Task unloading method for D2D-LEO full-dimensional collaborative edge computing network

Abstract

A task unloading method for a D2D-LEO full-dimensional collaborative edge computing network relates to the technical field of computer algorithms and comprises the following steps: 1, acquiring position information and task information of ground users and computing power information of edge nodes through a GEO satellite, and constructing a digital twin model; s2, a D2D-LEO full-dimensional cooperative computing unloading framework is constructed; s3, defining an action space, a state space and a reward function, and scheduling a calculation task and an execution mode of the ground user through a DSDDPG algorithm with battery power perception; and S4, in the digital twin model, defining a corresponding data deviation value to predict risks encountered by the system, and guiding the model to make an unloading decision. The terminals can perform task cooperative calculation through D2D horizontal cooperation, vertical cooperation between the terminals and the satellites, horizontal cooperation between the satellites and vertical cooperation between the satellites and the cloud, so that the calculation unloading efficiency is improved; and moreover, the digital twinning energization is realized, so that the stability of the system is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of computer algorithm technology, and in particular to a task offloading method for a D2D-LEO full-dimensional collaborative edge computing network. Background Art

[0002] With the launch of 6G technology and the growing popularity of artificial intelligence, edge computing has been rapidly adopted in various industries. However, some remote or disaster-stricken areas still lack sufficient ground-based communication infrastructure, making it difficult to achieve reliable network connectivity. Satellite communication technology can overcome geographical limitations and provide global coverage, especially in areas where ground networks cannot effectively reach. In this context, satellite edge computing (SEC) has become a key technology to meet the growing global demand for distributed computing. SEC shifts computing tasks to different layers of satellite nodes in the network. This optimizes bandwidth utilization, improves computing efficiency, reduces latency, and supports real-time data processing and resource allocation. Today, SEC is widely used in disaster recovery, navigation, and intelligent transportation systems.

[0003] Compared with terrestrial networks, satellite communication systems are mainly composed of geostationary earth orbit (GEO), medium earth orbit (MEO), and low earth orbit (LEO) satellites. Although GEO satellites provide wide coverage, they are far from the ground, which will result in higher propagation delays. MEO satellites are at a more moderate distance and are more suitable for task offloading. However, the high cost of MEO satellites makes them less suitable for large-scale deployment. In contrast, LEO satellites are cheaper and more suitable for large-scale deployment. In addition, since LEO satellites are closer to the earth, they can provide lower latency services. However, LEO satellites have the disadvantages of high mobility and small geographical coverage area, which will result in unstable connection between users and satellites. And the resources of a single LEO satellite are limited and cannot meet the needs of too many users at the same time. Therefore, multi-layer satellite edge computing architecture has become the focus of research.

[0004] Existing research mainly focuses on vertical task coordination mechanisms, while ignoring horizontal task coordination strategies. In addition, when the task data is small, frequent task offloading to satellite devices may lead to higher wireless communication overhead, thereby increasing latency and reducing energy efficiency. Device-to-device (D2D) communication allows nearby devices to transmit data directly without going through a base station, which makes it very suitable for small data packet exchanges in extreme or remote areas. Therefore, combining D2D with satellite communications to achieve high-performance task offloading has become a focus of attention.

[0005] In addition, due to the high cost of satellite construction and launch, any failure may lead to significant economic losses. Therefore, risk prediction and fault prevention of satellite systems are crucial. In a complex space environment, factors such as solar wind, cosmic radiation, and magnetic field fluctuations can affect the stability of satellites, which may lead to communication interruptions, mission failures, and even hardware damage. In order to effectively address these challenges, the introduction of digital twin technology is crucial. Digital twin technology creates a virtual satellite model that allows real-time monitoring and prediction of its performance in complex environments. By comparing with real-time data from actual satellites, digital twins can identify potential risks in advance and provide strong support for decision-making. It is able to take preventive measures before problems occur, optimize satellite resource allocation, and improve mission scheduling.

