A resource scheduling method, system, device and medium in an edge cloud computing platform

Abstract

A resource scheduling method, system, device and medium in an edge cloud computing platform, which relates to the field of resource scheduling. In this method, a corresponding digital twin model is constructed for each edge node according to the resource status and environmental parameters, the resource change situation within a preset future time is predicted according to the digital twin model, and a global resource status map is generated according to the resource change situation; the task requirements of each task are obtained, and the execution location and priority of each task are determined according to the task requirements and the global resource status map; a global allocation scheme is generated through a federated learning algorithm according to the global resource status map, the execution location and the priority; the resource allocation status is obtained in real time, the difference between the resource allocation status and the global allocation scheme is determined, and the global allocation scheme is corrected based on the difference through a hierarchical cooperation mechanism. Implementing the technical solution provided by this application achieves the effects of optimizing the resource allocation efficiency, reducing latency and energy consumption, and improving the task processing performance.

Description

Technical Field

[0001] This application relates to the technical field of resource scheduling, and particularly to a resource scheduling method, system, device and medium in an edge cloud computing platform. Background Art

[0002] As an emerging computing model, edge cloud computing has been widely used in recent years in fields such as the Internet of Things, intelligent transportation, and smart cities. Its core lies in deploying computing resources at edge nodes close to the data source, thereby reducing data transmission latency, improving response speed and bandwidth utilization. This distributed architecture can not only effectively relieve the pressure on traditional cloud computing centers, but also meet the requirements of application scenarios with high real-time requirements, greatly promoting the digital transformation and intelligent upgrade of all walks of life.

[0003] To solve the resource scheduling problem, existing technologies usually adopt resource allocation methods based on static rules, and allocate tasks through pre-set priority rules or fixed allocation strategies. However, when facing a large-scale and dynamically changing edge computing environment, the above methods often exhibit problems such as unbalanced resource allocation, low scheduling efficiency, and insufficient response to sudden tasks. Especially in the multi-node collaboration scenario, how to accurately predict resource changes and achieve efficient task scheduling has become a key technical problem to be solved urgently. Summary of the Invention

[0004] This application provides a resource scheduling method, system, device and medium in an edge cloud computing platform, achieving the effects of optimizing resource allocation efficiency, reducing latency and energy consumption, improving task processing performance, and at the same time ensuring data privacy and load balance, and is applicable to complex and changeable edge computing scenarios.

[0005] In the first aspect of this application, a resource scheduling method in an edge cloud computing platform is provided, which is applied to an edge cloud computing platform. The method includes:

[0006] Obtain the resource status and environmental parameters of each edge node, construct a corresponding digital twin model for each edge node according to the resource status and the environmental parameters, predict the resource change situation within a preset future time according to the digital twin model, and generate a global resource status map according to the resource change situation;

[0007] Obtain the task requirements of each task, and determine the execution location and priority of each task according to the task requirements and the global resource status map;

[0008] Generate a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location and the priority;

[0009] Obtain the resource allocation status in real time, determine the difference between the resource allocation status and the global allocation plan, and correct the global allocation plan based on the difference through a hierarchical cooperation mechanism.

[0010] Optionally, the generating the global resource status map according to the resource change situation includes:

[0011] Encode the resource change situations of each edge node into a multi-dimensional label vector, where the multi-dimensional label vector includes available resource labels, latency sensitivity labels, and energy consumption cost labels;

[0012] Calculate the node weights based on the weighted calculation of the multi-dimensional label vector, determine the edge weights from the network bandwidth and transmission latency between nodes, and construct a global resource status map based on the topological relationship between edge nodes, the node weights, and the edge weights. The topological relationship includes physical connection relationships, network hop counts, and geographical proximity.

[0013] Optionally, the determining the execution location and priority of each task according to the task requirements and the global resource status map includes:

[0014] Determine the task category of each task according to the task requirements, and extract the key features of each task. The key features include the computing resource requirements, network bandwidth requirements, expected execution time, and data privacy level of the task;

[0015] Determine the priority of each task according to the task category and the key features through predefined priority rules;

[0016] Match the most suitable execution location according to the priority, the key features, and the global resource status map.

[0017] Optionally, the generating the global allocation plan through a federated learning algorithm according to the global resource status map, the execution location, and the priority includes:

[0018] Calculate the weight of each node according to the total node resources and the historical task success rate, and weighted aggregate the resource allocation proposals of all nodes through the federated averaging algorithm to generate a preliminary global allocation plan;

[0019] Bind the target tasks with priorities higher than the threshold to the target edge nodes to adjust the preliminary global allocation plan;

[0020] Minimize the total latency and energy consumption of the adjusted preliminary global allocation plan according to a preset multi-objective optimization function, and use mixed-integer linear programming to determine the global allocation plan.

[0021] Optionally, calculating the weight of each node according to the total node resources and the historical task success rate, and aggregating the resource allocation proposals of all nodes weighted by the federated averaging algorithm according to the weight to generate a preliminary global allocation plan, including:

[0022] Statistical resource data of each edge node, and performing weighted summation on the resource data to obtain the total node resources. The resource data includes CPU, GPU, memory, storage, and network bandwidth, and performing weighted summation on the total node resources and the historical task success rate to obtain the node weight;

[0023] Generating a resource allocation proposal according to the resource status and task requirements of each edge node itself, and using the federated averaging algorithm to perform weighted averaging on the resource allocation proposal according to the node weight to obtain a preliminary global allocation plan.

[0024] Optionally, the method further includes:

[0025] For target tasks with a latency tolerance value lower than the threshold, skip the federated learning aggregation process, and directly match the execution location based on the local resource status map by the edge node;

[0026] When multiple tasks compete for the same node resources, allocate the resources already allocated to the low-priority tasks to the high-priority tasks;

[0027] Based on the consistent hashing algorithm, migrate the low-priority tasks to adjacent edge nodes or enter the waiting queue.

