A task-priority-based satellite-ground network end-edge cloud collaborative computing offloading method

Abstract

The application claims a task priority-based edge cloud collaborative computing offloading method for satellite-ground network, belonging to the technical field of wireless communication. In view of the problem that the unreasonable task priority setting caused by the unbalanced resource allocation of satellite-ground network leads to high task delay and low completion rate, a task priority-based edge cloud collaborative computing offloading method for satellite-ground network is proposed. First, an edge cloud collaborative computing framework considering idle devices is built. Second, a queue model is introduced, and a task priority scheduling algorithm based on remaining time is designed. Then, under the constraints of task maximum tolerance delay and communication resources, an edge cloud collaborative computing offloading problem model minimizing the total system delay is built, and the problem is decomposed into an offloading decision sub-problem and a communication resource allocation sub-problem. Finally, an improved grey wolf algorithm is proposed to solve the offloading decision, and a particle swarm algorithm is used to solve the communication resource allocation scheme, so as to obtain the edge cloud collaborative computing offloading scheme minimizing the total system delay.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology. Specifically, it relates to a method for offloading edge-cloud collaborative computing in satellite-ground networks based on task priority. Background Technology

[0002] Space-Ground converged networks, leveraging the complementary advantages of wide-area satellite coverage and high-bandwidth terrestrial networks, have become the core architecture for global communication and heterogeneous computing task processing. Computation offloading, a key technology in space-ground converged networks, effectively addresses issues such as insufficient terrestrial computing resources and lack of service coverage in remote areas by rationally offloading computing tasks from ground terminals or edge nodes to satellite nodes and cloud centers. However, the dynamic topology characteristics of low-Earth orbit satellites, the bandwidth and latency constraints of inter-satellite links, and the heterogeneity of computing tasks in terms of latency sensitivity, priority, and resource requirements make traditional computation offloading schemes difficult to adapt to space-ground converged network scenarios. High-priority tasks are prone to execution timeouts due to insufficient resource preemption, while the ineffective occupation of low-priority tasks wastes on-board computing power, severely restricting the overall efficiency and service quality of computation offloading. Against this backdrop, designing priority scheduling mechanisms based on task characteristics and formulating offloading strategies and resource allocation schemes that conform to network scenarios have become key paths to improve network resource utilization efficiency and system service quality.

[0003] Most current patents prioritize task scheduling based on task attributes. Patent CN121560419A proposes a dynamic priority function that accurately measures the actual urgency level of a task at different times, thus overcoming the shortcomings of the "static priority" model in handling real-time changes in emergency situations. Patent CN121614272A places tasks into a waiting queue according to their value priority within the task execution cycle, and the server base station determines whether to execute locally or offload to the task device based on the task value. However, under the hard constraint of the maximum tolerable latency of a task, some tasks still have the problem of timeout execution. To address this issue, this patent proposes a task priority scheduling algorithm based on deadlines, which reduces queuing latency by increasing the priority of timeout tasks, thereby increasing the task completion rate. In addition, this patent proposes a three-layer edge-cloud collaborative computing framework that considers idle equipment, which reduces the offloading of some tasks to LEO satellites by making full use of the computing resources of idle ground equipment, thereby reducing the overall system latency.

[0004] To address these issues, this invention seeks protection for a task priority-based end-edge-cloud collaborative computing offloading method for satellite-ground networks. Addressing the problem of unreasonable task priority settings caused by unbalanced allocation of satellite-ground network resources, resulting in high task latency and low completion rates, this invention proposes a task priority-based end-edge-cloud collaborative computing offloading method. First, an end-edge-cloud collaborative computing framework considering idle devices is established. Second, a queue model is introduced, and a task priority scheduling algorithm based on remaining time is designed. Then, under the constraints of maximum tolerable task latency and communication resources, an end-edge-cloud collaborative computing offloading problem model that minimizes the total system latency is constructed, decomposing the problem into offloading decision sub-problems and communication resource allocation sub-problems. Finally, an improved gray wolf algorithm is proposed to solve the offloading decision, and a particle swarm optimization algorithm is used to solve the communication resource allocation scheme, thereby obtaining an end-edge-cloud collaborative computing offloading scheme that minimizes the total system latency. Summary of the Invention

[0005] This invention aims to solve the problems of the prior art. It proposes a task-priority-based method for edge-cloud collaborative computing offloading in satellite-ground networks. The technical solution of this invention is as follows:

[0006] A method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority includes the following steps:

[0007] S1: Build a satellite-ground integrated network edge-cloud collaborative computing offloading model that takes into account idle equipment, and calculate the transmission rate of the communication link between each node;

[0008] S2: Establish a queue model and design a task priority scheduling algorithm based on remaining time; establish a latency model and calculate the latency of task unloading to each network node;

[0009] S3: Construct an edge-cloud collaborative computing offloading problem model that minimizes the total system latency, and decompose the problem into an offloading decision subproblem and a communication resource allocation subproblem;

[0010] S4: An improved gray wolf algorithm is proposed to solve the offloading decision, and the particle swarm algorithm is used to solve the communication resource allocation scheme, thereby obtaining an end-edge-cloud collaborative computing offloading scheme that minimizes the total system latency.

[0011] Furthermore, in step S1, a satellite-ground fusion network edge-cloud collaborative computing offloading model considering idle equipment is established to calculate the transmission rate of the communication link between each node, specifically including:

[0012] (1) Network model;

[0013] The satellite-ground integrated network backhaul architecture is mainly composed of user terminals, idle equipment, base stations (BS) equipped with edge servers, low-Earth orbit (LEO) satellites equipped with edge servers, and a cloud computing center, which together construct an end-edge-cloud collaborative computing architecture.