[0006] A Chinese patent (publication number: CN117555306A) discloses a method and system for adaptive scheduling of multi-production line tasks based on digital twins. In a digital twin environment, the patent performs offline pre-training of a task scheduling model based on a reinforcement learning algorithm, and then fine-tunes the parameters of the offline pre-trained task scheduling model in a real factory environment to obtain a trained task scheduling model, which can schedule the disjunctive graph features constructed according to multi-production line task orders, and determine the processing machines for each process of the multi-production line task orders and the order in which each processing machine processes the processes. However, the patent does not take into account the errors in the digital twin model and actual operations caused by errors and delays in the transmission of network status information. Summary of the invention

[0007] In order to overcome the shortcomings in the background technology, the present invention discloses a digital twin task offloading solution in a D2D-LEO full-dimensional collaborative edge computing network.

[0008] In order to achieve the above-mentioned invention object, the present invention adopts the following technical scheme:

[0009] A method for offloading tasks in a D2D-LEO full-dimensional collaborative edge computing network comprises the following steps:

[0010] S1. Obtain the location information, mission information, and computing power information of ground users through GEO satellites to build a digital twin model;

[0011] S2. Construct a D2D-LEO full-dimensional collaborative computing offloading framework, including one or more of local computing, D2D offloading, D2L offloading, L2L offloading, and L2G offloading;

[0012] S3, define the action space, state space and reward function, and schedule the computing tasks and execution modes of ground users through the DSDDPG algorithm with battery power awareness;

[0013] S4. In the digital twin model, define corresponding data deviation values ​​to predict the risks that the system will encounter, thereby guiding the model to make correct unloading decisions.

[0014] Preferably, in step S1, the network topology, channel state information and mission requirements are collected by MEO satellites to predict and guide the system to make correct decisions; then real-time interaction is performed with the physical entity; the parameter model of the physical layer is:

[0015]

[0016]

[0017]

[0018]

[0019] in, Respectively Band, Band and The channel bandwidth of the band; Respectively and The distance between and The distance between and The distance between Distance to MEO satellite. Respectively and The transmission power, and The transmission power, and The transmission power, Transmission power with MEO satellites.

[0020] Preferably, in step S2, three binary variables are used to represent the task offloading status;

[0021] They are:

[0022] Indicates the task user Whether the task is offloaded to the service user through D2D Link ;

[0023] Indicates the task user Whether the task is offloaded to satellite;

[0024] Indicates whether the mission is transmitted to other LEO satellites or MEO satellites;

[0025] definition γ m ∈ [ 0 ， 1 ] Indicates the ratio of task offloading;

[0026] when , the task will be executed locally. , the task will be completely offloaded; then , the tasks will be partially offloaded and executed in parallel locally and externally.

[0027] Preferably, in the step S3,

[0028] The action space defines the actions that the agent can perform at each time step; it consists of the following four dimensions: ;

[0029] in, Indicates whether the task is offloaded to , Indicates whether the mission is offloaded to the LEO satellite. Indicates whether the mission is offloaded to other LEO satellites or MEO satellites when the battery power of the LEO satellite is low. Indicates the offloading ratio of the task;

[0030] State Space describes the characteristics of the environment that the agent can observe at any time step; the state space Includes the following:

[0031]

[0032] in, Indicates the status of the task user, including: Current task user The amount of task data and the current task user The computing power of the current task user The location and current task user The delay constraint is:

[0033]

[0034] Indicates the status of service users, including service users The remaining computing power to serve users CPU remaining, service users Location:

[0035]

[0036] Indicates the status of the LEO satellite, including the remaining computing power of the LEO satellite , the remaining energy of LEO satellites and the position of LEO satellites :

[0037]

[0038] Indicates the environmental status, including the frequency band type , Channel bandwidth , noise power density :

[0039]

[0040] Reward Function Used to evaluate the effect of the agent after performing a certain action; the reward function is calculated in the following form:

[0041]

[0042] in is the total delay, is the total energy consumption, and is the weighting coefficient and .

