Network task processing method and device based on cloud-edge collaboration, equipment and medium

Abstract

The application discloses a cloud-edge collaboration-based network task processing method and device, equipment and medium. The method comprises the following steps: obtaining the task cost and resource requirement information of the network task to be processed in the cloud-edge collaboration architecture, combining the network load data and the equipment load data of the edge node, and constructing a dynamic decision mechanism; determining the comprehensive decision score according to the network task, the task cost and the network load data; calculating the dynamic allocation threshold according to the resource requirement information, the network load and the equipment load data; determining the target node from the cloud node or the edge node by comparing the comprehensive decision score with the dynamic allocation threshold, and calling the large model intelligent agent deployed in the target node to process the network task. The application can effectively reduce the response delay and resource consumption of the network task, improve the overall network service quality and optimize the operation and maintenance cost. The technical scheme of the application can be widely applied in the field of network technology.

Description

Technical Field

[0001] This application relates to the field of network technology, and in particular to a network task processing method, apparatus, device and medium based on cloud-edge collaboration. Background Technology

[0002] With the rapid development of communication technologies, network architecture is gradually evolving towards cloudification, virtualization, and distribution. Cloud-edge collaboration has become a key infrastructure supporting the Internet of Things and real-time services. Against this backdrop, the complexity and dynamism of network tasks (such as topology optimization, resource allocation, fault location, and traffic prediction) have increased significantly.

[0003] In related technologies, traditional network task processing methods have significant limitations when facing dynamic and complex network environments, making it difficult to meet the needs of intelligent development. For example, existing task processing methods often rely on static rules or manual intervention, resulting in low task processing efficiency and becoming a bottleneck restricting the improvement of network service quality and cost optimization. Summary of the Invention

[0004] This application provides a cloud-edge collaborative network task processing method, apparatus, device, and medium, which can effectively reduce network task response latency and resource consumption, improve overall network service quality, and optimize operation and maintenance costs.

[0005] One aspect of this application provides a network task processing method based on cloud-edge collaboration, the method comprising: Obtain the network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks; wherein, the cloud-edge collaborative architecture includes cloud nodes and edge nodes; Collect network load data of the cloud-edge collaborative architecture and device load data of the edge nodes; Based on the network task, the task cost, and the network load data, determine the comprehensive decision score corresponding to the network task; Based on the resource demand information, the network load data, and the device load data, determine the dynamic allocation threshold corresponding to the network task; Based on the comprehensive decision score and the dynamic allocation threshold, the target node for processing the network task is determined from the cloud-edge collaborative architecture, and the network task is processed by a large language model agent deployed on the target node.

[0006] For example, in some embodiments, determining the comprehensive decision score corresponding to the network task based on the network task, the task cost, and the network load data includes: Obtain a predefined processing strategy; wherein the processing strategy is used to indicate the type of task that the cloud node and the edge node are suitable for processing; Based on the processing strategy, a first prompt word is constructed; Based on the first prompt word, the applicable nodes of the network task are inferred and judged by a large language model to obtain the task type score corresponding to the network task; wherein, the task type score is used to indicate the matching score of the cloud node and the edge node as the applicable nodes of the network task. Based on the task type score, the task cost, and the network load data, a comprehensive decision score is determined for the network task.

[0007] For example, in some embodiments, the task cost includes task consumption cost and task latency cost; determining the comprehensive decision score corresponding to the network task based on the task type score, the task cost, and the network load data includes: Based on the network load data, determine the network load score of the cloud-edge collaborative architecture; The comprehensive decision score corresponding to the network task is obtained by weighted summing of the task consumption cost, the task delay cost, the task type score, and the network load score.

[0008] For example, in some embodiments, determining the dynamic allocation threshold corresponding to the network task based on the resource demand information, the network load data, and the device load data includes: Based on the resource requirement information, determine the latency sensitivity and resource requirements of the network task; Based on the network load data, determine the network load score of the cloud-edge collaborative architecture; Based on the device load data, determine the device load score of the edge node; The dynamic allocation threshold corresponding to the network task is determined based on the network load score, the device load score, the latency sensitivity, and the resource requirement.

[0009] For example, in some embodiments, the step of determining the target node for processing the network task from the cloud-edge collaborative architecture based on the comprehensive decision score and the dynamic allocation threshold is as follows: The comprehensive decision score and the dynamic allocation threshold are compared; If the comprehensive decision score is less than the dynamic allocation threshold, the edge node will be identified as the target node. If the comprehensive decision score is greater than or equal to the dynamic allocation threshold, the cloud node will be identified as the target node.

[0010] Exemplary, in some embodiments, the large language model agent is deployed through the following steps: Collect background knowledge datasets in the network field; Using the aforementioned background knowledge dataset, the base large language model is incrementally pre-trained to obtain a domain-adaptive model; For each node in the cloud-edge collaborative architecture, a corresponding toolset is configured for the domain adaptation model according to a predefined processing strategy to obtain the basic intelligent agent corresponding to each node; Obtain the prompt word template set for each of the basic intelligent agents, and load the prompt word template into the basic intelligent agent; wherein, the prompt word template includes at least one of capability definition, business background, business objective, business details and output requirements; The ReAct inference framework is embedded in the basic agent to obtain a deployed large language model agent.

[0011] For example, in some embodiments, processing the network task through a large language model agent deployed on the target node includes: The network task is parsed to obtain task information; The prompt word template is filled in according to the task information to obtain the second prompt word; Based on the second prompt word, the network task is processed by the large language model agent to obtain the corresponding processing result.

