Multi-type task adaptive calculation unloading method and system for satellite Internet of Things

Abstract

The invention provides a multi-type task adaptive calculation unloading method and system for the satellite Internet of Things, and the method is characterized in that the method comprises the following steps: S1 to S2, resource use conditions of the local Internet of Things device, each edge server corresponding to the local Internet of Things device and the satellite cloud are respectively collected as a current resource use condition of the device and a current resource use condition of other non-self devices; s3, taking the task attribute of the calculation task, the resource use condition of the current equipment and the resource use condition of other current non-self equipment as an input state space; s4, inputting the input state space into the unloading object decision model to obtain a current task unloading target; and S5, adopting the current task unloading target to process the calculation task. In a word, the method can optimize the long-term system overhead of the calculation task in the satellite Internet of Things.

Description

Technical Field

[0001] The present invention belongs to the field of satellite Internet of Things, and in particular relates to a method and system for adaptively offloading multi-type tasks of satellite Internet of Things. Background Art

[0002] With the development of satellite communication technology, satellite networks, as a complement and extension of the terrestrial Internet of Things (IoT), can achieve wide-area coverage across space, air, and land. This effectively addresses issues such as lack of network coverage in remote areas and the vulnerability of ground network facilities to disasters, enabling on-demand coverage in remote areas, at sea, and in the air. The Satellite Internet of Things (SIoT), a fusion of satellite communication and IoT technologies, can facilitate the interconnection of IoT devices across land, desert, and ocean regions worldwide.

[0003] In traditional satellite IoT architectures, when IoT devices generate computing tasks, they are difficult to process locally due to the limited computing resources of the local IoT devices themselves. Therefore, these tasks must be transmitted via satellite relay to ground-based cloud computing centers, resulting in unacceptable transmission delays. With the advancement of satellite and cloud computing technologies, satellite clouds are being formed by combining low-, medium-, and high-orbit satellites, bringing computing power closer to local IoT devices and reducing task transmission delays.

[0004] However, given the growing demand for satellite IoT and the increasing diversity of computing tasks, deploying a single offload strategy for each type of task would waste computing resources. Furthermore, due to the random nature of computing tasks in satellite IoT, focusing solely on short-term system optimization would lead to underutilization of computing device resources, further increasing task processing overhead.

[0005] In short, existing technologies still find it difficult to achieve optimal offloading of tasks under satellite Internet of Things and cannot further reduce system overhead. Summary of the Invention

[0006] To solve the above problems, the present invention provides a method and system for adaptive computing offloading of multiple types of tasks for satellite Internet of Things.

[0007] The present invention provides a multi-type task adaptive computing offloading method for satellite Internet of Things, which has the following characteristics: a local Internet of Things device group, a ground edge server group and a satellite cloud, wherein the local Internet of Things device group includes multiple local Internet of Things devices, the ground edge server group includes multiple edge servers, and each local Internet of Things device is respectively connected to the satellite cloud and at least one corresponding edge server in communication, wherein the computing task T received by the local Internet of Things device U i，j The processing process includes the following steps: Step S1, respectively collect local IoT devices U i , and local IoT devices Ui The resource usage of each corresponding edge server and satellite cloud; Step S2, the local IoT device U i The resource usage of the local IoT device U i The resource usage of all corresponding edge servers and satellite clouds is used as the current resource usage of other non-self-devices; Step S3, the calculation task T i，j The type of task, the current device's own resource usage and the current resource usage of other non-self devices are used as the input state space; step S4, input the input state space into the offloading object decision model to obtain the corresponding current task offloading target; step S5, use the current task offloading target to process the computing task T i，j ,Task types include delay-sensitive, computation-intensive and delay-tolerant.,The current task offloading target is the local IoT device U i Or edge servers or satellite clouds.