[0043] Preferably, in step S4, the digital twin layer It is expressed as:

[0044]

[0045] in, represents the channel bandwidth, represents the distance between the user and the satellite, is the transmission power, , and They represent the user’s bandwidth deviation, height deviation, and power deviation when the device uploads data, respectively, to describe the data deviation between the actual user and its digital model.

[0046] Preferably, in step S4, a cost function is defined based on delay and energy consumption, and the delay and energy consumption are normalized to a range of [0, 1] by a Sigmoid activation function.

[0047] Due to the adoption of the above-mentioned technical solution, the present invention has the following beneficial effects:

[0048] (1) The present invention discloses a task offloading method for a D2D-LEO full-dimensional collaborative edge computing network and proposes a D2D-LEO multi-dimensional collaborative computing architecture, in which users can not only achieve horizontal end-to-end collaboration through D2D communication, but also achieve vertical end-to-satellite collaboration through LEO satellites. In addition, LEO satellites can also achieve horizontal D2D to LEO collaboration or vertical LEO to MEO collaboration to improve computing collaboration efficiency.

[0049] (2) Digital twin technology is introduced into the satellite edge offloading system. Considering the inevitable errors and noise in the data acquisition and transmission process, a task offloading and resource allocation optimization model with deviation perception is established. In order to solve the heterogeneity of the size of the joint optimization objective function, the sigmoid function is used for normalization.

[0050] (3) In order to respond to the dynamic environment of the satellite edge computing system, a multi-agent deep reinforcement learning algorithm is proposed. The algorithm uses the real-time dynamic interaction between multiple agents and the edge environment to find the optimal strategy. Simulation results show that compared with other algorithms, this method improves network latency, reduces energy consumption, and improves task offloading efficiency by 10%, 5%, and 10%, respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of the method of the present invention;

[0052] Figure 2 This is a diagram of the D2D-LEO full-dimensional collaborative edge computing network architecture of the present invention;

[0053] Figure 3 It is a flow chart of the satellite edge computing partial unloading method based on DSDDPG of the present invention;

[0054] Figure 4 It is a digital twin network system architecture diagram of the present invention;

[0055] Figure 5 It is the experimental parameter diagram of the present invention;

[0056] Figure 6 It is a graph showing the impact of task size on different offloading schemes in the present invention;

[0057] Figure 7 It is a comparison chart of different data deviations in the present invention as the task size changes; DETAILED DESCRIPTION

[0058] The present invention can be explained in detail through the following embodiments. The purpose of disclosing the present invention is to protect all technical improvements within the scope of the present invention. In the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "front", "back", "left", "right", etc. indicating directions or positional relationships, they only correspond to the drawings of the present application for the convenience of describing the present invention, and do not indicate or imply that the device or element referred to must have a specific direction.

[0059] Combined with Figures 1 to 7 , a task offloading method for a D2D-LEO full-dimensional collaborative edge computing network, comprising the following steps:

[0060] S1. Obtain the location information, mission information, and computing power information of ground users through GEO satellites to build a digital twin model:

[0061] MEO satellites collect network topology, channel status information, and mission requirements to predict and guide the system to make correct decisions, and then interact with physical entities in real time. The parameter model of the physical layer is:

[0062]

[0063]

[0064]

[0065]

[0066] in, Respectively Band, Band and The channel bandwidth of the band; Respectively and The distance between and The distance between and The distance between Distance to MEO satellite. Respectively and The transmission power, and The transmission power, and The transmission power, Transmission power with MEO satellites.

[0067] S2. Construct a D2D-LEO full-dimensional collaborative computing offloading framework, including one or more of local computing, D2D offloading, D2L offloading, L2L offloading, and L2G offloading;

[0068] The SMECN (Satellite Mobile Edge Computing Network) framework constructed by the present invention includes a MEO layer, a LEO layer and a user layer. Figure 2 As shown in Figure 1, SMECN consists of a MEO satellite, LEO satellites and The MEO satellites are responsible for planning these LEO satellites, which have fairly powerful computing resources and slow movement speeds, and can be used as the location of cloud computing centers for traditional edge computing. LEO satellites move quickly, have a small coverage area, but have strong communication links, and can be used as edge nodes.

