Computing offloading strategy based on LEO satellite and HAP cooperation in large-scale disaster areas

Abstract

This invention discloses a computational offloading strategy based on LEO satellite and HAP collaboration in large-scale disaster areas. The strategy first allocates subcarrier resources to tasks using a cooperative game theory algorithm under the offloading decision-making scheme. Then, tasks offloaded to edge servers compete for resources with the goal of maximizing the system's QoE, achieving optimal allocation of computational resources. Finally, the offloading decision problem is proven to be a potential game process, and the multi-user game computing offloading (MUGCO) algorithm is used to find an offloading decision scheme under Nash equilibrium. Simulation results show that the proposed strategy can effectively improve the system's total QoE compared to contrasting strategies.

Description

Technical Field

[0001] This invention relates to wireless communication technology in large-scale disaster areas, specifically a computation offloading strategy based on LEO satellite and HAP collaboration in large-scale disaster areas. Background Technology

[0002] In recent years, large-scale disasters (such as earthquakes, floods, and hurricanes) have occurred frequently. During these disasters, communication networks are often paralyzed due to damage to ground communication infrastructure (such as base stations), leading to the interruption of wireless services in disaster areas. In such emergencies, disaster victims urgently need to broadcast their location and send out real-time updates via social media platforms on their user devices. Rescue personnel are dispatched to the disaster area and carry out rescue missions using intelligent rescue user devices (such as rescue robots). To process these data tasks running on ground user equipment (such as video streaming, big data analysis, and image processing) in a timely manner, ensuring smooth post-disaster relief efforts, mobile edge computing (MEC) is typically integrated into emergency communication networks to provide compute offloading services.

[0003] In traditional disaster relief communication networks, unmanned aerial vehicles (UAVs) play a crucial role in post-disaster relief due to their advantages of low cost, flexible deployment, and high mobility. Currently, Zhao, N, and others have proposed that UAVs collaborate with surviving base stations to reduce communication latency between disaster victims and UAVs by jointly optimizing UAV flight trajectories and scheduling. However, due to the limited payload and battery life of UAVs, MEC servers deployed on UAVs struggle to handle large numbers of intensive tasks. To overcome the challenge of limited UAV computing resources, Chen, W, and others proposed using UAV relay to offload GUE tasks to surviving base stations or UAVs deployed with edge servers. Xu, K, and others proposed a network architecture that integrates UAVs with remote wide area networks (WANs), using flexible UAV relay to offload tasks to WANs with abundant computing resources. While they can temporarily solve the problem of limited computing resources on UAVs, due to the inherent limitation of UAVs' small coverage area, the number of UAVs needed to achieve full wireless connectivity coverage in large disaster areas would increase exponentially, which is almost impossible to achieve.

[0004] Compared to UAVs, the new HAP has the capability to carry heavy 5G infrastructure and offers longer service life, larger coverage area, and wider spectrum resources, making HAP-integrated Mobile Edge Computing (MEC) ideal for large-scale disaster relief emergencies. However, for large disaster areas, offloading too many GUE tasks to HAPs can easily cause overload, leading to a decrease in the total QoE of GUEs. Expanding the network with cloud data centers that have abundant computing resources becomes a good option. Compared to HAPs, LEO satellites, with their large coverage area, have the ability to communicate directly with cloud data centers outside the disaster area. Therefore, in large disaster areas, LEO satellites can act as an edge node for aggregation and relay forwarding. Currently, such emerging space-air-ground integrated networks have become a research hotspot. To this end, L. Zhang et al. proposed a SAIC architecture and studied the joint user association and offloading decision problem in a two-layer network with the goal of maximizing efficiency and rate. Ding, J. et al. proposed a SAIECN architecture that minimizes the system's total energy consumption by jointly optimizing GUE association, MU-MIMO transport precoding, resource allocation, and offloading decisions. While these architectures primarily consider a two-layer structure involving HAP and LEO satellite collaboration, they all neglect the auxiliary role of cloud data centers in the computation offloading process. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a computational offloading strategy based on LEO satellite and HAP collaboration in large-scale disaster areas, aiming to improve the Quality of Experience (QoE) of GUE wireless services. The optimization problem is modeled as maximizing the weighted sum of the relative reduction in processing latency and energy consumption for all task offloading.

