can be solved using standard convex algorithms, such as the interior point method and the Lagrangian dual method. In addition, convex optimization tools, such as CVX, can also be used for solving. In this paper, CVX is used for solving, and the solution process of this problem is as follows:

[0319] (1) Initialize the parameters. and is the optimal solution of problem P1, is the CPU resources initially allocated by the CPU to the user, represents the initial position of the UAV.

[0320] The pseudo-code of the CVX solution algorithm is shown in Table 3 below:

[0321] Table 3: UAV Position Deployment Optimization Algorithm

[0322]

[0323] C. Resource Allocation Sub-problem

[0324] By solving the above two sub-problems, the optimized solution offloading matrix A * , the calculation matrix C * and the optimized deployment position Q * have been obtained. These values are used as the input of the resource allocation sub-problem to optimize the resource allocation. This sub-problem is described as follows:

[0325]

[0326]

[0327]

[0328]

[0329]

[0330]

[0331] Among them, the constraint F1 represents the task completion time constraint; the constraint F2 represents the total CPU resource limit of the UAV; the constraint F3 represents the total energy consumption constraint limit of the UAV, and the constraints F4 and F5 represent the CPU temperature limits of the user and the UAV, respectively.

[0332] In problem P3, except for the variable all others are constants or known variables. Therefore, this problem is a single-variable optimization problem regarding

[0333] Among them, in the constraint F1, as shown below, in this problem, the first, second, and fourth terms on the left side of the inequality are all constant terms:

[0334]

[0335] Let Then the above formula is transformed into the following:

[0336]

[0337] Transforming the above formula, we get the following:

[0338]

[0339] In the above formula, the right side of the inequality is a constant. Then this constraint is linear with respect to the variable Therefore, this constraint is a convex constraint.

[0340] Similarly, it can be proved that the objective function and the constraints F2 - F5 are all affine with respect to the variable The objective function and the constraints of this problem are all convex. Therefore, the resource allocation problem is a convex optimization problem and can be directly solved using the CVX tool.

[0341] D. Overall algorithm and complexity analysis

[0342] By iteratively optimizing P1, P2, and P3 until the tolerable accuracy is achieved in the outer loop, a sub-optimal solution of the original algorithm can be obtained. The pseudo-code of the algorithm is shown in Table 4 below:

[0343] Table 4: Pseudo-code of the algorithm for the user's minimum-maximum task total delay problem

[0344]

[0345] ​The complexity analysis is as follows: For the offloading decision and computing decision sub-problems, the GA is used for solution. Refer to the literature (Fengxian 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]. IEEE / ACM Transactions on Networking, 2018-10-12, 26(6): 2651-2664) for the GA time complexity analysis. Considering there are N users and M UAVs, and the program running termination condition is the genetic algebra I. First, initialize a population of K individuals, then the complexity is O(K(2N)). The time complexity when calculating the fitness is O(I(KNM)+IK), which is abbreviated as O(I(KNM)). The time complexity of the selection operation is O(IK), and the time complexity of the crossover operation is The time complexity of the mutation operation is O(IK). In summary, the time complexity of the GA algorithm is Also, because K < NM, therefore, For the UAV deployment sub-problem, the sequential convex approximation (SCA) is used to transform the problem into a convex problem, and finally CVX is used for solution. Then the complexity of its algorithm is O(l1(M(N + M)) 3 ). Therefore, the complexity of the overall algorithm is O(g(I(KNM)+l1(M(N + M)) 3 ), where g is the number of outer loop iterations and l1 is the number of inner loop iterations.

[0346] It shows that the complexity of the algorithm proposed by the method provided in this application is within an acceptable range. Compared with the direct use of the bnb solution method, the complexity of the proposed algorithm is lower.

[0347] Embodiment

[0348] This embodiment conducts simulations according to the method provided in this application above. The simulation parameter settings are as follows:

[0349] Multiple devices are randomly distributed in a 200m×200m area. Each device has a task computing requirement. The UAV can act as both an MEC server and a relay to help users calculate or forward tasks. Unless otherwise specified, the simulation parameters are summarized in Table 5.