[0028] Optionally, the migrating the low-priority tasks to adjacent edge nodes or entering the waiting queue based on the consistent hashing algorithm includes:

[0029] Allocating multiple virtual nodes to each edge node to construct a hash ring. When the low-priority task is bound to a virtual node, locate the target node to be migrated clockwise according to the hash ring;

[0030] Determine whether the low-priority task is a task restricted by data sovereignty. When the low-priority task is a task restricted by data sovereignty, determine a node with a load rate lower than the load threshold within a preset geographical area as the target node to be migrated;

[0031] When there is no node with a load rate lower than the load threshold within the preset geographical area, put the low-priority task into the waiting queue.

[0032] In the second aspect of the present application, a resource scheduling system in an edge cloud computing platform is provided, including a collection module, a task module, an allocation module, and an adjustment module, where:

[0033] The acquisition module is configured to obtain the resource status and environmental parameters of each edge node, build a corresponding digital twin model for each edge node according to the resource status and the environmental parameters, predict the resource change situation within a preset future time according to the digital twin model, and generate a global resource status map according to the resource change situation;

[0034] The task module is configured to obtain the task requirements of each task, and determine the execution location and priority of each task according to the task requirements and the global resource status map;

[0035] The allocation module is configured to generate a global allocation scheme through a federated learning algorithm according to the global resource status map, the execution location, and the priority;

[0036] The adjustment module is configured to obtain the resource allocation status in real time, determine the difference between the resource allocation status and the global allocation scheme, and correct the global allocation scheme based on the difference through a hierarchical cooperation mechanism.

[0037] In the third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory, so that the electronic device executes the method described in any one of the above.

[0038] In the fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions, and when the instructions are executed, the method described in any one of the above is executed.

[0039] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0040] 1. By building a digital twin model for each edge node, the resource status and environmental parameters of the node can be reflected in real time, and the future resource change situation can be predicted. This method enables resource scheduling to be based not only on the current state, but also to be able to predict in advance the changes in resource requirements, so as to allocate and schedule resources more proactively; encoding the resource change situations of each node into a multi-dimensional label vector and constructing a global resource status map in combination with the topological relationship between nodes provides a comprehensive resource view, which helps to make more accurate resource scheduling decisions;

[0041] 2. Determine the execution location and priority of the task according to the requirements of the task (such as computing resources, network bandwidth, latency tolerance, data privacy, etc.), which can meet the diverse needs of different tasks and improve the flexibility of task execution; generate a global allocation scheme through the federated learning algorithm, which not only protects the privacy of nodes but also can make full use of the resource information of each node to achieve more optimized resource allocation. Especially in a dynamic environment, this method can quickly adapt to changes in resources and tasks;

[0042] 3. When generating the global allocation scheme, consider the minimization of total latency and energy consumption, and determine the final scheme through mixed-integer linear programming to ensure the efficient use of resources and reduce the energy consumption of the system at the same time; monitor the resource allocation status in real time, and when there is a difference between the actual allocation and the global scheme, correct it through a hierarchical cooperation mechanism. This mechanism can adjust the resource allocation in a timely manner, avoid resource waste and task execution failures, and improve the overall performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a schematic flowchart of a resource scheduling method in an edge cloud computing platform disclosed in an embodiment of the present application;

[0044] Figure 2 is a schematic block diagram of a resource scheduling system in an edge cloud computing platform disclosed in an embodiment of the present application;

[0045] Figure 3 is a schematic structural diagram of an electronic device disclosed in an embodiment of the present application.

[0046] Description of reference numerals: 201, acquisition module; 202, task module; 203, allocation module; 204, adjustment module; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0048] In the description of the embodiments of the present application, words such as "for example" or "for instance" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "for example" or "for instance" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "for example" or "for instance" is intended to present relevant concepts in a specific manner.

[0049] In the description of the embodiments of the present application, the term "plural" means two or more. For example, plural systems mean two or more systems, and plural screen terminals mean two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "include", "comprise", "have" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0050] This embodiment discloses a resource scheduling method in an edge cloud computing platform, which is applied to the edge cloud computing platform. Figure 1 It is a schematic flowchart of the resource scheduling method in the edge cloud computing platform disclosed in the embodiments of the present application. As Figure 1 shown, the method includes the following steps:

[0051] S101. Obtain the resource status and environmental parameters of each edge node, construct a corresponding digital twin model for each edge node according to the resource status and the environmental parameters, predict the resource change situation within a preset future time according to the digital twin model, and generate a global resource status map according to the resource change situation;

[0052] S102. Obtain the task requirements of each task, and determine the execution location and priority of each task according to the task requirements and the global resource status map;

[0053] S103. Generate a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location and the priority;

[0054] S104. Obtain the resource allocation status in real time, determine the difference between the resource allocation status and the global allocation plan, and correct the global allocation plan based on the difference through a hierarchical cooperation mechanism.

