Earth-space integrated network edge computing offloading and resource allocation method and system

Abstract

This invention discloses a method and system for offloading and resource allocation in edge computing for integrated space-ground networks, belonging to the field of integrated space-ground networks and satellite edge computing technology. This invention constructs a satellite-ground-edge-cloud collaborative architecture, where the terminal layer generates cached tasks, the satellite edge layer provides near-end computing capabilities, and the ground cloud layer provides remote support. Addressing the challenges of dynamically changing network resources and high real-time task requirements, an information age index is introduced, establishing latency and energy consumption models for task arrival, information age, and three computing modes. An optimization problem is constructed with the goal of minimizing the weighted cost of information age and energy consumption; this problem is modeled as a Markov decision process, employing a near-end strategy optimization algorithm based on cost-aware priority sampling to jointly optimize discrete offloading decisions and continuous resource allocation. This invention can ensure information freshness, reduce system energy consumption, and improve task processing efficiency and resource utilization.

Description

Technical Field

[0001] This invention relates to the fields of integrated space-ground networks, satellite edge computing, task scheduling, and deep reinforcement learning, and specifically to a method and system for offloading and allocating resources for edge computing in integrated space-ground networks. Background Technology

[0002] With the rapid development of 5G, IoT, and intelligent sensing technologies, an increasing number of business scenarios, such as real-time sensing, remote monitoring, and marine operations, require task processing and status updates to be completed in areas without terrestrial cellular network coverage. Satellite edge computing, by deploying computing resources on low-Earth orbit satellites, enables tasks to be processed closer to users, providing a new solution for low-latency services. In integrated space-ground network scenarios, the highly dynamic and precise scheduling of satellite-to-ground links, inter-satellite links, and satellite resources is crucial for ensuring the freshness of task information and reducing overall system energy consumption, and has significant practical implications for improving the computing efficiency of integrated space-ground networks.

[0003] In existing technologies, some progress has been made in satellite edge computing task offloading and resource allocation schemes. However, there is still a significant lack of specialized research on comprehensively considering information freshness and hybrid resource scheduling in highly dynamic scenarios. For example, patent CN115022894A discloses a task offloading and computing resource allocation method and system for low-Earth orbit satellite networks. This method establishes a task offloading model based on the fusion of edge cloud and cloud edge low-Earth orbit satellite computing models, and uses a hybrid particle swarm optimization algorithm and analytical solutions to obtain the optimal strategy. However, its core focus is on the trade-off between latency, energy consumption, and platform funding costs, and it does not introduce information age (AoI) as a measure of information freshness to address the core requirements of real-time sensing services. At the same time, the heuristic particle swarm optimization algorithm it uses lacks deep joint optimization logic for discrete offloading decisions and continuous resource allocation. When faced with dynamic and time-varying complex network states, it is prone to getting trapped in local optima and cannot solve the problem of strong coupling in resource allocation in integrated space-ground scenarios.

[0004] Another patent, CN118250750A, discloses a satellite edge computing task offloading and resource allocation method based on deep reinforcement learning. This method is based on a three-layer computing architecture of terminal-satellite-cloud and uses the Deep Deterministic Policy Gradient (DDPG) algorithm to solve for the policy that meets the deterministic latency requirements and minimizes energy consumption. However, its optimized target parameters do not cover the key attribute reflecting the timeliness of the data at the receiving end (i.e., information age, AoI), and its model is not optimized for the indivisible full offloading mode and dynamic time slots, making it difficult to cope with highly time-varying environments. At the same time, this method lacks an effective extraction mechanism for cost awareness and high-value sample sampling. When facing a high-dimensional action space with a mixture of discrete and continuous parameters, the algorithm's exploration efficiency and training stability are limited, and it is prone to slow policy convergence.

[0005] Another patent, CN116633422A, discloses a multi-dimensional resource scheduling method for low-Earth orbit satellite networks oriented towards IoT task offloading. This method is based on satellite network visibility and distance calculation communication constraints, uses a heuristic algorithm for iterative solution, and decomposes it into two sub-problems: uplink power allocation and computing resource allocation, which are solved separately. However, it only addresses the specific static or quasi-static dimensionality reduction processing mode of resource decoupling, does not involve special scheduling for joint optimization of discrete and continuous variables, and does not construct a multi-dimensional collaborative system that includes multiple low-Earth orbit satellites sharing resources through inter-satellite links (ISL). It also lacks a long-term planning algorithm optimization mechanism based on Markov decision processes, and cannot meet the dynamic detection and real-time scheduling requirements of multi-mode, multi-state satellite-ground fusion networks.