[0350] Table 5: Simulation parameters

[0351]

[0352] Simulation results and analysis:

[0353] The numerical results obtained from the simulation verify the effectiveness and performance of the proposed multi-UAV cooperative computing (OFCS) scheme.

[0354] To evaluate the performance of the overall algorithm of the proposed multi-UAV cooperative computing OFCS, the following reference schemes are considered as benchmarks and named:

[0355] 1) The offloading scheduling and the computing scheduling are separated, and each UAV is given a dual identity as an MEC server and a relay. The UAV can not only assist in the user task computing, but also forward the task to other UAVs for computing when overloaded.

[0356] 2) The offloading scheduling and the computing scheduling are not separated, and the UAV only serves as an MEC server, and the user's task is computed where it is offloaded.

[0357] 3) Without temperature constraints, considering the CPU temperature limit of the on-board MEC server.

[0358] 4) Random allocation and optimization, that is, considering the random initialization of users and optimizing their offloading decisions.

[0359] Figure 2 Experiments are shown for different numbers of users, which shows that when the total resources of the system increase, the total delay decreases. Because, in this experiment, when the number of users remains stable and the allocation ratio is consistent, increasing the total resources of the system will result in an increase in the resources allocated to each user. Therefore, the corresponding delay will decrease.

[0360] For the same computing resources, as the number of users increases, the delay generally shows an upward trend. Because, when the total computing resources in the system remain unchanged, the increase in the number of users leads to a decrease in the computing resources allocated to each user. Therefore, the total delay for users to process tasks increases accordingly.

[0361] In addition Figure 3 shows the influence of considering temperature on the actual computing, but Figure 3 in, when N = 12, after the 3rd point, the delay tends to be stable, because at the current point, the maximum computing resources that can be allocated under the temperature constraint have been reached, while in the case without temperature constraints, more computing resources are continuously allocated, so that the total task delay still shows a decreasing trend. For the first two points, the delay with temperature is the same as that without temperature, because at this time the total CPU resources are relatively small. Even though the threshold for UAV computing resource allocation has been reached, this threshold is less than or equal to the maximum computing resources that can be allocated under the temperature constraint.

[0362] In addition, Figure 4Furthermore, it further shows that as the temperature threshold increases, the time delay shows a downward trend, indicating that at a higher temperature threshold, users can be allocated more computing resources. Secondly, for the same temperature threshold, the more users there are, the higher the demand for resources within the system. Compared with a smaller number of users, the more users there are, the earlier the threshold of the computing resources that can be allocated under the temperature constraint is reached. Because the more users there are, the higher the resource utilization rate, and the faster the temperature threshold is reached.

[0363] As Figure 5 shown, as the bandwidth resources increase, the overall user time delay shows a downward trend. This is because the total bandwidth resources are evenly distributed to each user, and the number of users remains unchanged. Therefore, the time delay decreases as the bandwidth resources increase. And in Figure 6 it shows the impact of different numbers of UAVs on the time delay. It can be seen from the figure that as the number of deployed UAVs increases, the maximum user processing time delay in the two comparison schemes shows a gradually decreasing trend. This is because the increase in the number of UAVs in the system means that the computing resources in the whole system increase accordingly.

[0364] As Figure 7 shown, generally speaking, on the one hand, the communication time delay will increase as the number of users increases. On the other hand, the relative positions of the UAVs and users and the density of the user distribution jointly determine the allocation of CPU resources, which in turn affects the computing time delay. Secondly, the communication time delay and the maximum user time delay of W3 where users are most concentrated are smaller than those in the other two cases. The main reason is that the UAVs can be deployed close to the user distribution, and the users are relatively concentrated. Therefore, the overall communication distance is relatively small. For the case of W1, the user distribution is polarized, and on the basis of ensuring the minimum safety distance, the probability of a long user transmission distance is greater than that of W2 and W3. Therefore, load imbalance is more likely to occur.