[0055] Resource status: including computing power (CPU / GPU utilization), memory occupancy, network bandwidth, storage capacity, etc. Environmental parameters: including node geographical location, mobile device trajectory, wireless channel quality, etc. A lightweight digital twin model is constructed for each edge node, and the input includes real-time resource status parameters, dynamic environmental parameters, and historical resource fluctuation data. The digital twin model is trained using a long short-term memory network to predict the resource change trend within a preset future time window. The time window is dynamically adjusted according to the node load, with a window of 10 minutes for low load and shortened to 2 minutes for high load. The prediction error is corrected through a sliding window mechanism. When the deviation between the actual resource utilization and the predicted value exceeds the threshold, incremental learning and updating of the digital twin model are triggered. The predicted resource status of each edge node is encoded as a multi-dimensional label vector. Based on the topological relationship between edge nodes (physical connection relationship, network hop count, geographical proximity) and the multi-dimensional label vector, a global resource status map is constructed. The update trigger conditions for the map include periodic updates (default interval of 5 minutes), node resource mutations (such as CPU utilization fluctuation > 20% for 1 minute), and task offloading decisions triggering local updates of the map. Tasks are classified into critical, compute-intensive, privacy-sensitive, etc. according to the QoS (Quality of Service) requirements of the tasks (such as latency upper limit, computational complexity). The computing resource requirements, network bandwidth requirements, expected execution time, and data privacy level of the tasks are extracted. According to the task category and key features, the priority of each task is determined through predefined priority rules. For example, critical tasks have the highest priority, compute-intensive tasks determine the priority based on computing resource requirements and expected execution time, and privacy-sensitive tasks determine the priority considering the data privacy protection level. The most suitable execution location is matched according to the task priority, key features, and global resource status map. For high-priority tasks, edge nodes with sufficient resources, low latency, and reasonable energy consumption costs are preferentially selected; for privacy-sensitive tasks, edge nodes with high security levels or local execution are selected; for compute-intensive tasks, the computing power and network bandwidth of the nodes are comprehensively considered. The resources such as CPU, GPU, memory, storage, and network bandwidth of each node are counted to form a total resource index. The success rate data of past task executions on the nodes is collected to reflect the stability and reliability of the nodes. Through weighted summation, the total resource index and historical task success rate are converted into node weights. Each node generates a resource allocation proposal based on its own resource status and task requirements. Using the federated averaging algorithm, the proposals of each node are weighted and averaged according to the node weights to obtain a preliminary global allocation plan. The preliminary global allocation plan is optimized to ensure that the plan meets the execution location and priority constraints of the tasks, while optimizing the resource allocation ratio to improve resource utilization. The resource allocation status is obtained in real time, including indicators such as the execution progress of the tasks, CPU / GPU utilization, memory occupancy, network bandwidth usage, etc. The difference between the actual resource allocation status and the global allocation plan is determined, and the reasons for the difference are analyzed.For minor differences, local adjustments are made by the edge autonomy layer, such as migrating some containers and adjusting resource quotas. For major differences, global coordination is carried out by the cloud coordination center, such as reallocating tasks to other available nodes. The adjusted data is fed back to the digital twin model for model update and optimization, forming a closed-loop optimization link of prediction-allocation-feedback.

[0056] Optionally, generating the global resource status map according to the resource change situation includes:

[0057] Encoding the resource change situation of each edge node into a multi-dimensional label vector, the multi-dimensional label vector including available resource labels, latency sensitivity labels, and energy consumption cost labels;

[0058] Calculating the node weights by weighted calculation according to the multi-dimensional label vector, determining the edge weights from the inter-node network bandwidth and transmission latency, and constructing a global resource status map based on the topological relationship between edge nodes, the node weights, and the edge weights, the topological relationship including physical connection relationship, network hop count, and geographical proximity.

[0059] Available resource label: Reflects the resource capabilities that a node can provide for new tasks, including remaining computing power (number of CPU cores / GPU video memory), allocable bandwidth, storage capacity, etc. Latency sensitivity label: Reflects the support capabilities of a node for latency-sensitive tasks, including the average round-trip delay (RTT) from the node to the user device, the maximum latency tolerance threshold, etc. Energy consumption cost label: Measures the energy consumption overhead of executing tasks on this node, including the energy consumption per unit computing power of the node (watts / core), network transmission energy consumption (joules / MB), etc. Obtain the resource change situation of each edge node in real time, including resource utilization rate, task execution situation, etc. Extract the eigenvalue of the above three types of labels from the collected data. Integrate the extracted eigenvalues into a multi-dimensional vector, with each dimension corresponding to a label. According to the system design and actual requirements, set weight factors for the available resource label, latency sensitivity label, and energy consumption cost label respectively, reflecting the importance of each label in the node weight calculation. Multiply each label value by its corresponding weight factor and then sum to obtain the comprehensive weight of the node. For example: Node weight = w1 × available resource label + w2 × latency sensitivity label + w3 × energy consumption cost label, where w1, w2, and w3 are weight factors, and w1 + w2 + w3 = 1. Network bandwidth: Reflects the data transmission ability between nodes. The higher the bandwidth, the faster the data transmission and the greater the edge weight. Transmission delay: Reflects the delay of data transmission between nodes. The lower the delay, the more timely the data transmission and the greater the edge weight. Combine the network bandwidth and transmission delay to calculate the edge weight. For example, the following formula can be used: Edge weight = α × network bandwidth + β × (1 / transmission delay), where α and β are weight coefficients used to balance the influence of network bandwidth and transmission delay. Represent each edge node as a vertex in the graph, and the attributes of the vertex include the node weight and the multi-dimensional label vector. Represent the connection relationship between nodes as an edge in the graph, and the attributes of the edge include the edge weight. Combine topological information such as physical connection relationship, network hop count, and geographical proximity to construct a complete graph structure. According to the preset update trigger conditions, such as periodic update, node resource mutation, task offloading decision, etc., update the node weight, edge weight, and topological relationship in the graph in a timely manner to ensure the real-time and accuracy of the graph.

[0060] By encoding the resource change situation of each edge node into a multi-dimensional label vector, including available resource labels, latency sensitivity labels, and energy consumption cost labels, the resource status of each node can be comprehensively and accurately reflected. This way of quantification and abstraction makes the complex resource status intuitive and easy to compare and calculate, providing a solid quantitative basis for subsequent resource scheduling decisions. The node weight is calculated by weighted calculation based on the multi-dimensional label vector, which can comprehensively consider multiple factors such as the resource availability, latency characteristics, and energy consumption efficiency of the node. This weight calculation method enables a reasonable evaluation of the node's resource status and provides a scientific reference for subsequent resource allocation. The edge weight is determined by the network bandwidth and transmission latency between nodes, and a global resource status map is constructed based on the topological relationship between edge nodes, fully considering the network connection characteristics between nodes. This helps to select nodes with better network conditions during resource scheduling, reduce data transmission latency, and improve the overall performance of the system. The update trigger conditions of the map include periodic updates, sudden changes in node resources, and local updates triggered by task offloading decisions, which can timely reflect the changes in resource status, ensure the real-time and accuracy of the global resource status map, and enable resource scheduling to make decisions based on the latest resource information.

