Dynamic resource scheduling method for edge network cloud
By constructing a multi-dimensional resource supply and demand matching evaluation model and hierarchical scheduling judgment, the problems of evaluation bias and insufficient flexibility in edge network cloud resource scheduling are solved, achieving efficient and secure resource scheduling to meet the needs of complex scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NAT UNIV OF DEFENSE TECH
- Filing Date
- 2025-12-24
- Publication Date
- 2026-05-05
AI Technical Summary
Existing edge network cloud resource scheduling methods suffer from problems such as a single evaluation dimension, static thresholds and weights, imperfect collaborative scheduling mechanisms, and lack of security and dynamic adjustment, resulting in resource matching deviations, insufficient scheduling flexibility, low resource utilization, and high task failure rates.
A multi-dimensional resource supply and demand matching evaluation model is constructed, adopting a hierarchical scheduling judgment logic. Data from edge nodes, tasks, and cloud centers are collected in real time, and parameter weights are dynamically adjusted. By combining edge collaboration and cloud center scheduling, security authentication and dynamic adjustment are achieved, thereby optimizing resource allocation.
It improves the matching degree of resource supply and demand, reduces the execution latency of time-sensitive tasks, enhances resource utilization and task success rate, adapts to load fluctuations and sudden scenarios, and ensures the flexibility and security of scheduling.
Smart Images
Figure CN121979668A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of edge computing, and in particular to a dynamic resource scheduling method for edge cloud networks. Background Technology
[0002] Existing edge network cloud resource scheduling methods have the following technical shortcomings: The evaluation dimension is too narrow: most solutions only use the CPU / memory utilization of edge nodes as the scheduling basis, ignoring key parameters such as task latency requirements and cloud-edge link bandwidth, which leads to "resource matching deviation". For example, scheduling latency-sensitive tasks to edge nodes with low load but far distance can cause execution latency to exceed the standard.
[0003] Static thresholds and weights: The thresholds for scheduling judgment (such as the upper limit of edge node load) and parameter weights (such as the importance ratio of latency requirements) are mostly fixed values, which cannot adapt to the fluctuation of edge node load (such as the difference of 40% between morning and evening peak loads) and the dynamic switching of latency-sensitive / computation-intensive tasks, resulting in insufficient scheduling flexibility.
[0004] The collaborative scheduling mechanism is imperfect: existing edge collaboration mostly considers only a single node and has not formed a clustered collaborative logic. Moreover, resource allocation lacks optimization algorithm support, which easily leads to the imbalance problem of "some nodes being overloaded and some nodes being idle". At the same time, cloud center scheduling does not take into account the idle time of the resource pool and the link bandwidth constraints, resulting in low resource utilization.
[0005] Lack of security and dynamic adjustment: Most solutions do not integrate identity authentication and data encryption, which poses a risk of scheduling instructions being tampered with and task data being leaked; and there is a lack of real-time monitoring and dynamic adjustment mechanisms after scheduling. When the load on edge nodes suddenly increases or the link bandwidth drops sharply, it is impossible to reschedule in time, resulting in an increased task failure rate. Summary of the Invention
[0006] The purpose of this invention is to provide a dynamic resource scheduling method for edge cloud networks to solve the problems mentioned in the background art.
[0007] To achieve the above objectives, the present invention provides the following technical solution: a dynamic resource scheduling method for edge cloud networks, comprising the following steps: S1. Data Acquisition: Real-time acquisition of resource status data of edge nodes in the edge network cloud architecture, demand data of tasks to be scheduled, and resource redundancy data of the cloud center; S2. Multi-dimensional evaluation model construction: Based on the data collected in step S1, a resource supply and demand matching evaluation model is constructed. The model takes edge node load rate, task latency requirements, and cloud-edge link bandwidth as core parameters and dynamically adjusts the weight of each parameter. S3. Layered scheduling judgment: Based on the evaluation result of step S2, execute the three-layer scheduling judgment logic; 3.1 First-level judgment: If the resource supply and demand matching degree of the edge node is greater than or equal to the first preset threshold, the task will be scheduled to be executed locally on the edge node; 3.2 Second-level judgment: If the resource supply and demand matching degree of the edge node is less than the first preset threshold, and the resource supply and demand matching degree of the surrounding edge node cluster is greater than or equal to the second preset threshold, then the collaborative scheduling between edge nodes is initiated, and tasks are assigned to the surrounding edge nodes; 3.3 Third-level judgment: If the resource supply and demand matching degree of the edge node and the surrounding cluster are both less than the corresponding preset threshold, and the resource supply and demand matching degree of the cloud center is greater than or equal to the third preset threshold, then the task will be scheduled to be executed in the cloud center. S4. Scheduling Execution and Dynamic Adjustment: Execute the scheduling scheme determined in step S3, and monitor the resource status and latency data during task execution in real time. If the monitored data exceeds the preset range, return to step S2 to re-execute the evaluation and scheduling.