[0012] On the other hand, embodiments of this application provide a network task processing device based on cloud-edge collaboration, the device comprising: The acquisition unit is used to acquire network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks; wherein, the cloud-edge collaborative architecture includes cloud nodes and edge nodes; The data acquisition unit is used to collect network load data of the cloud-edge collaborative architecture and device load data of the edge nodes. The first analysis unit is used to determine the comprehensive decision score corresponding to the network task based on the network task, the task cost, and the network load data. The second analysis unit is used to determine the dynamic allocation threshold corresponding to the network task based on the resource demand information, the network load data, and the device load data. An execution unit is configured to determine the target node for processing the network task from the cloud-edge collaborative architecture based on the comprehensive decision score and the dynamic allocation threshold, and to process the network task through a large language model agent deployed on the target node.

[0013] On the other hand, embodiments of this application provide an electronic device, including a processor and a memory; The memory is used to store computer programs; The processor executes the computer program to implement the aforementioned cloud-edge collaborative network task processing method.

[0014] On the other hand, embodiments of this application provide a computer-readable storage medium storing a computer program, which is executed by a processor to implement the aforementioned cloud-edge collaborative network task processing method.

[0015] On the other hand, embodiments of this application also provide a computer program product, which includes a computer program stored in a computer-readable storage medium. The processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the aforementioned cloud-edge collaborative network task processing method.

[0016] The embodiments of this application include at least the following beneficial effects: This application provides a network task processing method, apparatus, device, and medium based on cloud-edge collaboration. This application acquires the task cost and resource requirement information of the network task to be processed in the cloud-edge collaborative architecture, and combines this with network load data and edge node device load data to construct a dynamic decision-making mechanism. A comprehensive decision score is determined based on the network task, task cost, and network load data. This score quantifies the execution cost and adaptability of the task under the current network state. A dynamic allocation threshold is calculated based on resource requirement information, network load, and device load data, enabling the task allocation boundary to change in real time with the network environment and edge computing power. By comparing the comprehensive decision score with the dynamic allocation threshold, a target node is determined from cloud nodes or edge nodes, and a large-scale intelligent model deployed on the target node is invoked to process the network task. Through the above methods, this application overcomes the limitations of traditional static rules or manual intervention, achieving adaptive matching of cloud-edge resources and automated closed-loop execution of tasks. This effectively reduces network task response latency and resource consumption, improves overall network service quality, and optimizes operation and maintenance costs. Attached Figure Description

[0017] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0018] Figure 1 This is a system architecture diagram of a cloud-edge collaborative network task processing method provided in the embodiments of this application; Figure 2 This is a flowchart illustrating a method for determining the comprehensive decision score corresponding to a network task, as provided in an embodiment of this application. Figure 3 This is a schematic diagram of the architecture of a cloud-edge collaborative network task processing system provided in the embodiments of this application; Figure 4 This is a schematic diagram of the operation flow of a cloud-edge collaborative network task processing system provided in the embodiments of this application; Figure 5 This is a structural block diagram of a cloud-edge collaborative network task processing device provided in the embodiments of this application; Figure 6 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] It is understood that the terms “first,” “second,” etc., used in this application may be used to describe various concepts herein, but unless otherwise stated, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another.

[0021] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0022] With the rapid development of communication technologies, network architecture is gradually evolving towards cloudification, virtualization, and distribution. Cloud-edge collaboration has become a key infrastructure supporting the Internet of Things and real-time services. Against this backdrop, the complexity and dynamism of network tasks (such as topology optimization, resource allocation, fault location, and traffic prediction) have increased significantly.

[0023] In related technologies, traditional network task processing methods have significant limitations when facing dynamic and complex network environments, making it difficult to meet the needs of intelligent development. For example, existing task processing methods often rely on static rules or manual intervention, resulting in low task processing efficiency and becoming a bottleneck restricting the improvement of network service quality and cost optimization.

[0024] In view of this, this application provides a network task processing method, apparatus, device, and medium based on cloud-edge collaboration. This application acquires the task cost and resource requirement information of the network task to be processed in the cloud-edge collaborative architecture, and combines this with network load data and edge node device load data to construct a dynamic decision-making mechanism. A comprehensive decision score is determined based on the network task, task cost, and network load data. This score quantifies the execution cost and adaptability of the task under the current network state. A dynamic allocation threshold is calculated based on resource requirement information, network load, and device load data, enabling the task allocation boundary to change in real time with the network environment and edge computing power. By comparing the comprehensive decision score with the dynamic allocation threshold, a target node is determined from cloud nodes or edge nodes, and a large-scale intelligent model deployed on the target node is invoked to process the network task. Through this method, this application overcomes the limitations of traditional static rules or manual intervention, achieving adaptive matching of cloud-edge resources and automated closed-loop execution of tasks. This effectively reduces network task response latency and resource consumption, improves overall network service quality, and optimizes operation and maintenance costs.

[0025] General Description of Embodiments in this Application Please refer to Figure 1 , Figure 1 This illustration shows a flowchart of a cloud-edge collaborative network task processing method provided in an embodiment of this application. Figure 1 As shown, a cloud-edge collaborative network task processing method according to an embodiment of this application includes, but is not limited to, the following steps: Step 110: Obtain the network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks; wherein, the cloud-edge collaborative architecture includes cloud nodes and edge nodes; Step 120: Collect network load data and device load data of the cloud-edge collaborative architecture; Step 130: Determine the comprehensive decision score corresponding to the network task based on the network task, task cost, and network load data; Step 140: Determine the dynamic allocation threshold for network tasks based on resource demand information, network load data, and device load data; Step 150: Based on the comprehensive decision score and dynamic allocation threshold, determine the target node for processing network tasks from the cloud-edge collaborative architecture, and process the network tasks through a large language model agent deployed on the target node.