[0006] In summary, existing technologies have the following shortcomings: First, they lack a dedicated scheduling framework for ensuring information freshness. Existing solutions mostly focus on end-to-end latency or unilateral energy consumption, which is insufficient for adapting to the multi-parameter characteristics (AoI) that reflect the aging of real-time business information. Second, the system architecture and feature selection are incomplete, failing to cover core differentiated features such as inter-satellite link cooperation and resource reservation mechanisms, resulting in low accuracy of collaborative computing. Third, they lack a joint solution strategy that integrates discrete offloading decisions with continuous communication / computing resource allocation, making decoupling or fixed modes difficult to match the resources in integrated space-ground networks. The existing technologies suffer from several problems: First, they lack strong coupling in resource allocation; second, their dynamic learning and exploration mechanisms are imperfect and susceptible to interference from high-dimensional continuous state spaces, and they lack priority sampling mechanisms for high-cost state transition samples, resulting in limited convergence and stability in long-term scheduling; third, their algorithms lack adaptability, lacking optimization methods such as hybrid fitting for discrete actions and continuous resources (e.g., Beta distribution modeling), leading to limited scheduling accuracy; and fourth, their optimization objectives are mostly single-dimensional energy efficiency or latency, failing to provide a complete evaluation system that includes AoI and energy consumption weighted comprehensive costs, thus offering weak support for optimization in high-load, multi-user scenarios. Therefore, a highly adaptable and accurate method for offloading and allocating resources for edge computing in integrated space-ground networks is urgently needed to address these issues. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the existing technologies. This invention proposes a method and system for offloading and resource allocation of edge computing in integrated space-ground networks. It proposes an innovative solution to the problems of traditional satellite edge computing, such as difficulty in balancing wide-area coverage and information freshness, strong coupling between discrete offloading decisions and continuous resource allocation, which makes joint optimization difficult, and poor adaptability of dynamic high-dimensional scheduling.

[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0009] A method for offloading and allocating resources for edge computing in a space-ground integrated network is characterized by its application in a space-ground-edge-cloud collaborative computing system consisting of a terminal layer, a satellite edge layer, and a ground cloud layer, and includes the following steps:

[0010] S1. Construct a satellite-ground-edge-cloud collaborative computing architecture. The terminal layer includes multiple IoT terminals, and each IoT terminal maintains a task queue to cache dynamically arriving computing tasks. The satellite edge layer includes multiple low-orbit satellites equipped with edge computing servers, and the satellites share computing and communication resources through inter-satellite links. The ground cloud layer includes a ground cloud data center to handle tasks that the satellite edge layer cannot handle.

[0011] S2. Model the terminal-generated tasks and user information age;

[0012] S3. Establish latency and energy consumption models for three task execution modes: local computing, satellite edge computing, and terrestrial cloud computing.

[0013] S4. Establish a joint optimization problem with unloading decision variables and resource allocation variables as optimization objects and minimizing the weighted comprehensive cost of system energy consumption and information age as the objective.

[0014] S5. Model the joint optimization problem as a Markov decision process, and construct the state space, action space, reward function and state transition function;

[0015] S6. A cost-aware priority-based near-end policy optimization algorithm is adopted to jointly solve the discrete offloading decision and the allocation of continuous computing and communication resources to obtain the task scheduling strategy for each time slot.

[0016] S7. Execute task scheduling according to the task scheduling strategy, and update the task queue status, resource occupancy status, and user information age status.

[0017] Furthermore, the aforementioned satellite-ground-edge-cloud collaborative computing system adopts a fully offloaded mode with indivisible tasks and a resource reservation mechanism; when a task During time slot p scheduling, the duration of communication resources allocated to this task is... The computing resources allocated to this task will last for a duration of [duration]. The estimated communication time is The estimated calculation time is The length of a single time slot is ;

[0018] The trigger condition for the resource reservation mechanism is: when the task The decision variable is scheduled and unloaded in time slot p. When the available communication and computing resources of the target unloading node meet the estimated requirements of the task, a resource reservation operation is triggered; the estimated communication time... The calculation is based on the current time slot's channel gain, available satellite bandwidth, and the projected communication resource allocation ratio. The calculation formula is as follows: ,in The uplink transmission rate is estimated based on the current time slot channel state; the estimation time is... The calculation is based on the total computational load of the mission and the proportion of the satellite's estimated allocable computational resources. The calculation formula is as follows: ,in To estimate the amount of computing resources that can be allocated to this mission based on the satellite's currently available computing resources;

[0019] Once a task has completed all communication transmission and computation processing on the reserved resources, it automatically releases the occupied communication and computation resources and remarks the released resources as available resources on the target node. When multiple tasks compete for the limited resources of the same target node, they are sorted from high to low according to a comprehensive priority score based on information age urgency, task waiting time, and computational complexity, and resources are allocated in sequence. If resources are insufficient, low-priority tasks are not scheduled and are kept in the task queue to wait for the next time slot to re-participate in the competition.

