A game-theory-based edge computing framework for tactical self-organizing networks

Abstract

The present invention discloses a tactical self-organizing network edge computing framework based on game theory, including an air-ground collaborative network, which is interconnected with several edge nodes; the air-ground collaborative network includes at least one interconnected low-orbit satellite, several unmanned aerial vehicles, and a ground mobile edge computing server platform; the air-ground collaborative network initiates computing service requests from edge nodes to the ground mobile edge computing server platform, unmanned aerial vehicles, and low-orbit satellites according to different mission requirements and propagation environments, and the air-ground collaborative network offloads tasks to the optimal service provider for processing. The present invention discloses a tactical self-organizing network edge computing framework based on game theory, which improves the computing power of self-organizing networks in tactical environments through the network edge computing framework.

Description

Technical Field

[0001] The present invention relates to the field of communications, specifically to the field of network communications technology, and in particular to a tactical self-organizing network edge computing framework based on game theory. Background Art

[0002] In recent years, with the continuous maturity and development of network communication technologies, a variety of intelligent mobile terminal platforms have been widely used as edge nodes on the battlefield. Unmanned combat platforms, in particular, can replace soldiers in completing tedious and dangerous tasks such as data collection, surveillance, and intelligence reconnaissance. These mobile terminal nodes are interconnected via self-organizing networks to exchange information. Simultaneously, with advances in sensing technology, the amount of data collected by sensors on tactical edge platforms has skyrocketed, leading to the widespread adoption of more compute-intensive applications at the tactical edge. For example, AI-based image processing, target recognition, and terrain modeling are being used.

[0003] Therefore, tactical self-organizing networks require not only reliable link connectivity but also powerful computing capabilities. However, the data processing capabilities and energy consumption of tactical edge terminal platforms are limited. Relying solely on the limited computing power of edge nodes cannot meet the growing demand for edge computing. Mobile Edge Computing (MEC) networks are one of the emerging technologies in 5G systems and are considered a promising solution for reducing computing latency and energy consumption. MEC technology alleviates the computing pressure of mobile networks by deploying MEC servers near edge nodes.

[0004] Defects in the existing technology:

[0005] Existing research on MEC servers (Mobile Edge Computing servers) primarily focuses on friendly environments such as cities. In urban settings, users can offload some of their heavy computing tasks to more resource-rich mobile computing servers, overcoming the limited computing power of edge users and significantly improving the user experience. However, deploying MEC servers in harsh tactical environments presents many unresolved challenges. First, tactical environments lack fixed ground communication infrastructure, necessitating the use of tactical mobile ad hoc networks to provide network services to edge nodes. Due to node mobility and harsh terrain conditions, channels in ad hoc networks exhibit disconnection, intermittent connectivity, and low bandwidth. This negatively impacts data offloading. Second, offloading links are subject to persistent hostile electromagnetic interference. Finally, server computing resources are extremely limited in tactical environments, making the proper allocation of computing resources within the network crucial. Summary of the Invention

[0006] The present invention overcomes the shortcomings of the existing technology and provides a tactical self-organizing network edge computing framework based on game theory, which improves the computing power of the self-organizing network in a tactical environment through the network edge computing framework.

[0007] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is: a tactical self-organizing network edge computing framework based on game theory, including an air-ground-space collaborative network, which is interconnected with several edge nodes; the air-ground-space collaborative network includes at least one interconnected low-orbit satellite mobile edge computing server, several unmanned aerial vehicle mobile edge computing servers, and a ground mobile edge computing server platform; the ground mobile edge computing server platform includes several ground mobile edge computing servers; an interconnected transmission model, a computing model, an optimization target model, and a hierarchical Steinberg game model are established between the air-ground-space collaborative network and the edge nodes; the transmission model also includes a ground end-to-end link model, an air-ground link model, and a satellite-ground link model; the computing model also includes a local computing mode, a ground end-to-end link offloading mode, an air-ground link offloading mode, and a satellite-ground link offloading mode; the air-ground-space collaborative network initiates computing service requests from edge nodes to ground mobile edge computing server platforms, unmanned aerial vehicles, and low-orbit satellites according to different task requirements and propagation environments, and offloads at least part of the tasks to preferred service providers for processing through the air-ground-space collaborative network, and the service providers include one or more of low-orbit satellite mobile edge computing servers, several unmanned aerial vehicle mobile edge computing servers, and ground mobile edge computing server platforms.

[0008] Furthermore, it is defined that the low-orbit satellite mobile edge computing server l, U drone mobile edge computing servers and M ground mobile edge computing servers provide services for N computing-intensive edge nodes; N = {EN n |n=1,2,...,N},M={GS m |m=1,2,...,M} and U={UAV u |u=1,2,...,U} denotes the set of edge nodes, ground mobile edge computing servers, and UAV mobile edge computing servers respectively;

[0009] For edge nodes EN n The computational task characteristics can be expressed as where c n and x n They represent the number of CPU calculation cycles and the size of the calculation task required for the unit bit calculation task respectively; Indicates EN n The maximum mission completion time is defined as the UAV flying at a fixed height h and a constant speed v;

[0010] In the ground end-to-end link model, the path state function z is usedk (t) is used to describe the impact of scattering environment changes on the path, the ground mobile edge computing server MS m and edge nodes EN n The ground end-to-end channel expression between is:

[0011]

