A UAV edge network task offloading method based on SFC

Abstract

The present invention claims protection for a UAV edge network task offloading method based on SFC, which belongs to the field of communication technology. In view of the problem that a specific type of business in a low-altitude edge network breaks out in a local area, which may cause network load imbalance, deterioration of task offloading quality and degradation of service experience, a UAV edge network task offloading method based on SFC is proposed. According to the real-time location distribution and task offloading request of ground users, a deep reinforcement learning method is used to dynamically plan the cruising trajectory of the UAV swarm, and an SFC mapping decision is constructed according to the location distribution of the UAV and the type of airborne VNF. According to the status of airborne computing resources and the task offloading delay requirements, a resource allocation decision is constructed. By dynamically optimizing the SFC mapping decision and resource allocation decision during the cruising process of the UAV swarm, the network resource utilization is maximized, thereby effectively improving the task offloading completion rate and reducing the system unit task offloading delay.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and in particular relates to a method for unloading tasks from an edge network of a drone based on SFC. Background Art

[0002] With the rapid development of mobile communications technology, drones, due to their high mobility and flexibility, have become a crucial component of future networks. Establishing low-altitude edge networks using drones can quickly and efficiently provide essential communication and computing services to users on the ground. Network Function Virtualization (NFV) virtualizes and abstracts underlying network resources through the decoupling of software and hardware. By orchestrating virtual network functions (VNFs), various types of service function chains (SFCs) can be constructed, enabling flexible and efficient service provisioning and network resource management.

[0003] Most of the existing research on drone network task offloading focuses on the offloading of single functions and single drones, without considering the problem of multi-drone collaborative offloading of multi-functional tasks. Existing research on drone network SFC deployment is relatively rare, and mainly focuses on VNF mapping and SFC deployment in static network environments, without considering the demand for task offloading. Especially in real-world scenarios, local outbreaks of specific types of tasks may have a serious impact on the resource balance of drone networks in static environments, leading to a decrease in resource utilization and deterioration in the quality of task offloading. In response to the above problems, the present invention proposes a drone edge network task offloading method based on SFC, based on the rapid development of drone network technology and the potential demand for offloading of a large number of multi-functional tasks in the future. It can effectively improve the resource utilization of low-altitude edge networks, increase the task offloading completion rate, and reduce the system unit task delay. Summary of the Invention

[0004] The present invention aims to solve the above problems of the prior art. A method for offloading tasks from a drone edge network based on SFC is proposed. The technical solution of the present invention is as follows:

[0005] A method for offloading UAV edge network tasks based on SFC, comprising the following steps:

[0006] 101. The cruising time of the drone swarm is divided into K discrete time slots of equal length Δt, and the system state remains unchanged in any k-th time slot, k∈K. The double-delay deep deterministic policy gradient algorithm is used to construct a deep reinforcement learning model. Initialize k=0 and establish the network state vector s containing the ground user position and the drone swarm position in the k-th time slot. k, UAV swarm motion vector a k , reward function r k , UAV swarm action strategy generator Actor, UAV swarm state-action value generator Critic, SFC mapping and resource allocation decision generator Mapping-Allocation, experience sample storage area, random sample set Mini-Batch;

[0007] 102. If k < K, let k = k + 1, according to the task request of user i in time slot k, the network state vector s k and the distribution status of onboard computing resources, update the SFC deployment decision {X, Y, Z} of user i in time slot k through the SFC mapping and resource allocation decision generator Mapping-Allocation, where X is the node mapping decision, Y is the link mapping decision, and Z is the resource allocation decision. Jump to step 103, otherwise jump to step 105;

[0008] 103. Perform task offloading for user i in time slot k according to {X, Y, Z} and obtain the corresponding drone group action reward r k , obtain s through the drone swarm action strategy generator Actor k The motion vector of the drone group in the state a k , according to s k with a k Calculate to obtain s k+1 , will [s k ,a k ,r k ,s k+1 ] Added experience sample storage area;

[0009] 104. Randomly sample a Mini-Batch sample set from the experience sample storage area, and import the Mini-Batch sample set into the drone swarm action strategy generator Actor and the drone swarm state-action value generator Critic for training, and jump to step 102;

[0010] 105. End.