[0365] Figure 8 It shows that as the total computing resources in the system increase, in the three schemes, the total time delay of user task processing gradually decreases. Among them, the time delay obtained from the randomly allocated initial computing resources is slightly higher than that after optimization. One reason is that the total resources in the system are increasing, and on the other hand, there are scheduling at two levels of offloading and computing, which can effectively improve the utilization rate of system resources, and thus the total time delay of task processing is smaller.

[0366] Figure 9 It shows that the load of the UAVs is analyzed from the number of users calculated by the UAVs and the proportion of the computing task volume. It can be seen that when there is computing scheduling, the load gap between each UAV is relatively small.

[0367] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for task scheduling in a mobile edge computing network and UAV resource deployment, characterized in that It includes the following steps: Step S10: Apply the BCD method to decouple the optimal deployment location Q = {q m}, the optimal task offloading scheduling strategy optimal task scheduling strategy computing resource allocation strategy to obtain the user offloading scheduling strategy and the computing scheduling strategy P1, the UAV deployment location P2, and the resource allocation strategy P3, with the optimization goal of minimizing the user processing time; Step S20: For the user offloading scheduling policy and the computing scheduling policy P1: Through the given deployment location {Q}, calculate the resource allocation {F}, and use the genetic algorithm to obtain the optimal user offloading policy and task scheduling policy {A * , C *}; Step S30: For the UAV deployment location P2: According to the initial computing resource allocation {F}, the optimal user offloading strategy {A *}, and the optimal task scheduling strategy {C *}, use the first-order Taylor expansion to transform the non-convex objective function and constraints into convex objectives and constraints, and then convert the non-convex problem into a convex optimization problem. Optimize the UAV deployment location {Q} until convergence to a tolerable accuracy to obtain the optimal {Q *}; Step S40: For resource allocation policy P3: Under the given conditions {A * , C * , Q *}, P3 is a convex problem, and the optimal computing resource allocation policy {F *} is obtained by CVX iteration; Step S50: Sequentially iterate through P1, P2, and P3 obtained from the optimized steps S20 - S40, and update the corresponding variables to obtain the optimal deployment plan and resource allocation plan for the UAV; Step S20 includes the following steps: Adopt the GA method, subject to the constraint s.t.: Obtain the optimal solution {A * , C *}, where the constraint B1 represents the delay constraint for the task; the constraint B2 indicates that as an MEC server, the computing resources of the UAV are also limited, so the resource constraint allocated to the user; the constraint B3 represents the CPU chip temperature constraints of the UAV and the user, and the constraints B4 and B5 respectively represent the energy consumption resource constraints of the user and the UAV; the constraint B6 means that the user needs to be within the maximum communication distance range of the UAV, and the distance constraint between the user and the UAV needs to be satisfied; the constraint B7 represents the constraint between the offloading decision and the computing decision, and the constraints B8 and B9 indicate that the user tasks selected for offloading may be computed on the offloading UAV or may be forwarded to other UAVs for computing; the constraints B10 and B12 respectively represent the value ranges of the offloading decision constraint and the computing decision constraint.

2. The method for mobile edge computing network task scheduling and UAV resource deployment according to claim 1, wherein The GA method includes the following steps: Step S21: Initialize the offloading or calculation of each individual task in the population using binary coding mode; Step S22: Calculate the fitness of each individual in the initialized population; Step S23: Select individuals using the roulette wheel method; Step S24: Perform crossover and mutation with a certain probability to generate new individuals; Step S25: The multiple new individual sets obtained by repeating steps S23 and S24 multiple times form a new population. Repeat step S22 to calculate the fitness value of each individual in the obtained population; Step S26: Determine whether each individual in the various populations obtained in step S25 satisfies the constraint s.t. If it satisfies, output the highest fitness and its corresponding individual; if it does not satisfy, return to steps S22 - S25.