[0012] Among them, K and f c are the total number of multipaths and carrier frequency respectively; α k (t), f D,k and τ k Represent the kth path amplitude, Doppler shift and delay respectively; path amplitude α k (t) Communication distance d between the ground mobile edge computing server and the edge node n,m (t), carrier frequency f c , scattering environment and the transmit and receive antenna gains of both communicating parties Related, among them, Indicates the main beam direction of the antenna;

[0013] Edge Node EN n Transmitted to the ground mobile edge computing server through the ground end-to-end link, the link offload rate can be expressed as: Among them, B g is the bandwidth of the terrestrial end-to-end link, P n,m is the transmit power of the terrestrial end-to-end channel, n0 is the Gaussian white noise power density; I n,m (t) represents the interference experienced by the terrestrial end-to-end link, which can be expressed as: Among them, P J |g J (t)| 2 and represent the hostile interference and the co-channel interference of other ground end-to-end links; P J and g J (t) represents the jammer power and interference link channel gain; Q represents the total number of co-channel interference links; P q,m and h q,m (t) represent the power and channel gain of the qth co-frequency terrestrial end-to-end interference link, respectively.

[0014] Furthermore, in the air-ground link model, when the edge node EN n When computing services are requested from a drone, data is offloaded to the drone via an air-ground link; the air-ground link communication is modeled as a line-of-sight link;

[0015] The coordinates of the open space are {x u (t),y u (t),h}, edge node ENn The coordinates are {x n (t),y n (t),0}, edge node EN n and UAV u The channel gain between can be expressed as:

[0016]

[0017] Among them, β0 is the channel gain coefficient when the reference distance is one meter. The UAV adopts the access method of frequency division multiple access, and each UAV is allocated an available bandwidth B for the ground link. u , the unloading rate of the air-ground link can be expressed as: Among them, P n,u is the transmission power of the air-ground link, n u,0 It's a UAV u The noise power spectral density at .

[0018] Furthermore, in the satellite-to-ground link model, when the edge node EN n It tends to offload its computing tasks to the satellite server through the satellite-to-ground link for calculation. The offloading rate of its uplink can be expressed as:

[0019] Among them, B l ,P n,l and n l,0 Represent the bandwidth, transmission power and noise power spectrum density of the satellite-to-ground link respectively; g n,l (t) indicates EN n The satellite-to-ground link channel gain to satellite l is modeled using the shadow-Rician model, where Y n,l =|g n,l | 2 The probability density function is written as: Among them, 1F1(m n,l ;1;δ n,l x) is the confluent hypergeometric function, β n，l =1 / 2b n，l , δ n，l =Ω n，l / 2b n，l (2b n， l m n，l +Ω n，l );Ω n，l is the average power of the line-of-sight link component, 2b n，l is the average power of the multipath component, m n，l ∈(-∞, +∞) is the model parameter of the channel model.

[0020] Furthermore, in the local computing mode, edge nodes use their own limited computing resources to independently process tasks; n,0 The proportion of tasks is performed by edge nodes EN n Calculate by yourself, for a given computing power f n , at the edge node EN n The computational delay and energy consumption at can be expressed as: Among them, ξ n is the energy consumption coefficient; the local computing benefit is expressed as the weighted value of latency and energy: Here, ω1 is the set value of the importance of the delay indicator to the system performance, and ω2 is the set value of the importance of the energy consumption indicator to the system performance.

[0021] Furthermore, in the ground end-to-end link offloading mode, the ground end-to-end offloading mode is to offload the task data to the ground mobile edge computing server with stronger computing power through the ground end-to-end link through the edge node;

[0022] When the edge node EN n Choose to offload some tasks to the ground mobile edge computing server GS m Calculation, transmission delay and calculation delay are: Among them, ρ n,m is the task offloading ratio; f m For GS m Computing power; I m To offload tasks to GS m edge node set; in the ground end-to-end link offloading mode, the edge node EN n The energy consumption at is mainly due to task offloading, and its expression is: The edge node requests computing services from the server and also needs to pay for purchasing resources. The cost is: in, For GS m The price of 1Mbit computing resources; based on latency, energy consumption and resource cost, the edge node EN n The benefits of ground end-to-end link offloading mode are: Among them, w3 represents the impact of resource cost on node efficiency.

[0023] Furthermore, the air-ground link offloading mode is to transmit the task to the UAV carrying the micro server through the air-ground link by the edge node for processing; in the air-ground link offloading mode, the edge node EN n Let the ratio be ρ n,u The task is offloaded to the UAV mobile edge computing server through the air-ground link for calculation; nThe transmission delay and computation delay in the air-ground link offloading mode are: Among them, f u represents the computing power of the UAV mobile edge computing server, I u Indicates unloading to UAV u The energy loss in the air-ground link unloading mode can be written as: The edge node charge for the service request of the u-th drone is: in, is the price of 1Mbit computing resource of UAV; in air-ground link offloading mode, EN n The computation offloading benefit is:

[0024]

[0025] Furthermore, in the satellite-to-ground link offloading mode, the satellite-to-ground offloading mode is to transmit the task to the server's low-orbit satellite cloud server through the satellite-to-ground link through the edge node; the uplink delay and energy consumption are respectively: Among them, ρ n,l Indicates the offloading ratio in the satellite-to-ground link offloading mode, d n,l and c are EN n The distance to the low-orbit satellite l and the speed of light; f l represents the computing power of satellite l, I l A collection of nodes that offload tasks to low-orbit satellites; and They represent transmission delay, propagation delay, and computation delay respectively. The charges for service requests from low-orbit satellites are: in, The price of 1Mbit computing resources of the low-orbit satellite mobile edge computing server; in the satellite-to-ground link offloading mode, EN n The computation offloading benefit is: For edge nodes EN n , its offloading benefit value can be expressed as the sum of the benefits of all offloading modes:

[0026] Furthermore, in the optimization target model, the offloading link selection and task offloading ratio are jointly optimized to maximize the offloading benefit. The optimization problem is expressed as:

[0027] P1:

[0028] stC1:

[0029] C2:

[0030] C3:

[0031] C4:

[0032] C5:

[0033] C6:

[0034] Where st represents the constraint condition; a={a1,a2,...,a N} represents the action combination of all edge nodes; a n =(k n , ρ n，s ) represents the action of each edge node, where k n Represents the offloading link selection, and its action space is K n ={0}∪M∪U∪{l}, 0 means the task is computed locally; ρ n，s ∈[0, 1] represents the task offloading ratio; C1-C3 respectively indicate that the total computing offloading request of the edge node cannot exceed the maximum computing capacity of the low-orbit satellite mobile edge computing server, the drone mobile edge computing server, and the ground mobile edge computing server; C4 indicates that the satellite delay cannot exceed its coverage time; C5 ensures that the task delay cannot exceed the maximum delay constraint; C6 represents the task offloading relationship of the edge node.

[0035] Furthermore, in the interactive relationship modeling between the mobile edge computing server and the edge node in the hierarchical Steinberg game model, the mobile edge computing server is the leader and the edge node is the follower; the mobile edge computing server determines the charging price of computing resources, and the edge node selects the offloading link and task allocation ratio according to the given price to reduce delay, energy consumption and resource overhead; wherein, the mobile edge computing server includes one or more of a low-orbit satellite mobile edge computing server, a drone mobile edge computing server, and a ground mobile edge computing server; a hierarchical distributed iterative algorithm is adopted in the hierarchical Steinberg game model; a hierarchical distributed iterative optimization algorithm is used to find the Steinberg game equilibrium point of the ground mobile edge computing server and the edge node. At the Steinberg game equilibrium point, both parties in the game cannot act unilaterally to obtain higher benefits. At this time, the system reaches a stable state and the total utility of the system is maximized; in the hierarchical Steinberg game model, the service provider and the edge node are optimized separately.

[0036] The present invention solves the defects existing in the technical background, and the beneficial technical effects of the present invention are:

[0037] The present invention provides a tactical self-organizing network edge computing framework based on game theory, which improves the computing power of the self-organizing network in a tactical environment through the network edge computing framework.

[0038] 1. This paper proposes a tactical self-organizing network edge computing framework assisted by low-orbit satellites and drones. By deploying low-orbit satellites and drones, the negative impact of poor terrestrial channels on data offload is overcome, improving the service quality and user experience of edge nodes. This framework enhances the reliability and robustness of task offload, alleviating the problem of insufficient computing power and resources in tactical communication environments.

[0039] 2. This paper proposes a multi-leader and multi-follower Steinberg game to model the complex interactions between mobile edge computing servers and edge mobile devices. As the leader, the mobile edge computing server sets the price for computing resources. Based on the given price, the edge node selects the appropriate offload link and task offload ratio.

[0040] 3. To achieve Steinberg equilibrium, a hierarchical distributed iterative algorithm is used to improve the utility of mobile edge computing servers and edge nodes. At the Steinberg equilibrium, the optimal resource price, offloading link selection, and task offloading ratio in the air-ground collaborative system are obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below with reference to the accompanying drawings and examples.

[0042] Figure 1 This is a model diagram of the air-space-ground-space system in an embodiment of the present invention;

[0043] Figure 2 This is a flowchart of a hierarchical distributed algorithm according to an embodiment of the present invention;

[0044] Figure 3 is the convergence of the upper leader subgame in the embodiment of the present invention;

[0045] Figure 4 is the convergence of the lower follower subgame in the embodiment of the present invention;

[0046] Figure 5 2 is a schematic diagram comparing the performance of the game scheme in the embodiment of the present invention and the traditional scheme. DETAILED DESCRIPTION

[0047] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. These drawings are simplified schematic diagrams that only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0048] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, bottom, top, etc.), the directional indications are only used to explain the relative positional relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. The terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. Unless otherwise clearly specified and defined, the terms "set", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected, or indirectly connected through an intermediate medium, or it can be a communication between the internal parts of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0049] like Figures 1 to 5 As shown, a tactical self-organizing network edge computing framework based on game theory includes an air-space-ground collaborative network, which is interconnected with several edge nodes; the air-space-ground collaborative network includes at least one interconnected low-orbit satellite, several drones, and a ground mobile edge computing server platform; the ground mobile edge computing server platform includes several ground mobile edge computing servers; an interconnected transmission model, computing model, optimization target model, and hierarchical Steinberg game model are established between the air-space-ground collaborative network and the edge nodes; the air-space-ground collaborative network initiates computing service requests to the ground mobile edge computing server platform, drones, and low-orbit satellites through edge nodes according to different task requirements and propagation environments, and offloads at least part of the tasks to the optimal service provider through the air-space-ground collaborative network for processing, and the service provider includes one or more of the low-orbit satellite mobile edge computing server, several drone mobile edge computing servers, and the ground mobile edge computing server platform.