[0006] The technical solution to achieve the objective of this invention is as follows: First, under the offloading decision scheme, a cooperative game theory algorithm is used to allocate subcarrier resources to tasks; then, tasks offloaded to the edge server compete for resources with the goal of maximizing the system QoE value, achieving optimal allocation of computing resources; finally, the offloading decision problem is proven to be a potential game process, and the multi-user game computing offloading (MUGCO) algorithm is used to find an offloading decision scheme under Nash equilibrium. Simulation results show that the proposed strategy can effectively improve the overall QoE value of the system compared to the contrasting strategies.

[0007] The present invention provides a computational offloading strategy based on LEO satellite and HAP collaboration in large-scale disaster areas, including:

[0008] (1) Establish a communication model: The tasks to be offloaded are transmitted uplink to the HAP using Non-orthogonal Multiple Access (NOMA) technology. Assume there are K subcarriers in the system, and each HAP has full channel state information for these K subcarriers. The tasks to be offloaded are divided into K groups, and the GUEs in each group can share the same subcarrier; use the set S = {s1, s2, ... s...} k} represents these K groups, where s k Let Γ represent a group of tasks transmitted via subcarrier k, containing Q tasks, denoted by the set Γ = [1, q, ..., Q]. Furthermore, let Γ represent... Let represent the channel gain of task q uplink transmission via subcarrier k, where but It can be expressed by equation (1):

[0009]

[0010] In the formula X i,j Let GUE j be the horizontal distance relative to HAP i, η0 be the channel gain at the reference distance of 1m, and τ be the path loss exponent between GUE and HAP. The channel gain of the uplink transmission task via subcarrier k satisfies

[0011] According to the definition of NOMA technology, when task n,m The signal-to-interference-plus-noise ratio (SIR) transmitted to the HAP via subcarrier k uplink can be expressed as:

[0012]

[0013] In the formula Represents the subcarrier selection factor, when task n,m When selecting subcarrier k Conversely, the opposite is also true; σ represents the transmission power of user q who uploads data via subcarrier K. 2 This represents background noise; the task can be derived from the above formula. n,m The transmission rate from the ground uplink to the HAP is:

[0014]

[0015] In the formula B k This represents the transmission bandwidth of subcarrier k;

[0016] (2) Establishing a computational model: Under the SAGIEC disaster relief network, the tasks generated by GUE have three offloading processing modes, defined as o n,m ∈{0,1,2,3} is a taskn,m Unloading decision factors:

[0017] When o n,m When the value is 0, it indicates that the task is processed locally, with a latency of [missing information]. in For GUE's computing power; energy consumption Where κ is a constant related to the CPU architecture;

[0018] When o n,m When = 1, it indicates that the task is offloaded to the MEC server on the HAP for processing, with latency... in Assign tasks to MEC servers on HAP n,m Computing resources; energy consumption

[0019] When o n,m When the value is 2, it indicates that the task is offloaded to the satellite. MEC server processing, latency in Assign tasks to the MEC server on the LEO satellite n,m Computing resources, L n,s Where n is the distance from HAP to the LEO satellite, and c is the speed of light; energy consumption

[0020] When o n,m When the value is 3, it indicates that the task is offloaded to the cloud data center for processing, and the latency is [missing information]. in This indicates the latency during which a task is sent from a LEO satellite to a cloud data center for processing; energy consumption.

[0021] Get task n,m The latency and energy consumption after scheduling are:

[0022]

[0023]

[0024] In the formula Z {*} This means that if the expression * holds, then Z {*} =1, otherwise Z {*} =0;

[0025] (3) Optimization problem modeling: task n,m The scheduled QoE function is defined as follows:

[0026]

[0027] in These represent the weights for latency and energy consumption, respectively; the problem is modeled as maximizing the QoE function value for all GUEs, i.e.:

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036] In the formula, F represents the computing resource allocation scheme, S represents the subcarrier selection scheme, and W represents the offloading decision scheme; the constraints C1 represent the maximum tolerable delay of the task, C2 represent the offloading decision selection space of the task, C3 represent the subcarrier decision factor of the task, C4 to C5 represent that the total computing resources allocated to the offloading task cannot exceed the total computing resources of the corresponding MEC server, and C6 to C7 represent the minimum computing resources required for the task to be offloaded to the corresponding edge server.