[0011] Furthermore, in step 101, a network state vector s is established which includes the ground user position and the drone group position in the kth time slot. k , UAV swarm motion vector a k , reward function r k , as shown in formulas (1), (2), and (3) respectively:

[0012]

[0013]

[0014]

[0015] In formula (1), represents the three-dimensional coordinates of UAV j in time slot k, represents the three-dimensional coordinates of user i, J represents the set of drones, and I represents the set of users; in formula (2), α j,k ∈[0,2π] is the horizontal flight angle of UAV j, is the horizontal flight speed, β j,k ∈{-1,1} is the vertical flight direction of UAV j, is the vertical flight speed; in formula (3), ε1, ε2, ε3 are the weight factors of the drone action reward function, where 0≤ε1≤1, 0≤ε2≤1, 0≤ε3≤1, ψ represents the system task completion rate, represents the normalized system unit task delay, φ represents the user communication interruption rate;

[0016] Furthermore, the method for obtaining the SFC deployment decision {X, Y, Z} of user i through the SFC mapping and resource allocation decision generator Mapping-Allocation in step 102 is as follows:

[0017] 1) According to the task offloading request proposed by user i, the VNFs required by user i are sequentially placed into the set S, and the candidate mapping drone set is initialized Count variable m = 0, optimal SFC deployment fitness ξ * =0;

[0018] 2) According to S and the current drone network status, obtain access to drone j for user i a If successful, jump to step 3), otherwise, jump to step 10);

[0019] 3) If m<M max , where M max Indicates the maximum search range hop count of the drone node in the network. Let m = m + 1 and jump to step 4). Otherwise, jump to step 9.

[0020] 4) For distance j a Each drone j within the hop range of drone m will use its available computing resources c j >c0 drones are added to set J map , where c0 represents the basic computing resources required for VNF instantiation. If |J map |≥|S|, set the counting variable q=0 and jump to step 5), otherwise, jump to step 3);

[0021] 5) If q < Q max, where Q max Indicates that in set J map The maximum number of times to search for SFC, let q = q + 1, jump to step 6), otherwise, jump to step 3);

[0022] 6) According to set J map and access to drones a , build and update the mapping decision {X, Y} of the required SFC, if successful, jump to step 7), otherwise, jump to step 5);

[0023] 7) Based on the current UAV network status and mapping decision {X, Y}, construct and update the resource allocation decision {Z} of the required SFC. If successful, jump to step 8), otherwise, jump to step 5);

[0024] 8) Calculate the fitness ξ of the SFC deployment decision {X, Y, Z}, if ξ * ≤ξ, let ξ * =ξ, optimal SFC deployment decision {X * ,Y * ,Z *}={X,Y,Z}, jump to step 5), otherwise, jump to step 5);

[0025] 9) Output the optimal SFC deployment decision {X * ,Y * ,Z *};

[0026] 10) The algorithm ends.

[0027] Furthermore, in step 2), user i obtains access to drone j a The method is as follows:

[0028] 11) Establish a temporary set J′, add the drones in the drone set J that meet the communication distance condition of user i to the set J′, if Jump to step 12), otherwise, jump to step 15);

[0029] 12) Create a temporary set J″ and add the drones in J′ that meet the first VNF ​​type requirement of SFC to J″. If Jump to step 13), otherwise, jump to step 14);

[0030] 13) Based on the available computing resources c in each drone j j The value of c is used to sort the elements in the set J″ in descending order. According to the communication distance between user i and drone j, the elements with the same c are sorted. j Arrange the elements of the value in ascending order, take out the first drone j in J″, and let user i’s access drone ja =j, jump to step 15);

[0031] 14) According to the communication distance between user i and drone j, sort the elements in set J′ in ascending order, take out the first drone j in J′, and let user i’s access drone j a =j;

[0032] 15) The algorithm ends.