[0026] In this application embodiment, a network task processing method based on cloud-edge collaboration is provided. This method aims to improve the problem that the scheduling of network tasks in related technologies relies on static rules or manual intervention, resulting in insufficient resource utilization, low task processing efficiency and high operation and maintenance costs. By constructing a dynamic decision-making and intelligent agent execution framework based on multi-dimensional data fusion, adaptive matching of cloud-edge resources and automated closed-loop processing of network tasks can be achieved.

[0027] Specifically, this method first acquires the task cost and resource requirement information of the network tasks to be processed in the cloud-edge collaborative architecture, and simultaneously collects network load data and device load data of edge nodes. Based on this, a comprehensive decision score is calculated according to the network task, task cost, and network load data to quantify the task's adaptability to the current environment. Simultaneously, a dynamic allocation threshold is determined based on resource requirement information, network load, and device load data, allowing the task allocation boundary to change in real time with the network status. By comparing the comprehensive decision score with the dynamic allocation threshold, the target node is accurately determined from cloud or edge nodes, and the agent deployed on that node is invoked to automate the processing of the network task. This method combines dynamic threshold decision-making with large-model agent inference, effectively reducing task response latency and resource consumption, and improving network service quality and operational efficiency.

[0028] Below, in conjunction with Figure 1 This paper introduces and explains each step of the cloud-edge collaborative network task processing method in the embodiments of this application.

[0029] In step 110, obtain the network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks.

[0030] In this embodiment, network tasks refer to various management and control operations that need to be performed in the network, such as topology optimization, resource allocation, fault location, or traffic prediction. These network tasks typically have different business objectives and technical characteristics; some focus on long-term strategic planning from a global perspective, while others focus on immediate responses to local environments. Due to the diversity and complexity of network tasks, matching the most suitable computing resources to them based on their own characteristics (such as sensitivity to latency and computational power consumption requirements) is a challenge to ensuring network service quality.

[0031] The cloud-edge collaborative architecture is a distributed computing framework that combines the powerful computing capabilities of cloud computing with the low latency of edge computing. It primarily consists of cloud nodes and edge nodes. Cloud nodes typically possess massive computing and storage resources, making them suitable for handling non-real-time, globally optimized, heavy-duty tasks. Edge nodes, deployed close to the data source or user at the network edge, offer low network latency and high bandwidth, making them suitable for handling localized tasks with high real-time requirements. This application leverages the heterogeneous nature of this architecture and, through a dynamic scheduling mechanism, fully utilizes the respective advantages of the cloud and edge to achieve efficient processing of network tasks.

[0032] When processing network tasks using the methods described in this application, obtaining the relevant attribute information of the network task to be processed is the foundation for subsequent intelligent decision-making. Unlike traditional methods that only focus on the basic input and output of network tasks, this application divides task attributes into two dimensions: task cost and resource requirement information, in order to accurately profile the task from both consumption and demand perspectives. By analyzing the type, priority, and expected goals of the network task, the system can extract key parameters characterizing the difficulty of task execution, providing data support for subsequent differentiated scheduling.

[0033] Specifically, in this embodiment, the task cost typically includes the expected consumption of computing, storage, and bandwidth during the execution of the network task, while the resource requirement information focuses on the task's sensitivity to latency and its dependence on specific hardware environments. Here, this information can be read from the network management system or service orchestrator through a standardized interface and converted into unified quantitative indicators.

[0034] In step 120, network load data of the cloud-edge collaborative architecture and device load data of edge nodes are collected.

[0035] In this embodiment, by continuously collecting macroscopic network load data and microscopic edge device load data, a panoramic view of the current system resource status can be constructed. It is easy to understand that network load data reflects the congestion level of the current communication link, while edge node device load data reveals the remaining capacity of edge computing resources. The combination of the two can accurately reflect the external environment for task execution. This data can serve as a reference for subsequent scheduling of network tasks.

[0036] Specifically, in this embodiment, the bandwidth utilization, packet loss rate, and transmission latency of the core network and access network can be collected periodically using Simple Network Management Protocol (SNMP) or telemetry technology, serving as a quantitative basis for network load data. Simultaneously, for edge nodes, key performance indicators such as CPU utilization, memory usage, and disk I / O are monitored. This real-time collected data not only reflects the current resource bottlenecks but also provides crucial input parameters for subsequent dynamic allocation threshold calculations, enabling the system to adjust scheduling strategies promptly during network fluctuations or device overload, preventing task failures or performance degradation due to resource contention.

[0037] In step 130, the comprehensive decision score corresponding to the network task is determined based on the network task, task cost, and network load data.

[0038] In this embodiment of the application, the calculation of the comprehensive decision score aims to integrate multi-dimensional task characteristics and environmental characteristics into a quantitative decision index. This step can use a preset mapping relationship or reasoning model to correlate the task cost with the current network load data to obtain a score that can comprehensively evaluate the performance of the task in the current environment. This process effectively improves the problem of low task-resource matching in traditional methods.

[0039] Here, the overall decision score directly reflects the urgency or suitability of network tasks under the current network conditions. A higher score usually means that the task needs to be migrated to high-computing nodes or requires special scheduling strategies, thus providing core decision-making basis for subsequent cloud-edge allocation.

[0040] In step 140, the dynamic allocation threshold corresponding to the network task is determined based on the resource demand information, network load data, and device load data.