[0014] (2) Construct a channel model and calculate the transmission rate;

[0015] In a space-ground converged network, three main channel models are considered: the terrestrial channel model, the space-ground channel model, and the inter-satellite channel model. The terrestrial channel model is used for communication between users, idle equipment, and the BS; the space-ground channel model is used for communication between users and LEO satellites, and between LEO satellites and cloud computing centers; and the inter-satellite channel model is used for communication between LEO satellites.

[0016] Furthermore, the three channel models are as follows:

[0017] 1) Ground channel model:

[0018] In terrestrial communications, non-line-of-sight transmission dominates, and its channels are modeled as Rayleigh channels; users With BS The transmission rate between them is given by the following formula:

[0019]

[0020] in, BS Assigned to user bandwidth; User The transmission power; This represents the Gaussian white noise power of the link; User With BS Channel gain between , BS With users The distance; The path loss index; Represents a Rayleigh random variable;

[0021] Similarly, users With idle equipment The transmission rate between them is given by the following formula:

[0022]

[0023] in, Indicates idle equipment Assigned to user bandwidth; User The transmission power;

[0024] 2) Satellite-to-Ground Channel Model:

[0025] In satellite-to-ground communication, line-of-sight transmission dominates. Users communicate with LEO satellites via Ka-band satellite-to-ground links, and the corresponding channels are modeled as Ricean channels. With LEO satellite The transmission rate between them is expressed as:

[0026]

[0027] in, Indicates LEO satellite Assigned to user bandwidth; Indicates user The transmission power; Indicates user Antenna gain; Indicates LEO satellite The receiving antenna gain; Indicates the channel gain of the link;

[0028] When a user offloads a task to the cloud computing center, the task needs to be offloaded to a LEO satellite first, and then forwarded by the LEO satellite to the ground-based cloud computing center; therefore, the uplink transmission rate from user n to LEO satellite i is shown in equation (3), and the downlink transmission rate from LEO satellite i to the cloud computing center is expressed as:

[0029]

[0030] in, This indicates that the cloud computing center is allocated to LEO satellites. bandwidth; Indicates LEO satellite The transmission power; Indicates LEO satellite Antenna gain; Indicates the receiving antenna gain of the cloud computing center; Indicates the channel gain of the link;

[0031] 3) Inter-satellite channel model

[0032] Inter-satellite links are primarily for line-of-sight transmission; therefore, the transmission rate of an inter-satellite link is expressed as:

[0033]

[0034] in, This represents the bandwidth allocated to LEO satellite k by LEO satellite i; This indicates the transmission power of LEO satellite k; This represents the antenna gain of LEO satellite k; Indicates the receiving antenna gain of LEO satellite i; This indicates the channel gain of the link.

[0035] Furthermore, in step S2, a queue model is established, and a task priority scheduling algorithm based on remaining time is designed; a latency model is established to calculate the latency of task offloading to each network node, as detailed below:

[0036] (1) Queue model:

[0037] The queuing model follows the first-come, first-served principle, with task queues on idle equipment, BS, and LEO satellites represented as follows: , and For time slots Arrival of idle equipment The computational task can be performed in time slots. Internal processing, while time slots The task that has been completed requires waiting for a time slot. Arrival of idle equipment After the computational task is completed, the idle equipment is processed. The queue is represented as:

[0038]

[0039] in, Indicates arrival at idle equipment The unprocessed number For each task, the queuing delay can be expressed as:

[0040]

[0041] in, For idle equipment No. One unprocessed task; Indicates idle equipment Computing resources;

[0042] Similarly, for BS and LEO satellites, The queuing delay can be expressed as follows:

[0043]

[0044]

[0045] in, and They represent BS respectively Computing resources and LEO satellites Computing resources;

[0046] The task priority scheduling algorithm based on remaining time runs on each offload node, including idle equipment, BS and LEO satellites; the remaining time in the algorithm is represented as the maximum tolerable delay of the task minus the task transmission delay and propagation delay.

[0047] (2) Delay model:

[0048] 1) Local computation:

[0049] Local computing only needs to consider the latency of task processing, therefore user devices The latency for local task processing is:

[0050]

[0051] in, This indicates the proportion of local computing tasks; Indicates user equipment Computing resources; Indicates the size of the task data; Indicates the calculated strength;

[0052] 2) Unload to an idle device:

[0053] When users offload tasks to idle devices, the latency incurred in completing these tasks consists of three parts: the latency of user task transmission, the task queuing latency, and the latency of idle device task processing; Offload the task to an idle device The delay is:

[0054]

[0055] in, This indicates that the task will be unloaded to an idle device. The proportion; Indicates idle equipment Computing resources; This indicates a queuing delay, among which Indicates unloading to an idle device The order of tasks in the task queue;

[0056] 3) Uninstall to BS:

[0057] When a user offloads a task to the BS, only the latency of the user transmitting the task, the latency of the BS processing the task, and the latency of the task queuing need to be considered; Unload the task to BS The delay is:

[0058]

[0059] in, This indicates that the task is unloaded to the BS. The proportion; BS Computing resources; Indicates queuing delay;

[0060] 4) Unload to LEO satellite:

[0061] user Offload the mission to the LEO satellite The delay is:

[0062]

[0063] in, This indicates that the mission is offloaded to the LEO satellite. The proportion; Indicates LEO satellite Computing resources; , Representing users respectively With LEO satellite , The distance; , They represent LEO satellites. , With LEO satellite The distance; Represents the speed of light;

[0064] 5) Unload to the cloud computing center:

[0065] user The latency for offloading tasks to the cloud computing center is:

[0066]

[0067] in, This indicates the proportion of tasks that are offloaded to the cloud computing center; Indicates user n and LEO satellite The distance; Indicates LEO satellite Distance to the cloud computing center; This refers to the computing resources of a cloud computing center;

[0068] 6) Total system latency:

[0069] user The total latency of edge-cloud collaborative offloading can be expressed as:

[0070]

[0071] The total system delay can be expressed as:

[0072] .