[0033] Furthermore, the method for constructing and updating the required SFC mapping decision {X, Y} in step 6) is as follows:

[0034] 21) Initialize temporary variable j t =j a , temporary set J′=J map , Count variables n=0, l=0;

[0035] 22) Let n = n + 1, add the UAVs in J′ that meet the n-th VNF type requirement of SFC to the set J″;

[0036] 23) For each UAV j in the set J″, the minimum cost path algorithm is used to calculate the link cost between UAVs. t The minimum cost path to UAV j And remove the nodes that do not have the minimum cost path from the set J″;

[0037] 24) If Update the node mapping probability of each drone in J″ based on the available computing resources of the drone And according to the mapping probability distribution According to the probability, a drone j is randomly selected from J″ as the nth VNF mapping node to join the node mapping decision X, and the path As the lth virtual link, join the link mapping decision Y, and let j t = j, remove j from the set J′ and jump to step 25), otherwise, jump to step 26);

[0038] 25) If n<|S|, let l=l+1, Jump to step 22), otherwise, output the mapping decision {X, Y}, and jump to step 26);

[0039] 26) The algorithm ends.

[0040] Furthermore, the calculation method of the link cost between each UAV in step 23) is shown in formula (4):

[0041]

[0042] In formula (4), is the available bandwidth resource b of physical link e e The ratio of to the total bandwidth resources of physical link e represents the normalized available bandwidth resources of physical link e, and b represents the bandwidth requirement of task offloading;

[0043] In step 24), the node mapping probability of each drone in J″ is updated The method is shown in formula (5):

[0044]

[0045] In formula (5), c j represents the available computing resources of UAV j.

[0046] Furthermore, the method for constructing and updating the resource allocation decision {Z} of the required SFC in step 7) is as follows:

[0047] 31) According to the node mapping decision X, the mapping node is added to the temporary set J′, and the basic computing resources c0 required for VNF instantiation are allocated to each drone j in the set J′. According to the link mapping decision Y, the bandwidth b required by the virtual link is added to the resource allocation decision {Z}, and the total task transmission delay t of the offload path is obtained. tran , where t tran is the cumulative transmission delay of task data through each physical link on the offloading path;

[0048] 32) Let the temporary set J″=J′, according to the available computing resources c of drone j j The value of , sort the elements in the set J″ in ascending order;

[0049] 33) If Calculate the task computing resources that need to be allocated to the first UAV j in J″ And update the resource allocation decision {Z}, remove drone j from the set J″, jump to step 33), otherwise, jump to step 34);

[0050] 34) If the task calculation delay t com ≤T max -t tran , where T max If the task tolerates delay, output the resource allocation decision {Z} and jump to step 35), otherwise, jump to step 35);

[0051] 35) The algorithm ends.

[0052] Furthermore, in step 33), the task computing resources allocated by drone j to user i The calculation method is shown in formula (6):

[0053]

[0054] In formula (6), u represents the amount of task data of user i, w represents the complexity of task of user i, and t c Represents the remaining computation delay, t c The calculation method is shown in formula (7):

[0055]

[0056] Furthermore, the calculation method of the fitness ξ of the SFC deployment decision {X, Y, Z} in step 8) is as shown in formula (8):

[0057]

[0058] In formula (8), They represent the normalized computing resources and normalized bandwidth resources required for task offloading, in, The calculation method is shown in formulas (9) and (10):

[0059]

[0060]

[0061] In formula (9), C j represents the total computing resources of UAV j; in formula (10), E represents the set of all physical links in the UAV network, represents the bandwidth resources allocated by physical link e to user i, B e Indicates the total bandwidth resources of physical link e.

[0062] Furthermore, in step 103, according to s k with a k Calculate to obtain s k+1 The methods include:

[0063] Will Import the drone swarm action strategy generator Actor to obtain the k-time slot drone swarm action vector Get it again in,

[0064] The advantages and beneficial effects of the present invention are as follows:

[0065] The present invention discloses a method for offloading tasks in a drone edge network based on SFC. Existing research on drone network task offloading mostly focuses on the offloading of single functions and single drones, without considering the collaborative offloading of multi-function tasks by multiple drones. Existing research on SFC deployment in drone networks is relatively limited, and primarily focuses on VNF mapping and SFC deployment in static network environments, without considering task offloading requirements. In particular, in real-world scenarios, localized bursts of specific types of tasks can severely impact the resource balance of drone networks in static environments, leading to decreased resource utilization and deterioration in task offloading quality. To address the above issues, the present invention utilizes a dual-delay deep deterministic policy gradient algorithm based on the location distribution of ground users and task offloading requests to adaptively plan the cruising trajectory of a drone swarm. SFC mapping decisions are constructed based on the location distribution of drones and the type of onboard VNFs. Resource allocation decisions are constructed based on the state of onboard computing resources and the task offloading latency requirements. By dynamically optimizing SFC mapping and resource allocation decisions during the drone swarm's cruising process, network resource utilization is maximized, thereby effectively improving the task offloading completion rate and reducing the system unit task offloading latency. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 The present invention provides a flowchart of a method for unloading tasks from a drone edge network based on SFC in a preferred embodiment. DETAILED DESCRIPTION

[0067] The following will describe the technical solutions in the embodiments of the present invention in detail with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of the present invention.

