Task scheduling method based on dynamic cloud edge collaboration

Abstract

The invention belongs to the technical field of mobile edge computing, and discloses a task scheduling method based on dynamic cloud edge collaboration. According to the invention, through link-level modeling and graph neural network feature extraction, adaptive expansion of different numbers of terminals and servers is realized; according to the method, the problems of instability and over-estimation in training are effectively avoided by combining cutting updating of a near-end strategy optimization algorithm and a generalized advantage estimation method. According to the method, the elastic computing power of the cloud and the edge low-delay characteristic can be fully utilized, the dynamically changing network scale and terminal requirements can be adapted, flexible unloading of tasks and dynamic resource allocation can be realized, task processing delay can be effectively reduced in a dynamic environment, the resource utilization rate is improved, and the resource utilization rate is increased. And good expansibility and stability are maintained under the condition that the number of terminals and the scale of the server fluctuate, the problems of computing resource bottleneck and task scheduling stiffness existing in the virtual reality video service of the existing mobile edge computing are solved, and the interaction experience of the user is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application mainly relates to the technical field of mobile edge computing, in particular to a task scheduling method based on dynamic cloud-edge collaboration. BACKGROUND

[0002] In recent years, with the rapid development of virtual reality (VR) technology, the global shipment volume of VR devices has been growing continuously. According to the statistics of the International Data Corporation (IDC), the shipment volume of VR devices in the third quarter of 2024 increased by 12.8% year-on-year. At the same time, more and more end users use head-mounted display devices with a resolution of 4K or even 8K and a refresh rate of up to 120Hz. Although such high-performance terminals significantly improve the immersive experience, their demand for computing and network resources also grows exponentially.

[0003] Mobile edge computing (MEC) technology can shorten the transmission path of tasks, reduce the delay, and meet the needs of real-time and bandwidth-sensitive applications such as VR, AR, and cloud gaming by deploying computing resources at the network edge. However, due to the complex computing tasks involved in VR applications such as high-definition video encoding, real-time scene rendering, and user interaction synchronization, relying solely on edge servers often faces the problem of insufficient computing power, especially in high-concurrency and multi-user scenarios, which is more likely to cause resource congestion and degradation of service quality.

[0004] Most existing task scheduling methods are based on static assumptions, such as assuming that the number of terminal devices and server resources are fixed and unchanging. This approach has obvious shortcomings in actual dynamic network environments: when the number of terminals suddenly increases or the network link conditions fluctuate, static scheduling is easy to cause task queue overflow, sharp rise in average processing delay, and even cause users to experience severe dizziness and discomfort, seriously affecting the VR business experience. In addition, static models lack adaptability to dynamic changes in resources and are difficult to apply in edge-cloud collaborative systems of different scales.

[0005] Existing technologies attempt to combine deep reinforcement learning (DRL), graph neural networks (GNN), and other intelligent optimization methods to improve the flexibility of edge-cloud task scheduling. Traditional intelligent optimization methods such as simulated annealing and genetic algorithms are not suitable for systems with millisecond-level delay sensitivity. Reinforcement learning (RL) methods with environmental interaction capabilities have become an ideal alternative solution due to the lack of historical data required to support the training of supervised learning models.

[0006] Reinforcement learning (RL) is a machine learning method that learns decision-making strategies through interaction with the environment. Its core principle is that the agent selects an action in a given state, and the environment responds with a new state and corresponding reward according to the action. The agent iteratively adjusts the strategy according to this feedback to maximize the long-term cumulative reward. In reinforcement learning (RL) algorithms, proximal policy optimization (PPO) has been widely applied due to its stability and effectiveness. Proximal policy optimization (PPO) optimizes the policy network by limiting the size of the policy update during each iteration, thereby maintaining training stability and effectively overcoming the curse of dimensionality related to discrete action space in traditional Deep Q-Network (DQN) algorithms. The importance sampling mechanism in proximal policy optimization (PPO) significantly improves sampling efficiency, which is particularly important in cloud-edge collaborative environments that require real-time response to millisecond-level VR interactions.