[0061] Optionally, the determining the execution location and priority of each task according to the task requirements and the global resource status map includes:

[0062] Determine the task category of each task according to the task requirements, and extract the key features of each task. The key features include the computing resource requirements, network bandwidth requirements, expected execution time, and data privacy level of the task;

[0063] Determine the priority of each task according to the task category and the key features through predefined priority rules;

[0064] Match the most suitable execution location according to the priority, the key features, and the global resource status map.

[0065] Tasks are classified into different categories according to their QoS requirements (such as upper limit of latency, computational complexity), such as critical, computationally intensive, privacy-sensitive, etc. For example, industrial control tasks belong to the critical category, video analysis tasks belong to the computationally intensive category, and medical data processing tasks belong to the privacy-sensitive category. Extract the key features of each task, including the computational resource requirements of the task (such as the number of CPU / GPU cores, memory size), network bandwidth requirements, expected execution time, and data privacy level. These features will be used for subsequent priority evaluation and execution location matching. According to the task category and key features, determine the priority of each task through predefined priority rules. For example: Critical tasks are given the highest priority due to their extremely high requirements for real-time performance and reliability; computationally intensive tasks determine their priorities based on their computational resource requirements and expected execution time, with higher priorities for tasks with greater requirements and longer execution times; privacy-sensitive tasks determine their priorities considering the data privacy protection level, with higher priorities for tasks with higher privacy levels. Combine the task category and key features, and evaluate each task according to the predefined priority rules to assign corresponding priority values. For example, the priority value ranges from 1 to 100, and a higher value indicates a higher priority. According to the priority, key features of the task, and the global resource status map, match the most suitable execution location. When matching, mainly consider resource matching degree, latency requirements, privacy protection, and energy consumption cost. Resource matching degree: Whether the computational resource requirements and network bandwidth requirements of the task match the available resources of the node. Latency requirements: Whether the upper limit of the task's latency can be met by the node, especially for latency-sensitive tasks. Privacy protection: For privacy-sensitive tasks, ensure that the execution node has sufficient data security measures. Energy consumption cost: On the premise of meeting the task requirements, select a node with a lower energy consumption cost to save energy. In the global resource status map, traverse each edge node and find the optimal execution location for each task according to the above matching criteria. For example, for high-priority critical tasks, select edge nodes with sufficient resources, low latency, and reasonable energy consumption costs; for privacy-sensitive tasks, select edge nodes with a high security level or execute locally; for computationally intensive tasks, comprehensively consider the computational power and network bandwidth of the node and select the most suitable node.

[0066] Tasks are classified into different categories (such as critical, computationally intensive, privacy-sensitive) according to the QoS requirements of the tasks (such as upper limit of latency, computational complexity), which provides a basis for subsequent priority determination and execution location matching. Key features such as the computational resource requirements, network bandwidth requirements, expected execution time, and data privacy level of the tasks are extracted to comprehensively capture the resource requirements and execution constraints of the tasks, providing detailed information for accurate resource matching. The most suitable execution location is matched according to the task priority, key features, and the global resource status map. For example, high-priority tasks preferentially select edge nodes with sufficient resources, low latency, and reasonable energy consumption costs; privacy-sensitive tasks select edge nodes with high security levels or local execution; computationally intensive tasks comprehensively consider the computing power and network bandwidth of the nodes. This matching method can ensure that tasks are assigned to the most suitable nodes for execution, improving the success rate and efficiency of task execution. By using information such as multi-dimensional label vectors and node weights in the global resource status map, the resource status and network topology relationship of the nodes are comprehensively considered to find the optimal execution location for the tasks, avoiding task execution failures caused by insufficient local resources or poor network conditions. Through refined task classification and feature extraction, the diverse needs of different tasks can be met, improving the flexibility of resource scheduling. Whether it is real-time tasks sensitive to latency, data processing tasks with high requirements for computing resources, or tasks with strict requirements for data privacy, reasonable resource allocation and scheduling can be obtained.

[0067] Optionally, the generating of the global allocation plan according to the global resource status map, the execution location, and the priority through a federated learning algorithm includes:

[0068] Calculate the weight of each node according to the total node resources and the historical task success rate, and weighted aggregate the resource allocation proposals of all nodes through the federated averaging algorithm to generate a preliminary global allocation plan;

[0069] Bind target tasks with priorities higher than the threshold to target edge nodes to adjust the preliminary global allocation plan;

[0070] Minimize the total latency and energy consumption of the adjusted preliminary global allocation plan according to a preset multi-objective optimization function, and use mixed integer linear programming to determine the global allocation plan.

[0071] The weight of each node is calculated based on its total resource amount and historical task success rate. The total resource amount includes resources such as CPU, GPU, memory, storage, and network bandwidth, and the total resource amount indicator is obtained through weighted summation. The historical task success rate reflects the stability and reliability of the node. Each node generates a resource allocation proposal based on its own resource status and task requirements, and the proposal contains resource allocation suggestions for each task. Using the Federated Averaging algorithm (FedAvg), the proposals of each node are weighted and averaged according to the node weights to obtain a preliminary global allocation plan. The proposals of nodes with higher weights have a greater impact on the global plan. Optimize the preliminary global allocation plan to ensure that the plan meets the execution location and priority constraints of the tasks, and at the same time optimize the resource allocation ratio to improve resource utilization. According to the task priority, filter out the target tasks with priorities higher than the threshold. For example, the priority threshold is set to 80, and all tasks with priorities higher than 80 are filtered out. In the global resource status graph, select the most suitable target edge node for each high-priority task and bind the task to the node. When selecting, comprehensively consider the resource status of the node, task requirements, and network topology relationship. For example, for the high-priority task T1, select node A with sufficient resources, low latency, and reasonable energy consumption cost for binding. According to the binding relationship between tasks and nodes, adjust the preliminary global allocation plan to ensure that high-priority tasks obtain sufficient resource allocation to meet their QoS requirements. The adjustment process may involve operations such as reallocating resources and adjusting the task execution order. Define a multi-objective optimization function that comprehensively considers two objectives: total latency and energy consumption. For example, the multi-objective optimization function can be expressed as:

[0072] ;

[0073] where N represents the total number of tasks, T i and P i represent the latency and energy consumption of task i respectively, and λ is the weight coefficient used to balance latency and energy consumption. Adopt the Mixed Integer Linear Programming (MILP) method to optimize resource allocation under the condition of meeting the task execution location and priority constraints. The specific steps include: define the task allocation variable X ij indicating whether task i is allocated to node j (1 means allocated, 0 means not allocated); ensure that each task can only be allocated to one node and the node resources do not exceed its capacity; minimize the weighted sum of total latency and energy consumption; use the MILP solver to solve the optimization problem to obtain the final global allocation plan.

[0074] When calculating the node weight, resource data such as CPU, GPU, memory, storage, and network bandwidth are statistically analyzed and weighted and summed to obtain the total node resources, which comprehensively reflects the resource status of the node. Combining the historical task success rate, the node weight is obtained through weighted summation, which not only considers the richness of resources but also takes into account the stability of task execution, accurately reflecting the value of the node. Using the federated averaging algorithm to perform weighted averaging on the resource allocation proposals generated by each node, without uploading the original data, protecting data privacy, avoiding single-point failures, and improving the robustness of the system. Performing weighted averaging on the resource allocation proposals according to the node weight to generate a reasonable and efficient preliminary global allocation plan, improving the efficiency and effect of resource allocation. Screening target tasks with a priority higher than the threshold, binding them to the target edge nodes, and adjusting the preliminary plan to ensure that critical tasks obtain sufficient resource allocation to meet the QoS requirements. Defining a multi-objective optimization function, considering the total delay and energy consumption, and solving it through mixed-integer linear programming to obtain the final global allocation plan, achieving a balance between performance and energy consumption. Optimizing task allocation, reducing resource waste, and improving resource utilization and task execution success rate. Combining with the real-time update of the global resource status map, timely adjusting the preliminary global allocation plan to adapt to the dynamic changes of the resource environment and ensure the stable operation of the system. Reasonably allocating resources, reducing resource waiting and data transmission delays, improving task execution efficiency, and enhancing the overall performance and user experience of the system.

[0075] Optionally, calculating the weight of each node according to the total node resources and the historical task success rate, and using the federated averaging algorithm to perform weighted aggregation on the resource allocation proposals of all nodes according to the weight to generate a preliminary global allocation plan includes:

[0076] Statistically analyze the resource data of each edge node, and perform weighted summation on the resource data to obtain the total node resources. The resource data includes CPU, GPU, memory, storage, and network bandwidth. Perform weighted summation on the total node resources and the historical task success rate to obtain the node weight;

[0077] Generate resource allocation proposals according to the resource status and task requirements of each edge node itself, and use the federated averaging algorithm to perform weighted averaging on the resource allocation proposals according to the node weight to obtain a preliminary global allocation plan.

[0078] Collect the resource data of each edge node, including CPU, GPU, memory, storage, network bandwidth, etc. These data can be obtained through the node's resource monitoring tools or management interfaces. Perform a weighted sum on the collected resource data to obtain the total resource of each node. Collect the success rate data of past task executions on each node, which reflects the stability and reliability of the node. Perform a weighted sum on the total node resources and the historical task success rate to obtain the node weight. Each edge node generates a resource allocation proposal based on its own resource status and task requirements. The proposal includes resource allocation suggestions for each task on the node, such as the number of CPU cores allocated, memory size, network bandwidth, etc. Federated Averaging algorithm (FedAvg): FedAvg is a commonly used federated learning algorithm for aggregating model parameters among multiple nodes. Here, FedAvg is used to perform a weighted average on the resource allocation proposals of each node. Perform a weighted average on the resource allocation proposals of each node according to the node weight to obtain a preliminary global allocation plan. For example, assume that the weights of nodes A and B are wA and wB respectively, and their proposals are PA and PB respectively. Then the preliminary global allocation plan P_preliminary can be expressed as: P_preliminary = (wA × PA + wB × PB) / (wA + wB). This preliminary global allocation plan comprehensively considers the resource status and task requirements of each node, and at the same time performs a reasonable weighting according to the node weight, ensuring the rationality and fairness of resource allocation.

[0079] Statistically analyze the resource data of edge nodes, such as CPU, GPU, memory, storage, and network bandwidth, and perform weighted summation to obtain the total node resources. This multi-dimensional consideration method comprehensively reflects the resource status of nodes, avoiding the one-sidedness that may be caused by a single resource metric. The total node resources and the historical task success rate are weighted and summed to obtain the node weight, which not only considers the richness of node resources but also takes into account the task execution stability of nodes. This comprehensive evaluation method enables the node weight to more accurately reflect the value of nodes in resource scheduling. The federated averaging algorithm is used to perform weighted averaging on the resource allocation proposals generated by each node, without uploading the original data of the nodes to the central server, effectively protecting data privacy. At the same time, this distributed collaboration method avoids single-point failures and improves the robustness of the system. The resource allocation proposals are weighted and averaged according to the node weights, making the resource allocation scheme more reasonable. Nodes with high weights have a greater impact on the global scheme, and their proposals can better reflect the allocation strategies of resource-rich nodes, thereby improving the efficiency and effectiveness of resource allocation. After generating the preliminary global allocation scheme, it will be optimized to meet the execution location and priority constraints of tasks, and the resource allocation ratio will be optimized to further improve resource utilization and the success rate of task execution. Combining with the real-time update mechanism of the global resource status map, it can respond to changes in resource status in a timely manner and dynamically adjust the preliminary global allocation scheme. This dynamic adaptability enables resource scheduling to better cope with the dynamic changes of resources in the edge cloud computing platform and ensures the stable operation of the system.

