Management method and system for multi-UAV edge computing network
By optimizing cache placement, computing task loading and computing resource allocation in multi-UAV edge computing networks, the problems of collaborative caching and computing are solved, significantly reducing network latency and improving network performance.
Patent Information
- Application Number
- CN202411075327.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2025-02-11
- Estimated Expiration
- 2044-08-07
AI Technical Summary
In the existing multi-UAV edge computing network, the problems of collaborative caching and computing have not been effectively solved, resulting in insufficient network performance.
By determining the network architecture of multi-UAV edge computing network, establish communication models, service cache models and collaborative computing models, and build optimization problems to minimize the total network latency of computing tasks, optimize cache placement, computing task offloading and computing resource allocation .
It significantly reduces network latency and improves the performance of multi-UAV edge computing networks, which is better than traditional benchmark methods.
Smart Images
Figure CN118764886B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of mobile edge computing technology, and in particular to a management method and system for a multi-UAV edge computing network. Background Art
[0002] The advent of 5G networks marks a new phase in mobile communications, and at the same time, the data generated at the edge of the network has grown significantly. This growth is mainly driven by applications such as autonomous driving and virtual reality, which require high computing power and low-latency communication capabilities, triggering a transition from traditional cloud computing to local computing methods. Mobile Edge Computing (MEC) is a novel solution that reduces computing response latency and improves service quality by placing computing resources closer to the data source at the edge of the network, which is critical for the operation of latency-sensitive applications. Traditional ground-based MEC servers often face the challenge of multiple service overloads when handling dynamic and intensive network demands, especially in urban areas with high user density and fluctuating service demand. To address these issues, unmanned aerial vehicles (UAVs) have been integrated into the MEC framework. With their maneuverability and flexibility, UAVs take advantage of line-of-sight (LoS) links to enhance coverage and reduce latency by integrating edge computing capabilities, making them ideal for rapid deployment in emergency or sudden situations.
[0003] Integrating multiple drones into a MEC system to form a multi-drone edge computing network can achieve collaborative, distributed computing, and resource sharing, thereby improving system efficiency and coverage. A multi-drone collaborative computing framework is proposed, taking into account the mobility of ground terminals and the arrival of random computing tasks. Device-to-Device (D2D) communication also plays a key role in improving the efficiency of collaborative computing by allowing idle mobile devices to share resources directly, which reduces energy consumption and latency. A D2D-assisted MEC framework is proposed to manage independent computing tasks and reduce server load. A D2D-assisted MEC system is proposed to handle a large number of independent computing tasks, reducing energy consumption and latency. Incorporating D2D communication into a multi-drone network allows devices to exchange data directly, reducing the load on drones and optimizing resource usage, thereby improving the scalability of the system. However, as MEC systems evolve, service or cache placement is critical to improving performance. User-generated computing tasks require specific environments or content (such as software) to be executed effectively. By storing this important content on drones or user devices, the system can minimize repeated data transmission, thereby reducing latency and bandwidth usage. Effective caching involves strategies to determine what to store and where to store it based on user demand and network conditions. In multi-UAV and D2D-assisted MEC networks, this complex cache placement problem has not been addressed in previous research works. Summary of the invention
[0004] The present invention provides a management method and system for a multi-UAV edge computing network, and solves the technical problem of how to collaboratively cache and calculate a multi-UAV edge computing network to improve the performance of the edge computing network.
[0005] In order to solve the above technical problems, the present invention provides a management method for a multi-UAV edge computing network, comprising the steps of:
[0006] S1. Determine the network architecture of the multi-UAV edge computing network;
[0007] The network architecture consists of U unmanned aerial vehicles (UAVs) and G ground devices (GDs); G ground devices are organized into U clusters, each cluster is provided with computing services by a specific associated UAV; ground devices in the same cluster are connected to each other through D2D links; the g-th ground device (GD g) is associated with the a(g)-th UAV (UAV a(g), 1≤a(g)≤U, 1≤g≤G; the UAV maintains a constant altitude H; there are K environment files that need to be cached;
[0008] S2. Establishing a communication model, a service cache model and a collaborative computing model based on the network architecture;
[0009] S3. Based on the established communication model, service cache model and collaborative computing model, with minimizing the total network delay of all computing tasks as the optimization goal, an optimization problem is constructed to simultaneously optimize cache placement, computing task offloading and computing resource allocation;
[0010] S4. Solve the optimization problem to obtain cache placement, computing task offloading and computing resource allocation results.