[0008] The method for adaptive offloading of multi-type tasks for satellite Internet of Things provided by the present invention may also have the following characteristics: wherein the offloading object decision model is obtained by solving the objective function corresponding to the adaptive offloading strategy problem of multi-type tasks for satellite Internet of Things using the DDPG algorithm, and the expression of the objective function is:

[0009] P1: Minimize D, stC1: Where X i，j For the computation task T i，j The corresponding uninstallation decision, x i，j For the computation task T i，j Is it in the corresponding local IoT device U i Execute, y i，j For the computation task T i，j Whether to offload to the edge server for execution, z i，j For the computation task T i，j Whether to offload to the satellite cloud for execution, For local IoT devices U i Processing computing task T i，j The computational delay generated, To offload the latency to the edge server, For the computation task T i，j The delay caused by offloading to the satellite cloud, ρ i，j For a given computing task T i，jThe time interval from generation to execution completion, I is the total number of local IoT devices in the local IoT device group, J is the total number of computing tasks generated in the local IoT device group, K is the total number of edge servers in the ground edge server group, T i，j (f k )=1 to ensure that each computing task is assigned to only one edge server for processing.

[0010] The multi-task adaptive computing offloading method for satellite IoT provided by the present invention may also have the following features: wherein, in the DDPG algorithm solution, the reward function is expressed as: In the formula, reward i，j For the computation task T i，j The reward value, D i，j For the computation task T i，j The cost, ρ i，j For the computation task T i，j Processing maximum tolerable delay, d i，j According to the calculation task T i，j The actual delay of the task after taking the corresponding action.

[0011] The multi-task adaptive computing offloading method for satellite Internet of Things provided by the present invention may also have the following characteristics: wherein the cost D i，j The calculation expression is:

[0012]

[0013]

[0014] Where α is the preset weight value, For local IoT devices U i Processing computing task T i，j Energy consumption, κ is the energy factor, For local IoT devices U i The computing power of c i,j To process the computing task T i,j The number of CPU cycles required per bit, To process the computing task T i,j The data size, For the computation task T i，j Energy consumption of offloading to edge servers, is the computing power of the edge server, R e For local IoT device U i The transmission rate to the edge server, P eTo offload the transmission power to the edge server, The computing power of satellite cloud, R s For local IoT device U i Transmission rate to the satellite cloud, P s is the transmission power offloaded to the satellite cloud.

[0015] The multi-task adaptive computing offloading method for satellite Internet of Things provided by the present invention may also have the following features: wherein the computing delay The expression is: In the formula For local IoT devices U i Processing computing task T i，j The resulting queuing delay, delay The expression is: In the formula For the computation task T i，j The queuing delay waiting for execution on the edge server, delay The expression is:

[0016]

[0017] The multi-task adaptive computing offloading method for satellite IoT provided by the present invention may also have the following features: wherein the local IoT device is connected to the satellite cloud through a gateway station, and the transmission rate R s The expression is: Where ω s is the bandwidth of the satellite-to-ground link, σ is the noise power, and the transmission rate R e The expression is: Where ω e is the bandwidth of the terrestrial wireless link.

[0018] The multi-type task adaptive computing offloading method for satellite Internet of Things provided by the present invention may also have the following characteristics: wherein, in the DDPG algorithm solution, according to the preset decay rate N, the maximum value of the exploration factor ε max and the minimum value of the exploration factor ε min As well as the current number of training rounds n, the exploration factor ε corresponding to the current number of training rounds n is calculated. The calculation expression of the exploration factor ε is:

[0019] The present invention also provides a satellite Internet of Things system having the following characteristics: a local Internet of Things device cluster, a ground edge server cluster, and a satellite cloud, all of which are used to process computing tasks. The local Internet of Things device cluster includes multiple local Internet of Things devices and a decision module, the ground edge server cluster includes multiple edge servers, and the satellite cloud includes a satellite cloud. Each local Internet of Things device is respectively in communication with the satellite cloud and at least one corresponding edge server. The decision module includes: a data acquisition unit for respectively collecting resource usage of the local Internet of Things device, each edge server corresponding to the local Internet of Things device, and the satellite cloud; an input state space generation unit for constructing an input state space based on the task attributes of the computing task received by the local Internet of Things device, the resource usage of the local Internet of Things device corresponding to the computing task, and the resource usage of each edge server and the satellite cloud corresponding to the local Internet of Things device corresponding to the computing task; a decision generation unit including an offloading object decision model for inputting the input state space into the offloading object decision model to obtain a corresponding current task offloading target; and a decision execution unit for controlling the current task offloading target to process the computing task. Task types include delay-sensitive, computation-intensive, and delay-tolerant. The current task offloading target is the local Internet of Things device, edge server, or satellite cloud corresponding to the computing task.