[0008] Preferably, in step S1: The resource status data of the edge nodes includes CPU utilization, memory utilization, remaining storage capacity, and real-time network bandwidth. The requirement data for the task to be scheduled includes the task latency threshold, computing power requirement, and data transmission volume. The resource redundancy data of the cloud center includes CPU redundancy rate, memory redundancy rate, and available bandwidth of cloud-edge links.
[0009] Preferably, the method for dynamically adjusting the parameter weights in step S2 is as follows: When the task to be scheduled is a latency-sensitive task, increase the weight of the "task latency requirement" parameter to 30%-50%; When the load fluctuation of edge nodes exceeds 15%, the weight of the "edge node load rate" parameter is increased to 25%-40%, and the weight adjustment is optimized using the random forest algorithm.
[0010] Preferably, in step S3: The first preset threshold is 80%-90%, and this threshold is dynamically updated weekly based on the historical load fluctuation data of the edge nodes over the past 72 hours. The first layer of judgment also includes: if the estimated latency of the edge node executing the task locally is less than or equal to the task latency threshold, then local scheduling is confirmed; otherwise, proceed to the second layer of judgment.
[0011] Preferably, the specific process of the second-level judgment in step S3 is as follows: The K-nearest neighbor algorithm is used to select neighboring edge nodes that are ≤5km away from the current edge node to form an edge collaborative cluster. The second preset threshold is 65%-75%. The overall resource supply and demand matching degree of the cluster is calculated. If it is ≥ the second preset threshold, and the transmission delay between edge nodes + collaborative execution delay ≤ task delay threshold, then the task is assigned to the 2-3 edge nodes with the highest resource redundancy rate in the cluster.
[0012] Preferably, the specific process of the third-level judgment in step S3 is as follows: The third preset threshold is 70%-80%, and the judgment must satisfy the following conditions: available bandwidth of cloud-edge link ≥ task data transmission volume / task allowed transmission duration; If there are multiple available resource pools in the cloud center, the resource pool with an idle time of more than 2 hours will be prioritized for task execution.
[0013] Preferably, the dynamically adjusted triggering conditions in step S4 include: Real-time load rate of edge nodes > 90%; The actual execution latency of the task is greater than 110% of the task latency threshold; The cloud-edge link bandwidth suddenly drops by more than 30% and lasts for more than 5 minutes.
[0014] Preferably, the process also includes a security authentication step: before executing the scheduling in step S3, the task initiator is authenticated (using the OAuth2.0 protocol), and the task data and scheduling instructions are encrypted using the AES-256 algorithm. The scheduling is only executed after the authentication is successful and the decryption is successful.
[0015] Preferably, the resource allocation for collaborative scheduling among edge nodes adopts the particle swarm optimization algorithm, with "minimizing collaborative execution latency" as the objective function and the constraint that the load rate of each edge node is ≤85%.
[0016] Preferably, the process also includes a scheduling result feedback optimization step: recording the resource utilization rate, task completion rate, and delay compliance rate for each scheduling session, and using the gradient descent algorithm based on historical data to optimize the parameter weights of the evaluation model in step S2 and the preset thresholds in step S3, with the optimization cycle being once per day.