[0041] In this embodiment, the dynamic allocation threshold setting is a control mechanism that distinguishes whether tasks should flow to cloud nodes or edge nodes. Unlike the existing technology that uses a fixed threshold for a one-size-fits-all allocation method, this application introduces resource demand information, network load, and edge device load as joint variables, and generates a dynamic allocation threshold in real time through multi-objective optimization calculation. This dynamic adjustment mechanism enables the system to sensitively perceive changes in the network environment and device status, and adaptively adjust the allocation boundaries of cloud and edge tasks.

[0042] In step 150, the target node for processing network tasks is determined from the cloud-edge collaborative architecture based on the comprehensive decision score and the dynamic allocation threshold, and the network tasks are processed by the large language model agent deployed on the target node.

[0043] In this embodiment, the determination of the target node and subsequent processing constitute a closed loop for task scheduling. By comparing the comprehensive decision score calculated in the preceding steps with the dynamic allocation threshold, the final node selection can be made in a data-driven manner, ensuring that each task is assigned to the most suitable physical location for its execution. Subsequently, the large language model agent deployed on the target node is used for task processing, realizing a leap from traditional rule execution to intelligent and automated decision-making.

[0044] Specifically, based on the comprehensive decision score and the dynamic allocation threshold, a simple numerical comparison logic can be executed: if the comprehensive decision score is less than the dynamic allocation threshold, it is determined that the edge node is competent or the network task requires lower latency, and thus the edge node is selected; otherwise, it is determined that the network task requires strong computing power support from the cloud, and the cloud node is selected.

[0045] After identifying the target node, the context of the network task is passed to the large language model agent deployed on the target node. Based on its built-in domain knowledge and reasoning capabilities, this agent deeply understands and automatically executes the network task, completing complex network configuration, optimization, or fault repair operations without human intervention, significantly improving the efficiency and intelligence of network operation and maintenance.

[0046] It is understood that the cloud-edge collaborative network task processing method provided in this application embodiment obtains the task cost and resource requirement information of the network task to be processed in the cloud-edge collaborative architecture, and combines it with network load data and edge node device load data to construct a dynamic decision-making mechanism; determines a comprehensive decision score based on the network task, task cost, and network load data, which quantifies the execution cost and adaptability of the task in the current network state; calculates a dynamic allocation threshold based on resource requirement information, network load, and device load data, so that the task allocation boundary can change in real time with the network environment and edge computing power; by comparing the comprehensive decision score and the dynamic allocation threshold, a target node is determined from cloud nodes or edge nodes, and a large model intelligence agent deployed on the target node is invoked to process the network task. Through the above approach, this method overcomes the limitations of traditional static rules or manual intervention, realizes adaptive matching of cloud and edge resources and automated closed-loop execution of tasks, effectively reduces network task response latency and resource consumption, improves overall network service quality, and optimizes operation and maintenance costs.

[0047] Specifically, in some embodiments, reference is made to Figure 2 Based on network task, task cost, and network load data, a comprehensive decision score is determined for each network task, including: Obtain predefined processing strategies; whereby processing strategies are used to indicate the types of tasks that cloud nodes and edge nodes are suitable for processing. Based on the processing strategy, construct the first prompt word; Based on the first prompt word, the applicable nodes of the network task are inferred and judged by the large language model to obtain the task type score corresponding to the network task; among them, the task type score is used to indicate the matching score of cloud nodes and edge nodes as applicable nodes of the network task. Based on task type scoring, task cost, and network load data, a comprehensive decision score is determined for each network task.

[0048] In this embodiment of the application, in order to improve the problem that traditional scheduling algorithms have difficulty understanding the semantics of complex tasks, a large model discriminator is introduced to perform deep semantic analysis of network tasks, and combined with the idea of ​​multi-objective optimization, to achieve accurate matching of cloud and edge resources.

[0049] Specifically, in this embodiment, a predefined processing strategy is first obtained. This strategy clearly defines the task types applicable to cloud nodes and edge nodes. For example, it can be predefined that cloud nodes are mainly used for long-cycle, global optimization tasks (such as bandwidth planning and fault recovery), while edge nodes are mainly used for low-latency, local scheduling tasks (such as traffic shaping and status monitoring). Based on this processing strategy, a first prompt word containing task description and contextual information can be constructed. Preferably, a Few-Shot learning paradigm is adopted, incorporating several typical tasks and their classification labels as examples into the first prompt word to guide the large language model to more accurately understand task features.

[0050] Subsequently, the initial prompt is input into a large-scale model discriminator pre-trained with incremental network domain data. This model simulates the decision-making process of domain experts, reasoning and judging the appropriate nodes for the current network task and outputting a task type score. This score indicates the matching score between cloud nodes and edge nodes as suitable nodes for the network task. For example, a lower task type score can be set to indicate that edge nodes are more suitable, while a higher score indicates that cloud nodes are more suitable. For instance, the task type score can be set between 0 and 10, where 0-5 indicates that the task is more likely to be handled by edge nodes, and 6-10 indicates that it is more suitable to be assigned to cloud nodes. This process breaks through the limitations of traditional rule engines, giving the system the intelligence to understand the essence of the task.

[0051] After obtaining the semantic-level score (i.e., task type score) from the large model output, the multi-objective optimization calculation stage begins. In this embodiment, the task allocation process can be constructed as a multi-objective optimization problem, no longer relying solely on a single-dimensional discrimination result.

[0052] Specifically, in some embodiments, task cost (including task latency cost and resource consumption cost) and real-time network load data can be extracted and fused with the task type score output by the large model. Based on the network load data, the network load score of the cloud-edge collaborative architecture can be determined.