[0069] use To represent the collection of GEO satellites and LEO satellites, Indicates a GEO satellite.

[0070] The user who needs to uninstall the task is set by Indicates that the users available for service are composed of the set Indicates. Each It can be expressed as ,in Indicates the number of CPU cycles required to complete the task. Indicates the maximum delay that the device can withstand. Indicates the maximum energy consumption that the device can withstand. represents the transmission power, express CPU frequency.

[0071] In addition, three binary variables are defined to represent the task offloading status. Indicates the task user Whether the task is offloaded to the service user through D2D-Link , Indicates the task user Whether the task is offloaded to Satellite and Indicates whether the mission is transferred to other LEO satellites or MEO satellites. Because the mission is offloaded in parallel, define γ m ∈ [ 0 ， 1 ] Indicates the ratio of task offloading. , the task will be executed locally. , the task will be completely uninstalled. Then , the tasks will be partially offloaded and executed in parallel locally and externally.

[0072] S3. Design the corresponding action space, state space and reward function, and schedule the computing tasks and execution modes of ground users through the DSDDPG algorithm with battery power awareness;

[0073] like Figure 3 As shown, the present invention designs the corresponding action space, state space and reward function, and schedules the computing tasks and execution modes of ground users through the DSDDPG algorithm with battery power awareness. The action space defines the operations that the agent can perform in each time step.

[0074] These actions are composed of the following four dimensions: ; Indicates whether the task is offloaded to , Indicates whether the task is offloaded to LEO. Indicates whether the mission is offloaded to other LEO satellites and MEO satellites when the LEO satellite battery is low. Indicates the offloading ratio of the task.

[0075] The state space describes the characteristics of the environment that the agent can observe at any time step. Combined with the task offloading scenario in the code, the state space The main components include the following:

[0076]

[0077] in, Represents the status of the task user, including: the task data volume of the current task user, the computing power of the current task user, the location of the current task user, and the delay constraint of the current task user:

[0078]

[0079] Indicates the status of the service user, including the remaining computing capacity of the service user, the remaining CPU rate of the service user, and the location of the service user:

[0080]

[0081] Indicates the status of the LEO satellite, including the remaining computing power of the LEO satellite, the remaining energy of the LEO satellite, and the position of the LEO satellite:

[0082]

[0083] Indicates the environmental status, including the frequency band type, channel bandwidth, and noise power density:

[0084]

[0085] The reward function is used to evaluate the effect of the agent after performing a certain action. The rewards include: the total delay and total energy consumption required to complete the task. The total delay includes local computing delay, upload delay, transmission delay, etc. The lower the delay time, the higher the reward. The total energy consumption includes computing energy consumption, transmission energy consumption, etc. The lower the energy consumption, the higher the reward.

[0086] Specifically, the reward function is calculated in the following form:

[0087]

[0088] where is the total time delay, is the total energy consumption, and are the weighting coefficients and .

[0089] S4. As Figure 4 shown, in the digital twin model, corresponding data deviation values are designed to predict the risks that the system may encounter, so as to guide the model to make correct offloading decisions;

[0090] Use , and to represent the bandwidth deviation, altitude deviation, and power deviation when the device uploads data of the user respectively, to describe the data deviation between the actual user and its digital model. Therefore, the digital twin layer is expressed as:

[0091]

[0092] The corresponding data deviation is added to simulate the data fluctuations that may be encountered in the real physical layer, so as to guide the model to make correct offloading decisions.

[0093] The computing tasks and execution modes include local mode, D2D offloading, D2L offloading, L2L offloading, and L2G offloading:

[0094] Local mode: Since the model executes a parallel offloading mode, only a part of the tasks are executed locally. Therefore, the local computing delay can be expressed as :

[0095]

[0096] Correspondingly, the energy consumption of local computing is expressed as :

[0097]

[0098] in, Represents the energy coefficient, whose value depends on the device itself.