[0050] like Figure 1 As shown in Figure 1, the proposed space-ground collaborative network includes a low-orbit satellite mobile edge computing server l, U drone mobile edge computing servers, and M ground mobile edge computing servers to provide services for N computing-intensive edge nodes. n |n=1,2,...,N},M={GS m |m=1,2,...,M} and U={UAV u |u=1,2,...,U} respectively represent the set of edge nodes, ground mobile edge computing servers and drone mobile edge computing servers. n , its computing task characteristics can be expressed as {c n,x n}, where c n and x n The numbers of CPU cycles and task size required for a unit bit computation task are respectively represented. Ground edge nodes have very limited computing power, so the data they generate needs to be partially offloaded to MEC servers with greater computing power for auxiliary computation. A drone is defined as flying at a fixed altitude h and a constant speed v. Drones and low-orbit satellites can cover the entire ground network and provide resource access services to the ground network. Low-orbit satellites, drones, and ground mobile edge computing servers constitute the mobile service provider set S = {M∪U∪N}. Because the tasks under consideration are homogeneous and divisible, a complete task can be transferred simultaneously to multiple servers for parallel processing to improve computational efficiency.

[0051] It is worth noting that when multiple nodes request computing services from a server at the same time, the server will evenly distribute its computing resources.

[0052] The transmission model also includes the ground end-to-end link model, the air-ground link model, and the satellite-ground link model.

[0053] Ground end-to-end link model:

[0054] Using path state function z k (t)∈{0,1} is used to describe the impact of scattering environment changes on the path, where 0 means the kth link disappears and 1 means the kth link exists. m and edge nodes EN n The ground end-to-end channel gain expression between is: Where t represents time, and the link status changes over time; K and f c are the total number of multipaths and carrier frequency respectively; α k (t), f D,k and τ k Represent the kth path amplitude, Doppler shift and delay respectively; path amplitude α k (t) Communication distance d between the server and edge nodes n,m (t), carrier frequency f c , the transmit and receive antenna gains of both communicating parties in, Indicates the direction of the antenna main beam. Due to the shortage of bandwidth resources, ground links share spectrum resources. Therefore, different edge nodes may choose the same channel for transmission, causing co-channel interference; edge node EN n Transmitted to the ground mobile edge computing server through the ground end-to-end link, the link offload rate can be expressed as: Among them, Bg is the bandwidth of the terrestrial end-to-end link, P n,m is the transmit power of the terrestrial end-to-end channel, n0 is the Gaussian white noise power density; H n,m (t) is the changing link condition, Indicates the link status between the ground mobile edge computing server and the edge node; I n,m (t) represents the interference experienced by the terrestrial end-to-end link, which can be expressed as: Among them, P J |g J (t)| 2 and represent the hostile interference and the co-channel interference of other ground end-to-end links; P J and g J (t) represents the jammer power and interference link channel gain; Q represents the total number of co-channel interference links; P q,m and h q,m (t) represent the power and channel gain of the qth co-frequency terrestrial end-to-end interference link, respectively.

[0055] Air-ground link model:

[0056] When the edge node EN n When computing services are requested from a UAV, data is offloaded to the UAV via an air-ground link; the air-ground link communication is modeled as a line-of-sight link; the UAV u The coordinates are {x u (t),y u (t),h}, edge node EN n The coordinates are {x n (t),y n (t),0}, edge node EN n With UAV u The channel gain between can be expressed as: Among them, β0 is the channel gain coefficient when the reference distance is one meter. The UAV adopts the access method of frequency division multiple access, and each UAV is allocated an available bandwidth B for the ground link. u , the unloading rate of the air-ground link can be expressed as Among them, B u represents the bandwidth of the air-ground link, P n,u is the transmission power of the air-ground link, n u,0 It's a UAV u The noise power spectral density at .

[0057] Satellite-to-ground link model:

[0058] When the edge node EN nIt tends to offload its computing tasks to the satellite server through the satellite-to-ground link for calculation. The offloading rate of its uplink can be expressed as: Among them, B l , P n，l and n l，0 Represent the bandwidth, transmission power and noise power spectrum density of the satellite-to-ground link respectively; g n，l (t) represents the satellite-to-ground link channel gain from ENn to satellite l. The shadow-Rician model is used to model the satellite-to-ground link, where Y n，l =|g n，l | 2 The probability density function is written as: Among them, 1F1(m n，l ;1;δ n，l x) is the confluent hypergeometric function, β n，l =1 / 2b n，l , δ n，l =Ω n，l / 2b n，l (2b n，l m n，l +Ω n，l );Ω n，l is the average power of the line-of-sight link component (LOS component), 2b n，l is the average power of the multipath component, m n，l ∈(-∞, +∞) is the model parameter of the channel model, and the channel model adopts the Nakagami channel model.

[0059] The computing model also includes local computing mode, ground end-to-end link offloading mode, air-ground link offloading mode, and satellite-ground link offloading mode.

[0060] Local computing mode:

[0061] ρ n，0 The proportion of tasks is performed by edge nodes EN n Calculate by yourself, for a given computing power f n , at the edge node EN n The computational delay and energy consumption at can be expressed as: Among them, ξ n is the energy consumption coefficient; the local computing benefit is expressed as the weighted value of latency and energy: Here, ω1 is the set value of the importance of the delay indicator to the system performance, and ω2 is the set value of the importance of the energy consumption indicator to the system performance.