[0011] Further, in step S2, the communication model is established as follows: there are three types of communication links in the multi-UAV edge computing network: a ground D2D link between ground devices, through which the computing task of GD g is offloaded to other GDs in the same cluster; a ground-to-air link between ground devices and their associated UAVs, through which the computing task of GD g is offloaded to UAV a(g) via the ground-to-air link; an air-to-air link between an associated UAV and other UAVs, if UAV a(g) does not cache the computing service environment required by GD g, the UAV a(g) forwards the received computing task to a non-associated UAV that has cached the relevant content via the air-to-air link for execution;
[0012] In step S2, the service cache model is established as follows: each cache device can only cache a specific subset of available environment content, jointly covering all required computing environments, and the cache devices refer to UAV and GD; the total size of the content cached by each cache device cannot exceed the cache capacity of the cache device;
[0013] In step S2, the collaborative computing model is established as follows: each computing task is indivisible, and a computing task can only be offloaded to one UAV or GD.
[0014] Further, in step S3, the total network delay of the computing task is the sum of the delays for completing all computing tasks;
[0015] The delay in completing a computational task includes four types of delays:
[0016] The first type of delay is the delay caused by the local device completing the computing task;
[0017] The second type of delay is the delay from the local device unloading the computing task to other local devices completing the computing task;
[0018] The third type of delay is the delay from the local device unloading the computing task to its associated UAV completing the computing task;
[0019] The fourth type of delay is the delay from the local device offloading the computing task to its associated UAV, and from its associated UAV offloading the computing task to the non-associated UAV completing the computing task.
[0020] Furthermore, the total network delay of the computing task is calculated as follows:
[0021]
[0022] Among them, D g,k Indicates that the calculation task is completed The delay, Refers to the computational tasks generated from GD g that require environment file k;
[0023] D g,k Calculated by the following formula:
[0024]
[0025] in, Represents a computing task The decision variables executed by GD i, if the computation task Executed by GDI, then otherwise Indicates that GD i completes the calculation task Delay; x g,k,u represents the decision variable for the computational task to be performed by UAV u. If the computational task V k g is executed by UAV u, then otherwise D g,k,u Indicates that UAV u completes the computing task delay.
[0026] Furthermore, in step S3, the constraints of the optimization problem include:
[0027]
[0028] Among them, a k,u represents the decision variable for service k to be cached by UAV u, a k,u =1 means service k is cached by UAV u, otherwise a k,u =0; The decision variable indicating whether service k is cached by GD g, Indicates that service k is cached on GD g, otherwise C u and C g Represent the total cache capacity of UAV u and GD g, o g,k,u Indicates that UAV u is assigned to a computing task The proportion of computing resources; k represents the size of each specific computing requirement environment file k, Indicates any.
[0029] Furthermore, for the first type of delay, when the required service environment is cached in GD g and there are sufficient computing resources, the computing task is directly processed on GD g. The computing delay of GD g is expressed as f g is the computational power of GD g, Indicates that the calculation task is completed Required CPU computing resources;
[0030] For the second type of delay, if the local device cache near the GD g cluster has the corresponding service environment, GD g is unloaded to the corresponding GD i for computing task execution, and the corresponding computing delay Includes communication delay from GD g to GD i and GDI processing time Represents a computing task is the input data size, fi is the computing power of GD i;
[0031] For the third type of delay, if the associated UAV a(g) caches the required content, the associated UAV a(g) processes the corresponding offloaded computing task, and the associated UAV a(g) processes the computing task. The total delay D g,k,a(g) Includes the transmission delay from GD g to the associated UAV a(g) and the computational delay of UAV a(g) R g,a(g) is the rate at which GD g can send computing tasks to UAV a(g), o g,k,a(g) Assign computing tasks to UAV a(g) The proportion of computing resources, f a(g) is the computing capability of UAV a(g);
[0032] For the fourth type of delay, if neither the associated UAV a(g) nor the ground equipment in the cluster has the necessary computing service environment and sufficient computing resources, the computing task Offload to the non-associated UAV u, then the total delay D from GD g to UAV u g,k,u Includes communication delay from ground equipment to associated UAV a(g) Communication delay from associated UAV a(g) to non-associated UAV u and the computational latency of UAV u R a(g), u is the rate at which the associated UAV a(g) can send computing tasks to the unassociated UAV u, and f uis the computing capability of the non-associated UAV u.
[0033] Furthermore, the step S4 specifically includes the steps of:
[0034] S41, when the computing resource allocation O is fixed, simplify the original problem into a first sub-problem of optimizing computing task offloading X and cache placement A; when the cache placement A and computing task offloading X are given, simplify the original problem into a second sub-problem of optimizing computing resource allocation O;
[0035] S42. Solve the first subproblem and the second subproblem alternately and iteratively until the decrease rate of the total network delay of the computing task is lower than the threshold ε, and output the cache placement A, computing task offloading X and computing resource allocation O at this time.