[0020] Functions and effects of the invention

[0021] According to the method and system for adaptive computing offloading of multiple types of tasks for the satellite Internet of Things, the present invention first constructs the queuing delay based on queuing theory and introduces it into the construction of the objective function, so that the offloading object decision model obtained by solving the objective function can minimize long-term system overhead. Second, the state space and reward function of the reinforcement learning DDPG algorithm are designed according to the computing task type, improving the model's generalization ability when processing different types of tasks. Third, an improved exploration mechanism is used to further enhance the exploration ability of the DDPG algorithm, solving for an offloading object decision model with better performance. Therefore, the method and system for adaptive computing offloading of multiple types of tasks for the satellite Internet of Things can process different types of tasks while optimizing the long-term system overhead of computing tasks in the satellite Internet of Things. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a block diagram of a satellite Internet of Things system in an embodiment of the present invention;

[0023] Figure 2 Schematic diagram of communication connections among a satellite cloud, a ground edge server cluster, and a local IoT device cluster in an embodiment of the present invention;

[0024] Figure 32 is a schematic diagram comparing the model effects of the improved exploration mechanism and the traditional exploration mechanism in an embodiment of the present invention;

[0025] Figure 4 is a comparative schematic diagram of ablation experiment results of an offloading object decision model based on queuing theory in an embodiment of the present invention;

[0026] Figure 5 It is a flowchart of processing computing tasks in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the multi-type task adaptive computing offloading method and system for the satellite Internet of Things of the present invention.

[0028] This embodiment provides a satellite Internet of Things system for allocating and processing computing tasks.

[0029] Figure 1 1 is a block diagram of a satellite Internet of Things system in an embodiment of the present invention.

[0030] like Figure 1 As shown, the satellite IoT system 100 includes a satellite cloud 10, a ground edge server cluster 20 and a local IoT device cluster 30, all of which are used to process computing tasks.

[0031] The satellite cloud 10 includes a satellite cloud 101 .

[0032] The ground edge server cluster 20 includes multiple edge servers 201 .

[0033] The local IoT device cluster 30 includes a plurality of local IoT devices 301 and a decision module 302 .

[0034] Each local IoT device 301 is respectively communicated with the satellite cloud 101 and at least one corresponding edge server 201, that is, the satellite cloud 101 is communicated with each local IoT device 301, and each edge server 201 has its own corresponding service range and is respectively communicated with each local IoT device 301 within the service range.

[0035] Figure 2 Schematic diagram of the communication connection between the satellite cloud, the ground edge server cluster and the local IoT device cluster in an embodiment of the present invention.

[0036] like Figure 2As shown, the local IoT device group consisting of all local IoT devices 301, namely Local IoT Devices, is connected to the satellite cloud 101 consisting of multiple satellites, namely Satellite Cloud, and the ground edge server group consisting of all edge servers 201, namely Ground Edge Servers, through the gateway.

[0037] The decision module 302 includes a data acquisition unit 3021, an input state space generation unit 3022, a decision generation unit 3023 and a decision execution unit 3024, which is used to select a corresponding current task offloading target for each computing task and control the current task offloading target to process the computing task.

[0038] The data collection unit 3021 is used to collect resource usage of the local IoT device 301, each edge server 201 corresponding to the local IoT device 301, and the satellite cloud 101 respectively.

[0039] The input state space generation unit 3022 is used to construct the input state space according to the task type of the computing task received by the local IoT device 301, the resource usage of the local IoT device 301 corresponding to the computing task, and the resource usage of each edge server 201 and satellite cloud 101 corresponding to the local IoT device 301 corresponding to the computing task.

[0040] In this embodiment, the expression of the input state space is (Task, E, N), where Task is the task attribute, E is the resource usage of the local IoT device 301 corresponding to the computing task, and N is the resource usage of each edge server 201 and satellite cloud 101 corresponding to the local IoT device 301 corresponding to the computing task.

[0041] Among them, task types include delay-sensitive, computation-intensive, and delay-tolerant.