[0062] Ground end-to-end link offloading mode:

[0063] When the edge node EN nChoose to offload some tasks to the ground mobile edge computing server GS m Calculation, transmission delay and calculation delay are: Among them, ρ n，m The task offloading ratio. m For GS m Computing power. m To offload tasks to GS m edge node set; in the ground end-to-end link offloading mode, the edge node EN n The energy consumption at is mainly due to task offloading, and its expression is: The edge node requests computing services from the server and also needs to pay for purchasing resources. The cost is: in, For GS m The price of 1Mbit computing resources; based on latency, energy consumption and resource cost, the edge node EN n The benefits of ground end-to-end link offloading mode are: Among them, w3 represents the impact of resource cost on node efficiency.

[0064] Air-ground link offloading mode:

[0065] In the air-ground link offloading mode, the edge node EN n Let the ratio be ρ n,u The task is offloaded to the UAV mobile edge computing server through the air-ground link for calculation; n The transmission delay and computation delay in the air-ground link offloading mode are: Among them, f u represents the computing power of the UAV mobile edge computing server, I u Indicates unloading to UAV u The energy loss in the air-ground link unloading mode can be written as: The edge node charge for the service request of the u-th drone is: in, is the price of 1Mbit computing resource of UAV; in air-ground link offloading mode, EN n The computation offloading benefit is:

[0066] Satellite-to-ground link offloading mode:

[0067] The uplink delay and energy consumption are: Among them, ρ n,l Indicates the offloading ratio in the satellite-to-ground link offloading mode, d n,l and c are EN nThe distance to the low-orbit satellite l and the speed of light; f l represents the computing power of satellite l, I l A collection of nodes that offload tasks to low-orbit satellites; and They represent transmission delay, propagation delay, and computation delay respectively. The charges for service requests from low-orbit satellites are: in, The price of 1Mbit computing resources of the low-orbit satellite mobile edge computing server; in the satellite-to-ground link offloading mode, EN n The computation offloading benefit is: For edge nodes EN n , its offloading benefit value can be expressed as the sum of the benefits of all offloading modes:

[0068] Optimization target model:

[0069] The offloading link selection and task offloading ratio are jointly optimized to maximize the offloading benefit. The optimization problem is expressed as:

[0070] P1:

[0071] stC1:

[0072] C2:

[0073] C3:

[0074] C4:

[0075] C5:

[0076] C6:

[0077] Among them, st represents the constraint condition; a={a1,a2,....,a N} represents the action combination of all edge nodes; a n =(k n ,ρ n,s ) represents the action of each edge node, where k n Represents the offloading link selection, and its action space is K n ={0}∪M∪U∪{l}, 0 means the task is computed locally; ρ n,s∈[0,1] represents the task offloading ratio; C1-C3 respectively indicate that the total computing offloading request of the edge node cannot exceed the maximum computing capacity of the low-orbit satellite mobile edge computing server, the drone mobile edge computing server, and the ground mobile edge computing server; C4 indicates that the satellite delay cannot exceed its coverage time; C5 ensures that the task delay cannot exceed the maximum delay constraint; C6 represents the task offloading relationship of the edge node.

[0078] Hierarchical Steinberg game model:

[0079] In the hierarchical Steinberg game model, the complex interactive relationship between MEC servers and edge nodes is modeled. In this model, the ground mobile edge computing server is the leader, and the edge nodes are followers. The ground mobile edge computing server determines the price of computing resources, and the edge nodes select offload links and task allocation ratios based on the given price to reduce latency, energy consumption, and resource overhead.

[0080] The proposed Steinberg game model is defined as: S = {N, S, A N ,P S ,u N ,u S}; where N and S represent the set of edge nodes and MEC service providers respectively; A N and P S Represents the action combination of edge node and ground mobile edge computing server respectively; u N and u S They represent the game benefits of edge nodes and MEC service providers respectively.

[0081] For each MS s ∈S, the action is the pricing of the service provider's own resources 0<p s <p max , p max is the maximum price that can be selected; the action space of the service provider is R = {p s,1 ,p s,2 ,...,p s,R}, where R is the number of optional prices; P S =p={p1,p2,...,p S} is the mixed strategy of the service side. For each EN n ∈N, the action is a n =(ρ n,s ,k n ), where k n Indicates EN n The unloading mode selection, its action space can be expressed as K n ={0}∪M∪U∪{l};ρ n,s Represents the data offloading ratio, and its action space is Tn ={ρ n,s |0≤ρ n,s ≤1,s∈S}. Therefore EN n The action space is in Represents the Cartesian product. A N =a={a1,a2,...,a N} is the action combination of all edge nodes.

[0082] The optimization goal of the ground mobile edge computing server is to select the optimal resource price to maximize its own sales revenue. Its benefit function can be defined as: Among them, p -s =(p1,p2,...,p s-1 ,p s+1 ,...,p S ) represents the resource pricing of the remaining service providers except service provider s, and a is the combination of offloading strategies of all edge nodes in the follower layer.