[0073] Furthermore, the task priority scheduling algorithm based on remaining time is run on each unloading node, and the specific process is as follows: First, based on the unloading decision matrix... and communication resource allocation matrix Computation task queue The transmission and propagation delays for each task are calculated; then, the maximum tolerable delay for each task is determined. Calculate the task queue Remaining time for each task Finally, iterate through each task; if a task times out, i.e., the current queuing delay... and computational delay The sum is greater than the remaining time If the task does not time out, then iterate through the previous tasks; if the task does not time out after swapping the priorities of the previous and current tasks, then swap the priorities of the two tasks and stop the current loop; if the task still times out after iterating through the previous tasks, then place the task at the end of the task queue; finally, after iterating through all tasks, output the task queue.

[0074] Furthermore, step S3 constructs a model for the edge-cloud collaborative computing offloading problem that minimizes the total system latency, decomposing the problem into an offloading decision sub-problem and a communication resource allocation sub-problem, as detailed below:

[0075] (1) Construct a model for the edge-cloud collaborative computing offloading problem that minimizes the total system latency.

[0076] Under the constraints of maximum tolerable latency, partial offloading, and communication resources, an edge-cloud collaborative offloading problem model that minimizes the total system latency was constructed; and a user offloading decision matrix was established. , Indicates user Offloading decision; communication resource allocation matrix ,in, This represents the communication resource allocation matrix for idle devices. This represents the communication resource allocation matrix of the B and C. This represents the communication resource allocation matrix for LEO satellites. Let represent the communication resource allocation matrix of the cloud computing center. Then, the edge-cloud collaborative computing offloading problem model can be expressed as equation (17);

[0077] As shown in equation (17), the constraints in the problem are as follows: Indicates user The maximum tolerable latency for the task that generates the task; , , , , Five parameters represent the user's uninstallation decision, indicating uninstallation to local, idle device, BS, LEO satellite, and cloud computing center, respectively; , , , These represent idle equipment. BS LEO satellite The largest communication resource for cloud computing centers; Indicates user The total task latency is subject to the task's maximum tolerable latency. Constraints and This indicates that the user has opted for partial uninstallation. Tasks can be offloaded to local devices, idle devices, BS, LEO satellites, and cloud computing centers; - This means that the sum of communication resources allocated to users by idle equipment, BS, and LEO satellites cannot exceed the maximum value of their own communication resources. The sum of communication resources allocated to LEO satellites by the cloud computing center will not exceed the maximum value of their own communication resources, and the communication resources allocated to them are all greater than or equal to 0.

[0078]

[0079] (2) Problem model decomposition

[0080] Based on the idea of ​​block coordinate descent, the edge-cloud collaborative computing offloading optimization problem is decoupled into an offloading decision subproblem and a communication resource allocation subproblem, and solved by alternating iteration. For the offloading decision subproblem, under the premise of a fixed communication resource scheme, the task execution position is jointly optimized to determine the optimal offloading strategy. For the communication resource allocation subproblem, based on the fixed offloading decision scheme, communication resources are allocated and dynamically scheduled.

[0081] Furthermore, step S4 proposes an improved gray wolf algorithm to obtain the offloading decision, and uses the particle swarm optimization algorithm to obtain the communication resource allocation scheme, thereby obtaining an edge-cloud collaborative computing offloading scheme that minimizes the total system latency. Specifically, this includes:

[0082] (1) Unloading decision subproblem:

[0083] To address the unloading decision subproblem, a gray wolf algorithm based on differential evolution is proposed. This algorithm first initializes the gray wolf population and calculates the fitness value of each individual, while simultaneously selecting the wolf with the optimal fitness. suboptimal Compared to the third best The initial position of the wolf; then enter the main iteration loop, first traversing all individuals in the wolf pack to obtain the current position of the gray wolf. , and The candidate positions for wolf movement are used to generate new positions for gray wolves. After this round of individual position updates is completed, the positions are updated again. , and The wolf's location is determined; then, the population is traversed, first generating mutated individuals, then constructing experimental individuals, followed by a greedy selection operation to filter for better individuals, and then updating again. , and The wolf's position; if the fitness change rate is detected three consecutive times during each iteration. Less than or equal to the threshold If the condition is not met, the early termination mechanism is triggered to exit the loop. If the condition is not met, the iteration continues until the maximum number of iterations T is reached. Finally, the algorithm outputs the optimal unloading decision matrix and the corresponding optimal fitness value.

[0084] (2) Communication resource allocation sub-problem:

[0085] To address the communication resource allocation subproblem, a particle swarm optimization (PSO) algorithm is proposed. The algorithm first initializes the particle population and determines the local optimum and global optimum of the entire population by calculating particle fitness. Then, it enters the main iteration loop. In each iteration, all particles are traversed, the particle velocity vector is updated, and then the particle position vector is updated. If the updated position exceeds a preset boundary value, it is corrected to the boundary value. Finally, the fitness value of the particle is calculated, and its individual optimum is updated. After traversing and updating all particles, the global optimum of the population is updated. During each iteration, the algorithm monitors the fitness change rate. If the fitness change rate changes three times consecutively Less than or equal to the threshold If the condition is not met, the early termination mechanism is triggered to exit the loop. If the condition is not met, the iteration continues until the preset maximum number of iterations T is reached. Finally, the algorithm outputs the optimal communication resource allocation matrix and the corresponding optimal fitness value.

[0086] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the task priority-based edge-cloud collaborative computing offloading method for satellite-ground networks as described in any one of the claims.

[0087] A storage medium that internally stores a computer program, wherein when the computer program is read by a processor, it executes the aforementioned task priority-based satellite-ground network edge-cloud collaborative computing offloading method.