[0042] The decision generating unit 3023 includes an offloading object decision model, which is used to input the input state space into the offloading object decision model to obtain the corresponding current task offloading target.

[0043] Among them, the offloading object decision model is obtained by solving the objective function corresponding to the adaptive computing offloading strategy problem of multi-type tasks for satellite Internet of Things using the DDPG algorithm.

[0044] The modeling process of the adaptive computation offloading strategy problem for multi-type tasks in satellite IoT is as follows:

[0045] First, define the data. The local IoT device group is defined as U = {U i |1≤i≤I},Ui is the i-th local IoT device 301 in the local IoT device group, and I is the total number of local IoT devices 301. The computing task set generated in the local IoT device group U is defined as T={T i,j |1≤i≤I,1≤j≤J},T i,j is the jth computing task received by the i-th local IoT device 301, and J is the maximum number of tasks received by the local IoT device 301. Computation task T i,j It can be expressed as ( c i，j , ρ i，j , t i，j ), To process the computing task T i，j The data size in bits, c i，j To process the computing task T i，j The number of CPU cycles required per bit, in cycles / bit, ρ i，j For a given computing task T i，j The time interval from generation to execution completion, in s, t i，j For the computation task T i，j The ground edge server cluster is defined as E = {E k |1≤k≤K}, E k is the kth edge server 201 in the ground edge server group, and K is the total number of edge servers 201.

[0046] Second, a queuing model is constructed based on queuing theory. In this embodiment, the satellite cloud 101 can process computing tasks at any time, that is, it has sufficient computing resources, so there is no need to build a corresponding queuing model. However, the local IoT device group U and the ground edge server group E both have limited computing resources, so a corresponding queuing model needs to be constructed. i and edge server E i Queues and queues The received computing tasks are stored, and the computing tasks arriving at time slot t can be processed within time slot t+1. At the same time, the tasks arriving at time slot t+1 need to wait until the computing tasks arriving at time slot t are processed before being processed. That is, the local IoT device U i Processing computing task T i，j The resulting queuing delay and queue That is, the computing task T i，j The expression for the queue delay waiting to be executed on the edge server is:

[0047]

[0048]

[0049] In the formula To reach the local IoT device U i The processing delay required for the nth computing task that has not been processed includes computing delay and communication delay. To reach the edge server E i The processing delay required for the nth unprocessed computing task includes computing delay and communication delay.

[0050] Third, build a communication model. The transmission rate R from the gateway to the satellite cloud 101 s and the transmission rate R from the local IoT device to the edge server e The expression is:

[0051]

[0052]

[0053] Where ω s is the bandwidth of the satellite-to-ground link, σ is the noise power, ω e is the bandwidth of the ground wireless link, P e is the transmission power offloaded to the edge server, P s is the transmission power offloaded to the satellite cloud.

[0054] Fourth, construct the cost function. The computing task cannot be divided and processed by different devices, and since the local IoT device 301 receiving the computing task does not generate transmission delay when processing the computing task, the local IoT device U i Processing computing task T i，j The resulting computational delay The expression is:

[0055]

[0056] In the formula For local IoT devices U i Processing computing task T i，j The resulting queuing delay, For local IoT devices U i computing power.

[0057] Local IoT device U i Processing computing task T i，j Energy consumption The expression is:

[0058]

[0059] Where κ is the energy factor.

[0060] Then, the local IoT device U i Processing computing task T i，j Cost The expression is:

[0061]

[0062] Where α is the preset weight value.

[0063] Computational task T i，j Offload to edge server E k Execution delay The expression is:

[0064]

[0065] In the formula For the computation task T i，j On the edge server E k The queue delay waiting to be executed, For edge server E k computing power.

[0066] Computational task T i，j Offload to edge server E k Energy consumption of execution The expression is:

[0067]

[0068] Then, the edge server E k Processing computing task T i，j Cost The expression is:

[0069]

[0070] Computational task T i，j Delay caused by offloading to Satellite Cloud 101 The expression is:

[0071]

[0072] Where f s m The computing power of the satellite cloud.