[0099] D2D offloading mode: Considering the large number of idle users in the network, D2D offloading has become a focus of attention. Offload computational tasks to At this time Since the size of the calculation result is much smaller than the size of the calculation task, the delay and energy consumption of the calculation result transmission are not considered. Therefore, the total delay Denoted as:

[0100]

[0101] It includes transmission delay and calculation delay .

[0102] The calculation delay of the offloaded tasks is as follows:

[0103]

[0104] in, represents the computation frequency of the service user. Therefore, the task is offloaded to The transmission delay is:

[0105]

[0106] in, Indicates the transmission rate of the channel:

[0107]

[0108] in Indicates the bandwidth of the wireless channel, used by ground users Band, Indicates the transmission power of the data uploaded by the user. represents the wireless channel gain, represents the noise variance. Due to the data deviation between the digital twin and the actual network, it is assumed The bandwidth deviation of the band is ,

[0109]

[0110]

[0111] In this case, the deviation of the transmission delay is:

[0112]

[0113] Therefore, the actual transmission delay of the offloaded part and total delay It is expressed as follows:

[0114]

[0115]

[0116] Since tasks are executed in parallel, the total delay Is the local delay and uninstall part delay The larger part of the total delay is expressed as follows:

[0117]

[0118] For D2D task execution mode, the total energy consumption includes local computing energy consumption , transmission energy consumption , the computing energy consumption of the task offloading part Therefore, the total energy consumption of task execution is as follows.

[0119]

[0120] D2L Unload Mode: If If you choose to offload some of the computing tasks to the LEO satellite, you need to determine whether the satellite's computing resources and battery power meet the computing requirements. At this time, the current battery power of the LEO satellite is defined as If the computing energy consumption of the computing task that the user offloads to the LEO satellite is greater than the current battery power of the LEO satellite, the LEO satellite will not be able to complete the computing task. In this case, the computing task needs to be transferred to other LEO satellites with sufficient battery power. Otherwise, the LEO satellite can complete the computing task.

[0121] When the battery power of the LEO satellite is sufficient, the computation delay and transmission delay of the part of the mission offloaded to LEO are as follows:

[0122]

[0123]

[0124] in, Indicates the transmission rate of the channel; Indicates the CPU calculation frequency:

[0125]

[0126] in, Indicates the bandwidth of the wireless channel, used between users and LEO satellites Band, Represents the transmission power of the data uploaded by the user. Based on the previous analysis, the actual transmission delay can be obtained as follows.

[0127]

[0128]

[0129] Due to the long distance between users and LEO satellites, propagation delay is inevitable:

[0130]

[0131] in is the speed of light in the universe.

[0132] Since the shape of the earth is not a regular sphere, slight differences in altitude will also affect the propagation delay, so it is assumed that and The distance deviation between

[0133]

[0134] Therefore, the total delay at this time for:

[0135]

[0136] Since tasks are executed in parallel, the total delay Is the local delay and uninstall part delay The maximum value among:

[0137]

[0138] The total energy consumption at this time for:

[0139]

[0140] Among them, the total energy consumption includes local computing energy consumption , transmission energy consumption , the computing energy consumption of the task offloading part . represents the propagation energy consumption and is defined as follows:

[0141]

[0142] in, Represents the propagation power of the signal.

[0143] In satellite communication systems, the effective transmission of propagation power is affected by many factors, one of the most important of which is path loss. Path loss refers to the gradual attenuation of signal strength due to changes in propagation distance and signal frequency during signal propagation. According to the free space propagation model, path loss increases with the increase of propagation distance and signal frequency.