[0007] When the task involves graph-structured data with complex relationship dependencies, multilayer perceptron (MLP) often cannot capture the dependencies and topologies between nodes. In this case, graph neural network (GNN) provides obvious advantages. By utilizing the message passing mechanism, graph neural network (GNN) explicitly models the connections between nodes and extracts structural information feature representations. Therefore, graph neural network (GNN) performs superiorly in scenarios such as task offloading and resource allocation.

[0008] Currently, existing methods propose to reduce cold start overhead through deep reinforcement learning or optimize content caching based on graph neural networks. However, these methods mostly still remain at the level of node-level feature modeling, lacking fine-grained description of link-level dynamic relationships, making it difficult to fully cope with complex and variable VR scenarios. Existing technologies have limitations in resource scheduling staticization, insufficient system scalability, and excessive task delay. Therefore, there is an urgent need for a task scheduling method that can dynamically collaborate between the cloud and the edge. This method can fully utilize the elastic computing power of the cloud and the low latency characteristics of the edge, and adapt to dynamically changing network size and terminal demand to ensure the real-time and stability of VR video services. SUMMARY

[0009] To address the limitations of existing methods, which are mostly based on static assumptions and cannot adapt to dynamic fluctuations in the number of terminals and server resources, the present invention provides a task scheduling method based on dynamic cloud-edge collaboration. The method aims to achieve flexible task offloading and dynamic resource allocation, ensuring the real-time and stability of VR video services, and solving the computing resource bottleneck and task scheduling rigidity problems of existing mobile edge computing (MEC) in virtual reality (VR) video services, improving user interaction experience.

[0010] To achieve the above object, the application provides a task scheduling method based on dynamic cloud-edge cooperation, comprising the following steps:

[0011] Step 1: system modeling and heterogeneous graph construction:

[0012] Step 1.1: the server includes a cloud server , an edge server , a cloud-edge-end system model including a cloud server , an edge server and a terminal is established, and under the cloud-edge cooperation architecture, the cloud server node, the edge server node and the terminal node are modeled as a heterogeneous graph, the heterogeneous graph , wherein represents all possible terminal-server connections;

[0013] The terminal node features include task data volume , the edge server node features include computing power , and the link features between the terminal node and the server node include channel gain ;

[0014] Step 1.2: each terminal task is proportionally split into several subtasks, which are respectively offloaded to different servers, and transmission constraints and computing resource constraints are established;

[0015] Step 1.3: a total task delay model is established, including transmission delay and computing delay, and is stored in an initial scheduling table;

[0016] Step 2: generating a scheduling strategy based on a graph neural network GNN:

[0017] Step 2.1: heterogeneous graph feature extraction: using a graph neural network GNN to aggregate and update the terminal node features, the edge server node features and the link features between the terminal node and the server node, and extracting the relationship of task allocation, power allocation and resource allocation;

[0018] Step 2.2: scheduling strategy matrix output: through forward output of the graph neural network GNN, three types of scheduling strategy matrices are generated, including a task allocation matrix , a power allocation matrix and a resource allocation matrix , the task allocation matrix includes the offloading proportion between the terminal and the server, the power allocation matrix includes the link transmission power allocation, and the resource allocation matrix includes the server computing resource allocation proportion;

[0019] Step 2.3: Reward signal definition: Minimize the total delay as the optimization objective, and obtain the reward function after normalization;

[0020] Step 3: Update the scheduling policy using the Proximity-to-End (PPO) optimization algorithm:

[0021] Step 3.1: Model the scheduling problem as a Markov Decision Process (MDP). The state s includes the task data volume, channel gain, and computing power. The scheduling policy matrix a is a matrix of three types of scheduling policies, and the reward function is... This is the normalized time delay function;

[0022] Step 3.2: Scheduling strategy update: The near-end policy optimization algorithm (PPO) is used to constrain the iteration range of the scheduling strategy. Combined with the advantage function and the pruning loss function, the scheduling strategy is updated stably and iteratively.