[0073] Computational task T i，j Energy consumption offloaded to Satellite Cloud 101 The expression is:

[0074]

[0075] Then, the satellite cloud 101 processes the computing task T i，j Cost The expression is:

[0076]

[0077] In this embodiment, the computing task is assigned or offloaded to the local IoT device 301, edge server 201 or satellite cloud 101 that receives the task for processing, so the offloading decision X i，j The expression is:

[0078] X i，j ={x i，j ,y i，j , z i，j},

[0079] x i，j ∈{0, 1}, y i，j ∈{0, 1}, z i，j ∈{0, 1},

[0080] x i，j +y i，j +z i，j =1,

[0081] Where x i，j For the computation task T i，j Is it in the corresponding local IoT device U i Execute, y i， For the computation task T i，j Whether to offload to the edge server for execution, z i，j For the computation task T i，j Whether to offload to the satellite cloud for execution, x i，j 、y i，j and z i，j When the value is 1, it means that the corresponding device processes the computing task, x i，j 、y i，j and z i，j When the value is 0, it means that the corresponding device does not process the computing task. That is, the current task offloading target is the local IoT device 301 or edge server 201 or satellite cloud 101 corresponding to the computing task.

[0082] Then, the expression of the total cost function is:

[0083]

[0084] Fifth, construct the objective function. Based on the above model and function, the objective function expression corresponding to the problem of adaptive computing offloading strategy for multi-type tasks for satellite IoT can be obtained as follows:

[0085] P1: Minimize D,

[0086] stC1:

[0087] X i，j ={x i，j ,y i，j , z i，j}, x i，j ∈{0, 1}, y i，j ∈{0, 1}, z i，j ∈{0, 1},

[0088]

[0089]

[0090]

[0091] Where ρ i，j For a given computing task T i，j The time interval from generation to execution completion, T i，j (f k )=1 to ensure that each computing task is assigned to only one edge server for processing.

[0092] In this embodiment, when the DDPG algorithm is used to solve the objective function and construct the offloading object decision model, the expression of the reward function is:

[0093]

[0094]

[0095] In the formula, reward i，j For the computation task T i，j The reward value, D i，j For the computation task T i，j The cost, ρ i，j For the computation task T i，j Processing maximum tolerable delay, d i，j According to the calculation task T i，j The actual delay of the task after taking the corresponding action.

[0096] The DDPG algorithm is a popular method, but its exploration mechanism still has room for improvement. Existing exploration strategies, such as the ε-greedy method, often perform poorly in this area. Therefore, this embodiment, building on the ε-greedy method, improves the exploration mechanism so that it decreases with the number of training rounds. This improves the DDPG algorithm's exploration capabilities in unknown environments and enhances the performance of the offload object decision model solved by the DDPG algorithm.

[0097] Among them, the calculation expression of the improved exploration factor ε corresponding to the current number of training rounds is:

[0098]

[0099] Where N is the preset decay rate, ε max is the preset maximum value of the exploration factor, ε min is the preset minimum exploration factor ε min , n is the current number of training rounds.

[0100] In this embodiment, the performance of the improved exploration mechanism is compared with that of the conventional exploration mechanism in which the exploration factor is fixed at 0.1 in the existing method.

[0101] Figure 3 3 is a schematic diagram comparing the model effects of the improved exploration mechanism and the traditional exploration mechanism in an embodiment of the present invention.

[0102] like Figure 3 As shown in the figure, the horizontal axis is the number of training rounds of the model, and the vertical axis is the system overhead corresponding to the model. It can be seen that the model that converges after training with the improved exploration mechanism, that is, after the improved exploration mechanism, has less system overhead than the model that converges after training with the traditional exploration mechanism, that is, before the improved exploration mechanism, that is, the model results perform better.

[0103] In this embodiment, the performance of the unloading object decision model constructed based on queuing theory, that is, the model after adding queuing theory, is compared with the unloading object decision model without queuing theory, that is, the model before adding queuing theory, so as to verify the effectiveness of adding queuing theory.

[0104] Figure 4 3 is a comparative diagram of ablation experiment results of an offloading object decision model based on queuing theory in an embodiment of the present invention.