[0036] Furthermore, cache deployment Computational task offloading decision Resource Allocation Indicates definition;
[0037] In step S41, when the computing resource allocation O is fixed, the original problem is simplified to the first sub-problem of optimizing computing task offloading X and cache placement A, which specifically includes the following steps:
[0038] S411, when the computing resource allocation O is fixed, the original optimization problem is simplified into the sub-problems of optimizing computing task offloading X and cache placement A:
[0039]
[0040] stC1,C2,C4-C7
[0041] S412. Represent constraints C5 and C7 as the intersection of the following regions:
[0042]
[0043] Among them, the area intersection of constraint C5 is constraints C8 to C10, and the area intersection of constraint C7 is constraints C11 to C13;
[0044] S413. Two penalty terms are introduced into the objective function to penalize non-binary solutions of A and X, thereby obtaining the following optimization problem:
[0045]
[0046] stC1-C4,C6,C8,C11
[0047] Among them, η1, η2, λ1, λ2 are penalty factors greater than 0;
[0048] S414, yes and Using the first-order Taylor approximation we get:
[0049]
[0050] in, and is a in the rth iteration k,u , x g,k,u and
[0051] S415. Approximate problem (P2) as the first sub-problem:
[0052]
[0053] stC1-C4,C6,C8,C11;
[0054] In step S42, the first subproblem (P3) is solved using a standard convex optimization method or a CVX solver.
[0055] Further, in step S41, given the cache placement A and the computing task offloading X, the original problem is simplified to a second sub-problem of optimizing the computing resource allocation O, which specifically includes the steps of:
[0056] S416. Given cache placement A and computing task offloading X, the original problem (P0) is simplified to the second sub-problem by ignoring constant terms:
[0057]
[0058] stC3,C4
[0059] Then in step S42, the process of solving the second sub-problem includes the steps of:
[0060] S421, assume that the dual variable is The Lagrangian function of problem (P4) is expressed as:
[0061]
[0062] The corresponding dual function is expressed as Represents L({λ u},{o g,k,u}) Function about variable {o g,k,u}Find its infimum;
[0063] S422, the Lagrangian function is {o g,k,u}Find the first-order partial derivative and set it to zero to get the optimal solution:
[0064]
[0065] The present invention also provides a management system for a multi-UAV edge computing network, the key of which is that an intelligent agent is provided, and the intelligent agent is used to execute the management method of the multi-UAV edge computing network.
[0066] The management method and system of the multi-UAV edge computing network provided by the present invention first determine the network architecture of the multi-UAV edge computing network, then establish a communication model, a service cache model and a collaborative computing model based on the network architecture, and then, based on the established communication model, service cache model and collaborative computing model, take minimizing the total network delay of all computing tasks as the optimization goal, construct an optimization problem that simultaneously optimizes cache placement, computing task offloading and computing resource allocation; finally, solve the optimization problem to obtain the results of cache placement, computing task offloading and computing resource allocation. The present invention utilizes multi-UAV and D2D collaborative communication, and manages network resources in combination with cache placement, computing task offloading and computing resource allocation, promotes computing task and data sharing between devices, and improves the performance of the multi-UAV edge computing network. Simulation results show that compared with the existing benchmark scheme, the present invention can significantly reduce network delay. BRIEF DESCRIPTION OF THE DRAWINGS
[0067] Figure 1 is a flow chart of a management method for a multi-UAV edge computing network provided by an embodiment of the present invention;
[0068] Figure 2 is a content cache deployment diagram of a drone and ground equipment obtained by optimization according to an embodiment of the present invention;
[0069] Figure 3 is a diagram of the computing task unloading of the optimized GD 1 provided in an embodiment of the present invention;
[0070] Figure 4 This is a diagram showing the influence of the number of computing tasks on the total network delay provided by an embodiment of the present invention;
[0071] Figure 5 is a graph showing the influence of the number of ground devices on the total network delay provided by an embodiment of the present invention;
[0072] Figure 6 Graph showing the effect of drone computing capability on total delay provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0073] The following specifically illustrates the implementation mode of the present invention in conjunction with the accompanying drawings. The embodiments are provided for illustrative purposes only and cannot be understood as limiting the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute a limitation on the scope of patent protection of the present invention, because many changes can be made to the present invention without departing from the spirit and scope of the present invention.
[0074] The management method of the multi-UAV edge computing network provided by the embodiment of the present invention is as follows: Figure 1 As shown in the flowchart, the steps include:
[0075] S1. Determine the network architecture of the multi-UAV edge computing network;
[0076] S2, establishing a communication model, a service cache model and a collaborative computing model based on the network architecture;
[0077] S3. Based on the established communication model, service cache model and collaborative computing model, with minimizing the total network delay of all computing tasks as the optimization goal, an optimization problem is constructed to simultaneously optimize cache placement, computing task offloading and computing resource allocation;
[0078] S4. Solve the optimization problem to obtain cache placement, computing task offloading and computing resource allocation results.
[0079] The present invention proposes a collaborative edge computing framework, which utilizes multi-UAV and user device-to-device (D2D) collaborative communication, and promotes computing task and data sharing between devices by combining caching and computing task offloading technology to improve the edge computing network performance.