[0083] For edge nodes, the optimization goal is to minimize their own latency, energy consumption, and resource purchase costs by selecting the best offloading mode and offloading ratio. Inspired by marginal benefits, EN n The benefit function is defined as: Among them, a -n Indicates that except EN n The action combination of other nodes except V, p represents the action combination of the leader layer service party. n (a n ,a -n ,p) indicates EN n Its own benefits. j (a j ,a -j\n ,p) indicates EN n When no action is taken, the jth edge node EN in the network j The second term V j (a j ,a -j ,p)-V j (a j ,a -j\n ,p) indicates EN n When taking action and when not taking action, the edge node EN j The benefit difference; the physical meaning is EN n Impact on other users. Formula EN in n The physical meaning of the benefit function is EN n Impact on the entire network. Further sorting can be obtained: According to game theory, all game participants are rational and selfish, and their optimization goal is to maximize their own benefit function. s Hope to choose appropriate resource pricing to maximize benefits s ,Right now EN n Maximize u by choosing the appropriate computing mode and offloading ratio n ,Right now

[0084] The hierarchical distributed iterative algorithm is used in the hierarchical Steinberg game model:

[0085] A hierarchical distributed iterative optimization algorithm is used to find the Steinberg game equilibrium point between the mobile edge computing server and the edge node. At this point, neither player can unilaterally take action to achieve higher benefits. The system reaches a stable state and maximizes its total utility. In the hierarchical Steinberg game model, the server and edge node are optimized separately. The mobile edge computing server can include one or more of low-orbit satellite mobile edge computing servers, drone mobile edge computing servers, and ground mobile edge computing servers.

[0086] The update of computing resource price and offloading strategy is updated at different time scales. Assume that the ground end-to-end link, air-ground link, and satellite-ground link can be regarded as stationary channels in one cycle, and each cycle contains I max The number of iterations. In each cycle, the server's resource price is updated. In each iteration, each edge node maximizes its benefit by selecting an offloading strategy. After multiple iterations, the follower layer reaches a Nash equilibrium.

[0087] In order to avoid falling into the local optimum, a mixed strategy is used to set the price. Let the price space be R = {p s,1 ,p s,2 ,...,p s,R}, so the initial selection probability of each price is 1 / R. s,r (e) represents the MS in the e-th cycle s Select price p s,r The probability of all prices is Use the Boltzmann principle to update the price: Among them, u s,r (e) indicates MS s Select price p s,r, where η is the learning rate. To ensure algorithm convergence, the exploration probability needs to decrease with the number of cycles. Therefore, the learning parameter should be gradually reduced to shift the algorithm from an exploration state to an exploitation state, as shown in the following formula: η(e) = η0 - eΔη; where η0 and Δη represent the initial learning rate and the adjustment step size. Clearly, the leader will choose a more efficient pricing strategy with a higher probability. To explore the best performance of mobile edge computing servers, the Pareto criterion is used as the price update principle for the selected mobile edge computing servers. According to the Pareto criterion, a server will only update its price if the price update does not harm the utility of any other server. Therefore, when the selected server changes its price, the server's total utility will increase.

[0088] The algorithm flow is as follows Figure 2 As shown in Figure 1, the steps of the hierarchical distributed optimization algorithm include:

[0089] Step 1: Initialization: the service provider's resource pricing strategy and the edge node's offloading strategy, cycle number e = 1

[0090] Step 2: Randomly select a service provider MS s , choose its resource price as p s,r , the prices of other service providers’ resources remain unchanged.

[0091] Step 3: After the resource pricing is determined, the edge node acts as a follower, comprehensively considering the current price, its own resources and the data transmission link situation, and independently selects the optimal offloading mode and offloading ratio. In each iteration, each edge node updates its action in turn according to EN n For example, the action update expression can be written as: where a n '(i) is EN n An action is randomly selected from the action space, a n (i+1) is the updated action. n While updating their operations, other nodes keep their link selection and task offloading ratios unchanged.

[0092] Step 4: After multiple rounds of iterations, the follower-level subgame will converge to the Nash equilibrium point. s Estimated price at this time p s,r Its benefits s,r .

[0093] Step 5: Next, MS s Explore the benefits of other prices. s Select a new resource pricing in turn, such as p s,r',r'∈R, when the follower layer converges to NE, estimate the benefit u under this price s,r' ,r'. According to the Boltzmann exploration strategy in the formula, the price selection probability in the next iteration is calculated.

[0094] Step 6: MS s A new price is chosen based on the updated selection probabilities and the Pareto criterion.

[0095] Finally, after multiple rounds of interaction and iterative updates, the efficiency of the leadership and follower layers gradually improves, and each converges to a stable strategy. At this time, neither party in the game can improve its own utility by unilaterally changing its strategy. Finally, the Steinberg equilibrium is achieved, and the total utility of the system reaches its maximum value.

[0096] Working principle of the present invention:

[0097] like Figure 1-Figure 5 As shown in the figure, edge nodes with computing tasks and ground mobile edge computing servers are randomly distributed in a 1000×1000m 2 The mission area is located in the low-orbit satellite and drone flight altitudes of 1000 km and 100 m, respectively. In the transmission model, the transmission powers of the ground end-to-end link model, the air-ground link model, and the satellite-to-ground link model are 1 W, 0.05 W, and 0.5 W, respectively. The data size of the computational task of each edge node follows a uniform distribution between [5, 30] Mbits. The number of CPU cycles required for each bit task follows a uniform distribution between [100, 200] cycles / bit. The noise power spectral density of the ground end-to-end link model, the air-ground link model, and the satellite-to-ground link model is set to -174 dBm / Hz, -190 dBm / Hz, and -203 dBm / Hz, respectively.

[0098] Figure 3 The convergence of the mobile edge computing server in the leader upper subgame is shown, where the number of edge nodes is set to 5, 8, 10, and 15. The number of ground mobile edge computing servers and drone mobile edge computing servers is set to 4. Figure 3 As can be seen, the total utility of the mobile edge computing server gradually increases and finally converges to a stable point, which demonstrates the convergence of the proposed algorithm. In addition, the total utility increases with the number of edge nodes because more edge nodes generate more data, which needs to be offloaded to the mobile edge computing server, bringing greater resource sales benefits to the service provider.