[0020] Furthermore, in step S2, the number of tasks generated by user u within time slot t is: , Obtain the parameter as Poisson distribution:

[0021]

[0022] The rules for dynamically updating user information age are as follows:

[0023]

[0024] The average age of the system's information is:

[0025] .

[0026] Furthermore, in step S3, the task of user u in local computing mode End-to-end processing delay in time slot t and energy consumption They are respectively:

[0027]

[0028]

[0029] in, For the computing capacity of the user terminal, This represents the effective capacitance coefficient of the chip.

[0030] Furthermore, in step S3, the satellite edge computing mode is divided into a scenario without inter-satellite link (ISL) cooperation and a scenario with ISL cooperation.

[0031] Without ISL cooperation, the task is performed by the access satellite, and its latency is... and energy consumption They are respectively:

[0032]

[0033]

[0034] in, The communication resource allocation value for the access satellite in time slot t is assigned to the task. To access the computing resources allocated to the satellite for the task in time slot t, The distance between the user and the access satellite. At the speed of light, Transmit power for user terminals;

[0035] When ISL collaboration is involved, the mission is offloaded to the collaborating satellite via the X-hop inter-satellite link for processing, and its latency is reduced. and energy consumption They are respectively:

[0036]

[0037]

[0038] in, For inter-satellite link communication capacity, The distance between adjacent satellites. This refers to the inter-satellite relay power.

[0039] The cooperative satellites were selected from a set of feasible satellites that met constraints on link connectivity, computing resources, and information age, based on a multi-dimensional comprehensive score considering hop count, available computing resources, current load, and queue length; hop count... No greater than the maximum number of hops determined by the maximum tolerable latency of the task. If no alternative cooperating satellite is available, the system will switch to ground-based cloud computing first; otherwise, the mission will be reserved for scheduling in the next time slot.

[0040] Furthermore, in step S3, in the ground cloud computing mode, the task is transparently forwarded to the ground cloud center via the access satellite for execution, and its latency is reduced. and energy consumption They are respectively:

[0041]

[0042]

[0043] in, For the communication capacity of the link between satellite and ground cloud, For ground-based cloud computing capacity, The distance between the satellite and the ground-based cloud. This refers to the transmission power from the satellite to the ground-based cloud.

[0044] Furthermore, in step S4, the task The decision variable for unloading in time slot t is denoted as... Its set of values ​​is:

[0045]

[0046] Where 0 indicates that the current time slot is not scheduled, 1 to S correspond to the satellite number, S+1 indicates that it is executed locally, and S+2 indicates that it is offloaded to the ground cloud;

[0047] Task energy consumption Defined as:

[0048]

[0049] The joint optimization problem aims to minimize the weighted combined cost of the system's total energy consumption and average information age, and satisfies the following constraints:

[0050]

[0051] in, and These represent the available computing resources and available communication resources of satellite s in time slot t, respectively.

[0052] Furthermore, in step S5, in the state space of the Markov decision process, the state space of the Markov decision process... Defined as:

[0053]

[0054] in, The task queue state of user u in time slot t. The available computing resources for satellite s in time slot t. For satellite s in time slot t, the available communication resources and user information (age status) are considered; action space. Includes discrete unloading actions and continuous resource allocation actions , ,Right now:

[0055]

[0056] The reward function is defined as:

[0057]

[0058] in, For information age reward items, To complete the reward items ahead of schedule, As an energy consumption penalty item, This is a penalty item for illegal actions; when a task is not scheduled or is executed locally, the corresponding satellite computing and communication resource allocation values ​​are set to zero; when a task is offloaded to the ground cloud, the corresponding satellite computing resource allocation value is set to zero; the aforementioned penalty item for illegal actions. The triggering conditions include at least one of the following: the physical distance between the cooperating satellite and the access satellite exceeds the maximum communication line distance of the inter-satellite link; the computing resources requested by the mission cause the total computing power consumption of the satellite to exceed the battery power supply limit; and the rain attenuation value of the ground cloud link exceeds the preset threshold.

[0059] Furthermore, in step S6, the cost-aware priority-based proximal policy optimization algorithm adopts an actor-critic network structure, and the continuous action head is modeled using a Beta distribution, with parameters satisfying:

[0060]

[0061] in, , This is the weight matrix. , Let h be the bias vector and h be the state feature vector; the algorithm further includes a cost-aware priority sampling mechanism within the trajectory, where the priority of the i-th sample is determined by the bias vector. Defined as:

[0062]

[0063] in, and These are the weighting coefficients. and These represent the total system energy consumption in adjacent time slots. and These represent the average information age of the systems in adjacent time slots. To prevent small positive numbers with zero priority; the sampling probability of the i-th sample is:

[0064]

[0065] in, This is the priority strength parameter; and the reward is normalized online.