[0144] Therefore, long-distance transmission and high-frequency signals will cause significant signal attenuation. For example, in high-frequency signal transmission (such as band), path loss is more significant, especially over long distances, resulting in greater signal attenuation. This requires the transmitter to provide higher power. Therefore, when calculating propagation energy consumption When , the deviation value of user transmission power is introduced. At this time, the propagation energy consumption The true value of is:

[0145]

[0146] The total energy consumption at this time includes local computing energy consumption , transmission energy consumption , the computing energy consumption of the task offloading part and the propagation energy consumption of the unloading part :

[0147]

[0148] L2L unloading mode: If the battery power of the LEO satellite currently performing the unloading task is insufficient, the unloading task needs to be transferred to other LEO satellites with sufficient battery power. Its definition is as follows:

[0149]

[0150] When LEO satellite The current battery charge is greater than or equal to complete the computing task When the energy consumption is reduced, the task does not need to be offloaded to other LEO satellites. Otherwise, other LEO satellites are needed to assist in the calculation. Since it needs to be transmitted to other LEO satellites, the total delay and energy consumption will also increase accordingly. The increased part is the transmission and delay between LEO satellites. The total delay is:

[0151]

[0152] in represents the transmission delay between LEO satellites, represents the propagation delay between LEO satellites.

[0153]

[0154] in, Indicates the transmission rate of the channel:

[0155]

[0156] in Indicates the bandwidth of the wireless channel, used between MEO satellites and LEO satellites Band, Indicates the transmission power of LEO satellite uploading data, Represents the wireless channel gain. Based on the above analysis, the true value of the transmission delay can be obtained as:

[0157]

[0158] At the same time, the true value of the propagation delay between LEO satellites can be obtained as:

[0159]

[0160] Therefore, the true value of the total delay is:

[0161]

[0162] Since tasks are executed in parallel, the total latency is the maximum of the local latency and the latency of the offloaded portion:

[0163]

[0164] In addition to the total energy consumption of the LEO satellite part, the total energy consumption at this time It also includes the transmission energy consumption between LEO satellites and transmission energy consumption :

[0165]

[0166] in, represents the signal transmission power of the LEO satellite, represents the calculation frequency of LEO satellite, Indicates the signal propagation power of the LEO satellite.

[0167] L2G offloading mode: When the battery power of other satellites is also low, the task can be offloaded to the MEO satellite for execution. A variable is defined To represent this situation: when the battery power of other satellites is also insufficient, the task can be offloaded to the MEO satellite for execution. A variable is defined to represent this situation: when When , it means that the batteries of other LEO satellites are also insufficient and it is necessary to offload to the MEO satellite for calculation. , it means that the battery power of other LEO satellites is sufficient and does not need to be unloaded to the MEO satellite for calculation.

[0168] Since the MEO satellite moves slowly and is fully illuminated by the sun, it can be assumed that it has sufficient battery power and fairly powerful computing resources, and the calculation delay is relatively small. Similar to the LEO-LEO mode, the total delay increases the transmission and propagation energy consumption between the LEO satellite and the MEO satellite. The true value of the total delay is:

[0169]

[0170] in, is the transmission delay from LEO satellite to MEO satellite, is the propagation delay from LEO satellite to MEO satellite, The calculation delay of offloading tasks for MEO satellites. The transmission delay between LEO satellite and MEO satellite is:

[0171]

[0172] in, Indicates the transmission rate of the channel:

[0173]

[0174] in, represents the transmission power of data uploaded by LEO satellite, Represents the wireless channel gain. Based on the above analysis, the true value of the transmission delay can be obtained as:

[0175]

[0176] Likewise, the true value of the propagation delay between the LEO and MEO satellites is:

[0177]

[0178] The calculated delay for offloading tasks on a MEO satellite is:

[0179]

[0180] Since tasks are executed in parallel, the total latency is the maximum of the local latency and the latency of the offloaded portion:

[0181]

[0182] in, Represents the calculation frequency of the MEO satellite. Based on the above analysis, the true value of the total energy consumption at this time can be obtained as:

[0183]

[0184] in, Indicates the calculation frequency of MEO satellite.