3. The method for mobile edge computing network task scheduling and UAV resource deployment according to claim 2, wherein, Step S22 includes the following steps: Step S221: Determine whether each individual in the initial population satisfies the constraint s.t.; Step S222: If individual n meets the constraint, calculate its fitness according to the following formula: fit n = T-max(T n ) (0.1) where T is a constant used to ensure that the fitness is a positive value; If the individual does not meet the constraint, set its fitness to 0.

4. The method for mobile edge computing network task scheduling and UAV resource deployment according to claim 2, wherein, Step S23 is specifically: Calculate the fitness fit of each individual i , and calculate the probability of each individual being selected by the ratio of the fitness of each individual to the fitness of all individuals as follows: And calculate the cumulative probability of each individual according to the following formula: Pr s = {pr1, pr1 + pr2, pr1 + pr2 + pr3,..., pr1 + pr2 +... + pr I} Among them, Pr s can represent the cumulative probability of different intervals.

5. The method for mobile edge computing network task scheduling and UAV resource deployment according to claim 2, wherein Step S24 includes the following steps: Step S241: Crossover operation: The process of crossing two parents with a preset probability to generate new offspring. Assume the crossover probability is Pc. For each individual, there is a random number between 0 and 1. When the random number is less than Pc, cross the genes of two adjacent individuals to obtain two new individuals; Step S242: Mutation operation: Mutate the individual genes with a relatively small mutation probability.

6. The method for mobile edge computing network task scheduling and UAV resource deployment according to claim 1, wherein Step S3 includes the following steps: Step S31: Initialize parameters and set the number of iterations \(i\): After initialization, calculate the resource allocation \(\{F\}\), the optimal user offloading strategy \(\{A\}\) *}, the optimal task computing strategy \(\{C\}\) *}, and the initial position \(\{Q_0\}\) of the UAV. To minimize the maximum user processing time for the UAV deployment position \(P_2\) problem and satisfy the following constraint conditions s.t. as the iteration stop condition; Step S32: Calculate the communication rate and distance between the user and the UAV based on the initial value. By solving the UAV deployment location problem P2, obtain the UAV deployment location {Q *}, the communication rate between the user and the UAV The communication rate between UAVs Then update: Stop until the constraint conditions s.t. requirements shown below are met; X min ≤ X m ≤ X max (E7) Y min ≤ Y m ≤ Y max (E8) Constraint E1 represents the maximum delay requirement of the user, where {D n} represents the amount of tasks of the user, {θ n} represents the computing resource occupancy of the user. Constraint E2 represents the total energy consumption constraint of the UAV, where represents the hovering power of the UAV, represents the flight time of the UAV, {β m} represents the computing power constant. Constraint E3 represents the transmission rate constraint of the user, where represents the gradient of the user's upload rate, {B} represents the bandwidth size. Constraint E4 represents the transmission rate constraint between UAVs, where represents the gradient of the transmission rate between UAVs; Constraints E5 and E6 respectively represent the distance requirements between the user and the UAV and between UAVs; Constraints E7 and E8 represent the range requirements of the deployment positions between UAVs.

7. The method for mobile edge computing network task scheduling and UAV resource deployment according to claim 1, characterized in that, Step S40 specifically includes the following steps: Based on the obtained optimal user offloading strategy {A *}, optimal task computing strategy {C *}, and the optimal deployment location Q of the UAV * , under the constraints of CPU temperature, energy consumption, and computing resources, solve the resource allocation strategy P3 problem through CVX to minimize the maximum user processing time and ensure that the optimization result meets the constraint condition st; Among them, constraint F1 represents the task completion time constraint; constraint F2 represents the total CPU resource limit of the UAV; constraint F3 represents the total energy consumption constraint limit of the UAV, and constraints F4 and F5 respectively represent the CPU temperature limits of the user and the UAV.

Images