[0080] (1) Step S1: Network architecture of multi-UAV edge computing network
[0081] The present invention considers a D2D and UAV-enhanced MEC network, which consists of U UAVs and G ground devices (GD). The G ground devices are organized into U clusters, each cluster is provided with computing services by a specific associated UAV. The ground devices in the same cluster are connected to each other through D2D links. And the distance between the UAVs is less than d min , then the UAVs can communicate with each other. Without loss of generality, the present invention assumes that GD g (the gth ground device) is associated with UAV a(g) (the a(g)th UAV), where 1≤a(g)≤U, 1≤g≤G. The horizontal coordinate of GD g is represented by w g The present invention assumes that the UAV maintains a constant altitude H, and the horizontal position of each UAV u is q u , where 1≤u≤U.
[0082] The present invention assumes that the computing tasks generated by users require specific environments or content (such as software) to be effectively executed. Each UAV and GD caches a set of different resources that together cover all required computing environments. In order to optimize the completion of computing tasks across the entire network, the present invention considers the following four computing task processing strategies: 1) Perform local computing directly on ground devices; 2) Ground devices offload computing tasks to adjacent ground devices via D2D links; 3) Ground devices offload computing tasks to associated drones for computing; 4) Associated drones offload computing tasks to non-associated drones via links between drones. Suppose there are a total of K environment files that need to be cached, let z k represents the size of each specific computing requirement environment file k, 1≤k≤K. represents a computational task generated from GD g that requires environment file k, which is modeled as a tuple in Represents a computing task The input data size is Indicates that the calculation task is completed The CPU computing resources required.
[0083] (2) Step S2: Communication Model, Service Cache Model, and Collaborative Computing Model
[0084] A. Communication model
[0085] This invention assumes that the communication link adopts Orthogonal Frequency Division Multiple access (OFDMA) technology, and the interference between links can be ignored. There are three types of links in the multi-UAV edge computing network:
[0086] 1) Ground D2D link: computing tasks It can be offloaded to other GDs in the same cluster via terrestrial D2D links.
[0087] The channel gain from GD g to GD i is expressed as Indicates that, where d g,i =||w g -w i || (represents the distance between GD g and GDi), β0 represents the power gain of the channel at a standard distance of 1m (meter). α≥2 is the path loss component, and G0<1 represents the additional factor caused by scattering caused by building occlusion and NLoS link. Therefore, when GD g sends a computing task to GDi, the achievable rate is where p g is the transmission power of GD g, W0 is the available bandwidth from GD g to GD i, σ 2 is the noise power.
[0088] 2) Ground-to-air link: computing tasks Can be offloaded to an associated UAV via a ground-to-air link.
[0089] Given that the UAV is at high altitude, the LoS link dominates the ground-to-air communication channel. Therefore, the channel gain between GD g and the associated UAV UAVa(g) is expressed as in Therefore, when GD g sends computing tasks to UAV a(g), the achievable rate is Where W1 represents the available bandwidth between GD g and UAV a(g).
[0090] 3) Air-to-air link: If the associated UAV a(g) does not cache the computing service environment required by the associated GD g, it can forward the received computing task through the air-to-air link to the non-associated UAV u that has cached the relevant content for execution.
[0091] The channel gain between UAV a(g) and UAV u is expressed as where d a(g),u =||q a(g) -q u || represents the distance between UAV a(g) and UAV u. Therefore, when UAV a(g) sends a computing task to UAV u, the achievable rate is Where W2 represents the available wireless link bandwidth from UAV a(g) to UAV u, p u represents the transmit power of UAV u.
[0092] B. Service Cache Model
[0093] In this paper, UAVs and ground devices cache the basic computing environment required to perform specific computing tasks. Due to the limited cache capacity of these cache devices, each cache device can only cache a specific subset of the available environment content. The cache deployment decision of the environment service content k on UAV u is represented by a k,u Indicates that a k,u =1 means service k is cached by UAV u, otherwise a k,u = 0. Similarly, for ground equipment, Indicates that service k is cached on GD g, otherwise In order to optimize the efficiency of cache space utilization, the total size of the cache content cannot exceed the cache capacity of each cache device, that is, constraint C1 is satisfied:
[0094]
[0095] Among them, Cu and C g Represent the total cache capacity of UAV u and GD g respectively, Indicates any.
[0096] C. Collaborative Computing Model
[0097] The present invention assumes that each computing task is indivisible, and the computing task offloading decision is represented by two binary variables: and x g,k,u If the computation task Executed by GDI, then otherwise Among them, if i=g, The calculation task Executed locally by GD g. If the computation task Executed by UAV u, then x g,k,u =1, otherwise x g,k,u =0. When u=a(g),x g,k,a(g) =1, indicating a computing task Executed by the associated UAV a(g). Since a computing task can only be offloaded to one UAV or GD, the computing task offloading decision should satisfy constraint C2:
[0098]
[0099] There are four types of computing in a multi-drone edge computing network:
[0100] 1) Local computing: If the required service environment is cached locally and there are sufficient computing resources, the computing tasks can be processed directly on the local device.