[0185] Since the task user uses the parallel execution mode of partial offloading, that is, the task is executed simultaneously in the local and offload devices, the total delay of the task execution is determined by the maximum delay between the local execution, U2U offload execution and U2L offload execution. Therefore, the total delay of the task execution is:

[0186]

[0187] The total energy consumption of task execution depends on different offloading modes as follows:

[0188]

[0189] The present invention considers both latency and energy consumption to define the cost function. However, latency and energy consumption are indicators of two different dimensions, and their values ​​may vary greatly. For example, the latency of some systems may be very small, such as milliseconds, while the energy consumption of some systems may be very large, such as several joules. This difference may cause a certain indicator to have too much influence on the overall evaluation, especially in multi-objective optimization, where certain objectives may dominate. So they are normalized to the [0,1] range through the Sigmoid activation function, which eliminates the difference in units. Regardless of the unit of the original data, the normalized values ​​are in the same normalized range, which allows them to be compared, weighted, and integrated on the same scale:

[0190]

[0191]

[0192] Therefore, the following definitions are given:

[0193]

[0194]

[0195]

[0196] γ m ∈ [ 0 , 1 ] , ∀ m ∈ M

[0197]

[0198]

[0199]

[0200] Among them, constraint (a) means that the total delay must be less than the maximum delay that the system can tolerate. Constraint (b) means that the total energy consumption must be less than the maximum energy consumption that the system can bear. Constraint (c) means that the current battery power of the Leo satellite must be less than or equal to its maximum battery power. Constraint (d) means that the unloading rate must be within [ 0 , 1 ] Constraint (e) means that the task user can offload its computing task to at most one idle user or one satellite edge server. And a satellite edge server can collaborate with at most one adjacent LEO edge server or MEO cloud center. Constraint (f) means that if the user chooses to offload the task, it can only choose to offload to a LEO satellite or , but not at the same time. Constraint (g) means that when the LEO satellite’s current battery power is low, it can choose to offload to one of the MEO satellites and the other LEO satellite.

[0201] Experimental verification:

[0202] Assume there are 40 users, including 20 and 20 . 3 LEO satellites and 1 MEO satellite. For other parameters, see Figure 5 In order to verify the effectiveness of the above scheme, we first analyzed the performance of different offloading schemes, including the following four cases:

[0203] (1) All local computing: All computing tasks are performed on the local device.

[0204] (2) All uninstalled to : In this case, all computation tasks are performed by implement.

[0205] (3) All offloading to LEO satellites: In this case, all computing tasks are transmitted to LEO satellites for processing.

[0206] (4) Hybrid computing: Hybrid computing combines the advantages of local computing, user service computing, and Leo satellite computing, and can flexibly allocate computing tasks.

[0207] like Figure 6As shown in the figure, as the task scale increases, the total consumption of solution (a) increases the fastest, because the size of the data volume is its main influencing factor. The total consumption of solution (b) also grows rapidly, because when the task scale is large, although offloading to service users helps task users reduce computing pressure, it will generate additional transmission delay and transmission energy consumption. The total delay growth of solution (c) is more stable, because LEO satellites have relatively strong computing power, and the growth of task scale has little impact on it. The main factor affecting the increase in its total consumption is the transmission distance. When the task increases, the total consumption of solution (d) is the lowest. Since hybrid computing can give reasonable offloading decisions, the total consumption of the system is minimized.

[0208] like Figure 7 As shown in a, when the distance between the LEO satellite and the user deviates and transmission power deviation When constant, as the data deviation of the channel bandwidth continues to increase, the total overhead will gradually increase. Figure 7 As shown in b, when the channel bandwidth deviation and transmission power deviation When the altitude difference between the LEO satellite and the user increases, the total cost will gradually increase. Figure 7 As shown in c, when the channel bandwidth between the LEO satellite and the user deviates and height deviation When it remains unchanged, a reverse deviation value is given to the transmission power. As the transmission power deviation increases, the total overhead will gradually increase. Figure 7 As shown in Figure d, when these three factors change at the same time, the impact on the entire system is greater. Therefore, introducing data deviation is necessary to improve the stability of the system.