[0080] Optionally, the method further includes:

[0081] For target tasks with a latency tolerance value lower than the threshold, skip the federated learning aggregation process, and directly match the execution location by the edge node based on the local resource status map;

[0082] When multiple tasks compete for the same node resources, allocate the resources already allocated to low-priority tasks to high-priority tasks;

[0083] Based on the consistent hashing algorithm, migrate the low-priority tasks to adjacent edge nodes or enter the waiting queue.

[0084] For target tasks with a delay tolerance value lower than the threshold, there is no need to go through the federated learning aggregation process. Instead, the edge node directly matches the execution location based on the local resource status map. This can reduce latency and improve the response speed of the task. The edge node quickly finds a suitable execution location for the task according to the local resource status map, ensuring that the task can be executed in a timely manner. When multiple tasks compete for the resources of the same node, the resources already allocated to the low-priority task are allocated to the high-priority task. This can ensure that high-priority tasks can obtain sufficient resources to meet their QoS requirements. After the resources of the low-priority task are preempted, it is migrated to a neighboring edge node or enters the waiting queue based on the consistent hashing algorithm. Multiple virtual nodes are assigned to each edge node to construct a hash ring. When the low-priority task is bound to a virtual node, the target node to be migrated is located clockwise according to the hash ring. Determine whether the low-priority task is a task restricted by data sovereignty. If so, a node with a load rate lower than the load threshold within the preset geographical area is determined as the target node to be migrated. If there is no suitable node within the preset geographical area, the low-priority task is placed in the waiting queue.

[0085] For target tasks with a delay tolerance value lower than the threshold, skip the federated learning aggregation process to reduce scheduling latency. The edge node directly matches the execution location based on the local resource status map to improve scheduling efficiency, which is suitable for tasks with high real-time requirements. When multiple tasks compete for the resources of the same node, the resources already allocated to the low-priority task are allocated to the high-priority task to ensure that critical tasks can obtain sufficient resources and improve the overall efficiency of the system. By reallocating resources, resource idleness is avoided and resource utilization is improved. The low-priority task is migrated to a neighboring edge node or enters the waiting queue based on the consistent hashing algorithm to reduce system oscillations during task migration and improve system fault tolerance. The consistent hashing algorithm enables the system to quickly reallocate tasks when nodes are added or removed, improving scalability and meeting the requirements of large-scale edge cloud computing platforms. Through mechanisms such as skipping the federated learning aggregation process, resource reallocation, and task migration, the system can flexibly respond to different types of task requirements and dynamic resource environments. These mechanisms work together to enable the system to quickly adjust and maintain stable and efficient operation in the face of task priority changes, resource competition, and node dynamic changes.

[0086] Optionally, the migrating the low-priority task to a neighboring edge node or entering the waiting queue based on the consistent hashing algorithm includes:

[0087] Assign multiple virtual nodes to each edge node to construct a hash ring. When the low-priority task is bound to a virtual node, the target node to be migrated is located clockwise according to the hash ring;

[0088] Determine whether the low-priority task is a task restricted by data sovereignty. When the low-priority task is a task restricted by data sovereignty, determine a node with a load rate lower than the load threshold within a preset geographical area as the target node to be migrated;

[0089] When there is no node with a load rate lower than the load threshold within the preset geographical area, put the low-priority task into the waiting queue.

[0090] Allocate multiple virtual nodes to each edge node, and these virtual nodes are evenly distributed on the hash ring. The number of virtual nodes can be adjusted according to actual needs to improve the granularity and flexibility of the hash ring. Construct a hash ring and arrange all virtual nodes in a clockwise order on the ring according to their hash values. The hash function can use common hash algorithms such as SHA-1, MD5, etc. When a low-priority task needs to be migrated, bind it to a virtual node. The binding process can calculate the hash value by hashing the identifier or characteristics of the task, and then map the hash value to a virtual node on the hash ring. According to the hash ring, start from the bound virtual node and find the first actually existing edge node in the clockwise direction as the target node to be migrated. This clockwise search method ensures the certainty and efficiency of task migration. Determine whether the low-priority task is a task restricted by data sovereignty. Data sovereignty restriction means that the data of the task can only be processed in a specific geographical area or on specific nodes to meet the requirements of data privacy and security. If the task is restricted by data sovereignty, then find a node with a load rate lower than the load threshold within the preset geographical area as the target node to be migrated. The preset geographical area can be defined according to the data privacy policy or the requirements of the task. Within the preset geographical area, check whether the load rate of each node is lower than the load threshold. The load rate can be expressed as the ratio of the currently allocated resources of the node to the total resources. If a node with a load rate lower than the threshold is found, migrate the low-priority task to this node; if no suitable node is found within the preset geographical area, put the task into the waiting queue and wait for subsequent resource release or node load reduction.

[0091] Multiple virtual nodes are assigned to each edge node to construct a hash ring. When a low-priority task is bound to a virtual node, the target node to be migrated is located clockwise according to the hash ring. This mechanism ensures that tasks can be migrated to appropriate nodes quickly and efficiently, reducing the complexity and overhead of task migration. Determine whether the low-priority task is a task restricted by data sovereignty. If it is restricted, determine a node with a load rate lower than the load threshold within a preset geographical area as the target node to be migrated. This ensures data security and compliance while ensuring that tasks can be executed on appropriate nodes. The consistent hashing algorithm ensures that when nodes are added or removed, only some tasks need to be migrated, and the mapping relationships of the remaining tasks remain unchanged. This avoids large-scale task reallocation caused by node changes and improves the stability of the system. Through the introduction of virtual nodes, the distribution of nodes on the hash ring is more uniform, avoiding the situation where some nodes are overloaded while others are idle, achieving load balancing of the system and enhancing the fault tolerance of the system. Low-priority tasks are migrated to adjacent edge nodes or enter the waiting queue, enabling resources to be utilized more efficiently and avoiding waste of resources.