[0053] The final comprehensive decision score can be calculated using a pre-defined weighted summation formula, such as: Total Score = ω1·Latency Cost + ω2·Resource Usage Cost + ω3·Task Type Score + ω4·Network Load. In this formula, Total Score represents the comprehensive decision score corresponding to the network task; Latency Cost represents the task latency cost (low-latency tasks can be prioritized for edge devices); Resource Usage Cost represents the resource consumption cost; Task Type Score represents the task type score; and Network Load represents the network load score. ω1-ω4 represent the weighting factors, and their sum can be 1, and all are positive values.

[0054] It is understood that in the embodiments of this application, the comprehensive decision score takes into account the network task's tolerance for latency, its demand for computing power, the semantic judgment of its type by the large model, and the current network congestion status, taking into account the actual network operating status and resource utilization efficiency, and can achieve more accurate and adaptive task scheduling.

[0055] Specifically, in some embodiments, a dynamic allocation threshold for a network task is determined based on resource demand information, network load data, and device load data, including: Based on resource demand information, determine the latency sensitivity and resource requirements of network tasks; Determine the network load score of the cloud-edge collaborative architecture based on network load data; Determine the device load score of the edge node based on the device load data; Based on network load score, device load score, latency sensitivity, and resource requirements, determine the dynamic allocation threshold for network tasks.

[0056] In this embodiment, a dynamic boundary with a reverse adjustment mechanism is constructed to improve the problem that static thresholds cannot adapt to network transient changes, thereby achieving automatic traffic diversion and edge protection for highly sensitive services during network congestion.

[0057] Specifically, in this embodiment, key business characteristics are first extracted based on resource demand information to determine the task's latency sensitivity and resource demand. Latency sensitivity measures the severity of the business's demand for response time, while resource demand characterizes the task's consumption level of computing power such as computation and storage. Subsequently, a network load score is calculated based on current network load data (such as bandwidth utilization and packet loss rate), and a device load score is calculated based on edge node device load data (such as CPU and memory utilization).

[0058] Based on this, the system uses a multi-factor weighted model to calculate the dynamic allocation threshold (TS Threshold). For example, the calculation formula can be expressed as: TS Threshold = X α·Network Load β·Edge Load +γ·Latency Sensitivity δ·Resource Demand. In this formula, X is the basic offset constant, and α, β, γ, and δ are preset weighting coefficients used to quantify the influence of various environmental factors on scheduling decisions.

[0059] It is understood that the embodiments of this application implement intelligent load balancing and service assurance. When the network load or edge device load is high, due to the negative weight set in the formula ( α, The calculated dynamic allocation threshold (β) decreases accordingly. This makes it easier for the overall decision score to exceed the threshold, thus pushing the task to the resource-rich cloud and effectively preventing edge nodes from crashing under high load.

[0060] Conversely, for tasks with extremely high latency sensitivity, the positive weight (+γ) in the formula raises the threshold, making it more difficult for the task to be determined to be sent to the cloud, thus forcing it to be processed on edge nodes and ensuring a low-latency experience. Meanwhile, for resource-intensive tasks with huge demands, the negative weight (+γ)... δ) lowers the threshold, prompting tasks to flow to the cloud. This dynamic threshold adjustment algorithm, through its ingenious design, achieves adaptive scheduling effects such as congestion avoidance, edge priority, and heavy computation on the cloud, significantly improving the stability of the cloud-edge collaborative architecture.

[0061] Specifically, in some embodiments, the large language model agent is deployed through the following steps: Collect background knowledge datasets in the network field; By using the background knowledge dataset, the base language model is incrementally pre-trained to obtain a domain-adaptive model; For each node in the cloud-edge collaborative architecture, a corresponding toolset is configured for the domain adaptation model according to a predefined processing strategy to obtain the basic intelligent agent for each node. Obtain the prompt word templates set for each basic agent and load the prompt word templates into the basic agents; wherein, the prompt word templates include at least one of the following: capability definition, business background, business objective, business details and output requirements; By embedding the ReAct inference framework into the basic agent, a well-deployed large language model agent is obtained.

[0062] In this embodiment of the application, domain knowledge injection and reasoning framework enhancement are used to improve the problem that the general large language model has insufficient understanding ability in the vertical domain of the network and cannot perform actual operations, so as to endow the intelligent agent with deep business insight and automated execution capability.

[0063] Specifically, in this embodiment, when building the large language model agent, a network domain background knowledge dataset covering multi-dimensional information such as network topology, protocol specifications, traffic patterns, and device status is first collected. Using this high-quality data, a general-purpose base large language model with adapted parameters is incrementally pre-trained, transforming it from a general-purpose language model into a domain expert deeply familiar with network communication principles and operational logic, significantly improving its understanding of complex network tasks and decision-making accuracy.

[0064] Subsequently, considering the heterogeneous characteristics of the cloud-edge collaborative architecture, and based on predefined processing strategies (such as cloud focusing on global planning and edge focusing on real-time response), differentiated domain toolsets are mounted on the domain adaptation model. These tools endow the agent with the ability to perceive the environment and operate the network, enabling it to proactively call interfaces to read port status or adjust QoS parameters according to task requirements. Simultaneously, the system loads a structured prompt template for each basic agent. This template strictly adheres to the five-element paradigm of "capability definition, business background, business objectives, business details, and output requirements," ensuring a high degree of alignment between the agent's reasoning path and business needs by injecting the specific context and constraints of the current task.