[0101] Therefore, the computation delay of GD g can be expressed as where f g is the computing power of GD g.
[0102] 2) Calculation by other ground devices: If the local device nearby in the cluster has the corresponding service environment cached, the computing task can also be offloaded to the corresponding local device for execution.
[0103] As the computational tasks generated by GD g It is sent to GD i, which has service k cached, through the D2D link, and then the latter processes the computing task. The communication delay from GD g to GD i is Therefore, the computational task The total computational task delay includes both the communication delay from GD g to GD i and the processing time of GD i, and is calculated as in fi is the computing power of GD i.
[0104] 3) Associated UAV Computing: If the associated UAV has cached the required content, it can handle the corresponding offloaded computing tasks.
[0105] The transmission delay from GD g to the associated UAV a(g) is UAV u is assigned to computing tasks The proportion of computing resources is o g,k,u Indicates that all allocated computing resources are represented by O = {o g,k,u |1≤u≤U,1≤k≤K,1≤g≤G}. Then, the calculation delay of UAV a(g) is where f a(g) represents the total computing capacity of UAV a(g). Therefore, the associated UAV a(g) processes the computing task The total delay is
[0106] 4) Non-associated UAV computing: If neither the associated UAV nor the ground equipment in the cluster has the necessary computing service environment and sufficient computing resources, the computing tasks can be offloaded to the non-associated UAV.
[0107] Therefore, the computational task The total delay at the non-associated UAV u consists of three components: the communication delay from the ground equipment to the associated UAV a(g) The communication delay from the associated UAV a(g) to the unassociated UAV u, and the computational delay at UAV u. The communication delay from the associated UAV a(g) to the unassociated UAV u is The computational delay of UAV u is f u represents the total computing power of UAV u. Therefore, the total delay from GD g to UAV u is
[0108] According to the analysis of the above four different computing modes, the computing task The overall network delay can be expressed as
[0109] (3) Step S3: Target optimization problem
[0110] In this embodiment, the present invention considers optimizing cache deployment in D2D and drone-enhanced MEC networks. Computational task offloading decision and computing resource allocation The goal is to minimize the total network latency of all computational tasks. The optimization problem can be described as:
[0111]
[0112]
[0113] min means minimization, and st means to be satisfied. Among them, constraint C4 limits the allocation of computing resources to each drone. Constraint C6 ensures that each computing task is only offloaded to the ground equipment or drone that caches the required content. Due to the existence of non-convex objective function and non-convex constraints C4, C5, and C7, the optimization problem (P0) is a mixed integer non-convex optimization problem that is difficult to solve directly.
[0114] (4) Step S4: Solving the optimization problem
[0115] Next, the present invention decomposes the original problem (P0) into two sub-problems. The first sub-problem involves computing task offloading X and cache placement A, while the second sub-problem focuses on computing resource allocation O. The present invention proposes an efficient iterative algorithm to alternately optimize these sub-problems until convergence.
[0116] Therefore, step S4 specifically includes the steps of:
[0117] S41, when the computing resource allocation O is fixed, simplify the original problem into a first sub-problem of optimizing computing task offloading X and cache placement A; when the cache placement A and computing task offloading X are given, simplify the original problem into a second sub-problem of optimizing computing resource allocation O;
[0118] S42. Solve the first subproblem and the second subproblem alternately and iteratively until the decrease rate of the total network delay of the computing task is lower than the threshold ε, and output the cache placement A, computing task offloading X and computing resource allocation O at this time.
[0119] In step S41, when the computing resource allocation O is fixed, the original problem is simplified to the first sub-problem of optimizing computing task offloading X and cache placement A, which specifically includes the following steps:
[0120] S411, when the computing resource allocation O is fixed, the original problem is simplified to the sub-problem of optimizing computing task offloading X and cache placement A:
[0121]
[0122] stC1,C2,C4-C7
[0123] S412. To solve the binary constraints C5 and C7, the present invention represents the constraints C5 and C7 as the intersection of the following regions:
[0124]
[0125] The region intersection of constraint C5 is constraints C8 to C10, and the region intersection of constraint C7 is constraints C11 to C13. It can be shown that the solution that satisfies constraints C8 to C13 also satisfies the original binary constraints C5 and C7, and vice versa. Specifically, because 0≤a ku ≤1 and constraint C9 must hold, so a ku (1-a ku )=0, thereby A similar principle also applies to xg,k,u variables.