[0088] The advantages and beneficial effects of this invention are as follows:

[0089] This invention addresses the problem of high latency and low completion rates caused by unreasonable task priority settings due to unbalanced allocation of satellite-ground network resources. It proposes a satellite-ground network edge-cloud collaborative computing offloading method based on task priority. The main innovations of this invention are: 1) establishing a satellite-ground fusion network edge-cloud collaborative computing offloading framework that considers idle equipment; 2) designing a task priority scheduling algorithm based on remaining time; and 3) proposing a differential evolution-based gray wolf algorithm to solve the offloading decision.

[0090] The innovations include: 1) Incorporating idle devices into the collaborative system, breaking through the resource boundaries of the traditional three-layer architecture and providing richer computing power support for subsequent offloading and scheduling; 2) Dynamically adjusting priorities based on remaining task time to better align with task deadlines and reduce timeout risks from the scheduling source; and 3) Fusing differential evolution algorithms with the Grey Wolf algorithm to enhance global search capabilities, avoid local optima, and improve the accuracy and convergence efficiency of offloading decisions under complex constraints. Based on the established collaborative framework including idle devices, the system possesses more abundant distributed computing resources. Furthermore, by improving the Grey Wolf algorithm to precisely optimize offloading decisions, efficient matching of computing tasks with heterogeneous resources is achieved. Finally, the task priority scheduling algorithm based on remaining time can rationally arrange the task execution order, effectively alleviating resource competition and task congestion. These three aspects are progressively integrated and organically linked, fully utilizing idle network resources and improving scheduling and decision-making efficiency, ultimately achieving a comprehensive effect of significantly reduced system latency and simultaneous improvement in task completion rate and resource utilization.

[0091] In research on computation offloading in space-ground converged networks based on edge-cloud collaboration, most works often neglect the differentiated service quality requirements of different tasks in heterogeneous network environments and fail to establish an effective task priority scheduling mechanism. This leads to disordered competition between high-critical and low-value tasks, easily causing task timeouts or resource waste. Existing solutions, when dealing with resource-constrained scenarios, mostly rely on static indicators such as task type and value, or simply adopt a "first-come, first-served" strategy, ignoring the dynamic matching of task deadlines and system load status. This not only makes it difficult to guarantee the service quality of core tasks but also exacerbates the overall system latency and resource overhead due to ineffective scheduling. To address these issues, this paper proposes a task priority-based edge-cloud collaborative computation offloading method for space-ground networks. First, a space-ground converged network edge-cloud collaborative computation offloading framework considering idle equipment is established. By rationally utilizing the computing resources of idle ground equipment, the overall system latency can be effectively reduced. Second, a task priority scheduling algorithm based on remaining time is designed. Under the constraint of maximum task latency, the queuing latency is reduced by adjusting the task priority of timed-out tasks, thereby improving the task completion rate. Finally, a gray wolf algorithm based on differential evolution is proposed. The Grey Wolf algorithm is prone to getting stuck in local optima in the later stages of iteration due to population diversity decay, making it difficult to balance optimization accuracy and convergence speed. To address this, a differential evolution algorithm is introduced, using mutation and greedy selection mechanisms to enhance global exploration capabilities and effectively avoid premature convergence stagnation. This results in a hybrid optimization algorithm that converges faster, achieves higher accuracy, and demonstrates greater stability. This improved hybrid optimization algorithm exhibits higher accuracy and robustness in solving unloading decisions, providing strong algorithmic support for efficient computational unloading within the edge-cloud collaborative framework. Therefore, this invention demonstrates innovation and feasibility in its solution approach. Attached Figure Description

[0092] Figure 1 This is a network model diagram constructed according to a preferred embodiment of the present invention;

[0093] Figure 2 This is a flowchart of a satellite-ground network edge-cloud collaborative computing offloading method based on task priority, as described in this invention. Detailed Implementation

[0094] The technical solutions of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0095] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0096] A task-priority-based edge-cloud collaborative computing offloading method for satellite-ground networks. Addressing the problem of high latency and low completion rates caused by unreasonable task priority settings due to unbalanced resource allocation in satellite-ground networks, this method minimizes the total system latency under the constraints of maximum tolerable task latency and communication resources. The specific steps are as follows:

[0097] Step 1: Set network scenario parameters, including the number of nodes and resource information; import task information, including data size and maximum tolerable latency.

[0098] Step 2: Build a satellite-ground integrated network edge-cloud collaborative computing offloading model that takes into account idle equipment, and calculate the transmission rate of the communication link between each node based on network scenario parameters;

[0099] Step 3: Establish queue models for idle devices, BS, and LEO, and design a task priority scheduling algorithm based on remaining time; establish a latency model and calculate the latency of task offloading to each network node;

[0100] Step 4: Consider partial offloading. Under the constraints of maximum tolerable task latency and communication resources, construct an edge-cloud collaborative computing offloading problem model that minimizes the total system latency. Decompose the problem into an offloading decision subproblem and a communication resource allocation subproblem.

[0101] Step 5: First, an improved Grey Wolf algorithm is proposed to solve the offloading decision subproblem and obtain the current optimal offloading decision; then, the Particle Swarm Optimization algorithm is used to solve the communication resource allocation subproblem and obtain the current optimal communication resource allocation scheme; finally, the offloading decision and the communication resource allocation scheme are alternately optimized to obtain the end-edge-cloud collaborative computing offloading scheme that minimizes the total system latency.

[0102] The main symbols and parameters involved in this invention and their meanings are listed in Table 1.

[0103] Table 1. Main Symbols, Parameters, and Their Meanings

[0104]

[0105] Preferably, in the fourth step, partial offloading is considered, and a model for the end-edge-cloud collaborative computing offloading problem that minimizes the total system latency is constructed under the constraints of the maximum tolerable delay of the task and communication resources. This problem is decomposed into an offloading decision subproblem and a communication resource allocation subproblem.

[0106] (1) Construct a model for the edge-cloud collaborative computing offloading problem that minimizes the total system latency.