[0065] Specifically, the capabilities are defined as follows: **Capability Definition:** This section clarifies the capabilities and skills required of the agent to ensure correct task execution and achievement of expected results. **Business Context:** It provides the context of the task, such as network environment, resource constraints, and external conditions, helping the agent understand the actual background and requirements of the task. **Business Objectives:** The business objectives section specifies the concrete requirements and referable steps to ensure the agent follows an actionable path. **Business Details:** This section includes specific information and data related to the business scenario of this call. **Output Requirements:** It clarifies the required output format or result after the task is completed, such as network status adjustment schemes or resource allocation strategies, ensuring the task results meet expected standards.

[0066] Furthermore, in this embodiment, the ReAct (Reasoning and Acting) reasoning framework can be deeply embedded in the basic agent, enabling its decision-making process to move beyond simple text generation and instead simulate the closed-loop iterative logic of human expert thought → action → observation. Through the organic combination of multi-step reasoning and tool invocation, the agent can autonomously decompose complex tasks and gradually approach the optimal solution, ultimately outputting an executable network adjustment scheme.

[0067] Understandably, the deployment method provided in this application achieves professionalization of model knowledge through incremental pre-training; enables the leap from cognition to action through toolset configuration; and ensures that the agent maintains the accuracy of its goals and the consistency of its logic throughout multi-round interactions by combining five-element prompts with the ReAct framework. These measures collectively construct a highly autonomous network operation and maintenance agent, laying a solid intelligent foundation for subsequent automated task processing.

[0068] Specifically, in some embodiments, network tasks are processed by a large language model agent deployed on the target node, including: Analyze the network tasks to obtain task information; Fill in the prompt word template based on the task information to obtain the second prompt word; Based on the second prompt word, the network task is processed by a large language model agent to obtain the corresponding processing result.

[0069] In this embodiment, by using context-aware prompt word engineering and the autonomous reasoning ability of the agent, the problems of poor flexibility and inability to cope with complex and ever-changing network environments in traditional automated scripts can be improved, thereby achieving precise and intelligent execution of network tasks.

[0070] Specifically, in this embodiment, when the target node (cloud or edge) receives the task allocation instruction, it first parses the network task and extracts the structured and unstructured information contained therein, i.e., task information. This information typically includes specific business requirements (such as bandwidth expansion, troubleshooting), device identifiers involved (such as port numbers, IP addresses), real-time indicator data (such as current traffic, packet loss rate), and expected completion goals.

[0071] Subsequently, a prompt word template matching the node's function is invoked (this template predefines the agent's capability boundaries, business background, and output specifications), and the parsed task information (business details) is seamlessly filled into the corresponding position in the template, generating a second prompt word rich in the current scenario context. This process enables a general agent to instantly transform into a domain expert for specific network faults or optimization needs. The large model agent receives the second prompt word and initiates its core ReAct (Think-Action-Feedback) reasoning mechanism. The agent does not directly generate an answer, but follows a closed-loop process of thinking about the next operation → invoking domain tools to obtain real-time data → observing the returned results → correcting the reasoning path, until sufficient evidence is accumulated and the optimal solution is derived, outputting a processing result containing a specific instruction sequence (such as adjusting QoS parameters or switching routing paths) and verification metrics.

[0072] Understandably, this application achieves rapid switching from a general model to a scenario expert by dynamically filling in prompts, ensuring the agent's accurate perception of the current network situation. Simultaneously, relying on the multi-step reasoning and tool invocation capabilities of the large-model agent, it breaks through the limitation of traditional rule engines that can only handle preset scenarios, enabling it to cope with unknown faults and complex coupling problems. Thus, it can significantly improve the one-time resolution rate of network tasks, and the standardized output results ensure the traceability and security of network operations.

[0073] Please refer to Figure 3 , Figure 3 This paper illustrates an architecture diagram of a cloud-edge collaborative network task processing system provided in an embodiment of this application. Figure 3 As shown in the embodiments of this application, a network task processing system architecture based on cloud-edge collaboration is provided. This architecture is divided into four closely cooperating functional modules from bottom to top, realizing an intelligent closed loop from data acquisition to task execution.

[0074] The data acquisition module is responsible for real-time network environment perception, collecting multi-dimensional data including network load, device status, and resource usage to provide data support for upper-level decision-making. The central agent module is the core of the system. Based on the collected data, it trains models, builds and deploys domain-adaptive large-model agents for different cloud and edge nodes. These agents, through incremental pre-training and toolset configuration, possess the ability to understand network semantics and execute operations. The task allocation module acts as the scheduling hub. On one hand, it uses a cloud-edge network task allocation algorithm based on a large-model discriminator, combined with a dynamic threshold adjustment mechanism, to comprehensively score and allocate incoming network tasks. On the other hand, it uses task-customized agents to perform context-aware parsing and filling in of tasks. The automated task execution module receives the allocated instructions and drives cloud or edge nodes to complete specific network configuration, optimization, or fault handling operations, thereby translating intelligent decisions into actual network behavior. This layered, decoupled, and data-driven design achieves full-process automation and intelligence in network task processing.

[0075] Please refer to Figure 4 , Figure 4 This illustration shows a schematic diagram of the operation flow of a cloud-edge collaborative network task processing system provided in an embodiment of this application. In this embodiment, firstly, background knowledge related to the network is collected to construct a dataset, and the task types suitable for each cloud and edge node are defined, thus laying the data and rule foundation for subsequent model training and task allocation. Then, the large model intelligent agent construction stage is entered. The collected knowledge dataset is used to perform incremental pre-training on the base large model to obtain a domain-adaptive model. Next, intelligent agents are constructed for different nodes (cloud and edge). The required toolset is defined in the model and the ReAct inference framework is embedded. At the same time, prompt words containing capability definitions, business background, business objectives, business details and output requirements are customized to enable the intelligent agent to have domain understanding and autonomous action capabilities.

