Heterogeneous task collaborative unloading method for satellite Internet of Things

Abstract

The invention belongs to the technical field of satellite communication, and particularly relates to a satellite Internet of Things oriented heterogeneous task collaborative unloading method, which comprises the following steps: S1, establishing communication connection between a ground user and a low-orbit satellite, and uploading a task to be processed; s2, the low-orbit satellite judges whether the task can be locally processed in time or not, if yes, the task is added into a local processing queue, and the step S3 is executed; otherwise, adding a task unloading queue, and executing the step S4; s3, the low-orbit satellite sorts the tasks in the local processing queue, and after sorting is completed, the tasks which cannot be locally processed in time are transferred to a task unloading queue; s4, the low-orbit satellite constructs a task unloading optimization problem according to the local processing queue and the task unloading queue, the task unloading optimization problem is solved, and an optimal task unloading strategy is obtained; according to the method, the local task queue is optimized through the deadline perceived emergency priority strategy, the emergency task is ensured to be processed preferentially, and the probability that the task misses the deadline is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite communication, and particularly relates to a heterogeneous task collaborative offloading method for satellite Internet of Things. Background Art

[0002] With the rapid development of satellite Internet of Things, low-earth orbit satellites have become the core nodes to support the access of a large number of terminals due to their characteristics of low latency and high bandwidth. However, tasks in satellite networks exhibit significant heterogeneous characteristics, that is, different tasks have different data sizes, computing requirements, and deadline requirements. The traditional single task processing mode is difficult to meet the diverse quality of service requirements. For example, real-time tasks require extremely low processing latency, while compute-intensive tasks place higher demands on satellite computing resources. In addition, the collaborative offloading of heterogeneous tasks in a dynamic inter-satellite link (such as OISL) environment involves multi-dimensional coupled optimization of link status, resource load, and task priority. Existing strategies lack the fine-grained perception and intelligent decision-making ability for task type differences. Summary of the Invention

[0003] To solve the above problems of the prior art, the present invention adopts a heterogeneous task collaborative offloading method for satellite Internet of Things, which is characterized by including:

[0004] S1: A ground terminal user establishes a communication connection with a low-earth orbit satellite and uploads a task to be processed to the corresponding low-earth orbit satellite; each low-earth orbit satellite is provided with a local processing queue and a task offloading queue;

[0005] S2: The low-earth orbit satellite determines whether the currently received task can be processed locally in a timely manner. If so, the currently received task is added to its own local processing queue, and step S3 is executed; otherwise, the currently received task is added to its own task offloading queue, and step S4 is executed;

[0006] S3: The low-earth orbit satellite sorts the tasks in its own local processing queue. After the sorting is completed, the tasks that cannot be processed locally in a timely manner in the local processing queue are transferred to the task offloading queue;

[0007] S4: The low-earth orbit satellite constructs a task offloading optimization problem based on the local processing queue and the task offloading queue;

[0008] S5: The low-earth orbit satellite solves the task offloading optimization problem according to the reinforcement learning algorithm to obtain the optimal task offloading strategy.

[0009] Advantageous Effects:

[0010] 1. The present invention relies on the multi-node cooperation of low-earth orbit satellites and ground cloud centers to dynamically allocate computing tasks, balance the computing resource load, and improve the overall resource utilization rate. 2. The present invention optimizes the local task queue through an emergency priority strategy with deadline awareness to ensure that emergency tasks are processed first and reduce the probability of tasks missing the deadline. 3. The present invention constructs a task offloading optimization problem based on the current local processing queue and task offloading queue, and through deep reinforcement learning, combines task characteristics and link status to adaptively select the optimal offloading strategy, supports differential processing of heterogeneous tasks, and meets the low-latency requirements of real-time tasks and the resource requirements of compute-intensive tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the system model of the satellite Internet of Things provided by an embodiment of the present invention;

[0012] Figure 2 It is a flowchart of a heterogeneous task cooperative offloading method for satellite Internet of Things provided by an embodiment of the present invention;

[0013] Figure 3 It is a flowchart of the DNUF strategy provided by an embodiment of the present invention;

[0014] Figure 4 It is a structural diagram of a heterogeneous task cooperative offloading method for satellite Internet of Things provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] The framework of the satellite system model is as Figure 1 shown. The entire system consists of K low-earth orbit satellites, M ground terminal users, 1 geostationary orbit satellite, and 1 ground cloud center; among them, there is two-way communication between the ground terminal users and the low-earth orbit satellites, two-way communication between the low-earth orbit satellites through OISL, and a communication link is constructed between the low-earth orbit satellites and the ground cloud center with the help of the geostationary orbit satellite to realize two-way data transmission and information interaction.