[0126] S413. Since constraints C8 and C11 are continuous variable constraints, and the goal of the present invention is to obtain a binary solution of A and X in the optimization problem (P1), the present invention introduces two penalty terms in the objective function to penalize non-binary solutions of A and X, thereby obtaining the following optimization problem:
[0127]
[0128] stC1-C4,C6,C8,C11(11)
[0129] Where η1, η2, λ1, λ2 are penalty factors greater than 0. When the penalty factor is large enough, problem (P2) is equivalent to problem (P1).
[0130] S414. However, problem (P2) is still a non-convex problem because its objective function is non-convex. In order to solve this problem, the present invention uses the Successive Convex Approximation (SCA) technique to and Using the first-order Taylor approximation, we can get:
[0131]
[0132] in, and is the local point in the rth iteration.
[0133] S415. Therefore, problem (P2) can be approximated as:
[0134]
[0135] It can be verified that problem (P3) is a convex optimization problem, so in step S42, the subproblem can be solved efficiently using a standard convex optimization method or a solver such as CVX. and (xg,k,u ) 2 , the solution to problem (P3) must be the solution to (P2), and its objective value provides an upper bound for problem (P2).
[0136] In step S41, given cache placement A and computing task offloading X, the original problem is simplified to the second sub-problem of optimizing computing resource allocation O, which specifically includes the steps of:
[0137] S416. Given cache placement A and computing task offloading X, the original problem (P0) is simplified to the second sub-problem by ignoring constant terms:
[0138]
[0139] stC3,C4(14)
[0140] It can be proved that problem (P4) is a strongly dual convex optimization problem. Therefore, the optimal solution can be obtained by Lagrangian dual method.
[0141] Then in step S42, the process of solving the second sub-problem includes the steps of:
[0142] S421, assume that the dual variable is The Lagrangian function of problem (P4) is expressed as:
[0143]
[0144] Therefore, the corresponding dual function can be expressed as Represents L({λ u},{o g,k,u}) Function about variable {o g,k,u}Find its infimum;
[0145] S422, replace the Lagrangian function in (14) with {o g,k,u}Find the first-order partial derivative and set it to zero, the present invention can obtain the optimal solution:
[0146]
[0147] By substituting this optimal solution back into the equality constraint C4, we obtain:
[0148]
[0149] It can be proved is {λ u}>0. By applying the binary search method to find λ u The optimal value of , and then from formula (16) we can get Based on the above discussion, the present invention can use the Block Coordinate Descent (BCD) method to obtain a suboptimal solution to the original optimization problem (P0) by alternately iteratively optimizing the first subproblem (P3) and the second subproblem (P4). For details, see Algorithm 1 shown in Table 1.
[0150] Table 1
[0151]
[0152] Based on the above method, the present invention also provides a management system for a multi-UAV edge computing network, which is provided with an intelligent agent, and the intelligent agent is used to execute the management method of the multi-UAV edge computing network.
[0153] In this embodiment, the present invention conducts a comprehensive simulation experiment study to evaluate the effectiveness and performance characteristics of the proposed method and system. The present invention considers a scenario of a cache-enabled D2D and multi-UAV-assisted edge computing network, where 10 ground devices (G=10) are randomly deployed in x max =100m and y max =50m is the boundary of a rectangular area. Two drones (U=2) are deployed to the area to provide computing services. Each ground device can generate K=10 computing tasks, and each computing task K requires different cache content. The size of the cache content is z k ∈[50,100]kB, Due to limited computing power, the present invention assumes that each ground device can handle at most two computing tasks simultaneously. In addition, the size of the computing task generated by each ground device needs to be k ∈[10,100]kB, Processing cycle Unless otherwise specified, other system parameter settings are as follows:
[0154]
[0155] σ 2 =-110dBm; W1=20MHz; H=150m; W2=100MHz;
[0156]
[0157] In order to evaluate the performance of the method and system proposed in the present invention, the present invention gives the corresponding simulation results and compares them with the following three benchmark methods: 1) UAV-Only Offloading (UO) scheme: computing tasks can only be offloaded to associated or unassociated UAVs; 2) Cache-Oriented Offloading (COO) scheme: this scheme focuses on optimizing service caching and computing task offloading, without considering computing resource allocation; 3) Limited D2D Offloading (No D2D Offloading, NDO) scheme: computing tasks can be offloaded to associated or unassociated UAVs, or processed locally, but offloading through D2D is not allowed.
[0158] Figure 2 The optimized content cache deployment for drones and ground devices is shown. Figure 2 As shown, compared with ground devices, drones have larger storage capacity and can store more content, while ground devices can only cache content 7 and 10. Figure 3 The figure shows the optimization of GD 1 computational task offloading. Figure 3 As shown, computational tasks are only offloaded to devices that have cached the necessary services.