[0066] An edge computing offloading and resource allocation system for integrated space-ground networks is characterized by comprising: a task generation and caching module, an execution mode modeling module, an information age update module, a resource status management module, a joint optimization modeling module, a joint decision-making module, a strategy solving module, and a task execution and status update module. The task generation and caching module receives tasks generated by terminals and maintains a task queue. The execution mode modeling module establishes latency and energy consumption models under three task execution modes. The information age update module updates user information age. The resource status management module maintains satellite resource occupancy and release. The joint optimization modeling module constructs a joint optimization problem with the minimum weighted comprehensive cost. The joint decision-making module outputs offloading and resource allocation actions. The strategy solving module generates task scheduling strategies. The task execution and status update module executes task scheduling and updates the system status.

[0067] Compared with the prior art, the present invention, employing the above technical solution, has the following beneficial effects:

[0068] (1) The proposed method and system for offloading and allocating resources for edge computing in integrated space-ground networks constructs a three-layer collaborative architecture of terminal, satellite edge and ground cloud, and introduces a resource reservation mechanism: This invention enables computing tasks to be flexibly distributed among local, access satellite, cooperative satellite and ground cloud according to the current system communication and computing resource status, which fully adapts to complex scenarios with wide area coverage without ground cellular network coverage; at the same time, by estimating the occupation time and locking the corresponding communication and computing resources during task scheduling through the resource reservation mechanism, the latency jitter caused by resource preemption is effectively reduced, and the reliability of task processing is significantly improved.

[0069] (2) The proposed method and system for offloading and allocating resources for edge computing in integrated space-ground networks innovatively introduces information age (AoI) as a core optimization indicator and constructs a weighted comprehensive cost objective: Compared with the traditional evaluation indicator that only considers end-to-end latency, this invention models the dynamic update law of information age, which can more comprehensively and accurately reflect the freshness and aging of information received by users, greatly ensuring the stringent requirements of information timeliness for services such as real-time perception and remote monitoring; at the same time, the total energy consumption of the system is included in the joint optimization, which achieves the dynamic optimal balance of system energy efficiency while ensuring information freshness.

[0070] (3) The method and system for offloading and resource allocation of edge computing for integrated space-ground network proposed in this invention integrates discrete offloading decision and continuous resource allocation into the hybrid action space of Markov decision process (MDP) for joint optimization: This invention overcomes the problems of local optima and low resource utilization caused by the forced decoupling or step-by-step solution of offloading decision and resource allocation in traditional scheduling schemes. By constructing a joint action space that includes discrete offloading action and continuous computing / communication resource allocation action, the decision coordination and global scheduling optimization in a strongly coupled environment are significantly improved.

[0071] (4) The proposed method and system for edge computing offloading and resource allocation in the integrated space-ground network adopts the cost-aware priority-based proximal policy optimization algorithm (CAP-PPO), which significantly improves the convergence speed and stability in dynamic environments. The algorithm introduces a cost-aware priority sampling mechanism within the trajectory, enabling the model to focus on learning state transition samples with high cost impacts (such as a sharp deterioration in system energy consumption or average information age) during training, effectively overcoming the problem of low exploration efficiency in high-dimensional mixed action spaces. At the same time, online reward normalization processing is added to further enhance the stability of the algorithm in long-term deployment in the dynamic time-varying integrated space-ground network. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the satellite-ground-edge-cloud collaborative computing offloading system architecture of the present invention;

[0073] Figure 2 This is a schematic diagram of the CAP-PPO algorithm training framework of the present invention;

[0074] Figure 3 This is a flowchart illustrating the task scheduling and resource reservation process of the present invention.

[0075] Figure 4 This is a schematic diagram illustrating the evolution of the information age (AoI) of this invention;

[0076] Figure 5 This is a flowchart of the collaborative satellite screening process of the present invention;

[0077] Figure 6 This is a flowchart illustrating the overall execution process of the method of the present invention. Detailed Implementation

[0078] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] A method for offloading and allocating resources for edge computing in an integrated space-ground network is characterized by its application in a collaborative computing system consisting of a terminal layer, a satellite edge layer, and a ground cloud layer, such as... Figure 1 As shown, it includes the following steps:

[0080] S1. Construct a satellite-ground-edge-cloud collaborative computing architecture. The terminal layer includes multiple IoT terminals, and each IoT terminal maintains a task queue to cache dynamically arriving computing tasks. The satellite edge layer includes multiple low-orbit satellites equipped with edge computing servers, and the satellites share computing and communication resources through inter-satellite links. The ground cloud layer includes a ground cloud data center to handle tasks that the satellite edge layer cannot handle.

[0081] S2. Model the terminal-generated tasks and user information age;

[0082] S3. Establish latency and energy consumption models for three task execution modes: local computing, satellite edge computing, and terrestrial cloud computing.

[0083] S4. Establish a joint optimization problem with unloading decision variables and resource allocation variables as optimization objects and minimizing the weighted comprehensive cost of system energy consumption and information age as the objective.