[0068] The technical solution of the present invention to solve the above technical problems is:

[0069] The concepts and models involved in the present invention are as follows:

[0070] 1. System model:

[0071] The present invention assumes that the low-altitude network is composed of multiple drones, and the drones use direct communication or forwarding communication based on the effective communication distance constraint. Each drone carries a server with a certain computing power and can provide a limited number of VNF types. According to the user's task offloading requirements, the drone network can select a drone with the corresponding type of VNF for instantiation, and complete the task offloading by orchestrating the required SFC, where different VNFs in the same SFC need to be mapped to different drones. Suppose the ground user set I = {i}, the drone set J = {j}, the physical link set between drones is E = {e}, the set of all VNF types is F (f∈F), and the VNF type set of drone j is The cruising time of the drone swarm consists of K discrete time slots of equal length Δt, and the system state remains unchanged in any kth (∈K) time slot.

[0072] 2. Other symbols involved in the present invention are explained as follows:

[0073] k: number of time slots

[0074] K: total number of time slots

[0075] s k : state vector

[0076] a k : motion vector

[0077] r k : Reward function

[0078] Δt: time slot length

[0079] α j,k : horizontal flight angle of drone j

[0080] β j,k : The vertical flight direction of UAV j

[0081] Horizontal flight speed

[0082] Vertical flight speed

[0083] ψ: System task completion rate

[0084] Normalized system unit task delay

[0085] φ: User communication interruption rate

[0086] X: Node mapping decision

[0087] Y: Link mapping decision

[0088] Z: Resource allocation decision

[0089] τ e : Link cost of physical link e

[0090] Node mapping probability of drone j

[0091] ξ: fitness of SFC deployment decision

[0092] Computing resources allocated by drone j to user i

[0093] Physical link e is the bandwidth resource allocated to user i.

[0094] c0: Computing resources required for VNF instantiation

[0095] b: Bandwidth requirements for task offloading

[0096] S: The set of VNFs required for the task

[0097] u: task data volume

[0098] w: task complexity

[0099] t c : Remaining computation delay

[0100] t tran : Total transmission delay of the offload path

[0101] T max :Task tolerance delay

[0102] The technical solution of the present invention is described as follows:

[0103] 1. Network state vector, action vector, and reward function for the ground-air environment

[0104] Network state vector s k , UAV swarm motion vector a k , reward function r k , as shown in formulas (1), (2), and (3) respectively:

[0105]

[0106]

[0107]

[0108] In formula (1), represents the three-dimensional coordinates of UAV j in time slot k, represents the three-dimensional coordinates of user i, J represents the set of drones, and I represents the set of users; in formula (2), α j,k ∈[0,2π] is the horizontal flight angle of UAV j, is the horizontal flight speed, β j,k ∈{-1,1} is the vertical flight direction of UAV j, is the vertical flight speed; in formula (3), ε1, ε2, ε3 are the weight factors of the drone action reward function, where 0≤ε1≤1, 0≤ε2≤1, 0≤ε3≤1, ψ represents the system task completion rate, represents the normalized system unit task delay, φ represents the user communication interruption rate; 2. Method for calculating the fitness ξ of SFC deployment decision {X, Y, Z}

[0109] The fitness ξ of the SFC deployment decision {X, Y, Z} is shown in formula (4):

[0110]

[0111] In formula (4), They represent the normalized computing resources and normalized bandwidth resources required for task offloading, in, The calculation method is shown in formulas (5) and (6):

[0112]

[0113]

[0114] In formula (5), C j represents the total computing resources of UAV j; in formula (6), E represents the set of all physical links in the UAV network, represents the bandwidth resources allocated by physical link e to user i, B e represents the total bandwidth resource of physical link e;