[0017] The technical effects and advantages of this invention are as follows: By collecting multi-dimensional data from edge nodes, tasks, and cloud centers in real time, a dynamic weighted resource supply and demand matching evaluation model is constructed. The scheduling is then judged and arranged in layers of "edge local → edge collaboration → cloud center". This fundamentally solves the one-sidedness of existing solutions that only evaluate based on CPU / memory, reduces the error in resource supply and demand matching evaluation, improves the latency compliance rate of latency-sensitive tasks (such as remote surgery and autonomous driving), and meets the core scheduling requirements of "low latency and high accuracy" for edge network cloud architecture.
[0018] Based on the sovereign "dynamic adjustment" mechanism, the scheduling parameters are dynamically adapted to the scenario: when the load fluctuation of the edge node is greater than 15%, the weight of the "load rate" parameter is automatically increased; when the task type is switched from computing power intensive to latency sensitive, the weight of "latency requirement" can be increased to 30%-50%; and it can respond to sudden scenarios (such as a sudden drop in link bandwidth of greater than 30%), triggering rescheduling within 1 second, avoiding scheduling failure caused by fixed parameters, and adapting to complex scenarios where the load difference between morning and evening peak hours reaches 40%.
[0019] Based on the "edge collaboration" and "cloud center scheduling" logic in the hierarchical judgment, efficient resource allocation is achieved: within the edge collaboration cluster, the goal is to "minimize latency" and constrain the node load to ≤85%, thereby improving the cluster resource utilization rate; the cloud center prioritizes scheduling resource pools that have been idle for more than 2 hours, reducing the resource idle rate and balancing "scheduling performance" and "cost control". Attached Figure Description
[0020] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the hierarchical scheduling judgment steps of the present invention. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] This invention provides, for example Figures 1-2 The dynamic resource scheduling method for edge cloud networks shown includes the following steps: S1. Data Acquisition: Real-time acquisition of resource status data of edge nodes in the edge network cloud architecture, demand data of tasks to be scheduled, and resource redundancy data of the cloud center; S2. Multi-dimensional evaluation model construction: Based on the data collected in step S1, a resource supply and demand matching evaluation model is constructed. The model takes edge node load rate, task latency requirements, and cloud-edge link bandwidth as core parameters and dynamically adjusts the weight of each parameter. S3. Layered scheduling judgment: Based on the evaluation result of step S2, execute the three-layer scheduling judgment logic; 3.1 First-level judgment: If the resource supply and demand matching degree of the edge node is greater than or equal to the first preset threshold, the task will be scheduled to be executed locally on the edge node; 3.2 Second-level judgment: If the resource supply and demand matching degree of the edge node is less than the first preset threshold, and the resource supply and demand matching degree of the surrounding edge node cluster is greater than or equal to the second preset threshold, then the collaborative scheduling between edge nodes is initiated, and tasks are assigned to the surrounding edge nodes; 3.3 Third-level judgment: If the resource supply and demand matching degree of the edge node and the surrounding cluster are both less than the corresponding preset threshold, and the resource supply and demand matching degree of the cloud center is greater than or equal to the third preset threshold, then the task will be scheduled to be executed in the cloud center. S4. Scheduling Execution and Dynamic Adjustment: Execute the scheduling scheme determined in step S3, and monitor the resource status and latency data during task execution in real time. If the monitored data exceeds the preset range, return to step S2 to re-execute the evaluation and scheduling.
[0023] Furthermore, in step S1: The resource status data of the edge nodes includes CPU utilization, memory utilization, remaining storage capacity, and real-time network bandwidth. The requirement data for the task to be scheduled includes the task latency threshold, computing power requirement, and data transmission volume. The resource redundancy data of the cloud center includes CPU redundancy rate, memory redundancy rate, and available bandwidth of cloud-edge links.
[0024] Furthermore, the method for dynamically adjusting the parameter weights in step S2 is as follows: When the task to be scheduled is a latency-sensitive task, increase the weight of the "task latency requirement" parameter to 30%-50%; When the load fluctuation of edge nodes exceeds 15%, the weight of the "edge node load rate" parameter is increased to 25%-40%, and the weight adjustment is optimized using the random forest algorithm.