[0084] S5. Model the joint optimization problem as a Markov Decision Process (MDP), creating a state space, action space, reward function, and state transition function.

[0085] S6. Based on the MDP modeling results, the discrete offloading decision and continuous computing and communication resource allocation are jointly solved using the Cost-Aware Priority-Based Proximal Policy Optimization (CAP-PPO) algorithm to obtain the task scheduling strategy for each time slot; a schematic diagram of the CAP-PPO algorithm training framework is shown below. Figure 2 As shown;

[0086] S7. Execute task scheduling according to the task scheduling strategy, and update the task queue status, resource occupancy status, and user information age status.

[0087] Furthermore, the aforementioned satellite-ground-edge-cloud collaborative computing system adopts a fully offloaded mode with indivisible tasks and a resource reservation mechanism; the task scheduling and resource reservation flowchart is as follows. Figure 3 As shown, when the task During time slot p scheduling, the duration of communication resources allocated to this task is... The computing resources allocated to this task will last for a duration of [duration]. The estimated communication time is The estimated calculation time is The length of a single time slot is ;

[0088] The trigger condition for the resource reservation mechanism is: when the task The decision variable is scheduled and unloaded in time slot p. When the available communication and computing resources of the target unloading node meet the estimated requirements of the task, a resource reservation operation is triggered; the estimated communication time... The calculation is based on the current time slot's channel gain, available satellite bandwidth, and the projected communication resource allocation ratio. The calculation formula is as follows: ,in The uplink transmission rate is estimated based on the current time slot channel state; the estimation time is... The calculation is based on the total computational load of the mission and the proportion of the satellite's estimated allocable computational resources. The calculation formula is as follows: ,in To estimate the amount of computing resources that can be allocated to this mission based on the satellite's currently available computing resources;

[0089] Once a task has completed all communication transmission and computation processing on the reserved resources, it automatically releases the occupied communication and computation resources and remarks the released resources as available resources on the target node. When multiple tasks compete for the limited resources of the same target node, they are sorted from high to low according to a comprehensive priority score based on information age urgency, task waiting time, and computational complexity, and resources are allocated in sequence. If resources are insufficient, low-priority tasks are not scheduled and are kept in the task queue to wait for the next time slot to re-participate in the competition.

[0090] The task can choose from the following four execution paths:

[0091] (1) Execute locally on the user terminal;

[0092] (2) Unload to the currently accessed satellite for execution;

[0093] (3) The data is forwarded to a cooperating satellite via ISL for execution;

[0094] (4) The satellite is transparently forwarded to the ground cloud for execution.

[0095] Furthermore, in step S2, the number of tasks generated by user u within time slot t is: , Obtain the parameter as Poisson distribution:

[0096]

[0097] like Figure 4 As shown, the system uses AoI to describe the freshness of user information. The AoI for user u in time slot t is:

[0098]

[0099] The rules for dynamically updating user information age are as follows:

[0100]

[0101] The average age of the system's information is:

[0102]

[0103] At the initial moment of the system, it can be set as follows:

[0104]

[0105] This is used to indicate that user information has not been updated for a long time before the system starts.

[0106] Furthermore, in step S3, the task of user u in local computing mode End-to-end processing delay in time slot t and energy consumption They are respectively:

[0107]

[0108]

[0109] in, For the computing capacity of the user terminal, This represents the effective capacitance coefficient of the chip.

[0110] Furthermore, in step S3, the satellite edge computing mode is divided into a scenario without inter-satellite link (ISL) cooperation and a scenario with ISL cooperation.

[0111] Without ISL cooperation, the task is performed by the access satellite, and its latency is... and energy consumption They are respectively:

[0112]

[0113]

[0114] in, The communication resource allocation value for the access satellite in time slot t is assigned to the task. To access the computing resources allocated to the satellite for the task in time slot t, The distance between the user and the access satellite. At the speed of light, Transmit power for user terminals;

[0115] Collaborative satellite screening flowchart as follows Figure 5As shown, when there is ISL cooperation, the task is offloaded to the cooperating satellite for processing via the x-hop inter-satellite link, and its latency is... and energy consumption They are respectively:

[0116]

[0117]

[0118] in, For inter-satellite link communication capacity, The distance between adjacent satellites. This refers to the inter-satellite relay power. Furthermore, the total communication and computing resources allocated to all missions by a single satellite within a time slot must not exceed its currently available capacity.

[0119] The cooperative satellites were selected from a set of feasible satellites that met constraints on link connectivity, computing resources, and information age, based on a multi-dimensional comprehensive score considering hop count, available computing resources, current load, and queue length; hop count... No greater than the maximum number of hops determined by the maximum tolerable latency of the task. If no alternative cooperating satellite is available, the system will switch to ground-based cloud computing first; otherwise, the mission will be reserved for scheduling in the next time slot.