[0017] Based on the above communication scenario, the embodiment of the present invention adopts a heterogeneous task cooperative offloading method for satellite Internet of Things, as Figure 2 、 Figure 4 shown, including:

[0018] S1: The ground terminal user establishes a communication connection with the corresponding low Earth orbit (LEO) satellite and uploads the task to be processed to the corresponding LEO satellite.

[0019] The process of the user uploading the task includes:

[0020] To support simultaneous access of multiple terminals and avoid channel interference among multiple users, the user adopts the access mode of Frequency Division Multiple Access (FDMA). Each terminal is assigned an independent uplink frequency band resource to establish a reliable uplink communication link with the LEO satellite. For user m, its uplink bandwidth is B m , the channel gain is h m , the transmit power is P m , and the noise power spectral density is N0. Then its uplink data rate B m can be approximately expressed as:

[0021]

[0022] Then the time i required for the ground terminal user m to upload a task with a data size of D to the LEO satellite k is:

[0023]

[0024] Since the LEO satellite is relatively close to the Earth's surface, the wireless signal propagation delay is small. Therefore, the main time consumption of the uplink transmission comes from the data sending duration.

[0025] In addition, the ground user includes the following key parameters with clear characteristics of task i in the uploaded task i: the task data volume size D i , the number of CPU cycles C required for the task i and the task deadline

[0026] Each LEO satellite maintains two queues, one is the local processing queue and the other is the task offloading queue.

[0027] S2: The LEO satellite determines whether the currently received task can be processed locally in a timely manner. If so, it adds the currently received task to its own local processing queue and executes step S3; otherwise, it adds the currently received task to its own task offloading queue and executes step S4.

[0028] The LEO satellite's determination of whether the currently received task can be processed locally in a timely manner includes:

[0029] The LEO satellite calculates the total computing duration of all tasks in its local processing queue and the currently received task, and determines whether the total computing duration is greater than the deadline of the currently received task. If so, it cannot be processed locally in time; otherwise, it can be processed locally in time.

[0030] Specifically, the formula for determining whether task i will miss the deadline when processed locally is as follows:

[0031]

[0032] represents the requirement of the currently received task i, that is, the deadline of task i, f LEO represents the number of CPU cycles that the LEO satellite processor can provide per second, represents the local processing queue Q local the total computing duration of all tasks in it, and j is the index of the task in the local processing queue.

[0033] S3: The LEO satellite sorts the tasks in its local processing queue. After the sorting is completed, the tasks that cannot be processed locally in time in the local processing queue are transferred to the task offloading queue;

[0034] Such as Figure 3 shown, the specific steps for the LEO satellite to sort the tasks in the local processing queue include:

[0035] When a task is added to the local processing queue of the LEO satellite, the satellite will adopt the deadline-aware normalized urgency first (DNUF) strategy proposed in the present invention to manage the local processing queue, and give priority to processing more urgent tasks according to the deadline; specifically, calculate the urgency Δ of each task in the local processing queue i , defined as follows:

[0036]

[0037] Among them, is the total computing duration of task i and all tasks before it, Δ i represents the difference between the deadline and the expected processing time of the i-th task in the local processing queue, Δ i The smaller it is, the more urgent the task is.

[0038] At the same time, normalize the urgency of all tasks to obtain the normalized urgency

[0039]

[0040] Among them, Δ min, Δ max is the maximum and minimum urgency in the local processing queue, and ∈ is a small constant to prevent division by zero.

[0041] Then the DNUF priority index DNUF i is expressed as:

[0042]

[0043] Sort the tasks in the local processing queue according to the DNUF priority index.

[0044] Transfer the tasks in the local processing queue that cannot be processed locally in time to the task offloading queue, including:

[0045] Judge whether the tasks in the local processing queue can be processed locally in time according to the formula (3) in step S2. If not, transfer the tasks that cannot be processed locally in time to the task offloading queue to obtain the final local processing queue and task offloading queue. Specifically, for task i in the local processing queue, calculate the total computing duration of task i and all tasks before it in the local processing queue. If the deadline of task i is less than the total computing duration, then task i cannot be processed in time; otherwise, it can be processed in time.