[0159] Figure 4 It shows how the total network delay is affected by the number of computing tasks. Figure 4 As shown in the figure, the increase in the number of computing tasks leads to an increase in computing task processing time and transmission time, which leads to an increase in total delay, but the algorithm proposed in the present invention always maintains the lowest delay. Although the COO scheme focuses on optimizing cache and computing task offloading, it does not fully solve the problem of computing resource allocation, so its performance under high load is poor compared with the algorithm proposed in the present invention.
[0160] Figure 5 It shows that the total network delay is affected by the number of ground devices. Figure 5 As shown in Figure 1, as the number of ground devices increases, the total latency also increases. The increase in latency is mainly because the additional ground devices bring new computing tasks, and the processing capacity of each ground device is limited. Therefore, the increase in computing task requirements requires the system to offload more computing tasks to drones or other ground devices for processing, which results in longer computing task transmission time.
[0161] Figure 6 The effect of the computing power of the drone on the total latency is shown. Figure 6 As shown, with the computing power f u The total delay decreases as the value of f increases. This decrease occurs because higher fu The efficiency of computing task processing is improved, and the system can process computing tasks faster, thus effectively reducing the total delay. In contrast, the delay of the COO scheme first decreases and then increases, indicating that its resource allocation and computing task scheduling strategy is only effective within a specific computing power range. Once this range is exceeded, the efficiency of the strategy decreases due to the failure to adapt to higher levels of computing power, resulting in an increase in total delay.
[0162] In this embodiment, a management method and system for a multi-drone edge computing network provided by an embodiment of the present invention designs a cache-based D2D and multi-drone assisted MEC network collaborative computing framework, and solves the complex mixed integer non-convex optimization problem separately by decomposing it into content cache and computing task offloading optimization sub-problems and computing resource optimization sub-problems. On this basis, an effective iterative algorithm is designed. Simulation results show that the present invention can significantly reduce the total system delay, which is better than the traditional benchmark method.
[0163] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.
Claims
1. A management method for a multi-UAV edge computing network, characterized in that: Includes steps: S1. Determine the network architecture of the multi-UAV edge computing network; The network architecture consists of U unmanned aerial vehicles (UAVs) and G ground devices (GDs); G ground devices are organized into U clusters, each cluster is provided with computing services by a specific associated UAV; ground devices in the same cluster are connected to each other through D2D links; the g-th ground device (GD g) is associated with the a(g)-th UAV (UAV a(g), 1≤a(g)≤U, 1≤g≤G; the UAV maintains a constant altitude H; there are K environment files that need to be cached; S2. Establishing a communication model, a service cache model and a collaborative computing model based on the network architecture; The communication model is established as follows: there are three types of communication links in the multi-UAV edge computing network: ground D2D links between ground devices, through which the computational tasks of GD g are offloaded to other GDs in the same cluster; The ground-to-air link between the ground device and its associated UAV, the computing task of GD g is offloaded to UAV a(g) through the ground-to-air link; the air-to-air link between the associated UAV and other UAVs, if UAV a(g) does not cache the computing service environment required by GD g, the UAV a(g) will forward the received computing task to the non-associated UAV that has cached the relevant content through the air-to-air link for execution; The service cache model is established as follows: each cache device can only cache a specific subset of the available environment content, which together covers all required computing environments. The cache devices refer to UAV and GD; the total size of the content cached by each cache device cannot exceed the cache capacity of the cache device; The collaborative computing model is established as follows: each computing task is indivisible, and a computing task can only be offloaded to one UAV or GD; S3. Based on the established communication model, service cache model and collaborative computing model, with minimizing the total network delay of all computing tasks as the optimization goal, an optimization problem is constructed to simultaneously optimize cache placement, computing task offloading and computing resource allocation; In step S3, the total network delay of the computing task is the sum of the delays for completing all computing tasks; The delay in completing a computational task includes four types of delays: The first type of delay is the delay caused by the local device completing the computing task; The second type of delay is the delay from the local device unloading the computing task to other local devices completing the computing task; The third type of delay is the delay from the local device unloading the computing task to its associated UAV completing the computing task; The fourth type of delay is the delay from the local device unloading the computing task to its associated UAV, and from its associated UAV unloading the computing task to the non-associated UAV completing the computing task; The total network latency of a computing task is calculated as follows: Among them, D g,k Indicates that the calculation task is completed The delay, Refers to the computational tasks generated from GD g that require environment file k; D g,k Calculated by the following formula: in, Represents a computing task The decision variables executed by GD i, if the computation task Executed by GDI, then otherwise Indicates that GD i completes the calculation task Delay; x g,k,u Represents a computing task The decision variables