[0115] 3. Node mapping probability and link cost of drones

[0116] Node mapping probability of drones Link cost τ e , as shown in formulas (7) and (8) respectively:

[0117]

[0118]

[0119] In formula (7), c j represents the available computing resources of UAV j; in formula (8), is the available bandwidth resource b of physical link e e The ratio of to the total bandwidth resources of physical link e represents the normalized available bandwidth resources of physical link e, and b represents the bandwidth requirement of task offloading;

[0120] 4. According to s k with a k Calculate to obtain s k+1 Method

[0121] Will Import the drone swarm action strategy generator Actor to obtain the k-time slot drone swarm action vector Get it again in,

[0122] 5. Computing resources allocated by drone j to user i The calculation method is shown in formula (9):

[0123]

[0124] In formula (9), u represents the amount of task data of user i, w represents the complexity of task of user i, and t c Represents the remaining computation delay, t c The calculation method is shown in formula (10):

[0125]

[0126] 6. SFC Mapping and Resource Allocation Decision Generator Mapping-Allocation Method to obtain user i’s SFC deployment decision {X, Y, Z}

[0127] 1) According to the task offloading request proposed by user i, the VNFs required by user i are sequentially placed into the set S, and the candidate mapping drone set is initialized Count variable m = 0, optimal SFC deployment fitness ξ * =0;

[0128] 2) According to S and the current drone network status, obtain access to drone j for user i a If successful, jump to step 3), otherwise, jump to step 10);

[0129] 3) If m<M max , where M max Indicates the maximum search range hop count of the drone node in the network. Let m = m + 1 and jump to step 4). Otherwise, jump to step 9.

[0130] 4) For distance j a Each drone j within the hop range of drone m will use its available computing resources c j >c0 drones are added to set J map , where c0 represents the basic computing resources required for VNF instantiation. If |J map |≥|S|, set the counting variable q=0 and jump to step 5), otherwise, jump to step 3);

[0131] 5) If q < Q max , where Q max Indicates that in set J map The maximum number of times to search for SFC, let q = q + 1, jump to step 6), otherwise, jump to step 3);

[0132] 6) According to set J map and access to drones a , build and update the mapping decision {X, Y} of the required SFC, if successful, jump to step 7), otherwise, jump to step 5);

[0133] 7) Based on the current UAV network status and mapping decision {X, Y}, construct and update the resource allocation decision {Z} of the required SFC. If successful, jump to step 8), otherwise, jump to step 5);

[0134] 8) Calculate the fitness ξ of the SFC deployment decision {X, Y, Z}, if ξ * ≤ξ, let ξ * =ξ, optimal SFC deployment decision {X * ,Y * ,Z *}={X,Y,Z}, jump to step 5), otherwise, jump to step 5);

[0135] 9) Output the optimal SFC deployment decision {X * ,Y * ,Z *};

[0136] 10) The algorithm ends.

[0137] 7. Sub-algorithm 1: Get access to drones

[0138] 11) Establish a temporary set J′, add the drones in the drone set J that meet the communication distance condition of user i to the set J′, if Jump to step 12), otherwise, jump to step 15);

[0139] 12) Create a temporary set J″ and add the drones in J′ that meet the first VNF ​​type requirement of SFC to J″. If Jump to step 13), otherwise, jump to step 14);

[0140] 13) Based on the available computing resources c in each drone j j The value of c is used to sort the elements in the set J″ in descending order. According to the communication distance between user i and drone j, the elements with the same c are sorted. j Arrange the elements of the value in ascending order, take out the first drone j in J″, and let user i’s access drone j a =j, jump to step 15);

[0141] 14) According to the communication distance between user i and drone j, sort the elements in set J′ in ascending order, take out the first drone j in J′, and let user i’s access drone j a =j;

[0142] 15) The algorithm ends.