[0209] The parts of the present invention that are not described in detail are prior art. It is obvious to those skilled in the art that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention; therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and it is intended that all changes that fall within the meaning and scope of equivalent elements are included in the present invention.

Claims

1. A task offloading method for a D2D-LEO full-dimensional collaborative edge computing network, characterized in that: The following steps are involved: S1. Obtain the location information, mission information, and computing power information of ground users through GEO satellites to build a digital twin model; S2. Construct a D2D-LEO full-dimensional collaborative computing offloading framework, including one or more of local computing, D2D offloading, D2L offloading, L2L offloading, and L2G offloading; S3, define the action space, state space and reward function, and schedule the computing tasks and execution modes of ground users through the DSDDPG algorithm with battery power awareness; S4. In the digital twin model, define corresponding data deviation values ​​to predict the risks that the system will encounter, thereby guiding the model to make correct unloading decisions.

2. The task offloading method of the D2D-LEO full-dimensional collaborative edge computing network according to claim 1 is characterized by: In step S1, the network topology, channel state information and mission requirements are collected through MEO satellites to predict and guide the system to make correct decisions; then real-time interaction is performed with the physical entity; the parameter model of the physical layer is: in, Respectively represent the channel bandwidths of C-band, Ka-band and Ku-band; Respectively and The distance between and The distance between and The distance between Distance to MEO satellite. Respectively and The transmission power, and The transmission power, and The transmission power, Transmission power with MEO satellites.

3. The task offloading method of the D2D-LEO full-dimensional collaborative edge computing network as claimed in claim 1 is characterized by: In the step S2, three binary variables are used to represent the task offloading status; They are: Indicates the task user Whether the task is offloaded to the service user through D2D Link ; Indicates the task user Whether the task is offloaded to satellite; Indicates whether the mission is transmitted to other LEO satellites or MEO satellites; definition Indicates the proportion of task offloading; when The task will be executed locally when , the task will be completely offloaded; then , the tasks will be partially offloaded and executed in parallel locally and externally.

4. The task offloading method of the D2D-LEO full-dimensional collaborative edge computing network as claimed in claim 3 is characterized by: In the S3 step, The action space defines the actions that the agent can perform at each time step; It consists of the following four dimensions: ; in, Indicates whether the task is offloaded to , Indicates whether the mission is offloaded to the LEO satellite. Indicates whether the mission is offloaded to other LEO satellites and MEO satellites when the battery power of the LEO satellite is low. represents the offloading ratio of the task; State Space describes the characteristics of the environment that the agent can observe at any time step; the state space Includes the following: in, Indicates the status of the task user, including: Current task user The amount of task data and the current task user The computing power of the current task user The location and current task user The delay constraint is: Indicates the status of service users, including service users The remaining computing power to serve users CPU remaining, service users Location: Indicates the status of the LEO satellite, including the remaining computing power of the LEO satellite , the remaining energy of LEO satellites and the position of LEO satellites : Indicates the environmental status, including the frequency band type , Channel bandwidth , noise power density : Reward Function Used to evaluate the effect of the agent after performing a certain action; the reward function is calculated in the following form: in is the total delay, is the total energy consumption, and is the weighting coefficient and .

5. The task offloading method of the D2D-LEO full-dimensional collaborative edge computing network according to claim 1 is characterized by: In step S4, the digital twin layer It is expressed as: in, represents the channel bandwidth, represents the distance between the user and the satellite, is the transmission power, , and They represent the user’s bandwidth deviation, height deviation, and power deviation when the device uploads data, respectively, to describe the data deviation between the actual user and its digital model.

6. The task offloading method of the D2D-LEO full-dimensional collaborative edge computing network as claimed in claim 5 is characterized by: In the step S4, a cost function is defined based on delay and energy consumption, and the delay and energy consumption are normalized to the range of [0, 1] by a Sigmoid activation function.

Images