[0037] (4) Prioritize tasks based on two metrics: maximum task latency constraint and the number of CPU cycles required to compute the task. n,m The mathematical model for priority scoring can be expressed as:

[0038]

[0039] In the formula, 0≤λ n,m ≤1 indicates the weight of the two indicators, which can be obtained using the entropy method. After obtaining the score for each task, a sorting algorithm is used to divide them into four levels from high to low, and the scores are stored in matrix A = [A1, A2, A3, A4]. This yields the task scores. n,m The initialization pre-unloading scheme is as follows: when score n,m ∈A1, o n,m =1; when score n,m ∈A2, o n,m =2; when score n,m ∈A3, o n,m =3; when the task score n,m ∈A4, o n,m =0;

[0040] (5) Construct a game theory model Φ={Y,S,G} for the allocation of subcarrier resources, where Y represents the set of all tasks that need uplink transmission, and S={s1,s2,...s k} represents the packet set of subcarriers, G represents the utility function of the task, which is proportional to the uplink transmission rate, and s represents the subcarriers. k The formula for calculating the utility function in the equation is: q∈s k In the formula This indicates the transmission rate of task q through subcarrier k:

[0041] Definition 1: {s ι ,s k}→{s ι ∪q,s k {q} represents group s k Task q leaves and joins group s. ι , where s ι ∈S, and s ι ≠s k ;

[0042] Definition 2: This indicates that task q is paired with group s. k and s t The degree of preference, for example, group s k Task q preference grouping s t It can be represented as:

[0043] Definition 3: If group s k Task q and group s ι If task q' satisfies Definition 2, then a swapping action will occur, and the grouping will be updated to: {s} ι ,s k}→{s ι ∪{q}\{q'},s k ∪{q'}\{q}};

[0044] When GUE performs uplink transmission, it initializes the random allocation of subcarriers. After entering the iteration of cooperative game, it randomly selects a subcarrier that is different from the previous subcarrier to perform the grouping operation defined above until each group reaches stability, thereby achieving the highest efficiency of the NOMA system.

[0045] The mathematical model for allocating computing resources for edge servers can be derived from equation (9):

[0046]

[0047] s.tC4C5C6C7

[0048] (6) Model the unloading decision problem as a game Ψ={Z,W,H n,m}, where Z represents the GUE set, W = {(o n,m ):o n,m ∈{0,1,2,3},n∈I,m∈J} represents the selection policy space for each GUE, H n,m The QoE function denoted by Ψ is used to define a game Ψ. If a game Ψ has a function χ that satisfies the following conditions, then the game is called a potential game:

[0049]

[0050] And it was proven that the potential function in game Ψ is:

[0051]

[0052] The advantages of this invention are:

[0053] (1) Problem modeling: Taking the QoE value of all GUEs in the system as the optimization objective, the problem is planned as a weighted sum of the relative reduction in task offloading processing latency and energy consumption, and an offloading decision and resource allocation problem model based on the SAGIEC disaster relief network is established.

[0054] (2) Mathematical analysis: The optimization problem is decomposed into two subproblems. The first subproblem is the allocation of subcarrier resources and computing resources under the offloading scheme, which is solved by cooperative game algorithm and computing resource competition algorithm based on task priority, respectively; the second subproblem is the offloading decision problem under the optimal resource allocation. This problem is first modeled as a potential game process, and then the offloading decision scheme under Nash equilibrium is found by the MUGCO algorithm.

[0055] (3) Simulation Verification: The proposed unloading decision and resource allocation scheme was simulated and analyzed using MATLAB software, and compared with the reference scheme. Simulation results show that the proposed strategy can effectively improve the overall QoE value of the system compared with the reference strategy. Attached Figure Description

[0056] Figure 1 This is a system model diagram of an embodiment of the present invention;

[0057] Figure 2 This is a flowchart of the MUGCO algorithm according to an embodiment of the present invention;

[0058] Figure 3 This is a graph showing the relationship between the number of GUEs and the average transmission rate of GUEs in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram of the QoE values ​​of edge servers under different computing resource allocation algorithms in an embodiment of the present invention;

[0060] Figure 5 This is a graph showing the relationship between the number of GUEs and the total QoE value of the system in an embodiment of the present invention. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this is not intended to limit the scope of the invention.

[0062] Example:

[0063] The SAGIEC disaster relief network considered in this article is as follows: Figure 1 As shown. In the air layer, there are I HAPs and one LEO satellite, represented by a set. and They are all equipped with MEC servers to handle tasks running in GUE. HAP, in particular, was deployed over the disaster area. H Located at an altitude of L, it floats quasi-statically in the stratosphere, capable of handling GUE offloading tasks and serving as a relay node for communication with LEO satellites. Furthermore, LEO satellites equipped with enhanced MEC servers are deployed at an orbital altitude of L... s At this location, it can communicate with the cloud data center via a backhaul link. Assuming each high-altitude platform serves J GUEs randomly distributed throughout the disaster area, using a set... This indicates that each GUE has one data task to process, using the term "task". n,m This represents the task generated by the m-th GUE under the n-th HAP, where Use arrays They represent tasks respectively. n,m The attribute of d, where d n,m c represents the size of the task's data. n,m This indicates the number of CPU cycles required to process the task. This indicates the maximum tolerable delay for the task.