[0107] Under the constraints of maximum tolerable latency, partial offloading, and communication resources, an edge-cloud collaborative offloading model that minimizes the total system latency is constructed. This model allows users to make offloading decisions. , Indicates user Offloading decision; communication resource allocation matrix ,in, This represents the communication resource allocation matrix for idle devices. This represents the communication resource allocation matrix of the B and C. This represents the communication resource allocation matrix for LEO satellites. Let S represent the communication resource allocation matrix of the cloud computing center. Then, the end-edge-cloud collaborative computing offloading problem model in step S5 can be expressed as equation (1).

[0108] As shown in equation (1), the constraints in the problem are as follows: Indicates user The maximum tolerable latency for the task that generates the task; , , , , Five parameters represent the user's uninstallation decision, indicating uninstallation to local, idle device, BS, LEO satellite, and cloud computing center, respectively; , , , These represent idle equipment. BS LEO satellite The largest communication resource for cloud computing centers; Indicates user The total task latency is subject to the task's maximum tolerable latency. Constraints and This indicates that the user has opted for partial uninstallation. Tasks can be offloaded to local devices, idle devices, BS, LEO satellites, and cloud computing centers; - This means that the sum of communication resources allocated to users by idle equipment, BS, and LEO satellites cannot exceed the maximum value of their own communication resources. The sum of communication resources allocated to LEO satellites by the cloud computing center will not exceed the maximum value of their own communication resources, and the communication resources allocated to them are all greater than or equal to 0.

[0109]

[0110] (2) Problem model decomposition

[0111] The edge-cloud collaborative computing offloading problem is decomposed into an offloading decision subproblem and a communication resource allocation subproblem using the block coordinate descent method:

[0112]

[0113] The offloading decision subproblem is shown in equation (2). Assuming that the communication resource allocation is fixed, the optimal offloading decision is solved under the constraints of the offloading decision and the maximum tolerable delay of the task. - Only with the communication resource allocation matrix Related to the calculation of the unloading decision matrix This does not constitute a constraint, therefore it is not considered in this model.

[0114]

[0115] The communication resource allocation subproblem is shown in equation (3). Assuming the offloading decision is fixed, the optimal communication resource allocation scheme is solved under the constraints of communication resources and the maximum tolerable delay of the task. Due to the constraints... and Only with the unloading decision matrix This is relevant, therefore it is not considered in this model.

[0116] Preferably, in the fifth step, an improved gray wolf algorithm is first proposed to solve the offloading decision subproblem to obtain the current optimal offloading decision; the particle swarm algorithm is used to solve the communication resource allocation subproblem to obtain the current optimal communication resource allocation scheme; finally, the offloading decision and the communication resource allocation scheme are alternately optimized to obtain the end-edge-cloud collaborative computing offloading scheme that minimizes the total system latency.

[0117] (1) Unloading decision subproblem

[0118] To solve the optimal offloading decision, this section proposes a Differential Evolution-Grey Wolf Optimizer (DE-GWO) algorithm, which provides users with an offloading decision that minimizes the total system latency. The DE-GWO algorithm incorporates the Differential Evolution (DE) algorithm into the Grey Wolf Optimizer (GWO) algorithm, thus avoiding getting trapped in local optima.

[0119] Since the unloading method used in this paper is partial unloading, each individual in the DE-GWO algorithm is a microcontroller of size 1. A two-dimensional matrix represents the user's uninstallation decision, defined as shown in equation (4-19). Here, rows represent users, and columns represent uninstallation nodes. Let... equal , which represents the sum of all unloading decision variables. Indicates user The uninstallation ratio, and it is always greater than or equal to 0, and the sum of each line is equal to 1.

[0120]

[0121] To reduce the overall system latency, the fitness of each individual is defined as the sum of the latency at its current position. Then, the... In the nth iteration The fitness of an individual can be expressed as:

[0122]

[0123] The fitness change rate can be expressed as:

[0124]

[0125] 1) GWO algorithm updates individual positions

[0126] The GWO algorithm simulates the leadership hierarchy and cooperative hunting behavior of gray wolf packs in nature. The algorithm uses four types of gray wolves to model the social hierarchy of the pack, corresponding to different leadership roles within the population. In the mathematical modeling process, the solution with the optimal fitness is defined as... The wolf, with the second and third best fitness, is denoted as wolves, ... wolves and Wolves, and all other candidate solutions are uniformly considered as Wolf.

[0127] The target gray wolf is represented as:

[0128]

[0129] The current gray wolf individual is represented as:

[0130]

[0131] The current position of the gray wolf after moving towards the target gray wolf can be calculated using the following formula:

[0132]

[0133]

[0134] in, ; ; The value gradually decreases from 2 to 0 during the iteration process. and Values .

[0135] Gray wolves can identify and surround the location of their prey, and the hunting process is usually carried out by... Wolf-dominated wolves and Wolves also participate collaboratively. In optimizing the search space, the location of the optimal solution is often unknown. To simulate the hunting behavior of gray wolves, the algorithm assumes... , and Wolves possess a superior ability to detect the potential locations of prey. Therefore, the algorithm retains the three optimal solutions in the current iteration and drives the remaining search entities to update their own positions based on the positions of these three optimal solutions. The relevant mathematical model is shown below:

[0136]

[0137]

[0138]

[0139] Among them, equation (11) calculates the first... The number of individual gray wolves and the current best , and The distance between wolves; Equation (12) calculates the distance between the wolves. Individual gray wolves , and The candidate locations for wolf movement simulate the behavior of a gray wolf approaching its prey. Equation (13) will... , and The average of the three candidate positions for the wolf's movement is taken to obtain the first... The first gray wolf individual The new position in the next iteration.