[0076] Based on this, the system uses a large model discriminator to score and evaluate incoming network tasks (0-5 points are considered edge node tasks, 6-10 points are considered cloud node tasks), and performs cloud-edge network task allocation accordingly. At the same time, it calculates a dynamic threshold (TSThreshold) for task allocation by combining factors such as network load, edge device load, latency sensitivity, and resource requirements. When the TS value is less than the threshold, the task is allocated to the edge node; otherwise, it is allocated to the cloud node. Finally, the task enters the intelligent agent automated task processing stage, where the intelligent agent deployed on the target node executes the specific task by combining real-time context and pre-set logic, forming a complete closed loop from knowledge construction, intelligent agent deployment, task discrimination and allocation to automated execution.

[0077] It is understood that the technical solution of this application has at least the following technical effects: 1. This application incrementally pre-trains a large-scale foundational model using a network domain dataset and leverages the ReAct framework's "think-action-feedback" reasoning loop. Combined with customized prompts consisting of five elements—capability definition, business background, business objectives, business details, and output requirements—it constructs a highly specialized intelligent agent for heterogeneous tasks on the cloud and edge. This mechanism transforms network operations from a traditional passive "manual + fixed threshold" response model to a proactive governance model with self-driven and intelligent decision-making capabilities. Relying on the deep reasoning capabilities of the domain-adaptive model, the system can accurately understand the essential needs of complex tasks and perform distributed logical deduction. Simultaneously, the closed-loop automated steps introduced by the ReAct framework significantly reduce manual intervention, achieving high-precision execution tailored to specific tasks rather than general rules.

[0078] 2. This application abandons the traditional static allocation logic. By injecting task descriptions and few sample examples into a large model discriminator to output a quantified task adaptation score, this score, along with latency cost, resource consumption, and network load, is incorporated into a multi-objective optimization function. This achieves adaptive matching of cloud and edge resources, effectively avoiding task misallocation and resource idleness, and significantly optimizing the overall network load balancing and average latency.

[0079] 3. This application introduces a dynamic threshold adjustment algorithm based on real-time signals such as overall network load, edge device load, task latency sensitivity, and resource requirements. The system can dynamically correct the allocation boundary according to the current environment. In scenarios with local congestion but sufficient network capacity, it can reliably retain low-latency tasks at the edge for minute-level processing, while aggregating computationally intensive tasks to the cloud. This ensures continuous SLA (Service Level Agreement) protection in complex and ever-changing network environments, thereby comprehensively improving network service quality and operational efficiency.

[0080] Reference Figure 5 In this embodiment of the application, a cloud-edge collaborative network task processing device is also provided, which includes: The acquisition unit 510 is used to acquire network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks; wherein, the cloud-edge collaborative architecture includes cloud nodes and edge nodes; The acquisition unit 520 is used to collect network load data of the cloud-edge collaborative architecture and device load data of edge nodes. The first analysis unit 530 is used to determine the comprehensive decision score corresponding to the network task based on the network task, task cost and network load data. The second analysis unit 540 is used to determine the dynamic allocation threshold corresponding to the network task based on resource demand information, network load data and device load data. The execution unit 550 is used to determine the target node for processing network tasks from the cloud-edge collaborative architecture based on the comprehensive decision score and the dynamically allocated threshold, and to process the network tasks through a large language model agent deployed on the target node.

[0081] It is understandable that, such as Figure 1 The content shown in the cloud-edge collaborative network task processing method embodiment is applicable to the cloud-edge collaborative network task processing device embodiment. The specific functions implemented by the cloud-edge collaborative network task processing device embodiment are the same as those shown in the figure. Figure 1 The illustrated implementation of the cloud-edge collaborative network task processing method is the same, and the beneficial effects achieved are the same as those shown. Figure 1 The beneficial effects achieved by the cloud-edge collaborative network task processing method embodiment shown are also the same.

[0082] Reference Figure 6 This application also discloses an electronic device, including: At least one processor 610; At least one memory 620 is used to store at least one program; When at least one program is executed by at least one processor 610, such that at least one processor 610 performs as follows: Figure 1 The example shown is a network task processing method based on cloud-edge collaboration.

[0083] The electronic device in this application embodiment may be a mobile phone, a computer device, or a server device.

[0084] Understandable, Figure 1 The content of the cloud-edge collaborative network task processing method embodiments shown are all applicable to the embodiments of this electronic device. The specific functions implemented in the embodiments of this electronic device are the same as those in the embodiments of this electronic device. Figure 1 The illustrated implementation of the cloud-edge collaborative network task processing method is the same, and the beneficial effects achieved are also the same. Figure 1 The beneficial effects achieved by the cloud-edge collaborative network task processing method embodiment shown are also the same.

[0085] This application also discloses a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement, for example... Figure 1 The example shown is a network task processing method based on cloud-edge collaboration.

[0086] Understandable, Figure 1The content of the cloud-edge collaborative network task processing method embodiments shown herein is applicable to the embodiments of this computer-readable storage medium. The specific functions implemented in the embodiments of this computer-readable storage medium are the same as those in the embodiments of this computer-readable storage medium. Figure 1 The illustrated implementation of the cloud-edge collaborative network task processing method is the same, and the beneficial effects achieved are also the same. Figure 1 The beneficial effects achieved by the cloud-edge collaborative network task processing method embodiment shown are also the same.