[0023] Step 3.3: Calculate the dominance value using the generalized dominance estimation method (GAE) to avoid scheduling policy collapse and improve convergence speed;

[0024] Step 3.4: In each iteration, the scheduling policy matrix generated by the graph neural network (GNN) and the reward signal from step 2.3 are used together for optimization, ultimately resulting in a dynamically adjusted policy network;

[0025] Step 4: Dynamic cloud-edge collaborative execution:

[0026] Step 4.1: When edge server resources are sufficient, tasks should be executed on the local edge server first;

[0027] Step 4.2: When the edge server is overloaded, some subtasks can be offloaded to the cloud server;

[0028] Step 4.3: When there are multiple neighboring edge servers, tasks can be distributed through edge server-to-edge server collaboration to reduce the risk of single-point overload.

[0029] Furthermore, in step 1.2, the link Indicates terminal and The connection between them defines the task allocation ratio as follows: Transmission power is The computing resources allocated to the server are Then the transmission constraints must satisfy ,in, The representative will perform a summation operation. This represents the task allocation ratio between terminal i and the cloud server;

[0030] Computational resource constraints must be met ,in, Represents edge computing resources. Represents cloud computing resources. This represents the total resources of the server, ensuring that the total resources of the server do not exceed its maximum computing capacity.

[0031] Furthermore, step 1.3 establishes a total task latency model, including transmission latency and computation latency, and stores it in the initial scheduling table, including the following steps:

[0032] Step 1.3.1: Express the transmission delay as... ,in, With links Indicates terminal and The connection between them For the terminal and The task allocation ratio between them For the amount of task data, For the terminal and The link transmission rate between them For bandwidth, For the terminal and Transmission power between For noise power, For the terminal and Channel gain between For terminal k and Transmission power between For terminal k and The channel gain between them, where N is the number of terminals;

[0033] Step 1.3.2: Express the calculation delay as... ,in, The cost constant required to calculate per unit of data. Computing resources allocated to the server;

[0034] Step 1.3.3: Total delay is expressed as The total delay value serves as the core objective function for subsequent scheduling optimization and is also stored in the initial scheduling table.

[0035] Furthermore, the reward function in step 2.3 is expressed as follows: The reward value is in the range [0,1]. , To smooth out the delay, , Dynamically updated via exponential moving average. The value is 0.2, which is the time smoothing factor. The value is 0.1, which is the EMA update factor. and They represent time. The minimum and maximum values ​​of dynamically updated task processing latency.

[0036] Furthermore, in step 3.2, an objective function is adopted. To achieve stable iterative updates of the scheduling strategy, wherein, This represents the ratio of the probability of the new scheduling policy and the old scheduling policy being in the same state. Represents the dominance function. Represents the cutting parameters. The representative put The range is limited to [1−θ,1+θ]. min() represents the minimum value operation.

[0037] Beneficial Effects: This invention provides a task scheduling method based on dynamic cloud-edge collaboration. Through link-level modeling and graph neural network (GNN) feature extraction, it achieves adaptive scaling for different numbers of terminals and servers, overcoming the limitation of fixed input and output dimensions in traditional multilayer perceptron (MLP) methods. Combining the pruning update of the proximal policy optimization algorithm (PPO) and the generalized advantage estimation (GAE) method effectively avoids instability and overestimation problems during training, ensuring efficient and stable task scheduling. Experimental results show that the proposed dynamic edge-cloud collaborative scheduling method (DECS) reduces task processing latency by 8.52% and 20.54% compared to pure edge computing and pure cloud computing architectures, respectively, significantly improving the real-time interactive experience in VR services. Attached Figure Description

[0038] Figure 1 This is a basic structural diagram of the cloud-edge collaboration scenario involved in the embodiments of the present invention;

[0039] Figure 2 This is an algorithm structure diagram of the scalable dynamic edge-cloud collaborative scheduling method DECS in the cloud-edge collaborative scenario involved in the embodiments of the present invention;

[0040] Figure 3 This is a comparison chart of rewards under different models involved in the embodiments of the present invention;