[0140] 2) DE algorithm mutation generates new positions

[0141] The GWO algorithm suffers from population diversity decay and weakened global search capability in the later stages of iteration. When dealing with complex multimodal optimization problems, it is prone to premature convergence and getting stuck in local optima, and it struggles to achieve an effective balance between optimization accuracy and convergence speed. To address these issues, this section introduces the DE algorithm, which leverages its mutation, crossover, and greedy selection mechanisms to enhance population diversity and strengthen the algorithm's global exploration capability to avoid premature convergence and stagnation. Ultimately, this results in a hybrid optimization algorithm with faster convergence speed, higher optimization accuracy, and greater stability.

[0142] For each target vector The mutation operation will generate a corresponding mutation vector. The mutation vector can be represented as:

[0143]

[0144] in, , , These are three randomly selected, distinct vectors; It is the differential weight, a real constant.

[0145] Generate test vectors through cross operations. Among them, the first The crossover operation of dimensions is determined by the following rules:

[0146]

[0147] in, The range of values ​​is ; It is the crossover probability, and its value range is... ; Represent a A random number between two given numbers, therefore there must exist a number that is... equal.

[0148] The action selection determines whether to include the experimental vector. Individuals accepted as the next generation:

[0149]

[0150] in, express fitness value, express The fitness value.

[0151] 3) Iteration stopping condition

[0152] This algorithm employs a dual iteration termination strategy that combines a maximum iteration count limit with convergence detection: First, when the number of iterations reaches a preset upper limit, the algorithm triggers a forced termination mechanism to ensure the controllability of the computation process; second, a convergence detection mechanism is introduced, if the fitness change rate is less than a set threshold for three consecutive iterations, the algorithm is determined to have converged and an early termination operation is performed. This design avoids invalid iterations caused by the algorithm getting trapped in local optima and stops in time when the solution tends to stabilize, achieving a balance between solution accuracy and computational efficiency.

[0153] (2) Communication resource allocation sub-problem

[0154] For each user, the latency is determined by the maximum latency of unloading to each node. Since the objective function is discontinuous and non-convex, this section proposes the Particle Swarm Optimization (PSO) algorithm to solve the optimization problem shown in equation (3).

[0155] For the satellite-ground fusion network based on edge-cloud collaboration proposed in this chapter, a set of particles is generated using PSO to search for the optimal solution to the optimization problem. The position vector can be represented as:

[0156]

[0157] in, Indicates the first The iteration of the ... One particle.

[0158] particle velocity vector It can be represented as:

[0159]

[0160] Wherein, position vector Used to characterize the search direction and movement trend of particles; during the search iteration process, Record the individual optimal position for each particle; simultaneously, combine the search results of all particles to determine the global optimal position of the population. .

[0161] In each iteration, the particle's position and velocity are updated in the following way:

[0162]

[0163]

[0164] in, It is the number of iterations; It is the inertia weight, and its value range is... ; and It is the acceleration coefficient, which is usually taken as 2; and yes Random numbers within.

[0165] make Represents particles The total system delay. Then the particle... The local optimal position can be represented as:

[0166]

[0167] The global optimal position is updated in the following way:

[0168]

[0169] When updating the particle position using equations (19) and (20), it is necessary to ensure that the updated position is within the feasible region of the optimization problem. Feasible positions of the particles. The following constraints must be met:

[0170]

[0171] Assume that during the previous iterations, the current positions of each particle are... Individual optimal position and the global optimal position All are within the feasible region. Meanwhile, the velocities of each particle in the previous iteration satisfied the following constraints:

[0172]

[0173] The communication resource allocation algorithm based on PSO is as follows: First, initialize the particle population and obtain the local optimal position. and global optimal position Then, in each iteration, the position of each particle is updated using equations (19) and (20). and speed First, check if the new position exceeds the boundary value; then, the total system delay... The calculation can be performed using equation (5), and the individual optimality can be updated using equation (21). Finally, at the end of each iteration, the global optimum is updated using equation (22). At the same time, check whether the fitness change rate has changed three times consecutively. Less than or equal to the threshold .

[0174] The model involved in this invention is as follows:

[0175] 1. Network Model

[0176] The main application scenario of this invention is a satellite-ground fusion network based on edge-cloud collaboration, such as... Figure 1 As shown in the diagram, this network model consists of n users, j idle devices, m business units (BS), i LEO satellites, and a cloud computing center. The set of users can be represented as... The collection of idle equipment can be represented as Each BS is equipped with one edge server, and the BS set is represented as follows: Each LEO satellite is also equipped with an edge server, and the LEO satellite set is represented as follows: .

[0177] In this network scenario, the user's task can be represented by a triple. To indicate, among which Indicates the data size of the task. Indicates the calculated strength. This indicates the maximum tolerable delay for the task. Additionally, using... , , , , Five parameters represent the user's offloading decision, indicating the proportion of tasks offloaded to local storage, idle devices, BS, LEO satellites, and cloud computing centers, respectively. Since partial offloading is used, the values ​​of these parameters are... .

[0178] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions.

[0179] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0180] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0181] The above embodiments should be understood as illustrative only and not as limiting the scope of protection of the present invention. After reading the description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, characterized in that, Includes the following steps: S1: Build a satellite-ground integrated network edge-cloud collaborative computing offloading model that takes into account idle equipment, and calculate the transmission rate of the communication link between each node; S2: Establish a queue model and design a task priority scheduling algorithm based on remaining time; establish a latency model and calculate the latency of task unloading to each network node; S3: Construct an edge-cloud collaborative computing offloading problem model that minimizes the total system latency, and decompose the problem into an offloading decision subproblem and a communication resource allocation subproblem; S4: An improved gray wolf algorithm is proposed to solve the offloading decision, and the particle swarm algorithm is used to solve the communication resource allocation scheme, thereby obtaining an end-edge-cloud collaborative computing offloading scheme that minimizes the total system latency.