[0120] Furthermore, in step S3, in the ground cloud computing mode, the task is transparently forwarded to the ground cloud center via the access satellite for execution, and its latency is reduced. and energy consumption They are respectively:

[0121]

[0122]

[0123] in, For the communication capacity of the link between satellite and ground cloud, For ground-based cloud computing capacity, The distance between the satellite and the ground-based cloud. This refers to the transmission power from the satellite to the ground-based cloud.

[0124] Under the above three task execution modes, the energy consumption of ground-to-cloud computing is negligible, and the main consideration is the energy consumption during transmission.

[0125] Furthermore, in step S4, the task The decision variable for unloading in time slot t is denoted as... Its set of values ​​is:

[0126]

[0127] Where 0 indicates that the current time slot is not scheduled, 1 to S correspond to the satellite number, S+1 indicates that it is executed locally, and S+2 indicates that it is offloaded to the ground cloud;

[0128] Task energy consumption Defined as:

[0129]

[0130] The joint optimization problem aims to minimize the weighted combined cost of the system's total energy consumption and average information age, and satisfies the following constraints:

[0131]

[0132] in, and These represent the available computing resources and available communication resources of satellite s in time slot t, respectively.

[0133] Furthermore, in step S5, in the state space of the Markov decision process, the state space of the Markov decision process... Defined as:

[0134]

[0135] in, The task queue state of user u in time slot t. The available computing resources for satellite s in time slot t. For satellite s in time slot t, the available communication resources and user information (age status) are considered; action space. Includes discrete unloading actions and continuous resource allocation actions , ,Right now:

[0136]

[0137] The reward function is defined as:

[0138]

[0139] in, For information age reward items, To complete the reward item ahead of time, the specific formula is as follows:

[0140]

[0141] As an energy consumption penalty item, This is a penalty item for illegal actions; when a task is not scheduled or is executed locally, the corresponding satellite computing and communication resource allocation values ​​are set to zero; when a task is offloaded to the ground cloud, the corresponding satellite computing resource allocation value is set to zero; the aforementioned penalty item for illegal actions. The triggering conditions include at least one of the following: the physical distance between the cooperating satellite and the access satellite exceeds the maximum communication line distance of the inter-satellite link; the computing resources requested by the mission cause the total computing power consumption of the satellite to exceed the battery power supply limit; and the rain attenuation value of the ground cloud link exceeds the preset threshold.

[0142] This reward function is used to guide the system to reduce average AoI and system energy consumption while ensuring the legal scheduling of tasks.

[0143] Furthermore, in step S6, the cost-aware priority-based proximal policy optimization algorithm adopts an actor-critic network structure, and the continuous action head is modeled using a Beta distribution, with parameters satisfying:

[0144]

[0145] in, , This is the weight matrix. , h is the bias vector, and h is the state feature vector;

[0146] The policy objective function is:

[0147]

[0148] in:

[0149]

[0150] The algorithm further includes an in-trajectory cost-aware priority sampling mechanism, whereby the priority of the i-th sample is determined. Defined as:

[0151]

[0152] in, and These are the weighting coefficients. and These represent the total system energy consumption in adjacent time slots. and These represent the average information age of the systems in adjacent time slots. To prevent small positive numbers with zero priority;

[0153] The illegal action penalty item The triggering conditions include at least one of the following: the physical distance between the cooperating satellite and the access satellite exceeds the maximum communication line distance of the inter-satellite link; the computing resources requested by the mission cause the total computing power consumption of the satellite to exceed the battery power supply limit; and the rain attenuation value of the ground cloud link exceeds the preset threshold.

[0154] The sampling probability of the i-th sample is:

[0155]

[0156] in, The priority intensity parameter is used; and the reward is normalized online.

[0157]

[0158] It also monitors the approximate KL divergence, and terminates the current round of policy update in advance when it exceeds a preset threshold, so as to improve training stability.

[0159] like Figure 6 As shown, the method of the present invention can be performed according to the following steps:

[0160] Step S1: Initialize the user set, satellite set, task queue, system state, policy network parameters, and value network parameters;

[0161] Step S2: Receive newly arriving tasks in each time slot and write them into the task queue of each terminal;

[0162] Step S3: Based on the current state The CAP-PPO strategy network outputs discrete unloading actions and continuous resource allocation actions;

[0163] Step S4: Perform an action validity check and set invalid resource allocation components to zero;

[0164] Step S5: Determine the execution method of the task based on the action result: local, access satellite, cooperative satellite, or ground cloud;

[0165] Step S6: Calculate the corresponding processing latency, energy consumption, and reward value;

[0166] Step S7: Update the task queue status, satellite resource status, and user AoI status;

[0167] Step S8: Record the sample trajectory and calculate the sample priority based on the changes in system energy consumption and average AoI;

[0168] Step S9: Samples are extracted using a priority sampling method, the policy network is updated using the PPO-Clip objective function, and the value network is updated using the mean squared error.