executed by UAV u, if the computation task is executed by UAV u, then otherwise D g,k,u Indicates that UAV u completes the computing task Delays; For the first type of delay, when the required service environment is cached in GD g and there are sufficient computing resources, the computing task is processed directly on GD g. The computing delay of GD g is expressed as f g is the computational power of GD g, Indicates that the calculation task is completed Required CPU computing resources; For the second type of delay, if the local device cache near the GD g cluster has the corresponding service environment, GD g is unloaded to the corresponding GD i for computing task execution, and the corresponding computing delay Includes communication delay from GD g to GD i and GDI processing time Represents a computing task is the input data size, fi is the computing power of GD i, R g,i It represents the achievable rate when GD g sends computing tasks to GD i; For the third type of delay, if the associated UAV a(g) caches the required content, the associated UAV a(g) processes the corresponding offloaded computing task, and the associated UAV a(g) processes the computing task. The total delay D g,k,a(g) Includes the transmission delay from GD g to the associated UAV a(g) and the computational delay of UAV a(g) R g,a(g) Send a computation task to UAV a(g) for GD g The achievable rate when g,k,a(g) Assign computing tasks to UAV a(g) The proportion of computing resources, f a(g) is the computing capability of UAV a(g); For the fourth type of delay, if neither the associated UAV a(g) nor the ground equipment in the cluster has the necessary computing service environment and sufficient computing resources, the computing task Offload to the non-associated UAV u, then the total delay D from GD g to UAV u g,k,u Includes communication delay from ground equipment to associated UAV a(g) Communication delay from associated UAV a(g) to non-associated UAV u and the computational latency of UAV u R a(g),u is the rate at which the associated UAV a(g) can send computing tasks to the unassociated UAV u, f u is the computing capability of the unassociated UAV u, o g,k,u Indicates that UAV u is assigned to a computing task The proportion of computing resources; S4. Solve the optimization problem to obtain cache placement, computing task offloading and computing resource allocation results.
2. The management method of a multi-UAV edge computing network according to claim 1, characterized in that: In step S3, the constraints of the optimization problem include: Among them, a k,u represents the decision variable for service k to be cached by UAV u, a k,u =1 means service k is cached by UAV u, otherwise a k,u =0; The decision variable indicating whether service k is cached by GD g, Indicates that service k is cached on GD g, otherwise C u and C g Represents the total cache capacity of UAV u and GD g respectively; z k represents the size of each specific computing requirement environment file k, Indicates any.
3. The management method of a multi-UAV edge computing network according to claim 2, characterized in that: The step S4 specifically comprises the following steps: S41, when the computing resource allocation O is fixed, simplify the original problem to the first sub-problem of optimizing computing task offloading X and cache placement A; Given cache placement A and computational task offloading X, the original problem is reduced to the second subproblem of optimizing computational resource allocation O; S42. Solve the first subproblem and the second subproblem alternately and iteratively until the decrease rate of the total network delay of the computing task is lower than the threshold ε, and output the cache placement A, computing task offloading X and computing resource allocation O at this time.
4. The management method of a multi-UAV edge computing network according to claim 3, characterized in that: Cache deployment Computational task offloading decision Resource Allocation Indicates definition; In step S41, when the computing resource allocation O is fixed, the original problem is simplified to the first sub-problem of optimizing computing task offloading X and cache placement A, which specifically includes the following steps: S411, when the computing resource allocation O is fixed, the original optimization problem is simplified into the sub-problems of optimizing computing task offloading X and cache placement A: stC1,C2,C4-C7 S412. Represent constraints C5 and C7 as the intersection of the following regions: Among them, the area intersection of constraint C5 is constraints C8 to C10, and the area intersection of constraint C7 is constraints C11 to C13; S413. Two penalty terms are introduced into the objective function to penalize non-binary solutions of A and X, thereby obtaining the following optimization problem: stC1-C4,C6,C8,C11 Among them, η1, η2, λ1, λ2 are penalty factors greater than 0; S414, yes and Using the first-order Taylor approximation we get: in, and is a in the rth iteration k,u , x g,k,u and S415. Approximate problem (P2) as the first sub-problem: stC1-C4,C6,C8,C11; In step S42, the first subproblem (P3) is solved using a standard convex optimization method or a CVX solver.
5. The management method of a multi-UAV edge computing network according to claim 4, characterized in that: In step S41, given cache placement A and computing task offloading X, the original problem is simplified to the second sub-problem of optimizing computing resource allocation O, which specifically includes the steps of: S416. Given cache placement A and computing task offloading X, the original problem (P0) is simplified to the second sub-problem by ignoring constant terms: stC3,C4 Then in step S42, the process of solving the second sub-problem includes the steps of: S421, assume that the dual variable is The Lagrangian function of problem (P4) is expressed as: The corresponding dual function is expressed as Represents L({λ u },{o g,k,u }) Function about variable {o g,k,u }Find its infimum; S422, the Lagrangian function is {o g,k,u }Find the first-order partial derivative and set it to zero to get the optimal solution:
6. A management system for a multi-UAV edge computing network, characterized by: An intelligent agent is provided, and the intelligent agent is used to execute the management method of the multi-UAV edge computing network described in any one of claims 1 to 5.