[0092] This embodiment also discloses a resource scheduling system in an edge cloud computing platform. Figure 2 It is a schematic diagram of the modules of the resource scheduling system in the edge cloud computing platform disclosed in the embodiments of the present application. As Figure 2 shown, the system includes a collection module 201, a task module 202, an allocation module 203, and an adjustment module 204, where:

[0093] The collection module 201 is configured to obtain the resource status and environmental parameters of each edge node, construct a corresponding digital twin model for each edge node according to the resource status and the environmental parameters, predict the resource change situation within a preset future time according to the digital twin model, and generate a global resource status map according to the resource change situation;

[0094] The task module 202 is configured to obtain the task requirements of each task, and determine the execution location and priority of each task according to the task requirements and the global resource status map;

[0095] The allocation module 203 is configured to generate a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location, and the priority;

[0096] The adjustment module 204 is configured to obtain the resource allocation status in real time, determine the difference between the resource allocation status and the global allocation plan, and correct the global allocation plan based on the difference through a hierarchical cooperation mechanism.

[0097] Optionally, the collection module 201 is configured to:

[0098] Encode the resource change situation of each edge node into a multi-dimensional label vector, where the multi-dimensional label vector includes an available resource label, a latency sensitivity label, and an energy consumption cost label;

[0099] Calculate the node weight by weighted calculation according to the multi-dimensional label vector, determine the edge weight from the network bandwidth and transmission latency between nodes, and construct a global resource status map based on the topological relationship between edge nodes, the node weight, and the edge weight. The topological relationship includes a physical connection relationship, the number of network hops, and geographical proximity.

[0100] Optionally, the task module 202 is configured to:

[0101] Determine the task category of each task according to the task requirements, and extract the key features of each task. The key features include the computing resource requirements, network bandwidth requirements, expected execution time, and data privacy level of the task;

[0102] Determine the priority of each task according to the task category and the key features through a predefined priority rule;

[0103] Match the most suitable execution location according to the priority, the key features, and the global resource status map.

[0104] Optionally, the allocation module 203 is configured to:

[0105] Calculate the weight of each node according to the total node resources and the historical task success rate, and weighted aggregate all node resource allocation proposals through the federated averaging algorithm according to the weight to generate a preliminary global allocation plan;

[0106] Bind the target tasks with priorities higher than the threshold to the target edge nodes to adjust the preliminary global allocation plan;

[0107] Minimize the total latency and energy consumption of the adjusted preliminary global allocation plan according to a preset multi-objective optimization function, and use mixed integer linear programming to determine the global allocation plan.

[0108] Optionally, the allocation module 203 is configured to:

[0109] Statistically analyze the resource data of each edge node, and perform weighted summation on the resource data to obtain the total node resources. The resource data includes CPU, GPU, memory, storage, and network bandwidth, and perform weighted summation on the total node resources and the historical task success rate to obtain the node weight;

[0110] Generate a resource allocation proposal based on the resource status and task requirements of each edge node, and use the federated averaging algorithm to perform weighted averaging on the resource allocation proposal according to the node weights to obtain a preliminary global allocation plan.

[0111] Optionally, the system further includes a migration module configured to:

[0112] For target tasks with a latency tolerance value lower than the threshold, skip the federated learning aggregation process, and directly match the execution location based on the local resource status map by the edge node;

[0113] When multiple tasks compete for the same node resources, allocate the resources already allocated to the low-priority tasks to the high-priority tasks;

[0114] Migrate the low-priority tasks to neighboring edge nodes or enter the waiting queue based on the consistent hashing algorithm.

[0115] Optionally, the migration module is further configured to:

[0116] Allocate multiple virtual nodes to each edge node to construct a hash ring. When the low-priority task is bound to a virtual node, locate the target node to be migrated clockwise according to the hash ring;

[0117] Determine whether the low-priority task is a task restricted by data sovereignty. When the low-priority task is a task restricted by data sovereignty, determine a node with a load rate lower than the load threshold within a preset geographical area as the target node to be migrated;

[0118] When there is no node with a load rate lower than the load threshold within the preset geographical area, put the low-priority task into the waiting queue.

[0119] It should be noted that: when the device provided in the above embodiment realizes its functions, only the above-mentioned division of each functional module is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be elaborated here.

[0120] This embodiment also discloses an electronic device. Referring to Figure 3 , the electronic device may include: at least one processor 301, at least one communication bus 302, a user interface 303, a network interface 304, and at least one memory 305.

[0121] Among them, the communication bus 302 is used to realize the connection and communication between these components.

[0122] Among them, the user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may further include standard wired interfaces and wireless interfaces.

[0123] Among them, the network interface 304 may optionally include standard wired interfaces and wireless interfaces (such as WI-FI interfaces).

[0124] Among them, the processor 301 may include one or more processing cores. The processor 301 connects various parts within the entire server using various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 305, and by invoking data stored in the memory 305, it performs various functions of the server and processes data. Optionally, the processor 301 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 301 may integrate one or a combination of several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, and application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 301 and may be implemented separately by a single chip.

[0125] Among them, the memory 305 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 305 includes a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned method embodiments, etc.; the data storage area may store the data involved in the above-mentioned method embodiments. Optionally, the memory 305 may further be at least one storage device located far from the aforementioned processor 301. Such as Figure 3As shown, in the memory 305 serving as a computer storage medium, there may be included an operating system, a network communication module, a user interface module, and an application program of the resource scheduling method in the edge cloud computing platform.

[0126] In Figure 3 In the electronic device shown, the user interface 303 is mainly used to provide an interface for the user to obtain the data input by the user; while the processor 301 can be used to call the application program of the resource scheduling method in the edge cloud computing platform stored in the memory 305. When executed by one or more processors 301, the electronic device is caused to execute the method of one or more of the above embodiments.