[0143] 8. Sub-algorithm 2: Obtaining SFC mapping decision

[0144] 21) Initialize temporary variable j t =j a , temporary set J′=J map , Count variables n=0, l=0;

[0145] 22) Let n = n + 1, add the UAVs in J′ that meet the n-th VNF type requirement of SFC to the set J″;

[0146] 23) For each UAV j in the set J″, the minimum cost path algorithm is used to calculate the link cost between UAVs. t The minimum cost path to UAV j And remove the nodes that do not have the minimum cost path from the set J″;

[0147] 24) If Update the node mapping probability of each drone in J″ based on the available computing resources of the drone And according to the mapping probability distribution According to the probability, a drone j is randomly selected from J″ as the nth VNF mapping node to join the node mapping decision X, and the path As the lth virtual link, join the link mapping decision Y, and let j t = j, remove j from the set J′ and jump to step 25), otherwise, jump to step 26);

[0148] 25) If n<|S|, let l=l+1, Jump to step 22), otherwise, output the mapping decision {X, Y}, and jump to step 26);

[0149] 26) The algorithm ends.

[0150] 9. Sub-algorithm 3: Obtaining SFC resource allocation decisions

[0151] 31) According to the node mapping decision X, the mapping node is added to the temporary set J′, and the basic computing resources c0 required for VNF instantiation are allocated to each drone j in the set J′. According to the link mapping decision Y, the bandwidth b required by the virtual link is added to the resource allocation decision {Z}, and the total task transmission delay t of the offload path is obtained. tran , where t tran is the cumulative transmission delay of task data through each physical link on the offloading path;

[0152] 32) Let the temporary set J″=J′, according to the available computing resources c of drone j j The value of , sort the elements in the set J″ in ascending order;

[0153] 33) If Calculate the task computing resources that need to be allocated to the first UAV j in J″ And update the resource allocation decision {Z}, remove drone j from the set J″, jump to step 33), otherwise, jump to step 34);

[0154] 34) If the task calculation delay t com ≤T max -t tran , where T max If the task tolerates delay, output the resource allocation decision {Z} and jump to step 35), otherwise, jump to step 35);

[0155] 35) The algorithm ends.

[0156] A method for offloading UAV edge network tasks based on SFC, the specific implementation method of which includes the following steps:

[0157] Step 1: Divide the cruising time of the drone swarm into K discrete time slots of equal length Δt. In any kth (∈K) time slot, the system state remains unchanged. The double-delay deep deterministic policy gradient algorithm is used to construct a deep reinforcement learning model. Initialize k = 0 and establish the network state vector s containing the ground user position and the drone swarm position in the kth time slot. k , UAV swarm motion vector a k , reward function r k , UAV swarm action strategy generator Actor, UAV swarm state-action value generator Critic, SFC mapping and resource allocation decision generator Mapping-Allocation, experience sample storage area, random sample set Mini-Batch;

[0158] Step 2: If k < K, let k = k + 1, according to the task request of user i in time slot k, the network state vector s k Based on the distribution status of onboard computing resources, the SFC deployment decision {X, Y, Z} of user i in time slot k is updated through the SFC mapping and resource allocation decision generator Mapping-Allocation, where X is the node mapping decision, Y is the link mapping decision, and Z is the resource allocation decision. Jump to step 3, otherwise, jump to step 5;

[0159] Step 3: Execute the task offloading of user i in time slot k according to {X, Y, Z} and obtain the corresponding drone group action reward r k, obtain s through the drone swarm action strategy generator Actor k The motion vector of the drone group in the state a k , according to s k with a k Calculate to obtain s k+1 , will [s k ,a k ,r k ,s k+1 ] Added experience sample storage area;

[0160] Step 4: Randomly sample a Mini-Batch sample set from the experience sample storage area, and import the Mini-Batch sample set into the drone swarm action strategy generator Actor and the drone swarm state-action value generator Critic for training, then jump to step 2;

[0161] Step 5: The algorithm ends.

[0162] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0163] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0164] The above embodiments should be understood as merely illustrating the present invention and not as limiting the scope of protection of the present invention. After reading the contents of the present invention, technicians may make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for unloading tasks from a UAV edge network based on SFC, characterized in that: The following steps are involved:

101. The cruising time of the drone swarm is divided into K discrete time slots of equal length Δt, and the system state remains unchanged in any k-th time slot, k∈K. The double-delay deep deterministic policy gradient algorithm is used to construct a deep reinforcement learning model. Initialize k=0 and establish the network state vector s containing the ground user position and the drone swarm position in the k-th time slot. k , UAV swarm motion vector a k , reward function r k , UAV swarm action strategy generator Actor, UAV swarm state-action value generator Critic, SFC mapping and resource allocation decision generator Mapping-Allocation, experience sample storage area, random sample set Mini-Batch; 102. If k < K, let k = k + 1, according to the task request of user i in time slot k, the network state vector s k and the distribution status of onboard computing resources, update the SFC deployment decision {X, Y, Z} of user i in time slot k through the SFC mapping and resource allocation decision generator Mapping-Allocation, where X is the node mapping decision, Y is the link mapping decision, and Z is the resource allocation decision. Jump to step 103, otherwise jump to step 105; 103. Perform task offloading for user i in time slot k according to {X, Y, Z} and obtain the corresponding drone group action reward r k , obtain s through the drone swarm action strategy generator Actor k The motion vector of the drone group in the state a k , according to s k with a k Calculate to obtain s k+1 , will [s k ,a k ,r k ,s k+1 ] Added experience sample storage area; 104. Randomly sample a Mini-Batch sample set from the experience sample storage area, and import the Mini-Batch sample set into the drone swarm action strategy generator Actor and the drone swarm state-action value generator Critic for training, and jump to step 102; 105. End; The method for obtaining the SFC deployment decision {X, Y, Z} of user i through the SFC mapping and resource allocation decision generator Mapping-Allocation in step 102 is as follows: 1) According to the task offloading request proposed by user i, the VNFs required by user i are sequentially placed into the set S, and the candidate mapping drone set is initialized Count variable m = 0, optimal SFC deployment fitness ξ * =0; 2) According to S and the current drone network status, obtain access to drone j for user i a If successful, jump to step 3), otherwise, jump to step 10); 3) If m<M max , where M max Indicates the maximum search range hop count of the drone node in the network. Let m = m + 1 and jump to step 4). Otherwise, jump to step 9. 4) For distance j a Each drone j within the hop range of drone m will use its available computing resources c j >c0 drones are added to set J map , where c0 represents the basic computing resources required for VNF instantiation. If |J map |≥|S|, set the counting variable q=0 and jump to step 5), otherwise, jump to step 3); 5) If q < Q max , where Q max Indicates that in set J map The maximum number of times to search for SFC, let q = q + 1, jump to step 6), otherwise, jump to step 3); 6) According to set J map and access to drones a , build and update the mapping decision {X, Y} of the required SFC, if successful, jump to step 7), otherwise, jump to step 5); 7) Based on the current UAV network status and mapping decision {X, Y}, construct and update the resource allocation decision {Z} of the required SFC. If successful, jump to step 8), otherwise, jump to step 5); 8) Calculate the fitness ξ of the SFC deployment decision {X, Y, Z}, if ξ * ≤ξ, let ξ * =ξ, optimal SFC deployment decision {X * ,Y * ,Z * }={X,Y,Z}, jump to step 5), otherwise, jump to step 5); 9) Output the optimal SFC deployment decision {X * ,Y * ,Z * }; 10) The algorithm ends; The method for constructing and updating the resource allocation decision {Z} of the required SFC in step 7) is as follows: 31) According to the node mapping decision X, the mapping node is added to the temporary set J′, and the basic computing resources c0 required for VNF instantiation are allocated to each drone j in the set J′. According to the link mapping decision Y, the bandwidth b required by the virtual link is added to the resource allocation decision {Z}, and the total task transmission delay t of the offload path is obtained. tran , where t tran is the cumulative transmission delay of task data through each physical link on the offloading path; 32) Let the temporary set J″=J′, according to the available computing resources c of drone j j The value of , sort the elements in the set J″ in ascending order; 33) If Calculate the task computing resources that need to be allocated to the first UAV j in J″ And update the resource allocation decision {Z}, remove drone j from the set J″, jump to step 33), otherwise, jump to step 34); 34) If the task calculation delay t com ≤T max -t tran , where T max If the task tolerates delay, output the resource allocation decision {Z} and jump to step 35), otherwise, jump to step 35); 35) The algorithm ends; In step 33), drone j is the task computing resource allocated to user i. The calculation method is shown in formula (6): In formula (6), u represents the amount of task data of user i, w represents the complexity of task of user i, and t c Represents the remaining computation delay, t c The calculation method is shown in formula (7):