[0046] S4: The low-earth orbit satellite constructs a task offloading optimization problem based on the current local processing queue and task offloading queue;

[0047] Constructing the task offloading optimization problem includes:

[0048] S41: The low-earth orbit satellite collects the resource status of its one-hop and two-hop neighboring low-earth orbit satellites and the link characteristics with the neighboring low-earth orbit satellites, and determines the set of cooperative low-earth orbit satellites

[0049] Specifically, the low-earth orbit satellite broadcasts link status packets through OISL (inter-satellite optical communication link) to discover neighboring low-earth orbit satellites within one-hop and two-hop ranges and constructs an inter-satellite connection graph where is the low-earth orbit satellite node, and ε is the connected edge.

[0050] For each neighbor satellite k′ of the low-earth orbit satellite k, it is necessary to calculate its remaining available computing resource R k′ , and the formula is as follows:

[0051]

[0052] where f k′ is the maximum computing capacity of satellite k′ (such as the number of CPU cycles per second), Q k′ is the local processing queue of neighbor satellite k′, and C jIs the computing requirement of the tasks in the queue.

[0053] The LEO satellite also needs to send detection packets to measure the link characteristics with the neighboring satellite k'. The link characteristics include: link bandwidth, link transmission delay, and link channel quality.

[0054] The inter-satellite link bandwidth B k,k′ Is:

[0055]

[0056] Where P k,k′ Is the transmit power, h k,k′ Is the link gain, and N0 is the noise power spectral density.

[0057] The inter-satellite link transmission delay τ k,k′ Is:

[0058]

[0059] Where D k,k′ Is the physical distance between satellites k and k', and c is the speed of light.

[0060] The inter-satellite link channel quality SNR k,k′ Is:

[0061]

[0062] If the neighboring LEO satellite k' meets the following conditions, it is added to the set of cooperative LEO satellites

[0063]

[0064] Where Is the minimum link bandwidth, η tx Is the transmission time ratio coefficient, set to 0.2; Is the maximum link transmission delay, η τ Is the delay ratio coefficient, where η for a single-hop link τ = 0.15, and η for a two-hop link τ = 0.3; Is the minimum link channel quality, and B is the available link bandwidth.

[0065] S42: Calculate the computing delay of the LEO satellite k for locally processing task i, the computing delay of offloading task i to the cooperative satellite k', and the computing delay of offloading task i to the cloud center c;

[0066] If satellite k chooses to locally process the computing task i, the resulting computing delay Is:

[0067]

[0068] Among them, f k is the calculated frequency of the low-earth orbit satellite k, C i / f k is the calculation time of task i on satellite k, is the total waiting time of all tasks in the queue.

[0069] If satellite k cannot process task i and chooses to offload it to the cooperative satellite k′, the resulting computational delay is:

[0070]

[0071] Among them, f k′ is the calculated frequency of satellite k′, B k,k′ is the link bandwidth from the low-earth orbit satellite k to the cooperative satellite k′, τ k,k′ is the link transmission delay from satellite k to the cooperative satellite k′.

[0072] If satellite k cannot process task i and chooses to offload it to the cloud center c, the resulting computational delay is:

[0073]

[0074] Among them, B k,g is the link bandwidth from the low-earth orbit satellite k to the geostationary orbit satellite g, B g,c is the link bandwidth from the geostationary orbit satellite g to the cloud center c, f c is the calculated frequency of the cloud center c, τ k,g is the link transmission delay from the low-earth orbit satellite k to the geostationary orbit satellite g, τ g,c is the link transmission delay from the geostationary orbit satellite g to the cloud center c.

[0075] S43: Construct an optimization problem for task offloading according to the computational delay ;

[0076] The optimization problem for task offloading is:

[0077]

[0078] Among them, C1 is the constraint of the offloading strategy, indicating that only one offloading strategy can be selected. C2 is the constraint of the feasibility of satellite local processing. If a task is selected for local processing, the total computing time of all tasks (including this task) in the local queue cannot exceed its deadline. C3 is the constraint of the feasibility of cooperative satellite processing. If a task is selected to be offloaded to a cooperative satellite, the total computing time of the local processing queue of the cooperative satellite needs to meet the deadline requirement of this task. C4 is the constraint of the transmission delay of the cooperative satellite. The time for task data to be transmitted from the current satellite to the cooperative satellite plus the inter-satellite link transmission delay must be less than or equal to the task deadline. C5 is the constraint of the transmission delay of the cloud center. The cascaded transmission time for task data to be transmitted to the cloud center needs to meet the deadline requirement. C6 is the constraint of the offloading strategy weight, indicating that the sum of the weights of the computing time delays equals 1, x i,k is the decision to process task i locally, y i,k′ is the decision to offload task i to cooperative satellite k′, z i,c is the decision to offload task i to cloud center c, is the set of cooperative low-earth orbit satellites of low-earth orbit satellite k, α, β k′ , γ are weights, N is the number of tasks in the task offloading queue of low-earth orbit satellite k, and i is the index of the task in the task offloading queue of low-earth orbit satellite k.