[0025] Furthermore, in step S3: The first preset threshold is 80%-90%, and this threshold is dynamically updated weekly based on the historical load fluctuation data of the edge nodes over the past 72 hours. The first layer of judgment also includes: if the estimated latency of the edge node executing the task locally is less than or equal to the task latency threshold, then local scheduling is confirmed; otherwise, proceed to the second layer of judgment.
[0026] Furthermore, the specific process of the second-level judgment in step S3 is as follows: The K-nearest neighbor algorithm is used to select neighboring edge nodes that are ≤5km away from the current edge node to form an edge collaborative cluster. The second preset threshold is 65%-75%. The overall resource supply and demand matching degree of the cluster is calculated. If it is ≥ the second preset threshold, and the transmission delay between edge nodes + collaborative execution delay ≤ task delay threshold, then the task is assigned to the 2-3 edge nodes with the highest resource redundancy rate in the cluster.
[0027] Furthermore, the specific process of the third-level judgment in step S3 is as follows: The third preset threshold is 70%-80%, and the judgment must satisfy the following conditions: available bandwidth of cloud-edge link ≥ task data transmission volume / task allowed transmission duration; If there are multiple available resource pools in the cloud center, the resource pool with an idle time of more than 2 hours will be prioritized for task execution.
[0028] Furthermore, the dynamically adjusted triggering conditions in step S4 include: Real-time load rate of edge nodes > 90%; The actual execution latency of the task is greater than 110% of the task latency threshold; The cloud-edge link bandwidth suddenly drops by more than 30% and lasts for more than 5 minutes.
[0029] Furthermore, it also includes a security authentication step: before executing the scheduling in step S3, the task initiator is authenticated (using the OAuth2.0 protocol), and the task data and scheduling instructions are encrypted using the AES-256 algorithm. The scheduling is only executed after the authentication is successful and the decryption is successful.
[0030] Furthermore, the resource allocation for collaborative scheduling among edge nodes adopts the particle swarm optimization algorithm, with the objective function being "minimizing collaborative execution latency" and the constraint being that the load rate of each edge node is ≤85%.
[0031] Furthermore, it also includes a scheduling result feedback optimization step: recording the resource utilization rate, task completion rate, and delay compliance rate for each scheduling, and using the gradient descent algorithm based on historical data to optimize the parameter weights of the evaluation model in step S2 and the preset thresholds in step S3, with an optimization cycle of once per day.
[0032] By collecting multi-dimensional data from edge nodes, tasks, and cloud centers in real time, a dynamic weighted resource supply and demand matching evaluation model is constructed. Scheduling is then performed in a hierarchical manner, from "local edge → collaborative edge → cloud center," fundamentally solving the one-sidedness of existing solutions that only evaluate based on CPU / memory. This reduces the error in resource supply and demand matching evaluation, improves the latency compliance rate of latency-sensitive tasks (such as remote surgery and autonomous driving), and meets the core scheduling requirements of "low latency and high accuracy" for edge network cloud architecture.
[0033] Based on the sovereign "dynamic adjustment" mechanism, the scheduling parameters are dynamically adapted to the scenario: when the load fluctuation of the edge node is greater than 15%, the weight of the "load rate" parameter is automatically increased; when the task type is switched from computing power intensive to latency sensitive, the weight of "latency requirement" can be increased to 30%-50%; and it can respond to sudden scenarios (such as a sudden drop in link bandwidth of greater than 30%), triggering rescheduling within 1 second, avoiding scheduling failure caused by fixed parameters, and adapting to complex scenarios where the load difference between morning and evening peak hours reaches 40%.
[0034] Based on the "edge collaboration" and "cloud center scheduling" logic in the hierarchical judgment, efficient resource allocation is achieved: within the edge collaboration cluster, the goal is to "minimize latency" and constrain the node load to ≤85%, thereby improving the cluster resource utilization rate; the cloud center prioritizes scheduling resource pools that have been idle for more than 2 hours, reducing the resource idle rate and balancing "scheduling performance" and "cost control".