2. The method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, as described in claim 1, is characterized in that... Step S1 involves establishing a satellite-ground fusion network edge-cloud collaborative computing offloading model that considers idle equipment, and calculating the transmission rate of the communication links between nodes. Specifically, this includes: (1) Network model; The satellite-ground integrated network backhaul architecture is mainly composed of user terminals, idle equipment, base stations (BS) equipped with edge servers, low-Earth orbit (LEO) satellites equipped with edge servers, and a cloud computing center, which together construct an end-edge-cloud collaborative computing architecture. (2) Construct a channel model and calculate the transmission rate; In a space-ground converged network, three main channel models are considered: the terrestrial channel model, the space-ground channel model, and the inter-satellite channel model. The terrestrial channel model is used for communication between users, idle equipment, and the BS; the space-ground channel model is used for communication between users and LEO satellites, and between LEO satellites and cloud computing centers; and the inter-satellite channel model is used for communication between LEO satellites.

3. The method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, as described in claim 2, is characterized in that... The three channel models are as follows: 1) Ground channel model: In terrestrial communications, non-line-of-sight transmission dominates, and its channels are modeled as Rayleigh channels; users With BS The transmission rate between them is given by the following formula: ； in, BS Assigned to user bandwidth; User The transmission power; This represents the Gaussian white noise power of the link; User With BS Channel gain between , BS With users The distance; The path loss index; Represents a Rayleigh random variable; Similarly, users With idle equipment The transmission rate between them is given by the following formula: ； in, Indicates idle equipment Assigned to user bandwidth; User The transmission power; 2) Satellite-to-Ground Channel Model: In satellite-to-ground communication, line-of-sight transmission dominates. Users communicate with LEO satellites using Ka-band satellite-to-ground links, and the corresponding channels are modeled as Ricean channels. With LEO satellite The transmission rate between them is expressed as: ； in, Indicates LEO satellite Assigned to user bandwidth; Indicates user The transmission power; Indicates user Antenna gain; Indicates LEO satellite The receiving antenna gain; Indicates the channel gain of the link; When a user offloads a task to the cloud computing center, the task needs to be offloaded to a LEO satellite first, and then forwarded by the LEO satellite to the ground-based cloud computing center; therefore, the uplink transmission rate from user n to LEO satellite i is shown in equation (3), and the downlink transmission rate from LEO satellite i to the cloud computing center is expressed as: ； in, This indicates that the cloud computing center is allocated to LEO satellites. bandwidth; Indicates LEO satellite The transmission power; Indicates LEO satellite Antenna gain; Indicates the receiving antenna gain of the cloud computing center; Indicates the channel gain of the link; 3) Inter-satellite channel model Inter-satellite links are primarily for line-of-sight transmission; therefore, the transmission rate of an inter-satellite link is expressed as: ； in, This represents the bandwidth allocated to LEO satellite k by LEO satellite i; This indicates the transmission power of LEO satellite k; This represents the antenna gain of LEO satellite k; Indicates the receiving antenna gain of LEO satellite i; This indicates the channel gain of the link.

4. The method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, as described in claim 1, is characterized in that... In step S2, a queue model is established, and a task priority scheduling algorithm based on remaining time is designed; a latency model is established to calculate the latency of task unloading to each network node, as detailed below: (1) Queue model: The queuing model follows the first-come, first-served principle, with task queues on idle equipment, BS, and LEO satellites represented as follows: , and For time slots Arrival of idle equipment The computational task can be performed in time slots. Internal processing, while time slots The task that has been completed requires waiting for a time slot. Arrival of idle equipment After the computational task is completed, the idle equipment is processed. The queue is represented as: ； in, Indicates arrival at idle equipment The unprocessed number For each task, the queuing delay can be expressed as: ； in, For idle equipment No. One unprocessed task; Indicates idle equipment Computing resources; Similarly, for BS and LEO satellites, The queuing delay can be expressed as follows: ； ； in, and They represent BS respectively Computing resources and LEO satellites Computing resources; The task priority scheduling algorithm based on remaining time runs on each offload node, including idle equipment, BS and LEO satellites; the remaining time in the algorithm is represented as the maximum tolerable delay of the task minus the task transmission delay and propagation delay. (2) Delay model: 1) Local computation: Local computing only needs to consider the latency of task processing, therefore user devices The latency for local task processing is: ； in, This indicates the proportion of local computing tasks; Indicates user equipment Computing resources; Indicates the size of the task data; Indicates the calculated strength; 2) Unload to an idle device: When users offload tasks to idle devices, the latency incurred in completing these tasks consists of three parts: the latency of user task transmission, the task queuing latency, and the latency of idle device task processing; Offload the task to an idle device The delay is: ； in, This indicates that the task will be unloaded to an idle device. The proportion; Indicates idle equipment Computing resources; This indicates a queuing delay, among which Indicates unloading to an idle device The order of tasks in the task queue; 3) Uninstall to BS: When a user offloads a task to the BS, only the latency of the user transmitting the task, the latency of the BS processing the task, and the latency of the task queuing need to be considered; Unload the task to BS The delay is: ； in, This indicates that the task is unloaded to the BS. The proportion; BS Computing resources; Indicates queuing delay; 4) Unload to LEO satellite: user Offload the mission to the LEO satellite The delay is: ； in, This indicates that the mission is offloaded to the LEO satellite. The proportion; Indicates LEO satellite Computing resources; , Representing users respectively With LEO satellite , The distance; , They represent LEO satellites. , With LEO satellite The distance; Represents the speed of light; 5) Unload to the cloud computing center: user The latency for offloading tasks to the cloud computing center is: ； in, This indicates the proportion of tasks that are offloaded to the cloud computing center; Indicates user n and LEO satellite The distance; Indicates LEO satellite Distance to the cloud computing center; This refers to the computing resources of a cloud computing center; 6) Total system latency: user The total latency of edge-cloud collaborative offloading can be expressed as: ； The total system delay can be expressed as: 。 5. A method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, as described in claim 4, is characterized in that... The task priority scheduling algorithm based on remaining time runs on each unloading node, and the specific process is as follows: First, based on the unloading decision matrix... and communication resource allocation matrix Computation task queue The transmission and propagation delays for each task are calculated; then, the maximum tolerable delay for each task is determined. Calculate the task queue Remaining time for each task Finally, iterate through each task; if a task times out, i.e., the current queuing delay... and computational delay The sum is greater than the remaining time If the task does not time out, iterate through the preceding tasks; if the task does not time out after swapping the priorities of the preceding and current tasks, swap their priorities and stop the current loop; if the task still times out after iterating through the preceding tasks, place the task at the end of the task queue; finally, output the task queue after iterating through all tasks.