2. The SFC-based UAV edge network task offloading method according to claim 1, characterized in that: In step 101, a network state vector s is established in the kth time slot, which includes the ground user position and the drone group position. k , UAV swarm motion vector a k , reward function r k , as shown in formulas (1), (2), and (3) respectively: In formula (1), represents the three-dimensional coordinates of UAV j in time slot k, represents the three-dimensional coordinates of user i, J represents the set of drones, and I represents the set of users; in formula (2), α j,k ∈[0,2π] is the horizontal flight angle of UAV j, is the horizontal flight speed, β j,k ∈{-1,1} is the vertical flight direction of UAV j, is the vertical flight speed; in formula (3), ε1, ε2, ε3 are the weight factors of the drone action reward function, where 0≤ε1≤1, 0≤ε2≤1, 0≤ε3≤1, ψ represents the system task completion rate, represents the normalized system unit task delay, φ represents the user communication interruption rate; 3. The SFC-based UAV edge network task offloading method according to claim 1, characterized in that: In step 2), user i obtains access to drone j a The method is as follows: 11) Establish a temporary set J′, add the drones in the drone set J that meet the communication distance condition of user i to the set J′, if Jump to step 12), otherwise, jump to step 15); 12) Create a temporary set J″ and add the drones in J′ that meet the first VNF ​​type requirement of SFC to J″. If Jump to step 13), otherwise, jump to step 14); 13) Based on the available computing resources c in each drone j j The value of c is used to sort the elements in the set J″ in descending order. According to the communication distance between user i and drone j, the elements with the same c are sorted. j Arrange the elements of the value in ascending order, take out the first drone j in J″, and let user i’s access drone j a =j, jump to step 15); 14) According to the communication distance between user i and drone j, sort the elements in set J′ in ascending order, take out the first drone j in J′, and let user i’s access drone j a =j; 15) The algorithm ends.

4. The SFC-based UAV edge network task offloading method according to claim 1, characterized in that: The method for constructing and updating the mapping decision {X, Y} of the required SFC in step 6) is as follows: 21) Initialize temporary variable j t =j a , temporary set J′=J map , Count variables n=0, l=0; 22) Let n = n + 1, add the UAVs in J′ that meet the n-th VNF type requirement of SFC to the set J″; 23) For each UAV j in the set J″, the minimum cost path algorithm is used to calculate the link cost between UAVs. t The minimum cost path to UAV j And remove the nodes that do not have the minimum cost path from the set J″; 24) If Update the node mapping probability of each drone in J″ based on the available computing resources of the drone And according to the mapping probability distribution According to the probability, a drone j is randomly selected from J″ as the nth VNF mapping node to join the node mapping decision X, and the path As the lth virtual link, join the link mapping decision Y, and let j t =j, remove j from the set J′ and jump to step 25), otherwise, jump to step 26); 25) If n<|S|, let l=l+1, Jump to step 22), otherwise, output the mapping decision {X, Y}, and jump to step 26); 26) The algorithm ends.

5. The SFC-based UAV edge network task offloading method according to claim 4 is characterized in that: The calculation method of the link cost between each UAV in step 23) is shown in formula (4): In formula (4), is the available bandwidth resource b of physical link e e The ratio of to the total bandwidth resources of physical link e represents the normalized available bandwidth resources of physical link e, and b represents the bandwidth requirement of task offloading; In step 24), the node mapping probability of each drone in J″ is updated The method is shown in formula (5): In formula (5), c j represents the available computing resources of UAV j.

6. The SFC-based UAV edge network task offloading method according to claim 1, characterized in that: The calculation method of the fitness ξ of the SFC deployment decision {X, Y, Z} in step 8) is shown in formula (8): In formula (8), They represent the normalized computing resources and normalized bandwidth resources required for task offloading, in, The calculation method is shown in formulas (9) and (10): In formula (9), C j represents the total computing resources of UAV j; in formula (10), E represents the set of all physical links in the UAV network, represents the bandwidth resources allocated by physical link e to user i, B e Indicates the total bandwidth resources of physical link e.

7. The SFC-based UAV edge network task offloading method according to claim 1, characterized in that: In step 103, according to s k with a k Calculate to obtain s k+1 The methods include: Will Import the drone swarm action strategy generator Actor to obtain the k-time slot drone swarm action vector Get it again in,

Images