[0105] like Figure 4 As shown, the horizontal axis is the number of training rounds of the model, and the vertical axis is the system overhead corresponding to the model. It can be seen that the model after adding queuing theory has less system overhead than the model before queuing theory, that is, the model results perform better.

[0106] The decision execution unit 3024 is used to control the current task to offload the target processing computing task.

[0107] When the current task offloading target is the edge server 201 or the satellite cloud 101, the computing task is offloaded to the edge server 201 or the satellite cloud 101 through the local IoT device 301. When the current task offloading target is the local IoT device 301, the computing task is directly placed in the task processing queue of the local IoT device 301, waiting for the local IoT device 301 to process it.

[0108] The following describes the satellite IoT system 100 of this embodiment for the local IoT device U with reference to the accompanying drawings. i Received computing task T i，j The processing process is described.

[0109] Figure 5 It is a flowchart of processing computing tasks in an embodiment of the present invention.

[0110] like Figure 5 As shown, the local IoT device U i Received computing task T i，j The process being processed includes the following steps:

[0111] Step S1: Use the data collection unit 3021 to collect data from local IoT devices U i , and local IoT devices U i The resource usage of the corresponding edge servers 201 and satellite cloud 101.

[0112] Step S2: Connect the local IoT device U i The resource usage of the local IoT device U i The resource usage of all corresponding edge servers and satellite clouds is used as the current resource usage of other non-self-devices.

[0113] Step S3, using the input state space generation unit 3022 to calculate the task T i，j The task attributes, current device resource usage, and current resource usage of other non-devices are used as input state space.

[0114] Step S4: The decision generating unit 3023 is used to input the input state space into the offloading object decision model to obtain the corresponding current task offloading target.

[0115] Step S5, using the decision execution unit to control the current task to offload the target processing computing task T i，j .

[0116] Functions and Effects of the Embodiments

[0117] According to the method and system for adaptive computation offloading of multiple types of tasks for the satellite Internet of Things (IoT) described in this embodiment, first, queuing delay is constructed based on queuing theory and incorporated into the construction of the objective function, so that the offload object decision model obtained by solving the objective function can minimize long-term system overhead. Second, the state space and reward function of the reinforcement learning (DDPG) algorithm are designed based on the computational task type, improving the model's generalization ability when handling different types of tasks. Third, an improved exploration mechanism is used to further enhance the exploration capability of the DDPG algorithm, resulting in a better-performing offload object decision model. In summary, this method can handle different types of tasks while optimizing the long-term system overhead of computational tasks in the satellite IoT.

[0118] Those skilled in the art will appreciate that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-task adaptive computing offloading method for satellite Internet of Things, characterized by: include: Local IoT device clusters, ground edge server clusters, and satellite clouds, The local IoT device group includes multiple local IoT devices. The ground edge server group includes multiple edge servers, Each of the local IoT devices is respectively connected to the satellite cloud and the corresponding at least one edge server in communication. Among them, the local IoT device U i Received computing task T i，j The process being processed includes the following steps: Step S1: collect the local IoT devices U i , and the local IoT device U i The resource usage of each of the corresponding edge servers and the satellite cloud; Step S2: the local IoT device U i The resource usage of the local IoT device U i The resource usage of all the corresponding edge servers and the satellite cloud is used as the current resource usage of other non-self-devices; Step S3: the computing task T i，j The task attributes, the current device's own resource usage and the current other non-self device resource usage are used as input state space; Step S4, inputting the input state space into the offloading object decision model to obtain the corresponding current task offloading target; Step S5, using the current task offloading target to process the computing task T i，j , The task types include delay-sensitive, computation-intensive and delay-tolerant. The current task offloading target is the local IoT device U i Or the edge server or the satellite cloud.