[0035] Working principle: I. Basic Data Acquisition: Building the Data Foundation for Scheduling Decisions The scheduling process begins with real-time collection of data across all dimensions to ensure the accuracy and timeliness of decision-making. The data collected includes three core categories: Edge node resource status data: Real-time acquisition of hardware and network load of each edge node, including CPU utilization, memory utilization, remaining storage capacity, and real-time network bandwidth; Task requirements data to be scheduled: Clearly define the core resource requirements of the task, specifically including the task latency threshold (the maximum allowed execution latency of the task), computing power requirements (the computing resources required to complete the task), and data transmission volume (the scale of data to be transmitted for task execution). Cloud center resource redundancy data: Understand the available idle resources that can be scheduled on the cloud side, specifically including CPU redundancy rate, memory redundancy rate (the proportion of remaining resources in the cloud center to the total resources), and available bandwidth of cloud-edge links (real-time transmission bandwidth between the edge and the cloud center).
[0036] II. Construction of a Multi-Dimensional Evaluation Model: Establishing Matching Standards for Resources and Tasks Based on the collected full data, a resource supply and demand matching evaluation model is constructed. The core is to provide a quantitative basis for subsequent scheduling decisions by quantifying the degree of fit between "edge node / cloud center resources" and "task requirements". The key design points of the model are as follows: Core evaluation parameters: Three key dimensions are identified as the core of the evaluation: edge node load rate (reflecting the current busyness of edge nodes), task latency requirements (reflecting the sensitivity of tasks to low latency), and cloud-edge link bandwidth (reflecting cloud-edge data transmission capabilities). Dynamic weight adjustment: Based on task characteristics and edge node status, the weights of each parameter are flexibly adjusted to avoid a "one-size-fits-all" evaluation. If the task is latency-sensitive (such as real-time video analysis, industrial control), increase the weight of the "task latency requirement" parameter to 30%-50% to prioritize ensuring that latency meets the requirements. If the load fluctuation of the edge node is greater than 15% (such as sudden load increases or decreases caused by sudden tasks), increase the weight of the "edge node load rate" parameter to 25%-40% to prioritize avoiding node overload; The weight adjustment process is optimized using the random forest algorithm to ensure the rationality and objectivity of the weight allocation.
[0037] III. Security Authentication and Hierarchical Scheduling Decisions: Determining the Optimal Scheduling Path Before scheduling is executed, a security check is performed. Then, based on the evaluation model results, a three-layer progressive scheduling decision is made: "edge local → edge collaboration → cloud center" to ensure the security and optimality of the scheduling. (1) Pre-approval security authentication: ensuring the scheduling process is safe and controllable Before making a hierarchical scheduling decision, a double security check is required to prevent unauthorized tasks or data leaks: Identity authentication: The OAuth2.0 protocol is used to verify the identity of the task initiator (such as terminal device or application system) and confirm the legitimacy of the initiator; Data encryption: Task data (such as business data to be processed) and scheduling instructions (such as task allocation instructions) are encrypted using the AES-256 algorithm. Only after successful authentication and decryption will the subsequent scheduling judgment stage begin.
[0038] (2) First layer: Local scheduling judgment of edge nodes Prioritize assigning tasks to the edge node corresponding to the task initiator for local execution (to reduce cross-node transmission latency), based on the following criteria: The resource supply and demand matching degree of the edge node is greater than or equal to the first preset threshold (the threshold range is 80%-90%, and it is dynamically updated weekly based on the historical load fluctuation data of the edge node over the past 72 hours to avoid the threshold becoming rigid). The estimated latency of the edge node executing the task locally is less than or equal to the task latency threshold (even if the matching degree meets the standard, it is still necessary to ensure that the latency meets the task requirements). If all of the above conditions are met, the task is directly scheduled to be executed locally on the edge node; if any condition is not met, the second-level judgment is entered.