[0064] Establish a communication model

[0065] The tasks to be offloaded are transmitted uplink to the HAP based on NOMA technology. Assume there are K subcarriers in the system, and each HAP has full channel state information for these K subcarriers. We divide the tasks to be offloaded into K groups, where GUEs in each group can share the same subcarrier, denoted by a set S = {s1, s2, ... s...}. k} represents. Where, s k Let Γ represent a GUE packet transmitted via subcarrier k, containing Q tasks, denoted by the set Γ = [1, q, ..., Q]. We use... Let represent the channel gain of task q uplink transmission via subcarrier k, where but This can be expressed as a formula:

[0066]

[0067] Where X i,j Let be the horizontal distance between GUE j and HAP i, η0 be the channel gain at the reference distance of 1m, and τ be the path loss exponent between GUE and HAP. Without loss of generality, assume that the channel gain of the uplink transmission task via subcarrier k satisfies the following condition.

[0068] According to the definition of NOMA technology, when task n,m The signal-to-interference-plus-noise ratio (SIR) transmitted to the HAP via subcarrier k uplink can be expressed as:

[0069]

[0070] in Represents the subcarrier selection factor, when task n,m When selecting subcarrier k Conversely, the opposite is also true; σ represents the transmission power of user q who uploads data via subcarrier K. 2 This represents background noise. Based on the above, the task can be derived using Shannon's formula. n,m The transmission rate from the ground uplink to the HAP is:

[0071]

[0072] In the formula B k This represents the transmission bandwidth of subcarrier k. For ease of analysis, this paper sets up a task... n,m The transmission rate from HAP to LEO satellite is

[0073] Establish a computational model

[0074] Under the SAGIEC disaster relief network, tasks generated by GUE have three offloading processing modes, defined as o n,m ∈{0,1,2,3} is a task n,m The unloading decision factor. When o n,m When the value is 0, it indicates that the task is processed locally, with a latency of [missing information]. in For GUE's computing power; energy consumption Where κ is a constant related to the CPU architecture.

[0075] When o n,m When = 1, it indicates that the task is offloaded to the MEC server on the HAP for processing, with latency... in Assign tasks to MEC servers on HAP n,m Computing resources; energy consumption

[0076] When o n,m When the value is 2, it indicates that the task is offloaded to the satellite. MEC server processing, latency in Assign tasks to the MEC server on the LEO satellite n,m Computing resources, L n,s Where n is the distance from HAP to the LEO satellite, and c is the speed of light; energy consumption

[0077] When o n,m When the value is 3, it indicates that the task is offloaded to the cloud data center for processing, and the latency is [missing information]. in This indicates the latency during which a task is sent from a LEO satellite to a cloud data center for processing; energy consumption. Furthermore, we use matrix W = {o 1,1 ,o 1,2 ,...,o n,m} represents the unloading decision scheme for all GUEs.

[0078] In summary, we can obtain the task. n,m The latency and energy consumption after scheduling are:

[0079]

[0080]

[0081] In the formula Z {*} This means that if the expression * holds, then Z {*} =1, otherwise Z {*} =0.

[0082] Optimization problem modeling

[0083] Based on the above, task n,m The scheduled QoE function is defined as follows:

[0084]

[0085] in These represent the weights for latency and energy consumption, respectively. The goal of this paper is to maximize the QoE function value for all GUEs, i.e.:

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094] In the formula, F represents the computing resource allocation scheme, and S represents the subcarrier selection scheme. The constraints are: C1 represents the maximum tolerable delay of the task, C2 represents the task offloading decision selection space, C3 represents the subcarrier decision factor of the task, C4 to C5 represent that the total computing resources allocated to the offloading task cannot exceed the total computing resources of the corresponding MEC server, and C6 to C7 represent the minimum computing resources required for the task to be offloaded to the corresponding edge server.

[0095] Joint offloading decision and resource allocation strategy

[0096] The optimization variables in problem (7) include both integer and continuous variables, making it a mixed-integer nonlinear programming problem. Furthermore, due to the strong coupling of the optimization variables, this paper decomposes the problem into two subproblems for easier mathematical analysis. The first subproblem is a resource allocation problem with a pre-unloading scheme, including the allocation of subcarrier resources and edge server computing resources, solved using cooperative game theory and task-priority-based computing resource competition algorithms, respectively. The second subproblem is an unloading decision problem with optimal resource allocation. First, it is proven that the optimization problem is a potential function of the unloading decision; then, the MUGCO algorithm is used to find the unloading decision scheme under Nash equilibrium.