[0041] Figure 4 This is a schematic diagram illustrating the task processing latency of different models under different numbers of servers in embodiments of the present invention;

[0042] Figure 5 This is a schematic diagram illustrating task processing latency for different numbers of edge servers involved in embodiments of the present invention; Figure 6 This is a schematic diagram illustrating the task processing latency for different numbers of terminals involved in embodiments of the present invention. Detailed Implementation

[0043] like Figures 1 to 6 As shown, to address the problems of rigid resource allocation and inability to adapt to dynamic terminal scale and computing resource fluctuations in existing edge computing for virtual reality (VR) task scheduling, this invention provides a task scheduling method based on dynamic cloud-edge collaboration. The Dynamic Edge-Cloud Collaborative Scheduling (DECS) method is based on graph neural networks (GNNs) and near-end policy optimization (PPO). This method, based on a cloud-edge collaborative architecture and combined with a reinforcement learning policy update mechanism, can achieve flexible task offloading and optimized resource allocation under any number of terminals and edge nodes, thereby effectively reducing task processing latency for VR video services and improving the user interaction experience.

[0044] Example 1: As Figure 1 As shown, the system model of this invention consists of end users, edge servers, and cloud servers. End users are connected to edge servers via base stations, edge servers collaborate with each other via local area networks, and edge servers are connected to the cloud center via backhaul links. The specific implementation process is as follows: Figure 2 As shown.

[0045] Step 1: System Modeling and Task Graph Construction

[0046] In the initial task scheduling phase, the computing tasks generated by end users have not yet been assigned, and all tasks are processed through the cloud center by default. At this point, as the process begins, the task allocation relationship between the terminal and the server has not yet been established. Therefore, a system model and task graph must first be constructed for subsequent dynamic scheduling optimization, such as... Figure 3 As shown.

[0047] Step 1.1: Node and link initialization, denoted as terminal. Edge server Cloud server is denoted as For terminal nodes, the characteristic is the amount of task data. For edge server nodes, the characteristic is computing power. For cloud server nodes, the computing power is set to be significantly higher than that of the edge nodes, and the transmission link latency is set to be twice that of the edge links. Link characteristics include channel gain. This is used to represent the channel quality between the terminal and the server. Therefore, a heterogeneous graph is constructed. ,in This represents all possible terminal-server connections.

[0048] Step 1.2: Divide each terminal task into several sub-tasks according to the proportion, and offload them to different servers to establish transmission constraints and computing resource constraints;

[0049] With link Indicates terminal and The connection between them defines the task allocation ratio as follows: Transmission power is The computing resources allocated to the server are Then the transmission constraints must satisfy ,in, The representative will perform a summation operation. This represents the task allocation ratio between terminal i and the cloud server;

[0050] Computational resource constraints must be met ,in, Represents edge computing resources. Represents cloud computing resources. This represents the total resources of the server, ensuring that the total resources of the server do not exceed its maximum computing capacity.

[0051] Step 1.3: Establish a total task latency model, including transmission latency and computation latency, and store it in the initial scheduling table;

[0052] Includes the following steps:

[0053] Step 1.3.1: Express the transmission delay as... ,in, With links Indicates terminal and The connection between them For the terminal and The task allocation ratio between them For the amount of task data, For the terminal and The link transmission rate between them For bandwidth, For the terminal and Transmission power between For noise power, For the terminal and Channel gain between For terminal k and Transmission power between For terminal k and The channel gain between them, where N is the number of terminals;

[0054] Step 1.3.2: Express the calculation delay as... ,in, The cost constant required to calculate per unit of data. Computing resources allocated to the server;

[0055] Step 1.3.3: Total delay is expressed as The total latency value serves as the core objective function for subsequent scheduling optimization and is stored in the initial scheduling table for later GNN modeling and PPO policy training. Through the initialization in step 1, the system completes the definition of terminal nodes, server nodes, and their link relationships during the modeling phase, and establishes the task splitting ratio, resource constraints, and latency calculation model, laying the foundation for subsequent feature extraction and scheduling optimization based on graph neural networks.