2. The method for adaptively offloading multi-task computing for satellite IoT according to claim 1, characterized in that: in, The offloading object decision model is obtained by solving the objective function corresponding to the problem of adaptive computing offloading strategy for multi-type tasks oriented to satellite Internet of Things using DDPG algorithm. The expression of the objective function is: P1: Minimize D, stC1: X i，j ={x i，j ,y i，j ,z i，j },x i，j ∈{0,1},y i，j ∈{0,1},z i，j ∈{0,1}, C2： C3： Where X i，j For the computation task T i，j The corresponding uninstallation decision, x i，j For the computation task T i，j Is it in the corresponding local IoT device U i Execute, y i，j For the computation task T i，j Whether to offload to the edge server for execution, z i，j For the computation task T i，j Whether to offload to the satellite cloud for execution, For local IoT devices U i Processing computing task T i，j The computational delay generated, To offload the latency to the edge server, For the computation task T i，j The delay caused by offloading to the satellite cloud, ρ i，j For a given computing task T i，j The time interval from generation to execution completion, I is the total number of local IoT devices in the local IoT device group, J is the total number of computing tasks generated in the local IoT device group, K is the total number of edge servers in the ground edge server group, T i，j (f k )=1 to ensure that each computing task is assigned to only one edge server for processing.

3. The method for adaptively offloading multi-task computing for satellite IoT according to claim 2, characterized in that: in, In the DDPG algorithm solution, the reward function is expressed as: In the formula, reward i，j For the computation task T i，j The reward value, D i，j For the computation task T i，j The cost, ρ i，j For the computation task T i，j Processing maximum tolerable delay, d i，j According to the calculation task T i，j The actual delay of the task after taking the corresponding action.

4. The method for adaptively offloading multi-task computing for satellite IoT according to claim 3, characterized in that: in, Cost D i，j The calculation expression is: Where α is the preset weight value, For local IoT devices U i Processing computing task T i，j Energy consumption, κ is the energy factor, For local IoT devices U i The computing power of c i，j To process the computing task T i，j The number of CPU cycles required per bit, To process the computing task T i，j The data size, For the computation task T i，j Energy consumption of offloading to edge servers, is the computing power of the edge server, R e For local IoT device U i The transmission rate to the edge server, P e To offload the transmission power to the edge server, The computing power of satellite cloud, R s For local IoT device U i Transmission rate to the satellite cloud, P s is the transmission power offloaded to the satellite cloud.

5. The method for adaptively offloading multi-task computing for the satellite Internet of Things according to claim 4 is characterized in that: in, Calculating latency The expression is: In the formula For local IoT devices U i Processing computing task T i，j The resulting queuing delay, Latency The expression is: In the formula For the computation task T i，j The queuing delay waiting to be executed on the edge server, Latency The expression is:

6. The method for adaptively offloading multi-task computing for satellite IoT according to claim 4, characterized in that: in, The local IoT device is connected to the satellite cloud through a gateway station. Transmission rate R s The expression is: Where ω s is the bandwidth of the satellite-to-ground link, σ is the noise power, Transmission rate R e The expression is: Where ω e is the bandwidth of the terrestrial wireless link.

7. The method for adaptively offloading multi-task computing for satellite IoT according to claim 2, characterized in that: in, In the DDPG algorithm solution, according to the preset decay rate N, the maximum value of the exploration factor ε max and the minimum value of the exploration factor ε min And the current number of training rounds n, calculate the exploration factor ε corresponding to the current number of training rounds n, The calculation expression of the exploration factor ε is:

8. A satellite Internet of Things system, characterized in that: include: Local IoT device clusters, ground edge server clusters, and satellite clouds are all used to process computing tasks. The local IoT device group includes multiple local IoT devices and a decision module. The ground edge server group includes multiple edge servers. The satellite cloud comprises a satellite cloud, Each of the local IoT devices is respectively connected to the satellite cloud and the corresponding at least one edge server in communication. Wherein, the decision module includes: A data collection unit is used to collect resource usage of the local IoT device, each of the edge servers corresponding to the local IoT device, and the satellite cloud; An input state space generation unit is configured to construct an input state space based on the task attributes of the computing task received by the local IoT device, the resource usage of the local IoT device corresponding to the computing task, and the resource usage of each of the edge servers and the satellite cloud corresponding to the local IoT device corresponding to the computing task; A decision generating unit, comprising an offloading object decision model, is configured to input the input state space into the offloading object decision model to obtain a corresponding current task offloading target; A decision execution unit, configured to control the current task offloading target to process the computing task, The task types include delay-sensitive, computation-intensive and delay-tolerant. The current task offloading target is the local IoT device corresponding to the computing task, the edge server, or the satellite cloud.

Images