[0169] Step S10: Output the trained policy model when the stopping condition is met; otherwise, return to step S2.

[0170] An edge computing offloading and resource allocation system for integrated space-ground networks is characterized by comprising: a task generation and caching module, an execution mode modeling module, an information age update module, a resource status management module, a joint optimization modeling module, a joint decision-making module, a strategy solving module, and a task execution and status update module. The task generation and caching module receives tasks generated by terminals and maintains a task queue. The execution mode modeling module establishes latency and energy consumption models under three task execution modes. The information age update module updates user information age. The resource status management module maintains satellite resource occupancy and release. The joint optimization modeling module constructs a joint optimization problem with the minimum weighted comprehensive cost. The joint decision-making module outputs offloading and resource allocation actions. The strategy solving module generates task scheduling strategies. The task execution and status update module executes task scheduling and updates the system status.

[0171] The present invention also provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above-described computational offloading and resource allocation method.

[0172] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the above-described computation offloading and resource allocation method.

[0173] Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing specific embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing specific embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for offloading and allocating resources for edge computing in integrated space-ground networks, characterized in that: The application in a satellite-ground-edge-cloud collaborative computing system consisting of a terminal layer, a satellite edge layer, and a ground cloud layer includes the following steps: S1. Construct a satellite-ground-edge-cloud collaborative computing architecture. The terminal layer includes multiple IoT terminals, and each IoT terminal maintains a task queue to cache dynamically arriving computing tasks. The satellite edge layer includes multiple low-orbit satellites equipped with edge computing servers, and the satellites share computing and communication resources through inter-satellite links. The ground cloud layer includes a ground cloud data center to handle tasks that the satellite edge layer cannot handle. S2. Model the terminal-generated tasks and user information age; S3. Establish latency and energy consumption models for three task execution modes: local computing, satellite edge computing, and terrestrial cloud computing. S4. Establish a joint optimization problem with unloading decision variables and resource allocation variables as optimization objects and minimizing the weighted comprehensive cost of system energy consumption and information age as the objective. S5. Model the joint optimization problem as a Markov decision process, and construct the state space, action space, reward function and state transition function; S6. A cost-aware priority-based near-end policy optimization algorithm is adopted to jointly solve the discrete offloading decision and the allocation of continuous computing and communication resources to obtain the task scheduling strategy for each time slot. S7. Execute task scheduling according to the task scheduling strategy, and update the task queue status, resource occupancy status, and user information age status.

2. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, The satellite-ground-edge-cloud collaborative computing system adopts a fully unloaded mode and a resource reservation mechanism where tasks are indivisible; when tasks... During time slot p scheduling, the duration of communication resources allocated to this task is... The computing resources allocated to this task will last for a duration of [duration]. The estimated communication time is The estimated calculation time is The length of a single time slot is ; The trigger condition for the resource reservation mechanism is: when the task The decision variable is scheduled and unloaded in time slot p. When the available communication and computing resources of the target unloading node meet the estimated requirements of the task, a resource reservation operation is triggered; the estimated communication time... The calculation is based on the current time slot's channel gain, available satellite bandwidth, and the projected communication resource allocation ratio. The calculation formula is as follows: ,in The uplink transmission rate estimated based on the current time slot channel status; The estimated calculation time The calculation is based on the total computational load of the mission and the proportion of the satellite's estimated allocable computational resources. The calculation formula is as follows: ,in To estimate the amount of computing resources that can be allocated to this mission based on the satellite's currently available computing resources; Once the task has completed all communication transmission and computation processing on the reserved resources, it will automatically release the occupied communication and computation resources and remark the released resources as available resources on the target node. When multiple tasks compete for the limited resources of the same target node, they are sorted from high to low according to a comprehensive priority score based on information age urgency, task waiting time and computational complexity, and resources are allocated in sequence. When resources are insufficient, low-priority tasks are not scheduled and are kept in the task queue to wait for the next time slot to re-enter the competition.

3. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, In step S2, the number of tasks generated by user u within time slot t is: , Obtain the parameter as Poisson distribution: The rules for dynamically updating user information age are as follows: The average age of the system's information is: 。 4. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, In step S3, the task of user u in local computing mode End-to-end processing delay in time slot t and energy consumption They are respectively: in, For the computing capacity of the user terminal, This represents the effective capacitance coefficient of the chip.

5. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, In step S3, the satellite edge computing mode is divided into a scenario without inter-satellite link (ISL) cooperation and a scenario with ISL cooperation. Without ISL cooperation, the task is performed by the access satellite, and its latency is... and energy consumption They are respectively: in, The communication resource allocation value for the access satellite in time slot t is assigned to the task. To access the computing resources allocated to the satellite for the task in time slot t, The distance between the user and the access satellite. At the speed of light, Transmit power for user terminals; When ISL collaboration is involved, the mission is offloaded to the collaborating satellite via the X-hop inter-satellite link for processing, and its latency is reduced. and energy consumption They are respectively: in, For inter-satellite link communication capacity, The distance between adjacent satellites. This refers to the inter-satellite relay power. The cooperative satellites were selected from a set of feasible satellites that met constraints on link connectivity, computing resources, and information age, based on a multi-dimensional comprehensive score considering hop count, available computing resources, current load, and queue length; hop count... No greater than the maximum number of hops determined by the maximum tolerable latency of the task. If no alternative cooperating satellite is available, the system will switch to ground-based cloud computing first; otherwise, the mission will be reserved for scheduling in the next time slot.

6. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, In step S3, in the ground cloud computing mode, the task is transparently forwarded to the ground cloud center via the access satellite for execution, and its latency... and energy consumption They are respectively: in, For the communication capacity of the link between satellite and ground cloud, For ground-based cloud computing capacity, The distance between the satellite and the ground-based cloud. This refers to the transmission power from the satellite to the ground-based cloud.

7. The method for offloading and allocating resources for edge computing in an integrated space-ground network as described in claim 1, characterized in that, In step S4, the task The decision variable for unloading in time slot t is denoted as... Its set of values ​​is: Where 0 indicates that the current time slot is not scheduled, 1 to S correspond to the satellite number, S+1 indicates that it is executed locally, and S+2 indicates that it is offloaded to the ground cloud; Task energy consumption Defined as: The joint optimization problem aims to minimize the weighted combined cost of the system's total energy consumption and average information age, and satisfies the following constraints: in, and These represent the available computing resources and available communication resources of satellite s in time slot t, respectively.

8. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, In step S5, the state space of the Markov decision process... Defined as: in, The task queue status of user u in time slot t. The available computing resources for satellite s in time slot t. For satellite s in time slot t, the available communication resources and user information (age status) are considered; action space. Includes discrete unloading actions and continuous resource allocation actions , ,Right now: The reward function is defined as: in, For information age reward items, To complete the reward items ahead of schedule, As an energy consumption penalty item, This is a penalty item for illegal actions; when a task is not scheduled or is executed locally, the corresponding satellite computing and communication resource allocation values ​​are set to zero; when a task is offloaded to the ground cloud, the corresponding satellite computing resource allocation value is set to zero; the aforementioned penalty item for illegal actions. The triggering conditions include at least one of the following: the physical distance between the cooperating satellite and the access satellite exceeds the maximum communication line distance of the inter-satellite link; the computing resources requested by the mission cause the total computing power consumption of the satellite to exceed the battery power supply limit; and the rain attenuation value of the ground cloud link exceeds the preset threshold.

9. The method for offloading and allocating resources for edge computing in an integrated space-ground network according to claim 1, characterized in that, In step S6, the cost-aware priority-based proximal policy optimization algorithm adopts an actor-critic network structure, and the continuous action head is modeled using a Beta distribution, with parameters satisfying: in, , This is the weight matrix. , Let h be the bias vector and h be the state feature vector; the algorithm further includes a cost-aware priority sampling mechanism within the trajectory, where the priority of the i-th sample is determined by the bias vector. Defined as: in, and These are the weighting coefficients. and These represent the total system energy consumption in adjacent time slots. and These represent the average information age of the systems in adjacent time slots. To prevent small positive numbers with zero priority; the sampling probability of the i-th sample is: in, This is a priority intensity parameter; and the reward is normalized online. The weighting coefficient and As a dynamically adaptive parameter, when the total system energy consumption continuously increases, Automatically increase the size to enhance focus on energy-deteriorating samples; when the system's average information age approaches its upper limit... Automatically increase the size to enhance attention to samples with deteriorating information freshness.

10. A system for offloading and allocating resources for edge computing in an integrated space-ground network, applicable to the method for offloading and allocating resources for edge computing in an integrated space-ground network as described in any one of claims 1 to 9, characterized in that, include: The system comprises a task generation and caching module, an execution mode modeling module, an information age update module, a resource status management module, a joint optimization modeling module, a joint decision-making module, a strategy solving module, and a task execution and status update module. The task generation and caching module receives tasks generated by the terminal and maintains a task queue. The execution mode modeling module establishes latency and energy consumption models under three task execution modes. The information age update module updates user information age. The resource status management module maintains satellite resource occupancy and release. The joint optimization modeling module constructs a joint optimization problem with the minimum weighted comprehensive cost. The joint decision-making module outputs unloading and resource allocation actions. The strategy solving module generates task scheduling strategies. The task execution and status update module executes task scheduling and updates the system status.

Images