[0056] Step 2: Generate scheduling strategy based on graph neural network (GNN):

[0057] Step 2.1: Heterogeneous graph feature extraction: Graph neural network (GNN) is used to aggregate and update terminal node features, edge server node features, and link features between terminal nodes and server nodes, and to extract the relationships between task allocation, power allocation, and resource allocation.

[0058] Step 2.2: Scheduling Policy Matrix Output: Three types of scheduling policy matrices are generated through the forward output of the graph neural network (GNN), including the task allocation matrix. Power allocation matrix and resource allocation matrix Task allocation matrix Including the offloading ratio between terminals and servers, and the power allocation matrix. Including link transmission power allocation and resource allocation matrix This includes the allocation ratio of server computing resources;

[0059] Step 2.3: Reward signal definition: Minimize the total delay as the optimization objective, and obtain the reward function after normalization;

[0060] To guide scheduling optimization, the goal of minimizing latency is transformed into a reward function, expressed as follows: ,in, , To smooth out the delay, , Dynamically updated via exponential moving average. The value is 0.2, which is the time smoothing factor. The value is 0.1, which is the EMA update factor. and They represent time. The minimum and maximum values ​​of dynamically updated task processing latency.

[0061] Step 3: Update the scheduling policy using the Proximity-to-End (PPO) optimization algorithm:

[0062] Step 3.1: Model the scheduling problem as a Markov Decision Process (MDP). The state s includes the task data volume, channel gain, and computing power. The scheduling policy matrix a is a matrix of three types of scheduling policies, and the reward function is... This is the normalized time delay function;

[0063] Step 3.2: Scheduling Policy Update: The Nearest Policy Optimization (PPO) algorithm is used to constrain the iteration range of the scheduling policy. The objective function is expressed as follows: To achieve stable iterative updates of the scheduling strategy, wherein, This represents the ratio of the probability of the new scheduling policy and the old scheduling policy being in the same state. Represents the dominance function. Represents the cutting parameters. The representative put The objective function is limited to the range [1−θ,1+θ], and min() represents the minimum value operation. The objective function combines the advantage function and the pruning loss function to achieve stable iterative updates of the scheduling strategy.

[0064] Step 3.3: Calculate the dominance value using the Generalized Dominance Estimation (GAE) method, expressed as... ,in, As a discount factor, A factor that balances bias and variance. For timing difference error, if A value greater than 0 indicates that the current action is better than expected. <0 indicates that the situation is worse than expected, which helps to prevent the scheduling strategy from collapsing and improve the convergence speed.

[0065] Step 3.4: In each iteration, the scheduling policy matrix generated by the graph neural network (GNN) and the reward signal from step 2.3 are used together for optimization, ultimately resulting in a dynamically adjusted policy network;

[0066] Step 4: Dynamic cloud-edge collaborative execution:

[0067] Step 4.1: When edge server resources are sufficient, tasks should be executed on the local edge server first;

[0068] Step 4.2: When the edge server is overloaded, some subtasks can be offloaded to the cloud server;

[0069] Step 4.3: When there are multiple neighboring edge servers, tasks can be distributed through edge server-to-edge server collaboration to reduce the risk of single-point overload.

[0070] This process ensures efficient task allocation and minimizes latency in dynamic environments.

[0071] To verify the effectiveness of this invention, the proposed Dynamic Edge-Cloud Collaborative Scheduling (DECS) method for dynamic cloud-edge collaborative task scheduling is compared with traditional edge computing and cloud computing methods. The comparison methods include:

[0072] 1) Pure edge computing architecture: All tasks are processed by edge servers;

[0073] 2) Pure cloud computing architecture: All tasks are offloaded to cloud servers for processing;

[0074] 3) Traditional MLP+DQN scheduling method: Uses a neural network with fixed input and output dimensions for task allocation.

[0075] In the experimental scenario, the number of terminals was set to twice the number of servers, and the performance of different methods in terms of training convergence, task processing latency, and resource utilization was compared through simulation.