[0039] (3) Second layer: Coordinated scheduling judgment of peripheral edge nodes When a local edge node is unable to handle a task, it attempts to execute the task collaboratively through a cluster of surrounding edge nodes (avoiding direct reliance on the cloud center). The process is as follows: Cluster selection: The K-nearest neighbor algorithm is used to select peripheral edge nodes that are ≤5km away from the current edge node to form an "edge collaborative cluster" (controlling the transmission distance between nodes and reducing collaborative latency). Cluster matching degree judgment: Calculate the overall resource supply and demand matching degree of the cluster, which must meet ≥ the second preset threshold (threshold range 65%-75%). Latency verification: The "transmission latency between edge nodes + cluster collaborative execution latency" must meet the task latency threshold (ensuring that the total latency of the collaborative process meets the standard). Resource allocation: If the above conditions are met, the particle swarm optimization algorithm is used to allocate tasks (with the objective function of "minimizing collaborative execution latency" and the load rate of each edge node constrained to ≤85% to avoid overloading of some nodes). Finally, the tasks are allocated to the 2-3 edge nodes with the highest resource redundancy in the cluster.
[0040] (4) Third layer: Cloud center scheduling and judgment The task will only be scheduled to be executed in the cloud center when neither the edge (local + surrounding clusters) can handle the task. The judgment condition is: The resource supply and demand matching degree of the cloud center is greater than or equal to the third preset threshold (threshold range 70%-80%). Bandwidth verification: Available bandwidth of cloud-edge link ≥ task data transmission volume / task allowed transmission time (ensuring that task data can be transmitted to the cloud center within the allowed time); If there are multiple available resource pools in the cloud center, the resource pool with "idle time > 2 hours" will be selected first to execute the task (to reduce the overhead caused by frequent switching of resource pools).
[0041] IV. Scheduling Execution and Dynamic Adjustment: Responding to Real-Time Resource and Task Changes Once the scheduling plan is determined, the task execution process is initiated. Anomalies during execution are dynamically detected in real time. If adjustment conditions are triggered, the process immediately returns to the evaluation phase for re-decision, ensuring continuous and stable task execution. Real-time monitoring content includes: real-time resource status of edge nodes / cloud centers (such as load rate), actual task execution latency, and cloud-edge link bandwidth; Dynamically adjust trigger conditions (triggering occurs if any condition is met): The real-time load rate of edge nodes is >90% (the nodes are close to full load, and there is a risk of task lag). The actual execution latency of the task is greater than 110% of the task latency threshold (the latency exceeds the standard, affecting the task experience). A sudden drop in cloud-edge link bandwidth of more than 30% lasting for more than 5 minutes (insufficient bandwidth, resulting in data transmission obstruction). Adjustment logic: After the adjustment is triggered, return to step S2 (rebuild the evaluation model), and re-execute the evaluation and hierarchical scheduling based on the latest resource and task data to achieve "dynamic adaptation".
[0042] V. Optimization of Scheduling Result Feedback: Continuously Improve Scheduling Accuracy To avoid rigid scheduling strategies, historical data feedback is used to continuously optimize the evaluation model and scheduling thresholds, forming a closed-loop improvement: Data recording: After each scheduling is completed, record key result indicators, resource utilization (the actual usage ratio of edge / cloud center resources), task completion rate (the percentage of tasks successfully executed), and latency compliance rate (the percentage of tasks whose latency meets the threshold). Model and threshold optimization: Based on historical data, the gradient descent algorithm is used to optimize two parts: In step S2, the parameter weights of the model are evaluated; Each preset threshold in step S3; Optimization cycle: Optimization is performed once a day to ensure that the scheduling strategy is continuously iterated as business scenarios change.