[0097] Initialization processing

[0098] In the disaster area, different tasks have different service requirements. Therefore, priorities are established based on two indicators: the maximum latency constraint of each task and the number of CPU cycles required to compute the task. Among these, tasks... n,m The mathematical model for priority scoring can be expressed as:

[0099]

[0100] In the formula, 0≤λ n,m ≤1 indicates the weight of the two indicators, which can be obtained using the entropy method. After obtaining the score for each task, a sorting algorithm is used to divide them into four levels from high to low, and the results are stored in matrix A = [A1, A2, A3, A4]. This yields the task...n,m The initialization pre-unloading scheme is as follows: when score n,m ∈A1, o n,m =1; when score n,m ∈A2, o n,m =2; when score n,m ∈A3, o n,m =3; when the task score n,m ∈A4, o n,m =0.

[0101] Resource allocation

[0102] In the subcarrier resource allocation process, to efficiently group tasks from multiple GUEs and thus improve the uplink transmission rate, this paper employs a cooperative game theory algorithm, constructing a game model Φ = {Y, S, G}. Here, Y represents the set of all tasks requiring uplink transmission, and S = {s1, s2, ... s}. k Let} represent the set of subcarrier packets, and G represent the utility function of the task, which is proportional to the uplink transmission rate. For example, subcarrier s k The formula for calculating the utility function in the equation is: q∈s k In the formula This represents the transmission rate of task q via subcarrier k. The cooperative game process will be described in detail below:

[0103] Definition 1: {s ι ,s k}→{s ι ∪q,s k {q} represents group s k Task q leaves and joins group s. ι , where s ι ∈S, and s ι ≠s k .

[0104] Definition 2: This indicates that task q is paired with group s. k and s t The degree of preference, for example, group s k Task q preference grouping s t It can be represented as:

[0105] Definition 3: If group s k Task q and group s ι If tasks q' in the group satisfy Definition 2, then they will engage in swapping behavior and update the grouping to: {s ι ,s k}→{s ι ∪{q}\{q'},sk ∪{q'}\{q}}.

[0106] During uplink transmission of the GUE task, the initial subcarriers are randomly assigned. After entering the cooperative game iteration, a subcarrier different from the previous subcarrier is randomly selected for the grouping operation defined above, until each group reaches stability, thereby achieving the highest efficiency of the NOMA system. The specific cooperative game process is shown in Algorithm 1 below:

[0107]

[0108]

[0109] Lemma 1: Algorithm 1 converges to the final set S in a finite number of steps. end .

[0110] Proof: To optimize the utility function value G, users will continuously engage in grouping operations. In the k-th to k+1-th iterations, if the group set changes from S... k Update to S k+1 Then the utility function G in this game must strictly increase, that is: Therefore, in the game iteration, the utility function G always increases, that is: S start →S1→...→S end Since the number of users transmitting uplinks is finite, the packet set is also finite and based on the Bell number. Therefore, as the game iterations increase, the packet set will converge to S. end .

[0111] The mathematical model for allocating computing resources for edge servers can be derived from equation (7):

[0112]

[0113] s.tC4C5C6C7

[0114] To address the varying task service requirements of different GUEs in disaster areas, this paper designs a computational resource contention algorithm based on task priority. Under the premise of task priority, users offloaded to edge servers are allocated the minimum computational resources required to meet the maximum latency requirements in descending order of priority. Then, the idle computational resources on the servers are divided into several resource blocks. Offloaded tasks maximize the overall QoE value of the system by competing for these resource blocks. The specific steps are shown in Algorithm 2:

[0115]

[0116]

[0117] Uninstallation decision

[0118] For distributed computing users (GUEs) in disaster areas, it is necessary to design a low-complexity offloading decision scheme that satisfies all GUEs to maximize the overall QoE of the system. In game theory, potential games, due to their finite improvement property, have the advantage of fast convergence in studying distributed computing offloading strategies. Therefore, this paper, based on potential games, first models the offloading decision problem as a game Ψ={Z,W,H}. n,m}, where Z represents the GUE set, W = {(o n,m ):o n,m ∈{0,1,2,3},n∈I,m∈J} represents the selection policy space for each GUE, H n,m The QoE function is represented. Then we prove that game Ψ is a potential game. A detailed explanation follows:

[0119] Definition 4: If a game Ψ has a function χ that satisfies the following conditions, then the game is called a potential game:

[0120]

[0121] Lemma 2: The potential function in game Ψ is:

[0122]

[0123] Scenario 1: When task n,m The uninstallation decision is made by o n,m =0 updated to o' n,m When the value is greater than 0, some uninstalled tasks will be affected by the task... n,m This consumes computing resources from transmission subcarriers and edge servers, leading to a decrease in their transmission and computing performance. (Using tasks...) x,y This indicates the portion of the tasks that are affected. If H... n,m (o n,m ,o -(n,m) )-H n,m (o' n,m ,o -(n,m) If ) < 0, then we can deduce:

[0124]

[0125] In the formula, T x,y and E x,y Indicates task n,m Update uninstallation decision before task x,y The latency and energy consumption of T' x,y and E' x,y Indicates task n,m Update uninstallation decision task x,y The latency and energy consumption of T' n,mand E' n,m Indicates task n,m Latency and energy consumption after updating the uninstallation decision.

[0126]

[0127] Without loss of generality, let From the above formula, we can deduce that: χ(W)-χ(W')<0.

[0128] Scenario 2: When task n,m The uninstallation decision is made by o n,m >0 updated to o' n,m When the value is 0, some uninstalled tasks will be affected by the task. n,m This frees up computing resources for transmission subcarriers and edge servers, improving their transmission and computing performance. If using tasks... x,y This indicates the portion of the task that is affected, which can be represented by H. x,y (o x,y ,o -(x,y) ) < H x,y (o' x,y ,o -(x,y) If H n,m (o n,m ,o -(n,m) )-H n,m (o' n,m ,o -(n,m) If ) < 0, then we can deduce:

[0129]

[0130] Scenario 3: When task n,m The uninstallation decision is made by o n,m >0 updated to o' n,m >0, but o n,m ≠o' n,m At this point, we can further divide the discussion into three scenarios:

[0131] ①task n,m The unloading decision is made by 0 < o n,m <3 updated to 0<o' n,m <3. Without loss of generality, we assume task n,m The uninstallation decision is made by o n,m =1 updated to o' n,m =2, this update frees up the task. n,m Offloading the required computing resources to HAP competes with tasks offloaded to LEO satellites for computing resources, which also increases the number of tasks. n,m The transmission latency. Here, we use task. x1,y1 and These respectively indicate that because of task n,m Tasks affected by the release and occupation of computing resources.

[0132] If H n,m (o n,m ,o -(n,m) )-H n,m (o' n,m ,o -(n,m) If ) < 0, then we can deduce:

[0133]

[0134] In the formula and Indicates before being affected and Allotted computing resources; f' x1,y1 and f' x2,y2 Indicates that after being affected and The allocated computing resources.

[0135]

[0136] In the formula and They represent tasks respectively. n,m The processing latency of tasks affected before and after releasing computing resources. and They represent tasks respectively. n,m The processing latency of tasks affected before and after computing resources are used.

[0137] ②task n,m The uninstallation decision is made by o n,m =3 updated to 0<o' n,m <3. This update will cause a decrease in the computing performance of some tasks because it consumes the computing resources of the edge server. The proof method is similar to that of case one, and will not be repeated here.

[0138] ③task n,m The unloading decision is made by 0 < o n,m <3 updated to o' n,m =3. This update frees up the server's computing resources, which improves the transmission and computing performance for some users. The proof method is similar to that of Case 2, and will not be repeated here.

[0139] In summary, it is proven that the function χ(S) is the potential function of game Ψ, therefore game Ψ is a potential game. According to the finite improvement property of potential games, game Ψ can find the Nash equilibrium after a finite number of iterations. Based on this property, this paper proposes a MUGCO algorithm to achieve Nash equilibrium. The core idea of ​​the MUGCO algorithm is: to update the unloading decision of no more than one GUE at a time, and the update rule for each update is to find the best response of the GUE. If all GUEs find the best response, then the unloading decision scheme under Nash equilibrium can be obtained. The method for calculating the best response is shown in the following equation:

[0140]

[0141] Algorithm Complexity Analysis

[0142] To maximize the QoE of the system, this paper proposes a joint offload decision and resource allocation strategy. The allocation of subcarrier resources employs a cooperative game theory algorithm, performing K calculations in each iteration. When the Kth iteration is reached, the complexity is O(K^2). 2 The computational resource allocation uses a task priority-based computational resource contention algorithm with a complexity of O(I×J); therefore, the unloading decision algorithm has a complexity of O(K). 2 ×I×J).