[0079] S5: The low-earth orbit satellite solves the task offloading optimization problem according to the reinforcement learning algorithm to obtain the optimal task offloading strategy;

[0080] The low-earth orbit satellite solving the task offloading optimization problem includes:

[0081] Converting the task offloading optimization problem into a Markov decision process to obtain the state space, action space, and reward function;

[0082] The state space S = {s1, s2, …, s K}, and the state of satellite k Among them, D i , C i , respectively represent the data size, computing requirement, and deadline of task i received by low-earth orbit satellite k. ρ k , |Q k | respectively represent the computing resource idle degree and local processing queue length of low-earth orbit satellite k. f k , |Q k |, respectively represent the computing resource, local processing queue length, and the total number of CPU cycles required for all tasks in the local processing queue of cooperative satellite k′. B k,k′ , τ k,k′ ,, SNRk,k′ Indicates the link bandwidth, latency, and quality between the access satellite k and each cooperative satellite k'.

[0084] Action space The action a of satellite k k = {x i,k , y i,1 , y i,2 , …, y i,L , z i,c}, where {y i,1 , y i,2 , …, y i,L} represents the decision to offload task i to L different cooperative satellites k'.

[0085] Reward function: The purpose of the reward is to evaluate the quality of the action taken in the current time slot. The reward function is defined as follows:

[0086]

[0087] Use the DQN algorithm to solve the task offloading optimization problem based on the state space, action space, and reward function, and obtain the optimal task offloading strategy.

[0088] With the accumulation of positive and negative feedback of a large number of task decisions, the DQN agent will converge to an approximately optimal offloading decision strategy and adaptively select the best execution node for tasks in different network states.

[0089] After obtaining the optimal task offloading strategy, the task can be offloaded to the corresponding device for processing according to the optimal task offloading strategy. After the processing is completed, the device will return the processing result to the ground terminal user who initiated the task, completing a collaborative task offloading service.

[0090] If the task is completed locally on the access satellite or processed by the cooperative satellite, the satellite will directly send the result to the corresponding terminal user through the downlink. Downlink communication can adopt a frequency division multiple access or time division multiple access method symmetrical to the uplink to ensure that different terminals receive without interference during downlink reception.

[0091] Downlink transmission time is:

[0092]

[0093] where is the size of the result data, B k,m is the downlink bandwidth from satellite k to terminal m, and τ k,m is the link transmission delay.

[0094] If the task is processed by the ground cloud center, the task result will be forwarded to the terminal by the GEO satellite, and the return delay is:

[0095]

[0096] wherein, is the size of the result data, in B c,g is the downlink bandwidth of the transmission delay from the cloud center to GEO, τ c,g and τ g,m is the link transmission delay.

[0097] In the above-described embodiments, the object, technical solution, and advantages of the present invention have been further described in detail. It should be understood that the above-described embodiments are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made to the present invention within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for collaborative offloading of heterogeneous tasks for satellite Internet of Things, characterized in that: include: S1: The ground terminal user establishes a communication connection with the low-orbit satellite and uploads the tasks to be processed to the corresponding low-orbit satellite; each low-orbit satellite has a local processing queue and a task offloading queue; S2: The low-orbit satellite determines whether the currently received task can be processed locally in a timely manner. If so, the currently received task is added to its own local processing queue and step S3 is executed; otherwise, the currently received task is added to its own task unloading queue and step S4 is executed; S3: The low-orbit satellite sorts the tasks in its local processing queue. After the sorting is completed, the tasks in the local processing queue that cannot be processed locally in time are transferred to the task offloading queue; S4: The low-orbit satellite constructs the task offloading optimization problem based on the local processing queue and the task offloading queue; S5: The low-orbit satellite solves the task offloading optimization problem based on the reinforcement learning algorithm and obtains the optimal task offloading strategy.

2. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 1 is characterized in that: The low-orbit satellite determines whether the currently received task can be processed locally in a timely manner, including: the low-orbit satellite calculates the total calculation time of all tasks in its own local processing queue and the currently received task, and determines whether the total calculation time is greater than the deadline of the currently received task. If so, it cannot be processed locally in a timely manner; otherwise, it can be processed locally in a timely manner.

3. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 1 is characterized in that: The low-orbit satellite sorts the tasks in its local processing queue including: The urgency of each task in the local processing queue is calculated, the urgency of each task is normalized, the priority index of the task is calculated according to the normalized urgency of the task, and the tasks in the local processing queue are sorted according to the priority index.

4. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 3 is characterized in that: Calculating the urgency of task i in the local processing queue includes: calculating the total computing time of task i and all previous tasks in the local processing queue, subtracting the total computing time from the deadline of task i, and obtaining the urgency of task i.

5. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 1 is characterized in that: The low-orbit satellite constructs task offloading optimization problems based on the local processing queue and task offloading queue, including: Among them, C1 is the constraint of the offloading strategy, C2 is the constraint of the feasibility of local processing of low-orbit satellites, C3 is the constraint of the feasibility of processing of cooperative low-orbit satellites, C4 is the constraint of transmission delay of cooperative low-orbit satellites, C5 is the constraint of transmission delay of cloud center, C6 is the constraint of offloading strategy weight, N is the number of tasks in the task offloading queue of low-orbit satellite k, i is the index of the task in the task offloading queue of low-orbit satellite k, is the computational delay of low-orbit satellite k processing task i locally, is the computational delay of offloading task i to the cooperating satellite k′, is the computational latency of offloading task i to cloud center c, x i,k is the decision to process task i locally, y i,k′ is the decision to offload task i to the cooperative satellite k′, z i,c The decision to offload task i to cloud center c is: is the set of cooperative LEO satellites of LEO satellite k, α, β k′ , γ is the weight, C i is the number of CPU cycles required for task i, D i is the data size of task i, f k 、f k′ are the calculation frequencies of low-orbit satellite k and its cooperative satellite k′, is the deadline of task i, Q k , Q k′ are the task offloading queues of low-orbit satellite k and its cooperative satellite k′, respectively. k,k′ is the link bandwidth from low-orbit satellite k to cooperative satellite k′, B k,g is the link bandwidth from low-orbit satellite k to geostationary satellite g, B g,c is the link bandwidth from the geostationary satellite g to the cloud center c, τ k,k′ is the link transmission delay from low-orbit satellite k to cooperative satellite k′, τ k,g is the link transmission delay from the low-orbit satellite k of access mission i to the geostationary orbit satellite g, τ g,c is the link transmission delay from the geostationary orbit satellite g to the cloud center c.

6. The method for collaborative unloading of heterogeneous tasks for satellite Internet of Things according to claim 5 is characterized in that: Get the cooperative low-orbit satellite set of low-orbit satellite k include: The low-orbit satellite k discovers the neighboring low-orbit satellite k′ within one hop and two hops by broadcasting link status packets; Calculate the remaining available computing resources R of each neighboring low-orbit satellite k′ k′ and the link bandwidth B between LEO satellite k and its neighbor LEO satellite k′ k,k′ , link transmission delay τ k,k′ And the link channel quality SNR k,k′ ; According to the remaining available computing resources R k′ , link bandwidth B k,k′ , link transmission delay τ k,k′ And the link channel quality SNR k,k′ A cooperative low-orbit satellite is selected from the neighboring low-orbit satellites k′ of the low-orbit satellite k to obtain a cooperative low-orbit satellite set of the low-orbit satellite k.

7. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 5, characterized in that: The computational delay of low-orbit satellite k in local processing of task i is:

8. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 5, characterized in that: The computational delay of offloading task i to the cooperative satellite k′ is:

9. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 5, characterized in that: The computational delay of offloading task i to cloud center c is:

10. The method for collaborative offloading of heterogeneous tasks for satellite Internet of Things according to claim 5, characterized in that: The low-orbit satellite solves the task offloading optimization problem according to the reinforcement learning algorithm, including: converting the task offloading optimization problem into a Markov decision process to obtain the state space, action space and reward function; The reinforcement learning algorithm is used to solve the task offloading optimization problem based on the state space, action space and reward function to obtain the optimal task offloading strategy.

Images