[0076] like Figure 3 As shown, the reward value changes for different methods as the number of training iterations increases. It can be seen that the traditional MLP+DQN method experiences significant oscillations in the initial stages, has a slower convergence speed, and ultimately yields a lower reward value than the DECS method proposed in this invention. In contrast, the DECS method gradually stabilizes after 300 iterations, with the reward value approaching the optimal level, indicating that this invention is more stable in terms of scheduling policy learning.

[0077] like Figure 4 and Figure 5 As shown, the task processing latency of different methods is compared under different server scales. The results show that: when the number of servers is limited, the average task processing latency of the pure edge computing architecture is relatively high due to insufficient single-point computing power; although the pure cloud computing architecture has strong computing power, the additional latency caused by remote link transmission significantly increases the total latency; the DECS method proposed in this invention, under the cloud-edge collaborative architecture, can adaptively adjust the offloading ratio and resource allocation, and always maintain a low processing latency. Compared with the pure edge computing and pure cloud computing methods, the average task processing latency is reduced by 8.52% and 20.54%, respectively.

[0078] like Figure 6 As shown, the latency trends of different methods under a surge in terminals are compared. When the number of terminals increases rapidly, the latency of the pure edge computing method increases linearly, while the performance of the pure cloud computing method degrades due to increased link load. The DECS method of this invention, by dynamically adjusting the cloud-edge offloading ratio, significantly slows down the latency growth curve and remains superior to the comparative methods in large-scale scenarios.

[0079] As shown in Table 1, the comparison results of CPU resource utilization show that, under the same scale, pure edge computing has excessively high utilization due to the limited single-machine capability, while pure cloud computing has idle resources. However, the DECS method of this invention can reasonably distribute the load between the cloud and the edge, so that the edge side utilization is reduced to about 68% and the cloud side utilization is stabilized at about 30%, thereby effectively avoiding the problems of single-point overload and resource waste.

[0080] .

[0081] This invention provides a task scheduling method based on dynamic cloud-edge collaboration, shifting resource allocation from node-level features to link-level features. Each link between the terminal and the server is assigned an independent feature representation. Because the model processes features at the link level, rather than producing fixed-dimensional outputs at the node level, additional terminals and servers can be naturally accommodated by increasing the number of links without modifying the model's structure or parameters. This invention fully utilizes the elastic computing power of the cloud and the low latency of the edge, adapting to dynamically changing network scale and terminal demands. It enables flexible task offloading and dynamic resource allocation, effectively reducing task processing latency and improving resource utilization in dynamic environments. It maintains good scalability and stability even with fluctuations in the number of terminals and server scale. This solves the computational resource bottlenecks and rigid task scheduling problems existing in current mobile edge computing (MEC) for virtual reality (VR) video services, significantly improving the user's interactive experience.