6. The method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, as described in claim 1, is characterized in that... Step S3 constructs a model for the edge-cloud collaborative computing offloading problem that minimizes the total system latency. This problem is decomposed into an offloading decision sub-problem and a communication resource allocation sub-problem, as detailed below: (1) Construct a model for the edge-cloud collaborative computing offloading problem that minimizes the total system latency; Under the constraints of maximum tolerable latency, partial offloading, and communication resources, an edge-cloud collaborative offloading problem model that minimizes the total system latency was constructed; and a user offloading decision matrix was established. , Indicates user Offloading decision; communication resource allocation matrix ,in, This represents the communication resource allocation matrix for idle devices. This represents the communication resource allocation matrix of the B and C. This represents the communication resource allocation matrix for LEO satellites. Let the communication resource allocation matrix of the cloud computing center be represented; then the end-edge-cloud collaborative computing offloading problem model can be expressed as equation (17); As shown in equation (17), the constraints in the problem are as follows: Indicates user The maximum tolerable latency for the task that generates the task; , , , , Five parameters represent the user's uninstallation decision, indicating uninstallation to local, idle device, BS, LEO satellite, and cloud computing center, respectively; , , , These represent idle equipment. BS LEO satellite The largest communication resource for cloud computing centers; Indicates user The total task latency is subject to the task's maximum tolerable latency. Constraints and This indicates that the user has opted for partial uninstallation. Tasks can be offloaded to local devices, idle devices, BS, LEO satellites, and cloud computing centers; - This means that the sum of communication resources allocated to users by idle equipment, BS and LEO satellites cannot exceed the maximum value of their own communication resources. The sum of communication resources allocated to LEO satellites by the cloud computing center will not exceed the maximum value of their own communication resources, and the communication resources allocated to them are all greater than or equal to 0. ； (2) Problem model decomposition; Based on the idea of ​​block coordinate descent, the edge-cloud collaborative computing offloading optimization problem is decoupled into an offloading decision subproblem and a communication resource allocation subproblem, and solved by alternating iteration. For the offloading decision subproblem, under the premise of a fixed communication resource scheme, the task execution position is jointly optimized to determine the optimal offloading strategy. For the communication resource allocation subproblem, based on the fixed offloading decision scheme, communication resources are allocated and dynamically scheduled.

7. A method for offloading edge-cloud collaborative computing in a satellite-ground network based on task priority, as described in claim 1, is characterized in that... Step S4 proposes an improved gray wolf algorithm to obtain the offloading decision, and uses the particle swarm optimization algorithm to obtain the communication resource allocation scheme, thereby obtaining an end-edge-cloud collaborative computing offloading scheme that minimizes the total system latency. Specifically, this includes: (1) Unloading decision subproblem: To address the unloading decision subproblem, a gray wolf algorithm based on differential evolution is proposed. This algorithm first initializes the gray wolf population and calculates the fitness value of each individual, while simultaneously selecting the wolf with the optimal fitness. suboptimal Compared to the third best The initial position of the wolf; then enter the main iteration loop, first traversing all individuals in the wolf pack to obtain the current position of the gray wolf. , and The candidate positions for wolf movement are used to generate new positions for gray wolves. After this round of individual position updates is completed, the positions are updated again. , and The wolf's location is determined; then, the population is traversed, first generating mutated individuals, then constructing experimental individuals, followed by a greedy selection operation to filter for better individuals, and then updating again. , and The wolf's position; if the fitness change rate is detected three consecutive times during each iteration. Less than or equal to the threshold If the condition is not met, the early termination mechanism is triggered to exit the loop. If the condition is not met, the iteration continues until the maximum number of iterations T is reached. Finally, the algorithm outputs the optimal unloading decision matrix and the corresponding optimal fitness value. (2) Communication resource allocation sub-problem: To address the communication resource allocation subproblem, a particle swarm optimization (PSO) algorithm is proposed. The algorithm first initializes the particle population and determines the local optimum and global optimum of the entire population by calculating particle fitness. Then, it enters the main iteration loop. In each iteration, all particles are traversed, the particle velocity vector is updated, and then the particle position vector is updated. If the updated position exceeds a preset boundary value, it is corrected to the boundary value. Finally, the fitness value of the particle is calculated, and its individual optimum is updated. After traversing and updating all particles, the global optimum of the population is updated. During each iteration, the algorithm monitors the fitness change rate. If the fitness change rate changes three times consecutively Less than or equal to the threshold If the condition is not met, the early termination mechanism is triggered to exit the loop. If the condition is not met, the iteration continues until the preset maximum number of iterations T is reached. Finally, the algorithm outputs the optimal communication resource allocation matrix and the corresponding optimal fitness value.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the task priority-based edge-cloud collaborative computing offloading method for satellite-ground networks as described in any one of claims 1 to 7.

9. A storage medium that internally stores a computer program, characterized in that, When the computer program is read by the processor, it executes the task priority-based edge-cloud collaborative computing offloading method for satellite-ground network as described in any one of claims 1 to 7.

Images