[0143] Simulation Analysis

[0144] This paper presents a simulation experiment analysis of the proposed solution based on the SAGIEC disaster relief scenario. Within a large disaster area with a diameter of 100 km, GUEs are randomly distributed, and 10 HAPs and 1 LEO satellite are deployed above the disaster area at an altitude of L. H =20km and orbital altitude is L s =500km. The edge server computing resources on the HAP are 8GHz, and the edge server computing resources on the LEO satellite are 20GHz; the remaining simulation parameters are shown in Table 1:

[0145] Table 1 Simulation Parameters

[0146]

[0147] This paper, based on NOMA technology, employs a cooperative game theory (Cooperative Game-NOMA) algorithm for the reasonable subcarrier pairing task, maximizing the overall efficiency of the NOMA system and thus improving the average uplink transmission rate of the GUE. A comparison is made with random allocation schemes (Random-NOMA) and Orthogonal Multiple Access (OMA). Figure 3As shown, the average transmission rate of GUEs decreases continuously with the increase in the number of uplink transmissions. This is because the subcarrier reuse frequency increases, leading to increased interference experienced by GUEs during transmission. However, compared with the other two schemes, the cooperative game-NOMA algorithm achieves the highest average rate for GUEs.

[0148] In edge server computing resource allocation schemes, this paper proposes a task priority-based computing resource contention algorithm. This algorithm first sorts tasks offloaded to the edge server according to their priorities from highest to lowest. Then, while satisfying the maximum latency requirement of GUE, it allocates the minimum computing resources needed for task processing. Finally, it maximizes the overall system QoE by allocating the remaining computing resources according to multi-user resource contention rules. The algorithm is compared with convex optimization schemes and minimum resource allocation schemes. Figure 4 As shown in the figure, when J=17, the QoE values ​​obtained by the edge servers according to different schemes for resource allocation are as follows: edge servers 1-10 belong to HAP, and 11 belongs to LEO satellite. It can be seen from the figure that the QoE value obtained by the algorithm in this paper is better than the other two schemes.

[0149] In the offloading decision scheme, to maximize the overall QoE of the system, this paper uses the MUGCO algorithm based on potential game theory to find the Nash equilibrium in the system, and compares it with the Greedy algorithm, the Rand algorithm, and the All-In-HAP algorithm. Figure 5 As shown, when the number of GUEs is small, offloading tasks can achieve higher transmission rates and more computing resources, leading to a rapid increase in the total QoE value of the system. However, as the number of GUEs increases, the transmission rate decreases and competition for computing resources on edge servers intensifies, resulting in a reduction in the QoE obtained by GUE task offloading, causing the total QoE value of the system to grow slowly. Nevertheless, compared to the other three offloading decision algorithms, the MUGCO algorithm achieves the highest system QoE value.

[0150] in conclusion

[0151] This paper studies the resource allocation and offloading decision-making problem of GUEs based on the SAGIEC network in a large-scale disaster area. An optimization problem model is established with the total QoE of the system as the objective. Since the QoE requirements of different GUEs in the disaster area vary, this paper sets priorities based on the attributes of the GUE tasks and initializes the offloading decision scheme according to the priority levels. Then, a cooperative game theory algorithm is used to rationally allocate subcarrier resources for uplink GUEs, and a task priority-based computational resource contention algorithm is used to allocate computational resources for GUE tasks offloaded to edge servers. Finally, the MUGCO algorithm is used to obtain the offloading decision scheme under Nash equilibrium, thereby effectively improving the total QoE of the system.