[0099] Figure 4The convergence of the follower subgame is described. To improve utility, edge nodes adjust their offload link selection and task offload ratio to reduce latency, energy consumption, and resource overhead. As the number of iterations increases, each edge node selects better actions, and their individual utility increases. When the number of iterations exceeds 23, the utility of all edge nodes stabilizes, demonstrating the convergence of the underlying follower subgame.

[0100] Figure 5 The performance of the game scheme is compared with traditional schemes such as full local computing, full ground offloading, and random ratio offloading. Figure 5 It can be seen that the total utility of edge nodes decreases with the increase in the average task data size, because larger data sizes lead to greater delays and energy consumption. In addition, the game scheme outperforms the other three schemes at different average task data sizes. The offloading strategy can dynamically adjust link selection and task offloading ratios, thereby effectively utilizing communication and computing resources across the entire network. The fully local computing scheme relies on limited local computing power to complete tasks, which will result in higher latency and energy consumption. The poor performance of full ground offloading is due to the fact that poor wireless channel conditions lead to greater transmission delays and energy consumption. The random ratio offloading scheme does not optimize offloading link selection and offloading ratios, and therefore its performance is worse than the proposed game scheme.

[0101] The above specific implementation methods are specific support for the scheme ideas proposed in the present invention, and cannot be used to limit the scope of protection of the present invention. Any equivalent changes or equivalent modifications made on the basis of this technical scheme in accordance with the technical ideas proposed in the present invention still fall within the scope of protection of the technical scheme of the present invention.

Claims

1. A tactical self-organizing network edge computing framework based on game theory, characterized by: The invention comprises an air-ground collaborative network, wherein the air-ground collaborative network is interconnected with a plurality of edge nodes; the air-ground collaborative network comprises at least one interconnected low-orbit satellite mobile edge computing server, a plurality of unmanned aerial vehicle mobile edge computing servers, and a ground mobile edge computing server platform; the ground mobile edge computing server platform comprises a plurality of ground mobile edge computing servers; An interconnected transmission model, a calculation model, an optimization target model, and a hierarchical Steinberg game model are established between the air-ground-space collaborative network and the edge nodes; The transmission model also includes a ground end-to-end link model, an air-to-ground link model, and a satellite-to-ground link model; The computing model also includes a local computing mode, a ground end-to-end link offloading mode, an air-ground link offloading mode, and a satellite-ground link offloading mode; The air-ground collaborative network initiates computing service requests from edge nodes to ground mobile edge computing server platforms, drones, and low-orbit satellites based on different task requirements and propagation environments, and offloads at least part of the tasks to preferred service providers through the air-ground collaborative network. The service providers include one or more of the low-orbit satellite mobile edge computing servers, several drone mobile edge computing servers, and ground mobile edge computing server platforms. Defining Low Earth Orbit Satellite Mobile Edge Computing Servers , U drone mobile edge computing servers and M ground mobile edge computing servers provide services for N computing-intensive edge nodes; , and They represent the set of edge nodes, ground mobile edge computing servers, and UAV mobile edge computing servers respectively; For edge nodes The computational task characteristics can be expressed as ,in and They represent the number of CPU calculation cycles and the size of the calculation task required for the unit bit calculation task respectively; express The maximum mission completion time is defined as the UAV flying at a fixed height h and a constant speed v; In the ground end-to-end link model, the path state function is used To describe the impact of scattering environment changes on the path, the ground mobile edge computing server and edge nodes The ground end-to-end channel expression between is: ； in, and are the total number of multipaths and carrier frequency respectively; , and Respectively represent path amplitude, Doppler shift and delay; path amplitude Communication distance between ground mobile edge computing servers and edge nodes , carrier frequency , scattering environment and the transmit and receive antenna gains of both communicating parties Related, among them, Indicates the main beam direction of the antenna; edge nodes Transmitted to the ground mobile edge computing server through the ground end-to-end link, the link offload rate can be expressed as: ;in, is the bandwidth of the terrestrial end-to-end link, is the transmit power of the terrestrial end-to-end channel, is the Gaussian white noise power density; represents the interference to the terrestrial end-to-end link, which can be expressed as: ;in, and They represent the hostile interference and the co-channel interference of other ground end-to-end links respectively; and represents the jammer power and interference link channel gain; Q represents the total number of co-channel interference links; and represent the power and channel gain of the qth co-frequency terrestrial end-to-end interference link, respectively.

2. The game theory-based tactical self-organizing network edge computing framework according to claim 1, characterized in that: In the air-ground link model, when the edge node When computing services are requested from a drone, data is offloaded to the drone via an air-ground link; the air-ground link communication is modeled as a line-of-sight link; The coordinates of the open space are , edge nodes The coordinates are , edge nodes With drones The channel gain between can be expressed as: ; in, The reference distance is one meter, and the UAV uses frequency division multiple access. Each UAV is allocated an available bandwidth for the ground link. , the unloading rate of the air-ground link can be expressed as: ;in, is the transmission power of the air-ground link, yes The noise power spectral density at .