[0082] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. However, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A task scheduling method based on dynamic cloud-edge collaboration, characterized in that, Includes the following steps: Step 1: System Modeling and Heterogeneous Graph Construction: Step 1.1: Servers include cloud servers Edge servers Establish cloud servers Edge servers and terminal The cloud-edge-device system model, under the cloud-edge collaborative architecture, models cloud server nodes, edge server nodes, and terminal nodes as a heterogeneous graph. ,in This represents all possible terminal-server connections; Terminal node characteristics include task data volume Edge server node characteristics include computing power The link characteristics between the terminal node and the server node include channel gain. ; Step 1.2: Divide each terminal task into several sub-tasks according to the proportion, and offload them to different servers to establish transmission constraints and computing resource constraints; Step 1.3: Establish a total task latency model, including transmission latency and computation latency, and store it in the initial scheduling table; Step 2: Generate scheduling strategy based on graph neural network (GNN): Step 2.1: Heterogeneous graph feature extraction: Graph neural network (GNN) is used to aggregate and update terminal node features, edge server node features, and link features between terminal nodes and server nodes, and to extract the relationships between task allocation, power allocation, and resource allocation. Step 2.2: Scheduling Policy Matrix Output: Three types of scheduling policy matrices are generated through the forward output of the graph neural network (GNN), including the task allocation matrix. Power allocation matrix and resource allocation matrix The task allocation matrix Including the offloading ratio between terminals and servers, the power allocation matrix Including link transmission power allocation, the resource allocation matrix This includes the allocation ratio of server computing resources; Step 2.3: Reward signal definition: Minimize the total delay as the optimization objective, and obtain the reward function after normalization; Step 3: Update the scheduling policy using the Proximity-to-End (PPO) optimization algorithm: Step 3.1: Model the scheduling problem as a Markov Decision Process (MDP). The state s includes the task data volume, channel gain, and computing power. The scheduling policy matrix a is a matrix of three types of scheduling policies, and the reward function is... This is the normalized time delay function; Step 3.2: Scheduling strategy update: The near-end policy optimization algorithm (PPO) is used to constrain the iteration range of the scheduling strategy. Combined with the advantage function and the pruning loss function, the scheduling strategy is updated stably and iteratively. Step 3.3: Calculate the dominance value using the generalized dominance estimation method (GAE) to avoid scheduling policy collapse and improve convergence speed; Step 3.4: In each iteration, the scheduling policy matrix generated by the graph neural network (GNN) and the reward signal from step 2.3 are used together for optimization, ultimately resulting in a dynamically adjusted policy network; Step 4: Dynamic cloud-edge collaborative execution: Step 4.1: When edge server resources are sufficient, tasks should be executed on the local edge server first; Step 4.2: When the edge server is overloaded, some subtasks can be offloaded to the cloud server; Step 4.3: When there are multiple neighboring edge servers, tasks can be distributed through edge server-to-edge server collaboration to reduce the risk of single-point overload.

2. The task scheduling method based on dynamic cloud-edge collaboration according to claim 1, characterized in that, In step 1.2, using the link Indicates terminal and The connection between them defines the task allocation ratio as follows: Transmission power is The computing resources allocated to the server are Then the transmission constraints must satisfy ,in, The representative will perform a summation operation. This represents the task allocation ratio between terminal i and the cloud server; Computational resource constraints must be met ,in, Represents edge computing resources. Represents cloud computing resources. This represents the total resources of the server, ensuring that the total resources of the server do not exceed its maximum computing capacity.

3. The task scheduling method based on dynamic cloud-edge collaboration according to claim 1, characterized in that, Step 1.3 Establish the total task latency model, including transmission latency and computation latency, and store it in the initial scheduling table, including the following steps: Step 1.3.1: Express the transmission delay as... ,in, With links Indicates terminal and The connection between them For the terminal and The task allocation ratio between them For the amount of task data, For the terminal and The link transmission rate between them For bandwidth, For the terminal and Transmission power between For noise power, For the terminal and Channel gain between For terminal k and Transmission power between For terminal k and The channel gain between them, where N is the number of terminals; Step 1.3.2: Express the calculation delay as... ,in, The cost constant required to calculate per unit of data. Computing resources allocated to the server; Step 1.3.3: Total delay is expressed as The total delay value serves as the core objective function for subsequent scheduling optimization and is also stored in the initial scheduling table.

4. The task scheduling method based on dynamic cloud-edge collaboration according to claim 1, characterized in that, The reward function in step 2.3 is expressed as follows: The reward value is in the range [0,1]. , To smooth out the delay, , Dynamically updated via exponential moving average. The value is 0.2, which is the time smoothing factor. The value is 0.1, which is the EMA update factor. and They represent time. The minimum and maximum values ​​of dynamically updated task processing latency.

5. The task scheduling method based on dynamic cloud-edge collaboration according to claim 1, characterized in that, In step 3.2, the objective function is adopted. To achieve stable iterative updates of the scheduling strategy, wherein, This represents the ratio of the probability of the new scheduling policy and the old scheduling policy being in the same state. Represents the dominance function. Represents the cutting parameters. The representative put The range is limited to [1−θ,1+θ]. min() represents the minimum value operation.

Images