Claims

1. A computational offloading strategy based on LEO satellite and HAP collaboration in large-scale disaster areas, characterized by: including: (1) Establish a communication model: The tasks that need to be offloaded are transmitted to the HAP via non-orthogonal multiple access (NOMA) technology. Assume that the system has Each HAP has 1 subcarrier and 1 HAP has 1 subcarrier. The full channel state information of each subcarrier is used to divide the tasks that need to be offloaded into... Groups, where GUEs within each group can share the same subcarrier; using sets This means There are 1 group, among which... Indicates via subcarrier The upload task is grouped, and the number of tasks in this group is... One, using sets In addition, using Indicates task via subcarrier Uplink transmission channel gain, where , ;but It can be expressed by formula (1): （1） In the formula For GUE With HAP The horizontal distance is the baseline. For the channel gain at a reference distance of 1m, The path loss index between GUE and HAP is calculated using subcarriers. The channel gain of the uplink transmission task satisfies I represents the i-th HAP, J represents the randomly distributed GUE in the disaster area, and L H The indicated feedback is the altitude at which the HAP is deployed in the air; According to the definition of NOMA technology, when via subcarrier The signal-to-interference-plus-noise ratio (SIR) transmitted to the HAP via uplink can be expressed as: （2） In the formula Represents the subcarrier selection factor, task n,m This represents the edge computing offloading task generated by the m-th GUE under the n-th HAP, when Select subcarrier hour, Conversely, the opposite is also true; This represents the transmission power of user q who uploads data via subcarrier K. This is background noise; according to the above formula, we can obtain... The transmission rate from the ground uplink to the HAP is: （3） In the formula Indicates subcarrier The transmission bandwidth; (2) Establishing a computational model: Under the SAGIEC disaster relief network, the tasks generated by GUE have three offloading processing modes, defined as follows: for Unloading decision factors: when "Time" indicates that the task is processed locally, with a latency of [time value missing]. ,in For GUE's computing power; energy consumption ,in These are CPU architecture-related constants, C n,m This indicates the number of CPU cycles required for the computation task; when When, it indicates that the task is offloaded to the MEC server on the HAP for processing, with a latency of [time value missing]. ,in Assign MEC server on HAP Computing resources; energy consumption R h n,m Indicates task n,m Uplink transmission rate to HAP; when At that time, it indicates that the mission is unloaded onto the satellite. MEC server processing, latency ,in Assign MEC servers on LEO satellites Computing resources For HAP Distance to LEO satellite, Speed ​​of light; energy consumption d n,m R represents the size of the task data. s n,m Indicates task n,m Uplink transmission rate to LEO satellite when "Time" indicates that the task is offloaded to the cloud data center for processing, and the latency is... ,in This indicates the latency during which a task is sent from a LEO satellite to a cloud data center for processing; energy consumption. ; get The latency and energy consumption after scheduling are: （4） （5） In the formula If expression If it is established, then ,otherwise ; (3) Optimization problem modeling: The scheduled QoE function is defined as follows: （6） in , , representing the weights of latency and energy consumption, respectively; the problem is modeled as maximizing the QoE function value of all GUEs, i.e.: （7） ， ； ， ； ， ； ， ； ， ； ， ； ， ； In the formula Representative resource allocation plan, Represents the subcarrier selection scheme. Indicate the unloading decision scheme; constraints This indicates the maximum tolerable latency for the task. This represents the task unloading decision space. The subcarrier decision factor represents the task. This means that the total computing resources allocated to the unloading tasks cannot exceed the total computing resources of the corresponding MEC server. T represents the minimum computing resources required to offload a task to its corresponding edge server. max n.m Indicates the maximum tolerable latency for the task; (4) Construct priorities based on two metrics: the maximum latency constraint of the task and the number of CPU cycles required to compute the task. The mathematical model for priority scoring can be expressed as: （8） In the formula The weights of the two indicators can be calculated using the entropy method. After obtaining the score for each task, a sorting algorithm is used to divide them into four levels from high to low, and the scores are stored in a matrix. In the middle; then we can obtain The initialization pre-unloading scheme is as follows: when , ;when , ;when , When the task , ; (5) Construct a game theory model for the allocation of subcarrier resources. ,in, This represents the set of all tasks that require uplink transmission. Represents the packet set of subcarriers. The utility function representing the task is proportional to the uplink transmission rate, and the subcarrier... The formula for calculating the utility function in the equation is: , In the formula Indicates task via subcarrier Transmission rate: Definition 1: Indicates grouping Tasks in Leave and join a group. ,in ,and ; Definition 2: Indicates task Grouping and The degree of preference, such as grouping Tasks in Preference grouping It can be represented as: ; Definition 3: If grouping Tasks in Grouping Tasks in If Definition 2 is satisfied, a swap will occur, and the grouping will be updated as follows: ; When GUE performs uplink transmission, it initializes the random allocation of subcarriers. After entering the iteration of cooperative game, it randomly selects a subcarrier that is different from the previous subcarrier to perform the grouping operation defined above until each group reaches stability, thereby achieving the highest efficiency of the NOMA system. The mathematical model for allocating computing resources for edge servers can be derived from equation (9): （9） (6) Model the unloading decision problem as a game. ,in Represents the GUE set, This represents the selection strategy space for each GUE. The QoE function represents the state of a game. There exists a function A game is called a potential game if the following conditions are met: （10） And proved the game theory The potential function in is: （11）。

Images