3. The game theory-based tactical self-organizing network edge computing framework according to claim 2, characterized in that: In the satellite-to-ground link model, when the edge node It tends to offload its computing tasks to the satellite server through the satellite-to-ground link for calculation. The offloading rate of its uplink can be expressed as: ； in, , and Represent the bandwidth, transmission power and noise power spectral density of the satellite-to-ground link respectively; express to satellite The satellite-to-ground link channel gain is , and the shadow-Rician model is used to model the satellite-to-ground link, where The probability density function is written as: ;in, is the confluent hypergeometric function, , , ; is the average power of the line-of-sight link component, is the average power of the multipath component, are the model parameters of the channel model.

4. The game theory-based tactical self-organizing network edge computing framework according to claim 3, characterized in that: In the local computing mode, edge nodes use their own limited computing resources to independently process tasks; Proportion of tasks performed by edge nodes Self-calculation, for a given computing power , at the edge node The computational delay and energy consumption at can be expressed as: ;in, is the energy consumption coefficient; the local computing benefit is expressed as the weighted value of latency and energy: ;in, It is the set value of the importance of the delay indicator to the system performance. It is the set value of the importance of energy consumption index to system performance.

5. The game theory-based tactical self-organizing network edge computing framework according to claim 4, characterized in that: In the ground end-to-end link offloading mode, the ground end-to-end offloading mode is to offload the task data to the ground mobile edge computing server with stronger computing power through the ground end-to-end link through the edge node for calculation; When the edge node Choose to offload some tasks to the ground mobile edge computing server Calculation, transmission delay and calculation delay are: ;in, The task offloading ratio; for computing power; To offload tasks to edge node set; in the ground end-to-end link offloading mode, the edge node The energy consumption at is mainly due to task offloading, and its expression is: ； The edge node requests computing services from the server and also needs to pay for purchasing resources. The cost is: ;in, for The price of 1Mbit computing resources; based on latency, energy consumption and resource cost, edge nodes The benefits of ground end-to-end link offloading mode are: ;in, Indicates the impact of resource cost on node efficiency.

6. The game theory-based tactical self-organizing network edge computing framework according to claim 5, characterized in that: In the air-ground link offloading mode, the air-ground offloading mode is to transmit the task to the drone carrying the micro server through the air-ground link by the edge node for processing; In the air-ground link offloading mode, the edge node The ratio is The task is offloaded to the UAV mobile edge computing server through the air-ground link for calculation; The transmission delay and computation delay in the air-ground link offloading mode are: ;in, Indicates the computing power of the UAV mobile edge computing server, Indicates unloading to the drone The set of edge nodes; The energy loss in the air-ground link unloading mode can be written as: ; No. The edge node charges for service requests from drones are: ; in, The price of 1Mbit of computing resources for a drone; In air-ground link offloading mode, The computation offloading benefit is: 。 7. The game theory-based tactical self-organizing network edge computing framework according to claim 6, characterized in that: In the satellite-to-ground link offloading mode, the satellite-to-ground offloading mode is a low-orbit satellite cloud server processing in which the task is transmitted to the server via the satellite-to-ground link by the edge node; The uplink delay and energy consumption are: ; ;in, Indicates the offloading ratio in the satellite-to-ground link offloading mode. and c are To low-orbit satellite distance and the speed of light; Indicates satellite computing power, A collection of nodes that offload tasks to low-orbit satellites; , and They represent transmission delay, propagation delay, and computation delay respectively; The charges for low-orbit satellite service requests are: ;in, The price of 1Mbit computing resources for a low-orbit satellite mobile edge computing server; In satellite-to-ground link offloading mode, The computation offloading benefit is: ; For edge nodes , its offloading benefit value can be expressed as the sum of the benefits of all offloading modes: .

8. The game theory-based tactical self-organizing network edge computing framework according to claim 7, characterized in that: In the optimization objective model, the offloading link selection and task offloading ratio are jointly optimized to maximize the offloading benefit. The optimization problem is expressed as: ; ; ; ; ; ; ; Among them, st represents the constraint condition; Represents the action combination of all edge nodes; represents the action of each edge node, where represents the offloading link selection, and its action space is ,0 means the task is calculated locally; Indicates the task offloading ratio; The total computing offload request of the edge node cannot exceed the maximum computing capacity of the low-orbit satellite mobile edge computing server, the UAV mobile edge computing server, and the ground mobile edge computing server respectively; It means that the satellite’s delay cannot exceed its coverage time; It ensures that the task delay cannot exceed the maximum delay constraint; Indicates the edge node task offloading relationship.

9. The game theory-based tactical self-organizing network edge computing framework according to claim 1, characterized in that: In the hierarchical Steinberg game model, the mobile edge computing server is the leader and the edge node is the follower in modeling the interactive relationship between the mobile edge computing server and the edge node; The mobile edge computing server determines the charging price of computing resources, and the edge node selects the offload link and task allocation ratio based on the given price to reduce latency, energy consumption and resource overhead. The mobile edge computing server includes one or more of low-orbit satellite mobile edge computing servers, drone mobile edge computing servers, and ground mobile edge computing servers. A hierarchical distributed iterative algorithm is adopted in the hierarchical Steinberg game model; the Steinberg game equilibrium point of the ground mobile edge computing server and the edge node is found through the hierarchical distributed iterative optimization algorithm. At the Steinberg game equilibrium point, both parties in the game cannot act unilaterally to obtain higher benefits. At this time, the system reaches a stable state and the total utility of the system is maximized; in the hierarchical Steinberg game model, the service provider and the edge node are optimized separately.

Images