[0043] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing 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. 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 dynamic resource scheduling method for edge cloud networks, characterized in that, Includes the following steps: S1. Data Acquisition: Real-time acquisition of resource status data of edge nodes in the edge network cloud architecture, demand data of tasks to be scheduled, and resource redundancy data of the cloud center; S2. Multi-dimensional evaluation model construction: Based on the data collected in step S1, a resource supply and demand matching evaluation model is constructed. The model takes edge node load rate, task latency requirements, and cloud-edge link bandwidth as core parameters and dynamically adjusts the weight of each parameter. S3. Layered scheduling judgment: Based on the evaluation result of step S2, execute the three-layer scheduling judgment logic; 3.1 First-level judgment: If the resource supply and demand matching degree of the edge node is greater than or equal to the first preset threshold, the task will be scheduled to be executed locally on the edge node; 3.2 Second-level judgment: If the resource supply and demand matching degree of the edge node is less than the first preset threshold, and the resource supply and demand matching degree of the surrounding edge node cluster is greater than or equal to the second preset threshold, then the collaborative scheduling between edge nodes is initiated, and tasks are assigned to the surrounding edge nodes; 3.3 Third-level judgment: If the resource supply and demand matching degree of the edge node and the surrounding cluster are both less than the corresponding preset threshold, and the resource supply and demand matching degree of the cloud center is greater than or equal to the third preset threshold, then the task will be scheduled to be executed in the cloud center. S4. Scheduling Execution and Dynamic Adjustment: Execute the scheduling scheme determined in step S3, and monitor the resource status and latency data during task execution in real time. If the monitored data exceeds the preset range, return to step S2 to re-execute the evaluation and scheduling.
2. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, In step S1: The resource status data of the edge nodes includes CPU utilization, memory utilization, remaining storage capacity, and real-time network bandwidth. The requirement data for the task to be scheduled includes the task latency threshold, computing power requirement, and data transmission volume. The resource redundancy data of the cloud center includes CPU redundancy rate, memory redundancy rate, and available bandwidth of cloud-edge links.
3. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, The method for dynamically adjusting parameter weights in step S2 is as follows: When the task to be scheduled is a latency-sensitive task, increase the weight of the "task latency requirement" parameter to 30%-50%; When the load fluctuation of edge nodes exceeds 15%, the weight of the "edge node load rate" parameter is increased to 25%-40%, and the weight adjustment is optimized using the random forest algorithm.
4. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, In step S3: The first preset threshold is 80%-90%, and this threshold is dynamically updated weekly based on the historical load fluctuation data of the edge nodes over the past 72 hours. The first layer of judgment also includes: if the estimated latency of the edge node executing the task locally is less than or equal to the task latency threshold, then local scheduling is confirmed; otherwise, proceed to the second layer of judgment.
5. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, The specific process of the second-level judgment in step S3 is as follows: The K-nearest neighbor algorithm is used to select neighboring edge nodes that are ≤5km away from the current edge node to form an edge collaborative cluster. The second preset threshold is 65%-75%. The overall resource supply and demand matching degree of the cluster is calculated. If it is ≥ the second preset threshold, and the transmission delay between edge nodes + collaborative execution delay ≤ task delay threshold, then the task is assigned to the 2-3 edge nodes with the highest resource redundancy rate in the cluster.
6. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, The specific process of the third-level judgment in step S3 is as follows: The third preset threshold is 70%-80%, and the judgment must satisfy the following conditions: available bandwidth of cloud-edge link ≥ task data transmission volume / task allowed transmission duration; If there are multiple available resource pools in the cloud center, the resource pool with an idle time of more than 2 hours will be prioritized for task execution.
7. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, The triggering conditions dynamically adjusted in step S4 include: Real-time load rate of edge nodes > 90%; The actual execution latency of the task is greater than 110% of the task latency threshold; The cloud-edge link bandwidth suddenly drops by more than 30% and lasts for more than 5 minutes.
8. The dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, It also includes a security authentication step: before executing the scheduling in step S3, the task initiator is authenticated, and the task data and scheduling instructions are encrypted using the AES-256 algorithm. The scheduling is only executed after the authentication is successful and the decryption is successful.
9. A dynamic resource scheduling method for edge cloud networks according to claim 5, characterized in that, The resource allocation for collaborative scheduling among edge nodes adopts the particle swarm optimization algorithm, with the objective function being "minimize collaborative execution latency" and the constraint being that the load rate of each edge node is ≤85%.
10. A dynamic resource scheduling method for edge cloud networks according to claim 1, characterized in that, It also includes a scheduling result feedback optimization step: recording the resource utilization rate, task completion rate and delay compliance rate for each scheduling, and using the gradient descent algorithm based on historical data to optimize the parameter weights of the evaluation model in step S2 and the preset thresholds in step S3, with an optimization cycle of once a day.