[0087] This application also discloses a computer program product or computer program, which includes computer instructions stored in the aforementioned computer-readable storage medium. Figure 6 The processor of the illustrated electronic device can read the computer instructions from the aforementioned computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The example shown is a network task processing method based on cloud-edge collaboration.

[0088] Understandable, Figure 1 The content shown in the cloud-edge collaborative network task processing method embodiments is applicable to this computer program product or computer program embodiment. The specific functions implemented by this computer program product or computer program embodiment are the same as those shown in the embodiments. Figure 1 The illustrated implementation of the cloud-edge collaborative network task processing method is the same, and the beneficial effects achieved are also the same. Figure 1 The beneficial effects achieved by the cloud-edge collaborative network task processing method embodiment shown are also the same.

[0089] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0090] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0091] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0093] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0094] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0095] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0096] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0097] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A network task processing method based on cloud-edge collaboration, characterized in that, The method includes: Obtain the network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks; wherein, the cloud-edge collaborative architecture includes cloud nodes and edge nodes; Collect network load data of the cloud-edge collaborative architecture and device load data of the edge nodes; Based on the network task, the task cost, and the network load data, determine the comprehensive decision score corresponding to the network task; Based on the resource demand information, the network load data, and the device load data, determine the dynamic allocation threshold corresponding to the network task; Based on the comprehensive decision score and the dynamic allocation threshold, the target node for processing the network task is determined from the cloud-edge collaborative architecture, and the network task is processed by a large language model agent deployed on the target node.

2. The network task processing method based on cloud-edge collaboration according to claim 1, characterized in that, The step of determining the comprehensive decision score corresponding to the network task based on the network task, the task cost, and the network load data includes: Obtain a predefined processing strategy; wherein the processing strategy is used to indicate the type of task that the cloud node and the edge node are suitable for processing; Based on the processing strategy, a first prompt word is constructed; Based on the first prompt word, the applicable nodes of the network task are inferred and judged by a large language model to obtain the task type score corresponding to the network task; wherein, the task type score is used to indicate the matching score of the cloud node and the edge node as the applicable nodes of the network task. Based on the task type score, the task cost, and the network load data, a comprehensive decision score is determined for the network task.

3. The network task processing method based on cloud-edge collaboration according to claim 2, characterized in that, The task cost includes task consumption cost and task latency cost; determining the comprehensive decision score corresponding to the network task based on the task type score, the task cost, and the network load data includes: Based on the network load data, determine the network load score of the cloud-edge collaborative architecture; The comprehensive decision score corresponding to the network task is obtained by weighted summing of the task consumption cost, the task delay cost, the task type score, and the network load score.

4. The network task processing method based on cloud-edge collaboration according to claim 1, characterized in that, The step of determining the dynamic allocation threshold corresponding to the network task based on the resource demand information, the network load data, and the device load data includes: Based on the resource requirement information, determine the latency sensitivity and resource requirements of the network task; Based on the network load data, determine the network load score of the cloud-edge collaborative architecture; Based on the device load data, determine the device load score of the edge node; The dynamic allocation threshold corresponding to the network task is determined based on the network load score, the device load score, the latency sensitivity, and the resource requirement.

5. The network task processing method based on cloud-edge collaboration according to claim 1, characterized in that, The target node for processing the network task is determined from the cloud-edge collaborative architecture based on the comprehensive decision score and the dynamic allocation threshold. The comprehensive decision score and the dynamic allocation threshold are compared; If the comprehensive decision score is less than the dynamic allocation threshold, the edge node will be identified as the target node. If the comprehensive decision score is greater than or equal to the dynamic allocation threshold, the cloud node will be identified as the target node.

6. The network task processing method based on cloud-edge collaboration according to any one of claims 1-5, characterized in that, The large language model agent is deployed through the following steps: Collect background knowledge datasets in the network field; Using the aforementioned background knowledge dataset, the base large language model is incrementally pre-trained to obtain a domain-adaptive model; For each node in the cloud-edge collaborative architecture, a corresponding toolset is configured for the domain adaptation model according to a predefined processing strategy to obtain the basic intelligent agent corresponding to each node; Obtain the prompt word template set for each of the basic intelligent agents, and load the prompt word template into the basic intelligent agent; wherein, the prompt word template includes at least one of capability definition, business background, business objective, business details and output requirements; The ReAct inference framework is embedded in the basic agent to obtain a deployed large language model agent.

7. The network task processing method based on cloud-edge collaboration according to claim 6, characterized in that, The process of processing the network task through a large language model agent deployed on the target node includes: The network task is parsed to obtain task information; The prompt word template is filled in according to the task information to obtain the second prompt word; Based on the second prompt word, the network task is processed by the large language model agent to obtain the corresponding processing result.

8. A network task processing device based on cloud-edge collaboration, characterized in that, The device includes: The acquisition unit is used to acquire network tasks to be processed in the cloud-edge collaborative architecture, as well as the task cost and resource requirements corresponding to the network tasks; wherein, the cloud-edge collaborative architecture includes cloud nodes and edge nodes; The data acquisition unit is used to collect network load data of the cloud-edge collaborative architecture and device load data of the edge nodes. The first analysis unit is used to determine the comprehensive decision score corresponding to the network task based on the network task, the task cost, and the network load data. The second analysis unit is used to determine the dynamic allocation threshold corresponding to the network task based on the resource demand information, the network load data, and the device load data. An execution unit is configured to determine the target node for processing the network task from the cloud-edge collaborative architecture based on the comprehensive decision score and the dynamic allocation threshold, and to process the network task through a large language model agent deployed on the target node.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the cloud-edge collaborative network task processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the cloud-edge collaborative network task processing method according to any one of claims 1 to 7.

Images