[0127] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be in other sequences or performed simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0128] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0129] In several embodiments provided in this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0130] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit exists physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0132] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 305 and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned memory 305 includes: various media such as USB flash drives, mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0133] The foregoing are only exemplary embodiments of the present disclosure and should not be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. After considering the disclosure of the specification, those skilled in the art will readily think of other implementation manners of the present disclosure. The present application aims to cover any variations, uses, or adaptive changes of the present disclosure, and these variations, uses, or adaptive changes follow the general principles of the present disclosure and include the common general knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A resource scheduling method in an edge cloud computing platform, characterized in that, Applied to an edge cloud computing platform, the method includes: Obtain the resource status and environmental parameters of each edge node, construct a corresponding digital twin model for each edge node according to the resource status and the environmental parameters, predict the resource change situation within a preset future time according to the digital twin model, and generate a global resource status map according to the resource change situation; Obtain the task requirements of each task, and determine the execution location and priority of each task according to the task requirements and the global resource status map; Generate a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location, and the priority; Obtain the resource allocation status in real time, determine the difference between the resource allocation status and the global allocation plan, and correct the global allocation plan based on the difference through a hierarchical cooperation mechanism. The generating a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location, and the priority includes: Calculate the weight of each node according to the total node resources and the historical task success rate, and weighted aggregate the resource allocation proposals of all nodes through the federated averaging algorithm to generate a preliminary global allocation plan; Bind the target tasks with priorities higher than the threshold to the target edge nodes to adjust the preliminary global allocation plan; Minimize the total latency and energy consumption of the adjusted preliminary global allocation plan according to a preset multi-objective optimization function, and use mixed integer linear programming to determine the global allocation plan.

2. The resource scheduling method in the edge cloud computing platform according to claim 1, wherein, The generating a global resource status map according to the resource change situation includes: Encode the resource change situations of each edge node into a multi-dimensional label vector, where the multi-dimensional label vector includes available resource labels, latency sensitivity labels, and energy consumption cost labels; Calculate the node weights by weighted calculation according to the multi-dimensional label vector, determine the edge weights based on the inter-node network bandwidth and transmission latency, and construct a global resource status map based on the topological relationship between edge nodes, the node weights, and the edge weights. The topological relationship includes physical connection relationships, network hop counts, and geographical proximity.

3. The resource scheduling method in the edge cloud computing platform according to claim 1, wherein The determining the execution location and priority of each task according to the task requirements and the global resource status map includes: Determine the task category of each task according to the task requirements, and extract the key features of each task. The key features include the computing resource requirements, network bandwidth requirements, expected execution time, and data privacy level of the task; Determine the priority of each task through predefined priority rules according to the task category and the key features; Match the most suitable execution location according to the priority, the key features, and the global resource status map.

4. The resource scheduling method in the edge cloud computing platform according to claim 1, characterized in that, The calculating the weight of each node according to the total node resources and the historical task success rate, and weighted aggregating the resource allocation proposals of all nodes through the federated averaging algorithm to generate a preliminary global allocation plan includes: Statistically analyze the resource data of each edge node, and perform weighted summation on the resource data to obtain the total node resources. The resource data includes CPU, GPU, memory, storage, and network bandwidth. Perform weighted summation on the total node resources and the historical task success rate to obtain the node weight; Generate a resource allocation proposal based on the resource status and task requirements of each edge node. Use the federated averaging algorithm to perform weighted averaging on the resource allocation proposal according to the node weight to obtain a preliminary global allocation plan.

5. The resource scheduling method in the edge cloud computing platform according to claim 1, wherein, The method further includes: For target tasks with a latency tolerance value lower than the threshold, skip the federated learning aggregation process, and directly match the execution location based on the local resource status map by the edge node; When multiple tasks compete for the same node resources, allocate the resources already allocated to the low-priority tasks to the high-priority tasks; Migrate the low-priority tasks to neighboring edge nodes or enter the waiting queue based on the consistent hashing algorithm.

6. The resource scheduling method in the edge cloud computing platform according to claim 5, characterized in that, The migrating the low-priority tasks to neighboring edge nodes or entering the waiting queue based on the consistent hashing algorithm includes: Allocate multiple virtual nodes to each edge node to construct a hash ring. After the low-priority task is bound to a virtual node, locate the target node to be migrated clockwise according to the hash ring; Determine whether the low-priority task is a task restricted by data sovereignty. When the low-priority task is a task restricted by data sovereignty, determine a node with a load rate lower than the load threshold within a preset geographical area as the target node to be migrated; When there is no node with a load rate lower than the load threshold within the preset geographical area, put the low-priority task into the waiting queue.

7. A resource scheduling system in an edge cloud computing platform, characterized in that, It includes a collection module, a task module, an allocation module, and an adjustment module, where: The collection module is configured to obtain the resource status and environmental parameters of each edge node, construct a corresponding digital twin model for each edge node according to the resource status and the environmental parameters, predict the resource change situation within a preset future time according to the digital twin model, and generate a global resource status map according to the resource change situation; The task module is configured to obtain the task requirements of each task, and determine the execution location and priority of each task according to the task requirements and the global resource status map; The allocation module is configured to generate a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location, and the priority; The adjustment module is configured to obtain the resource allocation status in real time, determine the difference between the resource allocation status and the global allocation plan, and correct the global allocation plan based on the difference through a hierarchical cooperation mechanism. The generating a global allocation plan through a federated learning algorithm according to the global resource status map, the execution location, and the priority includes: Calculate the weight of each node according to the total node resources and the historical task success rate, and generate a preliminary global allocation plan by weighted aggregation of the resource allocation proposals of all nodes according to the weight through the federated averaging algorithm; Bind the target tasks with a priority higher than the threshold to the target edge nodes to adjust the preliminary global allocation plan; Minimize the total latency and energy consumption of the adjusted preliminary global allocation scheme according to a preset multi-objective optimization function, and use mixed-integer linear programming to determine the global allocation scheme.

8. An electronic device, characterized in that, It includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. Both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory, so that the electronic device executes the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, execute the method according to any one of claims 1-6.

Images